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;
444 case IVHD_DEV_SELECT:
445 case IVHD_DEV_RANGE_END:
447 case IVHD_DEV_EXT_SELECT:
448 /* all the above subfield types refer to device ids */
449 update_last_devid(dev->devid);
454 p += ivhd_entry_length(p);
463 * Iterate over all IVHD entries in the ACPI table and find the highest device
464 * id which we need to handle. This is the first of three functions which parse
465 * the ACPI table. So we check the checksum here.
467 static int __init find_last_devid_acpi(struct acpi_table_header *table)
470 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
471 struct ivhd_header *h;
474 * Validate checksum here so we don't need to do it when
475 * we actually parse the table
477 for (i = 0; i < table->length; ++i)
480 /* ACPI table corrupt */
483 p += IVRS_HEADER_LENGTH;
485 end += table->length;
487 h = (struct ivhd_header *)p;
490 find_last_devid_from_ivhd(h);
502 /****************************************************************************
504 * The following functions belong to the code path which parses the ACPI table
505 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
506 * data structures, initialize the device/alias/rlookup table and also
507 * basically initialize the hardware.
509 ****************************************************************************/
512 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
513 * write commands to that buffer later and the IOMMU will execute them
516 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
518 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
519 get_order(CMD_BUFFER_SIZE));
524 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
530 * This function resets the command buffer if the IOMMU stopped fetching
533 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
535 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
537 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
538 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
540 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
544 * This function writes the command buffer address to the hardware and
547 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
551 BUG_ON(iommu->cmd_buf == NULL);
553 entry = (u64)virt_to_phys(iommu->cmd_buf);
554 entry |= MMIO_CMD_SIZE_512;
556 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
557 &entry, sizeof(entry));
559 amd_iommu_reset_cmd_buffer(iommu);
560 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
563 static void __init free_command_buffer(struct amd_iommu *iommu)
565 free_pages((unsigned long)iommu->cmd_buf,
566 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
569 /* allocates the memory where the IOMMU will log its events to */
570 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
572 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
573 get_order(EVT_BUFFER_SIZE));
575 if (iommu->evt_buf == NULL)
578 iommu->evt_buf_size = EVT_BUFFER_SIZE;
580 return iommu->evt_buf;
583 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
587 BUG_ON(iommu->evt_buf == NULL);
589 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
591 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
592 &entry, sizeof(entry));
594 /* set head and tail to zero manually */
595 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
596 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
598 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
601 static void __init free_event_buffer(struct amd_iommu *iommu)
603 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
606 /* allocates the memory where the IOMMU will log its events to */
607 static u8 * __init alloc_ppr_log(struct amd_iommu *iommu)
609 iommu->ppr_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
610 get_order(PPR_LOG_SIZE));
612 if (iommu->ppr_log == NULL)
615 return iommu->ppr_log;
618 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
622 if (iommu->ppr_log == NULL)
625 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
627 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
628 &entry, sizeof(entry));
630 /* set head and tail to zero manually */
631 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
632 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
634 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
635 iommu_feature_enable(iommu, CONTROL_PPR_EN);
638 static void __init free_ppr_log(struct amd_iommu *iommu)
640 if (iommu->ppr_log == NULL)
643 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
646 static void iommu_enable_gt(struct amd_iommu *iommu)
648 if (!iommu_feature(iommu, FEATURE_GT))
651 iommu_feature_enable(iommu, CONTROL_GT_EN);
654 /* sets a specific bit in the device table entry. */
655 static void set_dev_entry_bit(u16 devid, u8 bit)
657 int i = (bit >> 6) & 0x03;
658 int _bit = bit & 0x3f;
660 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
663 static int get_dev_entry_bit(u16 devid, u8 bit)
665 int i = (bit >> 6) & 0x03;
666 int _bit = bit & 0x3f;
668 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
672 void amd_iommu_apply_erratum_63(u16 devid)
676 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
677 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
680 set_dev_entry_bit(devid, DEV_ENTRY_IW);
683 /* Writes the specific IOMMU for a device into the rlookup table */
684 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
686 amd_iommu_rlookup_table[devid] = iommu;
690 * This function takes the device specific flags read from the ACPI
691 * table and sets up the device table entry with that information
693 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
694 u16 devid, u32 flags, u32 ext_flags)
696 if (flags & ACPI_DEVFLAG_INITPASS)
697 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
698 if (flags & ACPI_DEVFLAG_EXTINT)
699 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
700 if (flags & ACPI_DEVFLAG_NMI)
701 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
702 if (flags & ACPI_DEVFLAG_SYSMGT1)
703 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
704 if (flags & ACPI_DEVFLAG_SYSMGT2)
705 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
706 if (flags & ACPI_DEVFLAG_LINT0)
707 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
708 if (flags & ACPI_DEVFLAG_LINT1)
709 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
711 amd_iommu_apply_erratum_63(devid);
713 set_iommu_for_device(iommu, devid);
716 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
718 struct devid_map *entry;
719 struct list_head *list;
721 if (type == IVHD_SPECIAL_IOAPIC)
723 else if (type == IVHD_SPECIAL_HPET)
728 list_for_each_entry(entry, list, list) {
729 if (!(entry->id == id && entry->cmd_line))
732 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
733 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
735 *devid = entry->devid;
740 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
745 entry->devid = *devid;
746 entry->cmd_line = cmd_line;
748 list_add_tail(&entry->list, list);
753 static int __init add_early_maps(void)
757 for (i = 0; i < early_ioapic_map_size; ++i) {
758 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
759 early_ioapic_map[i].id,
760 &early_ioapic_map[i].devid,
761 early_ioapic_map[i].cmd_line);
766 for (i = 0; i < early_hpet_map_size; ++i) {
767 ret = add_special_device(IVHD_SPECIAL_HPET,
768 early_hpet_map[i].id,
769 &early_hpet_map[i].devid,
770 early_hpet_map[i].cmd_line);
779 * Reads the device exclusion range from ACPI and initializes the IOMMU with
782 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
784 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
786 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
791 * We only can configure exclusion ranges per IOMMU, not
792 * per device. But we can enable the exclusion range per
793 * device. This is done here
795 set_dev_entry_bit(devid, DEV_ENTRY_EX);
796 iommu->exclusion_start = m->range_start;
797 iommu->exclusion_length = m->range_length;
802 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
803 * initializes the hardware and our data structures with it.
805 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
806 struct ivhd_header *h)
809 u8 *end = p, flags = 0;
810 u16 devid = 0, devid_start = 0, devid_to = 0;
811 u32 dev_i, ext_flags = 0;
813 struct ivhd_entry *e;
817 ret = add_early_maps();
822 * First save the recommended feature enable bits from ACPI
824 iommu->acpi_flags = h->flags;
827 * Done. Now parse the device entries
829 p += sizeof(struct ivhd_header);
834 e = (struct ivhd_entry *)p;
838 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
839 " last device %02x:%02x.%x flags: %02x\n",
840 PCI_BUS_NUM(iommu->first_device),
841 PCI_SLOT(iommu->first_device),
842 PCI_FUNC(iommu->first_device),
843 PCI_BUS_NUM(iommu->last_device),
844 PCI_SLOT(iommu->last_device),
845 PCI_FUNC(iommu->last_device),
848 for (dev_i = iommu->first_device;
849 dev_i <= iommu->last_device; ++dev_i)
850 set_dev_entry_from_acpi(iommu, dev_i,
853 case IVHD_DEV_SELECT:
855 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
857 PCI_BUS_NUM(e->devid),
863 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
865 case IVHD_DEV_SELECT_RANGE_START:
867 DUMP_printk(" DEV_SELECT_RANGE_START\t "
868 "devid: %02x:%02x.%x flags: %02x\n",
869 PCI_BUS_NUM(e->devid),
874 devid_start = e->devid;
881 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
882 "flags: %02x devid_to: %02x:%02x.%x\n",
883 PCI_BUS_NUM(e->devid),
887 PCI_BUS_NUM(e->ext >> 8),
888 PCI_SLOT(e->ext >> 8),
889 PCI_FUNC(e->ext >> 8));
892 devid_to = e->ext >> 8;
893 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
894 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
895 amd_iommu_alias_table[devid] = devid_to;
897 case IVHD_DEV_ALIAS_RANGE:
899 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
900 "devid: %02x:%02x.%x flags: %02x "
901 "devid_to: %02x:%02x.%x\n",
902 PCI_BUS_NUM(e->devid),
906 PCI_BUS_NUM(e->ext >> 8),
907 PCI_SLOT(e->ext >> 8),
908 PCI_FUNC(e->ext >> 8));
910 devid_start = e->devid;
912 devid_to = e->ext >> 8;
916 case IVHD_DEV_EXT_SELECT:
918 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
919 "flags: %02x ext: %08x\n",
920 PCI_BUS_NUM(e->devid),
926 set_dev_entry_from_acpi(iommu, devid, e->flags,
929 case IVHD_DEV_EXT_SELECT_RANGE:
931 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
932 "%02x:%02x.%x flags: %02x ext: %08x\n",
933 PCI_BUS_NUM(e->devid),
938 devid_start = e->devid;
943 case IVHD_DEV_RANGE_END:
945 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
946 PCI_BUS_NUM(e->devid),
951 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
953 amd_iommu_alias_table[dev_i] = devid_to;
954 set_dev_entry_from_acpi(iommu,
955 devid_to, flags, ext_flags);
957 set_dev_entry_from_acpi(iommu, dev_i,
961 case IVHD_DEV_SPECIAL: {
967 handle = e->ext & 0xff;
968 devid = (e->ext >> 8) & 0xffff;
969 type = (e->ext >> 24) & 0xff;
971 if (type == IVHD_SPECIAL_IOAPIC)
973 else if (type == IVHD_SPECIAL_HPET)
978 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
984 ret = add_special_device(type, handle, &devid, false);
989 * add_special_device might update the devid in case a
990 * command-line override is present. So call
991 * set_dev_entry_from_acpi after add_special_device.
993 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1001 p += ivhd_entry_length(p);
1007 /* Initializes the device->iommu mapping for the driver */
1008 static int __init init_iommu_devices(struct amd_iommu *iommu)
1012 for (i = iommu->first_device; i <= iommu->last_device; ++i)
1013 set_iommu_for_device(iommu, i);
1018 static void __init free_iommu_one(struct amd_iommu *iommu)
1020 free_command_buffer(iommu);
1021 free_event_buffer(iommu);
1022 free_ppr_log(iommu);
1023 iommu_unmap_mmio_space(iommu);
1026 static void __init free_iommu_all(void)
1028 struct amd_iommu *iommu, *next;
1030 for_each_iommu_safe(iommu, next) {
1031 list_del(&iommu->list);
1032 free_iommu_one(iommu);
1038 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1040 * BIOS should disable L2B micellaneous clock gating by setting
1041 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1043 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1047 if ((boot_cpu_data.x86 != 0x15) ||
1048 (boot_cpu_data.x86_model < 0x10) ||
1049 (boot_cpu_data.x86_model > 0x1f))
1052 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1053 pci_read_config_dword(iommu->dev, 0xf4, &value);
1058 /* Select NB indirect register 0x90 and enable writing */
1059 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1061 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1062 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1063 dev_name(&iommu->dev->dev));
1065 /* Clear the enable writing bit */
1066 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1070 * This function clues the initialization function for one IOMMU
1071 * together and also allocates the command buffer and programs the
1072 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1074 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1078 spin_lock_init(&iommu->lock);
1080 /* Add IOMMU to internal data structures */
1081 list_add_tail(&iommu->list, &amd_iommu_list);
1082 iommu->index = amd_iommus_present++;
1084 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1085 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1089 /* Index is fine - add IOMMU to the array */
1090 amd_iommus[iommu->index] = iommu;
1093 * Copy data from ACPI table entry to the iommu struct
1095 iommu->devid = h->devid;
1096 iommu->cap_ptr = h->cap_ptr;
1097 iommu->pci_seg = h->pci_seg;
1098 iommu->mmio_phys = h->mmio_phys;
1100 /* Check if IVHD EFR contains proper max banks/counters */
1101 if ((h->efr != 0) &&
1102 ((h->efr & (0xF << 13)) != 0) &&
1103 ((h->efr & (0x3F << 17)) != 0)) {
1104 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1106 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1109 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1110 iommu->mmio_phys_end);
1111 if (!iommu->mmio_base)
1114 iommu->cmd_buf = alloc_command_buffer(iommu);
1115 if (!iommu->cmd_buf)
1118 iommu->evt_buf = alloc_event_buffer(iommu);
1119 if (!iommu->evt_buf)
1122 iommu->int_enabled = false;
1124 ret = init_iommu_from_acpi(iommu, h);
1128 ret = amd_iommu_create_irq_domain(iommu);
1133 * Make sure IOMMU is not considered to translate itself. The IVRS
1134 * table tells us so, but this is a lie!
1136 amd_iommu_rlookup_table[iommu->devid] = NULL;
1138 init_iommu_devices(iommu);
1144 * Iterates over all IOMMU entries in the ACPI table, allocates the
1145 * IOMMU structure and initializes it with init_iommu_one()
1147 static int __init init_iommu_all(struct acpi_table_header *table)
1149 u8 *p = (u8 *)table, *end = (u8 *)table;
1150 struct ivhd_header *h;
1151 struct amd_iommu *iommu;
1154 end += table->length;
1155 p += IVRS_HEADER_LENGTH;
1158 h = (struct ivhd_header *)p;
1160 case ACPI_IVHD_TYPE:
1162 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1163 "seg: %d flags: %01x info %04x\n",
1164 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1165 PCI_FUNC(h->devid), h->cap_ptr,
1166 h->pci_seg, h->flags, h->info);
1167 DUMP_printk(" mmio-addr: %016llx\n",
1170 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1174 ret = init_iommu_one(iommu, h);
1190 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1192 u64 val = 0xabcd, val2 = 0;
1194 if (!iommu_feature(iommu, FEATURE_PC))
1197 amd_iommu_pc_present = true;
1199 /* Check if the performance counters can be written to */
1200 if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val, true)) ||
1201 (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2, false)) ||
1203 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1204 amd_iommu_pc_present = false;
1208 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1210 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1211 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1212 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1215 static ssize_t amd_iommu_show_cap(struct device *dev,
1216 struct device_attribute *attr,
1219 struct amd_iommu *iommu = dev_get_drvdata(dev);
1220 return sprintf(buf, "%x\n", iommu->cap);
1222 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1224 static ssize_t amd_iommu_show_features(struct device *dev,
1225 struct device_attribute *attr,
1228 struct amd_iommu *iommu = dev_get_drvdata(dev);
1229 return sprintf(buf, "%llx\n", iommu->features);
1231 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1233 static struct attribute *amd_iommu_attrs[] = {
1235 &dev_attr_features.attr,
1239 static struct attribute_group amd_iommu_group = {
1240 .name = "amd-iommu",
1241 .attrs = amd_iommu_attrs,
1244 static const struct attribute_group *amd_iommu_groups[] = {
1249 static int iommu_init_pci(struct amd_iommu *iommu)
1251 int cap_ptr = iommu->cap_ptr;
1252 u32 range, misc, low, high;
1254 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1255 iommu->devid & 0xff);
1259 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1261 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1263 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1266 iommu->first_device = PCI_DEVID(MMIO_GET_BUS(range),
1267 MMIO_GET_FD(range));
1268 iommu->last_device = PCI_DEVID(MMIO_GET_BUS(range),
1269 MMIO_GET_LD(range));
1271 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1272 amd_iommu_iotlb_sup = false;
1274 /* read extended feature bits */
1275 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1276 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1278 iommu->features = ((u64)high << 32) | low;
1280 if (iommu_feature(iommu, FEATURE_GT)) {
1285 pasmax = iommu->features & FEATURE_PASID_MASK;
1286 pasmax >>= FEATURE_PASID_SHIFT;
1287 max_pasid = (1 << (pasmax + 1)) - 1;
1289 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1291 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1293 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1294 glxval >>= FEATURE_GLXVAL_SHIFT;
1296 if (amd_iommu_max_glx_val == -1)
1297 amd_iommu_max_glx_val = glxval;
1299 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1302 if (iommu_feature(iommu, FEATURE_GT) &&
1303 iommu_feature(iommu, FEATURE_PPR)) {
1304 iommu->is_iommu_v2 = true;
1305 amd_iommu_v2_present = true;
1308 if (iommu_feature(iommu, FEATURE_PPR)) {
1309 iommu->ppr_log = alloc_ppr_log(iommu);
1310 if (!iommu->ppr_log)
1314 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1315 amd_iommu_np_cache = true;
1317 init_iommu_perf_ctr(iommu);
1319 if (is_rd890_iommu(iommu->dev)) {
1322 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1326 * Some rd890 systems may not be fully reconfigured by the
1327 * BIOS, so it's necessary for us to store this information so
1328 * it can be reprogrammed on resume
1330 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1331 &iommu->stored_addr_lo);
1332 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1333 &iommu->stored_addr_hi);
1335 /* Low bit locks writes to configuration space */
1336 iommu->stored_addr_lo &= ~1;
1338 for (i = 0; i < 6; i++)
1339 for (j = 0; j < 0x12; j++)
1340 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1342 for (i = 0; i < 0x83; i++)
1343 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1346 amd_iommu_erratum_746_workaround(iommu);
1348 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1349 amd_iommu_groups, "ivhd%d",
1352 return pci_enable_device(iommu->dev);
1355 static void print_iommu_info(void)
1357 static const char * const feat_str[] = {
1358 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1359 "IA", "GA", "HE", "PC"
1361 struct amd_iommu *iommu;
1363 for_each_iommu(iommu) {
1366 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1367 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1369 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1370 pr_info("AMD-Vi: Extended features: ");
1371 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1372 if (iommu_feature(iommu, (1ULL << i)))
1373 pr_cont(" %s", feat_str[i]);
1378 if (irq_remapping_enabled)
1379 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1382 static int __init amd_iommu_init_pci(void)
1384 struct amd_iommu *iommu;
1387 for_each_iommu(iommu) {
1388 ret = iommu_init_pci(iommu);
1393 init_device_table_dma();
1395 for_each_iommu(iommu)
1396 iommu_flush_all_caches(iommu);
1398 ret = amd_iommu_init_api();
1406 /****************************************************************************
1408 * The following functions initialize the MSI interrupts for all IOMMUs
1409 * in the system. It's a bit challenging because there could be multiple
1410 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1413 ****************************************************************************/
1415 static int iommu_setup_msi(struct amd_iommu *iommu)
1419 r = pci_enable_msi(iommu->dev);
1423 r = request_threaded_irq(iommu->dev->irq,
1424 amd_iommu_int_handler,
1425 amd_iommu_int_thread,
1430 pci_disable_msi(iommu->dev);
1434 iommu->int_enabled = true;
1439 static int iommu_init_msi(struct amd_iommu *iommu)
1443 if (iommu->int_enabled)
1446 if (iommu->dev->msi_cap)
1447 ret = iommu_setup_msi(iommu);
1455 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1457 if (iommu->ppr_log != NULL)
1458 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1463 /****************************************************************************
1465 * The next functions belong to the third pass of parsing the ACPI
1466 * table. In this last pass the memory mapping requirements are
1467 * gathered (like exclusion and unity mapping ranges).
1469 ****************************************************************************/
1471 static void __init free_unity_maps(void)
1473 struct unity_map_entry *entry, *next;
1475 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1476 list_del(&entry->list);
1481 /* called when we find an exclusion range definition in ACPI */
1482 static int __init init_exclusion_range(struct ivmd_header *m)
1487 case ACPI_IVMD_TYPE:
1488 set_device_exclusion_range(m->devid, m);
1490 case ACPI_IVMD_TYPE_ALL:
1491 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1492 set_device_exclusion_range(i, m);
1494 case ACPI_IVMD_TYPE_RANGE:
1495 for (i = m->devid; i <= m->aux; ++i)
1496 set_device_exclusion_range(i, m);
1505 /* called for unity map ACPI definition */
1506 static int __init init_unity_map_range(struct ivmd_header *m)
1508 struct unity_map_entry *e = NULL;
1511 e = kzalloc(sizeof(*e), GFP_KERNEL);
1519 case ACPI_IVMD_TYPE:
1520 s = "IVMD_TYPEi\t\t\t";
1521 e->devid_start = e->devid_end = m->devid;
1523 case ACPI_IVMD_TYPE_ALL:
1524 s = "IVMD_TYPE_ALL\t\t";
1526 e->devid_end = amd_iommu_last_bdf;
1528 case ACPI_IVMD_TYPE_RANGE:
1529 s = "IVMD_TYPE_RANGE\t\t";
1530 e->devid_start = m->devid;
1531 e->devid_end = m->aux;
1534 e->address_start = PAGE_ALIGN(m->range_start);
1535 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1536 e->prot = m->flags >> 1;
1538 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1539 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1540 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1541 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1542 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1543 e->address_start, e->address_end, m->flags);
1545 list_add_tail(&e->list, &amd_iommu_unity_map);
1550 /* iterates over all memory definitions we find in the ACPI table */
1551 static int __init init_memory_definitions(struct acpi_table_header *table)
1553 u8 *p = (u8 *)table, *end = (u8 *)table;
1554 struct ivmd_header *m;
1556 end += table->length;
1557 p += IVRS_HEADER_LENGTH;
1560 m = (struct ivmd_header *)p;
1561 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1562 init_exclusion_range(m);
1563 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1564 init_unity_map_range(m);
1573 * Init the device table to not allow DMA access for devices and
1574 * suppress all page faults
1576 static void init_device_table_dma(void)
1580 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1581 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1582 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1586 static void __init uninit_device_table_dma(void)
1590 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1591 amd_iommu_dev_table[devid].data[0] = 0ULL;
1592 amd_iommu_dev_table[devid].data[1] = 0ULL;
1596 static void init_device_table(void)
1600 if (!amd_iommu_irq_remap)
1603 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1604 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1607 static void iommu_init_flags(struct amd_iommu *iommu)
1609 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1610 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1611 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1613 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1614 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1615 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1617 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1618 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1619 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1621 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1622 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1623 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1626 * make IOMMU memory accesses cache coherent
1628 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1630 /* Set IOTLB invalidation timeout to 1s */
1631 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1634 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1637 u32 ioc_feature_control;
1638 struct pci_dev *pdev = iommu->root_pdev;
1640 /* RD890 BIOSes may not have completely reconfigured the iommu */
1641 if (!is_rd890_iommu(iommu->dev) || !pdev)
1645 * First, we need to ensure that the iommu is enabled. This is
1646 * controlled by a register in the northbridge
1649 /* Select Northbridge indirect register 0x75 and enable writing */
1650 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1651 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1653 /* Enable the iommu */
1654 if (!(ioc_feature_control & 0x1))
1655 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1657 /* Restore the iommu BAR */
1658 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1659 iommu->stored_addr_lo);
1660 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1661 iommu->stored_addr_hi);
1663 /* Restore the l1 indirect regs for each of the 6 l1s */
1664 for (i = 0; i < 6; i++)
1665 for (j = 0; j < 0x12; j++)
1666 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1668 /* Restore the l2 indirect regs */
1669 for (i = 0; i < 0x83; i++)
1670 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1672 /* Lock PCI setup registers */
1673 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1674 iommu->stored_addr_lo | 1);
1678 * This function finally enables all IOMMUs found in the system after
1679 * they have been initialized
1681 static void early_enable_iommus(void)
1683 struct amd_iommu *iommu;
1685 for_each_iommu(iommu) {
1686 iommu_disable(iommu);
1687 iommu_init_flags(iommu);
1688 iommu_set_device_table(iommu);
1689 iommu_enable_command_buffer(iommu);
1690 iommu_enable_event_buffer(iommu);
1691 iommu_set_exclusion_range(iommu);
1692 iommu_enable(iommu);
1693 iommu_flush_all_caches(iommu);
1697 static void enable_iommus_v2(void)
1699 struct amd_iommu *iommu;
1701 for_each_iommu(iommu) {
1702 iommu_enable_ppr_log(iommu);
1703 iommu_enable_gt(iommu);
1707 static void enable_iommus(void)
1709 early_enable_iommus();
1714 static void disable_iommus(void)
1716 struct amd_iommu *iommu;
1718 for_each_iommu(iommu)
1719 iommu_disable(iommu);
1723 * Suspend/Resume support
1724 * disable suspend until real resume implemented
1727 static void amd_iommu_resume(void)
1729 struct amd_iommu *iommu;
1731 for_each_iommu(iommu)
1732 iommu_apply_resume_quirks(iommu);
1734 /* re-load the hardware */
1737 amd_iommu_enable_interrupts();
1740 static int amd_iommu_suspend(void)
1742 /* disable IOMMUs to go out of the way for BIOS */
1748 static struct syscore_ops amd_iommu_syscore_ops = {
1749 .suspend = amd_iommu_suspend,
1750 .resume = amd_iommu_resume,
1753 static void __init free_on_init_error(void)
1755 free_pages((unsigned long)irq_lookup_table,
1756 get_order(rlookup_table_size));
1758 if (amd_iommu_irq_cache) {
1759 kmem_cache_destroy(amd_iommu_irq_cache);
1760 amd_iommu_irq_cache = NULL;
1764 free_pages((unsigned long)amd_iommu_rlookup_table,
1765 get_order(rlookup_table_size));
1767 free_pages((unsigned long)amd_iommu_alias_table,
1768 get_order(alias_table_size));
1770 free_pages((unsigned long)amd_iommu_dev_table,
1771 get_order(dev_table_size));
1775 #ifdef CONFIG_GART_IOMMU
1777 * We failed to initialize the AMD IOMMU - try fallback to GART
1785 /* SB IOAPIC is always on this device in AMD systems */
1786 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1788 static bool __init check_ioapic_information(void)
1790 const char *fw_bug = FW_BUG;
1791 bool ret, has_sb_ioapic;
1794 has_sb_ioapic = false;
1798 * If we have map overrides on the kernel command line the
1799 * messages in this function might not describe firmware bugs
1800 * anymore - so be careful
1805 for (idx = 0; idx < nr_ioapics; idx++) {
1806 int devid, id = mpc_ioapic_id(idx);
1808 devid = get_ioapic_devid(id);
1810 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1813 } else if (devid == IOAPIC_SB_DEVID) {
1814 has_sb_ioapic = true;
1819 if (!has_sb_ioapic) {
1821 * We expect the SB IOAPIC to be listed in the IVRS
1822 * table. The system timer is connected to the SB IOAPIC
1823 * and if we don't have it in the list the system will
1824 * panic at boot time. This situation usually happens
1825 * when the BIOS is buggy and provides us the wrong
1826 * device id for the IOAPIC in the system.
1828 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
1832 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1837 static void __init free_dma_resources(void)
1839 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1840 get_order(MAX_DOMAIN_ID/8));
1846 * This is the hardware init function for AMD IOMMU in the system.
1847 * This function is called either from amd_iommu_init or from the interrupt
1848 * remapping setup code.
1850 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1853 * 1 pass) Find the highest PCI device id the driver has to handle.
1854 * Upon this information the size of the data structures is
1855 * determined that needs to be allocated.
1857 * 2 pass) Initialize the data structures just allocated with the
1858 * information in the ACPI table about available AMD IOMMUs
1859 * in the system. It also maps the PCI devices in the
1860 * system to specific IOMMUs
1862 * 3 pass) After the basic data structures are allocated and
1863 * initialized we update them with information about memory
1864 * remapping requirements parsed out of the ACPI table in
1867 * After everything is set up the IOMMUs are enabled and the necessary
1868 * hotplug and suspend notifiers are registered.
1870 static int __init early_amd_iommu_init(void)
1872 struct acpi_table_header *ivrs_base;
1873 acpi_size ivrs_size;
1877 if (!amd_iommu_detected)
1880 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1881 if (status == AE_NOT_FOUND)
1883 else if (ACPI_FAILURE(status)) {
1884 const char *err = acpi_format_exception(status);
1885 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1890 * First parse ACPI tables to find the largest Bus/Dev/Func
1891 * we need to handle. Upon this information the shared data
1892 * structures for the IOMMUs in the system will be allocated
1894 ret = find_last_devid_acpi(ivrs_base);
1898 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1899 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1900 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1902 /* Device table - directly used by all IOMMUs */
1904 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1905 get_order(dev_table_size));
1906 if (amd_iommu_dev_table == NULL)
1910 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1911 * IOMMU see for that device
1913 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1914 get_order(alias_table_size));
1915 if (amd_iommu_alias_table == NULL)
1918 /* IOMMU rlookup table - find the IOMMU for a specific device */
1919 amd_iommu_rlookup_table = (void *)__get_free_pages(
1920 GFP_KERNEL | __GFP_ZERO,
1921 get_order(rlookup_table_size));
1922 if (amd_iommu_rlookup_table == NULL)
1925 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1926 GFP_KERNEL | __GFP_ZERO,
1927 get_order(MAX_DOMAIN_ID/8));
1928 if (amd_iommu_pd_alloc_bitmap == NULL)
1932 * let all alias entries point to itself
1934 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1935 amd_iommu_alias_table[i] = i;
1938 * never allocate domain 0 because its used as the non-allocated and
1939 * error value placeholder
1941 amd_iommu_pd_alloc_bitmap[0] = 1;
1943 spin_lock_init(&amd_iommu_pd_lock);
1946 * now the data structures are allocated and basically initialized
1947 * start the real acpi table scan
1949 ret = init_iommu_all(ivrs_base);
1953 if (amd_iommu_irq_remap)
1954 amd_iommu_irq_remap = check_ioapic_information();
1956 if (amd_iommu_irq_remap) {
1958 * Interrupt remapping enabled, create kmem_cache for the
1962 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
1963 MAX_IRQS_PER_TABLE * sizeof(u32),
1964 IRQ_TABLE_ALIGNMENT,
1966 if (!amd_iommu_irq_cache)
1969 irq_lookup_table = (void *)__get_free_pages(
1970 GFP_KERNEL | __GFP_ZERO,
1971 get_order(rlookup_table_size));
1972 if (!irq_lookup_table)
1976 ret = init_memory_definitions(ivrs_base);
1980 /* init the device table */
1981 init_device_table();
1984 /* Don't leak any ACPI memory */
1985 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1991 static int amd_iommu_enable_interrupts(void)
1993 struct amd_iommu *iommu;
1996 for_each_iommu(iommu) {
1997 ret = iommu_init_msi(iommu);
2006 static bool detect_ivrs(void)
2008 struct acpi_table_header *ivrs_base;
2009 acpi_size ivrs_size;
2012 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
2013 if (status == AE_NOT_FOUND)
2015 else if (ACPI_FAILURE(status)) {
2016 const char *err = acpi_format_exception(status);
2017 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2021 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2023 /* Make sure ACS will be enabled during PCI probe */
2029 /****************************************************************************
2031 * AMD IOMMU Initialization State Machine
2033 ****************************************************************************/
2035 static int __init state_next(void)
2039 switch (init_state) {
2040 case IOMMU_START_STATE:
2041 if (!detect_ivrs()) {
2042 init_state = IOMMU_NOT_FOUND;
2045 init_state = IOMMU_IVRS_DETECTED;
2048 case IOMMU_IVRS_DETECTED:
2049 ret = early_amd_iommu_init();
2050 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2052 case IOMMU_ACPI_FINISHED:
2053 early_enable_iommus();
2054 register_syscore_ops(&amd_iommu_syscore_ops);
2055 x86_platform.iommu_shutdown = disable_iommus;
2056 init_state = IOMMU_ENABLED;
2059 ret = amd_iommu_init_pci();
2060 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2063 case IOMMU_PCI_INIT:
2064 ret = amd_iommu_enable_interrupts();
2065 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2067 case IOMMU_INTERRUPTS_EN:
2068 ret = amd_iommu_init_dma_ops();
2069 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2072 init_state = IOMMU_INITIALIZED;
2074 case IOMMU_INITIALIZED:
2077 case IOMMU_NOT_FOUND:
2078 case IOMMU_INIT_ERROR:
2079 /* Error states => do nothing */
2090 static int __init iommu_go_to_state(enum iommu_init_state state)
2094 while (init_state != state) {
2096 if (init_state == IOMMU_NOT_FOUND ||
2097 init_state == IOMMU_INIT_ERROR)
2104 #ifdef CONFIG_IRQ_REMAP
2105 int __init amd_iommu_prepare(void)
2109 amd_iommu_irq_remap = true;
2111 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2114 return amd_iommu_irq_remap ? 0 : -ENODEV;
2117 int __init amd_iommu_enable(void)
2121 ret = iommu_go_to_state(IOMMU_ENABLED);
2125 irq_remapping_enabled = 1;
2130 void amd_iommu_disable(void)
2132 amd_iommu_suspend();
2135 int amd_iommu_reenable(int mode)
2142 int __init amd_iommu_enable_faulting(void)
2144 /* We enable MSI later when PCI is initialized */
2150 * This is the core init function for AMD IOMMU hardware in the system.
2151 * This function is called from the generic x86 DMA layer initialization
2154 static int __init amd_iommu_init(void)
2158 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2160 free_dma_resources();
2161 if (!irq_remapping_enabled) {
2163 free_on_init_error();
2165 struct amd_iommu *iommu;
2167 uninit_device_table_dma();
2168 for_each_iommu(iommu)
2169 iommu_flush_all_caches(iommu);
2176 /****************************************************************************
2178 * Early detect code. This code runs at IOMMU detection time in the DMA
2179 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2182 ****************************************************************************/
2183 int __init amd_iommu_detect(void)
2187 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2190 if (amd_iommu_disabled)
2193 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2197 amd_iommu_detected = true;
2199 x86_init.iommu.iommu_init = amd_iommu_init;
2204 /****************************************************************************
2206 * Parsing functions for the AMD IOMMU specific kernel command line
2209 ****************************************************************************/
2211 static int __init parse_amd_iommu_dump(char *str)
2213 amd_iommu_dump = true;
2218 static int __init parse_amd_iommu_options(char *str)
2220 for (; *str; ++str) {
2221 if (strncmp(str, "fullflush", 9) == 0)
2222 amd_iommu_unmap_flush = true;
2223 if (strncmp(str, "off", 3) == 0)
2224 amd_iommu_disabled = true;
2225 if (strncmp(str, "force_isolation", 15) == 0)
2226 amd_iommu_force_isolation = true;
2232 static int __init parse_ivrs_ioapic(char *str)
2234 unsigned int bus, dev, fn;
2238 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2241 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2245 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2246 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2251 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2253 cmdline_maps = true;
2254 i = early_ioapic_map_size++;
2255 early_ioapic_map[i].id = id;
2256 early_ioapic_map[i].devid = devid;
2257 early_ioapic_map[i].cmd_line = true;
2262 static int __init parse_ivrs_hpet(char *str)
2264 unsigned int bus, dev, fn;
2268 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2271 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2275 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2276 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2281 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2283 cmdline_maps = true;
2284 i = early_hpet_map_size++;
2285 early_hpet_map[i].id = id;
2286 early_hpet_map[i].devid = devid;
2287 early_hpet_map[i].cmd_line = true;
2292 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2293 __setup("amd_iommu=", parse_amd_iommu_options);
2294 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2295 __setup("ivrs_hpet", parse_ivrs_hpet);
2297 IOMMU_INIT_FINISH(amd_iommu_detect,
2298 gart_iommu_hole_init,
2302 bool amd_iommu_v2_supported(void)
2304 return amd_iommu_v2_present;
2306 EXPORT_SYMBOL(amd_iommu_v2_supported);
2308 /****************************************************************************
2310 * IOMMU EFR Performance Counter support functionality. This code allows
2311 * access to the IOMMU PC functionality.
2313 ****************************************************************************/
2315 u8 amd_iommu_pc_get_max_banks(u16 devid)
2317 struct amd_iommu *iommu;
2320 /* locate the iommu governing the devid */
2321 iommu = amd_iommu_rlookup_table[devid];
2323 ret = iommu->max_banks;
2327 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2329 bool amd_iommu_pc_supported(void)
2331 return amd_iommu_pc_present;
2333 EXPORT_SYMBOL(amd_iommu_pc_supported);
2335 u8 amd_iommu_pc_get_max_counters(u16 devid)
2337 struct amd_iommu *iommu;
2340 /* locate the iommu governing the devid */
2341 iommu = amd_iommu_rlookup_table[devid];
2343 ret = iommu->max_counters;
2347 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2349 int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2350 u64 *value, bool is_write)
2352 struct amd_iommu *iommu;
2356 /* Make sure the IOMMU PC resource is available */
2357 if (!amd_iommu_pc_present)
2360 /* Locate the iommu associated with the device ID */
2361 iommu = amd_iommu_rlookup_table[devid];
2363 /* Check for valid iommu and pc register indexing */
2364 if (WARN_ON((iommu == NULL) || (fxn > 0x28) || (fxn & 7)))
2367 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2369 /* Limit the offset to the hw defined mmio region aperture */
2370 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2371 (iommu->max_counters << 8) | 0x28);
2372 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2373 (offset > max_offset_lim))
2377 writel((u32)*value, iommu->mmio_base + offset);
2378 writel((*value >> 32), iommu->mmio_base + offset + 4);
2380 *value = readl(iommu->mmio_base + offset + 4);
2382 *value = readl(iommu->mmio_base + offset);
2387 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);