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 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);
230 static inline void update_last_devid(u16 devid)
232 if (devid > amd_iommu_last_bdf)
233 amd_iommu_last_bdf = devid;
236 static inline unsigned long tbl_size(int entry_size)
238 unsigned shift = PAGE_SHIFT +
239 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
244 /* Access to l1 and l2 indexed register spaces */
246 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
250 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
251 pci_read_config_dword(iommu->dev, 0xfc, &val);
255 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
257 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
258 pci_write_config_dword(iommu->dev, 0xfc, val);
259 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
262 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
266 pci_write_config_dword(iommu->dev, 0xf0, address);
267 pci_read_config_dword(iommu->dev, 0xf4, &val);
271 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
273 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
274 pci_write_config_dword(iommu->dev, 0xf4, val);
277 /****************************************************************************
279 * AMD IOMMU MMIO register space handling functions
281 * These functions are used to program the IOMMU device registers in
282 * MMIO space required for that driver.
284 ****************************************************************************/
287 * This function set the exclusion range in the IOMMU. DMA accesses to the
288 * exclusion range are passed through untranslated
290 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
292 u64 start = iommu->exclusion_start & PAGE_MASK;
293 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
296 if (!iommu->exclusion_start)
299 entry = start | MMIO_EXCL_ENABLE_MASK;
300 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
301 &entry, sizeof(entry));
304 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
305 &entry, sizeof(entry));
308 /* Programs the physical address of the device table into the IOMMU hardware */
309 static void iommu_set_device_table(struct amd_iommu *iommu)
313 BUG_ON(iommu->mmio_base == NULL);
315 entry = virt_to_phys(amd_iommu_dev_table);
316 entry |= (dev_table_size >> 12) - 1;
317 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
318 &entry, sizeof(entry));
321 /* Generic functions to enable/disable certain features of the IOMMU. */
322 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
326 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
328 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
331 static void iommu_feature_disable(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_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
344 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
345 ctrl &= ~CTRL_INV_TO_MASK;
346 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
347 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
350 /* Function to enable the hardware */
351 static void iommu_enable(struct amd_iommu *iommu)
353 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
356 static void iommu_disable(struct amd_iommu *iommu)
358 /* Disable command buffer */
359 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
361 /* Disable event logging and event interrupts */
362 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
363 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
365 /* Disable IOMMU hardware itself */
366 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
370 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
371 * the system has one.
373 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
375 if (!request_mem_region(address, end, "amd_iommu")) {
376 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
378 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
382 return (u8 __iomem *)ioremap_nocache(address, end);
385 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
387 if (iommu->mmio_base)
388 iounmap(iommu->mmio_base);
389 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
392 /****************************************************************************
394 * The functions below belong to the first pass of AMD IOMMU ACPI table
395 * parsing. In this pass we try to find out the highest device id this
396 * code has to handle. Upon this information the size of the shared data
397 * structures is determined later.
399 ****************************************************************************/
402 * This function calculates the length of a given IVHD entry
404 static inline int ivhd_entry_length(u8 *ivhd)
406 return 0x04 << (*ivhd >> 6);
410 * This function reads the last device id the IOMMU has to handle from the PCI
411 * capability header for this IOMMU
413 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
417 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
418 update_last_devid(PCI_DEVID(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
424 * After reading the highest device id from the IOMMU PCI capability header
425 * this function looks if there is a higher device id defined in the ACPI table
427 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
429 u8 *p = (void *)h, *end = (void *)h;
430 struct ivhd_entry *dev;
435 find_last_devid_on_pci(PCI_BUS_NUM(h->devid),
441 dev = (struct ivhd_entry *)p;
443 case IVHD_DEV_SELECT:
444 case IVHD_DEV_RANGE_END:
446 case IVHD_DEV_EXT_SELECT:
447 /* all the above subfield types refer to device ids */
448 update_last_devid(dev->devid);
453 p += ivhd_entry_length(p);
462 * Iterate over all IVHD entries in the ACPI table and find the highest device
463 * id which we need to handle. This is the first of three functions which parse
464 * the ACPI table. So we check the checksum here.
466 static int __init find_last_devid_acpi(struct acpi_table_header *table)
469 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
470 struct ivhd_header *h;
473 * Validate checksum here so we don't need to do it when
474 * we actually parse the table
476 for (i = 0; i < table->length; ++i)
479 /* ACPI table corrupt */
482 p += IVRS_HEADER_LENGTH;
484 end += table->length;
486 h = (struct ivhd_header *)p;
489 find_last_devid_from_ivhd(h);
501 /****************************************************************************
503 * The following functions belong to the code path which parses the ACPI table
504 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
505 * data structures, initialize the device/alias/rlookup table and also
506 * basically initialize the hardware.
508 ****************************************************************************/
511 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
512 * write commands to that buffer later and the IOMMU will execute them
515 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
517 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
518 get_order(CMD_BUFFER_SIZE));
523 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
529 * This function resets the command buffer if the IOMMU stopped fetching
532 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
534 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
536 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
537 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
539 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
543 * This function writes the command buffer address to the hardware and
546 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
550 BUG_ON(iommu->cmd_buf == NULL);
552 entry = (u64)virt_to_phys(iommu->cmd_buf);
553 entry |= MMIO_CMD_SIZE_512;
555 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
556 &entry, sizeof(entry));
558 amd_iommu_reset_cmd_buffer(iommu);
559 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
562 static void __init free_command_buffer(struct amd_iommu *iommu)
564 free_pages((unsigned long)iommu->cmd_buf,
565 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
568 /* allocates the memory where the IOMMU will log its events to */
569 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
571 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
572 get_order(EVT_BUFFER_SIZE));
574 if (iommu->evt_buf == NULL)
577 iommu->evt_buf_size = EVT_BUFFER_SIZE;
579 return iommu->evt_buf;
582 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
586 BUG_ON(iommu->evt_buf == NULL);
588 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
590 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
591 &entry, sizeof(entry));
593 /* set head and tail to zero manually */
594 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
595 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
597 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
600 static void __init free_event_buffer(struct amd_iommu *iommu)
602 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
605 /* allocates the memory where the IOMMU will log its events to */
606 static u8 * __init alloc_ppr_log(struct amd_iommu *iommu)
608 iommu->ppr_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
609 get_order(PPR_LOG_SIZE));
611 if (iommu->ppr_log == NULL)
614 return iommu->ppr_log;
617 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
621 if (iommu->ppr_log == NULL)
624 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
626 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
627 &entry, sizeof(entry));
629 /* set head and tail to zero manually */
630 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
631 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
633 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
634 iommu_feature_enable(iommu, CONTROL_PPR_EN);
637 static void __init free_ppr_log(struct amd_iommu *iommu)
639 if (iommu->ppr_log == NULL)
642 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
645 static void iommu_enable_gt(struct amd_iommu *iommu)
647 if (!iommu_feature(iommu, FEATURE_GT))
650 iommu_feature_enable(iommu, CONTROL_GT_EN);
653 /* sets a specific bit in the device table entry. */
654 static void set_dev_entry_bit(u16 devid, u8 bit)
656 int i = (bit >> 6) & 0x03;
657 int _bit = bit & 0x3f;
659 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
662 static int get_dev_entry_bit(u16 devid, u8 bit)
664 int i = (bit >> 6) & 0x03;
665 int _bit = bit & 0x3f;
667 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
671 void amd_iommu_apply_erratum_63(u16 devid)
675 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
676 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
679 set_dev_entry_bit(devid, DEV_ENTRY_IW);
682 /* Writes the specific IOMMU for a device into the rlookup table */
683 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
685 amd_iommu_rlookup_table[devid] = iommu;
689 * This function takes the device specific flags read from the ACPI
690 * table and sets up the device table entry with that information
692 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
693 u16 devid, u32 flags, u32 ext_flags)
695 if (flags & ACPI_DEVFLAG_INITPASS)
696 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
697 if (flags & ACPI_DEVFLAG_EXTINT)
698 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
699 if (flags & ACPI_DEVFLAG_NMI)
700 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
701 if (flags & ACPI_DEVFLAG_SYSMGT1)
702 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
703 if (flags & ACPI_DEVFLAG_SYSMGT2)
704 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
705 if (flags & ACPI_DEVFLAG_LINT0)
706 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
707 if (flags & ACPI_DEVFLAG_LINT1)
708 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
710 amd_iommu_apply_erratum_63(devid);
712 set_iommu_for_device(iommu, devid);
715 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
717 struct devid_map *entry;
718 struct list_head *list;
720 if (type == IVHD_SPECIAL_IOAPIC)
722 else if (type == IVHD_SPECIAL_HPET)
727 list_for_each_entry(entry, list, list) {
728 if (!(entry->id == id && entry->cmd_line))
731 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
732 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
734 *devid = entry->devid;
739 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
744 entry->devid = *devid;
745 entry->cmd_line = cmd_line;
747 list_add_tail(&entry->list, list);
752 static int __init add_early_maps(void)
756 for (i = 0; i < early_ioapic_map_size; ++i) {
757 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
758 early_ioapic_map[i].id,
759 &early_ioapic_map[i].devid,
760 early_ioapic_map[i].cmd_line);
765 for (i = 0; i < early_hpet_map_size; ++i) {
766 ret = add_special_device(IVHD_SPECIAL_HPET,
767 early_hpet_map[i].id,
768 &early_hpet_map[i].devid,
769 early_hpet_map[i].cmd_line);
778 * Reads the device exclusion range from ACPI and initializes the IOMMU with
781 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
783 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
785 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
790 * We only can configure exclusion ranges per IOMMU, not
791 * per device. But we can enable the exclusion range per
792 * device. This is done here
794 set_dev_entry_bit(devid, DEV_ENTRY_EX);
795 iommu->exclusion_start = m->range_start;
796 iommu->exclusion_length = m->range_length;
801 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
802 * initializes the hardware and our data structures with it.
804 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
805 struct ivhd_header *h)
808 u8 *end = p, flags = 0;
809 u16 devid = 0, devid_start = 0, devid_to = 0;
810 u32 dev_i, ext_flags = 0;
812 struct ivhd_entry *e;
816 ret = add_early_maps();
821 * First save the recommended feature enable bits from ACPI
823 iommu->acpi_flags = h->flags;
826 * Done. Now parse the device entries
828 p += sizeof(struct ivhd_header);
833 e = (struct ivhd_entry *)p;
837 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
838 " last device %02x:%02x.%x flags: %02x\n",
839 PCI_BUS_NUM(iommu->first_device),
840 PCI_SLOT(iommu->first_device),
841 PCI_FUNC(iommu->first_device),
842 PCI_BUS_NUM(iommu->last_device),
843 PCI_SLOT(iommu->last_device),
844 PCI_FUNC(iommu->last_device),
847 for (dev_i = iommu->first_device;
848 dev_i <= iommu->last_device; ++dev_i)
849 set_dev_entry_from_acpi(iommu, dev_i,
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 /* Initializes the device->iommu mapping for the driver */
1007 static int __init init_iommu_devices(struct amd_iommu *iommu)
1011 for (i = iommu->first_device; i <= iommu->last_device; ++i)
1012 set_iommu_for_device(iommu, i);
1017 static void __init free_iommu_one(struct amd_iommu *iommu)
1019 free_command_buffer(iommu);
1020 free_event_buffer(iommu);
1021 free_ppr_log(iommu);
1022 iommu_unmap_mmio_space(iommu);
1025 static void __init free_iommu_all(void)
1027 struct amd_iommu *iommu, *next;
1029 for_each_iommu_safe(iommu, next) {
1030 list_del(&iommu->list);
1031 free_iommu_one(iommu);
1037 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1039 * BIOS should disable L2B micellaneous clock gating by setting
1040 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1042 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1046 if ((boot_cpu_data.x86 != 0x15) ||
1047 (boot_cpu_data.x86_model < 0x10) ||
1048 (boot_cpu_data.x86_model > 0x1f))
1051 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1052 pci_read_config_dword(iommu->dev, 0xf4, &value);
1057 /* Select NB indirect register 0x90 and enable writing */
1058 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1060 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1061 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1062 dev_name(&iommu->dev->dev));
1064 /* Clear the enable writing bit */
1065 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1069 * This function clues the initialization function for one IOMMU
1070 * together and also allocates the command buffer and programs the
1071 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1073 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1077 spin_lock_init(&iommu->lock);
1079 /* Add IOMMU to internal data structures */
1080 list_add_tail(&iommu->list, &amd_iommu_list);
1081 iommu->index = amd_iommus_present++;
1083 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1084 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1088 /* Index is fine - add IOMMU to the array */
1089 amd_iommus[iommu->index] = iommu;
1092 * Copy data from ACPI table entry to the iommu struct
1094 iommu->devid = h->devid;
1095 iommu->cap_ptr = h->cap_ptr;
1096 iommu->pci_seg = h->pci_seg;
1097 iommu->mmio_phys = h->mmio_phys;
1099 /* Check if IVHD EFR contains proper max banks/counters */
1100 if ((h->efr != 0) &&
1101 ((h->efr & (0xF << 13)) != 0) &&
1102 ((h->efr & (0x3F << 17)) != 0)) {
1103 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1105 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1108 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1109 iommu->mmio_phys_end);
1110 if (!iommu->mmio_base)
1113 iommu->cmd_buf = alloc_command_buffer(iommu);
1114 if (!iommu->cmd_buf)
1117 iommu->evt_buf = alloc_event_buffer(iommu);
1118 if (!iommu->evt_buf)
1121 iommu->int_enabled = false;
1123 ret = init_iommu_from_acpi(iommu, h);
1127 ret = amd_iommu_create_irq_domain(iommu);
1132 * Make sure IOMMU is not considered to translate itself. The IVRS
1133 * table tells us so, but this is a lie!
1135 amd_iommu_rlookup_table[iommu->devid] = NULL;
1137 init_iommu_devices(iommu);
1143 * Iterates over all IOMMU entries in the ACPI table, allocates the
1144 * IOMMU structure and initializes it with init_iommu_one()
1146 static int __init init_iommu_all(struct acpi_table_header *table)
1148 u8 *p = (u8 *)table, *end = (u8 *)table;
1149 struct ivhd_header *h;
1150 struct amd_iommu *iommu;
1153 end += table->length;
1154 p += IVRS_HEADER_LENGTH;
1157 h = (struct ivhd_header *)p;
1159 case ACPI_IVHD_TYPE:
1161 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1162 "seg: %d flags: %01x info %04x\n",
1163 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1164 PCI_FUNC(h->devid), h->cap_ptr,
1165 h->pci_seg, h->flags, h->info);
1166 DUMP_printk(" mmio-addr: %016llx\n",
1169 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1173 ret = init_iommu_one(iommu, h);
1189 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1191 u64 val = 0xabcd, val2 = 0;
1193 if (!iommu_feature(iommu, FEATURE_PC))
1196 amd_iommu_pc_present = true;
1198 /* Check if the performance counters can be written to */
1199 if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val, true)) ||
1200 (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2, false)) ||
1202 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1203 amd_iommu_pc_present = false;
1207 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1209 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1210 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1211 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1214 static ssize_t amd_iommu_show_cap(struct device *dev,
1215 struct device_attribute *attr,
1218 struct amd_iommu *iommu = dev_get_drvdata(dev);
1219 return sprintf(buf, "%x\n", iommu->cap);
1221 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1223 static ssize_t amd_iommu_show_features(struct device *dev,
1224 struct device_attribute *attr,
1227 struct amd_iommu *iommu = dev_get_drvdata(dev);
1228 return sprintf(buf, "%llx\n", iommu->features);
1230 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1232 static struct attribute *amd_iommu_attrs[] = {
1234 &dev_attr_features.attr,
1238 static struct attribute_group amd_iommu_group = {
1239 .name = "amd-iommu",
1240 .attrs = amd_iommu_attrs,
1243 static const struct attribute_group *amd_iommu_groups[] = {
1248 static int iommu_init_pci(struct amd_iommu *iommu)
1250 int cap_ptr = iommu->cap_ptr;
1251 u32 range, misc, low, high;
1253 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1254 iommu->devid & 0xff);
1258 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1260 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1262 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1265 iommu->first_device = PCI_DEVID(MMIO_GET_BUS(range),
1266 MMIO_GET_FD(range));
1267 iommu->last_device = PCI_DEVID(MMIO_GET_BUS(range),
1268 MMIO_GET_LD(range));
1270 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1271 amd_iommu_iotlb_sup = false;
1273 /* read extended feature bits */
1274 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1275 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1277 iommu->features = ((u64)high << 32) | low;
1279 if (iommu_feature(iommu, FEATURE_GT)) {
1284 pasmax = iommu->features & FEATURE_PASID_MASK;
1285 pasmax >>= FEATURE_PASID_SHIFT;
1286 max_pasid = (1 << (pasmax + 1)) - 1;
1288 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1290 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1292 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1293 glxval >>= FEATURE_GLXVAL_SHIFT;
1295 if (amd_iommu_max_glx_val == -1)
1296 amd_iommu_max_glx_val = glxval;
1298 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1301 if (iommu_feature(iommu, FEATURE_GT) &&
1302 iommu_feature(iommu, FEATURE_PPR)) {
1303 iommu->is_iommu_v2 = true;
1304 amd_iommu_v2_present = true;
1307 if (iommu_feature(iommu, FEATURE_PPR)) {
1308 iommu->ppr_log = alloc_ppr_log(iommu);
1309 if (!iommu->ppr_log)
1313 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1314 amd_iommu_np_cache = true;
1316 init_iommu_perf_ctr(iommu);
1318 if (is_rd890_iommu(iommu->dev)) {
1321 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1325 * Some rd890 systems may not be fully reconfigured by the
1326 * BIOS, so it's necessary for us to store this information so
1327 * it can be reprogrammed on resume
1329 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1330 &iommu->stored_addr_lo);
1331 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1332 &iommu->stored_addr_hi);
1334 /* Low bit locks writes to configuration space */
1335 iommu->stored_addr_lo &= ~1;
1337 for (i = 0; i < 6; i++)
1338 for (j = 0; j < 0x12; j++)
1339 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1341 for (i = 0; i < 0x83; i++)
1342 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1345 amd_iommu_erratum_746_workaround(iommu);
1347 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1348 amd_iommu_groups, "ivhd%d",
1351 return pci_enable_device(iommu->dev);
1354 static void print_iommu_info(void)
1356 static const char * const feat_str[] = {
1357 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1358 "IA", "GA", "HE", "PC"
1360 struct amd_iommu *iommu;
1362 for_each_iommu(iommu) {
1365 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1366 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1368 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1369 pr_info("AMD-Vi: Extended features: ");
1370 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1371 if (iommu_feature(iommu, (1ULL << i)))
1372 pr_cont(" %s", feat_str[i]);
1377 if (irq_remapping_enabled)
1378 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1381 static int __init amd_iommu_init_pci(void)
1383 struct amd_iommu *iommu;
1386 for_each_iommu(iommu) {
1387 ret = iommu_init_pci(iommu);
1392 ret = amd_iommu_init_devices();
1399 /****************************************************************************
1401 * The following functions initialize the MSI interrupts for all IOMMUs
1402 * in the system. It's a bit challenging because there could be multiple
1403 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1406 ****************************************************************************/
1408 static int iommu_setup_msi(struct amd_iommu *iommu)
1412 r = pci_enable_msi(iommu->dev);
1416 r = request_threaded_irq(iommu->dev->irq,
1417 amd_iommu_int_handler,
1418 amd_iommu_int_thread,
1423 pci_disable_msi(iommu->dev);
1427 iommu->int_enabled = true;
1432 static int iommu_init_msi(struct amd_iommu *iommu)
1436 if (iommu->int_enabled)
1439 if (iommu->dev->msi_cap)
1440 ret = iommu_setup_msi(iommu);
1448 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1450 if (iommu->ppr_log != NULL)
1451 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1456 /****************************************************************************
1458 * The next functions belong to the third pass of parsing the ACPI
1459 * table. In this last pass the memory mapping requirements are
1460 * gathered (like exclusion and unity mapping ranges).
1462 ****************************************************************************/
1464 static void __init free_unity_maps(void)
1466 struct unity_map_entry *entry, *next;
1468 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1469 list_del(&entry->list);
1474 /* called when we find an exclusion range definition in ACPI */
1475 static int __init init_exclusion_range(struct ivmd_header *m)
1480 case ACPI_IVMD_TYPE:
1481 set_device_exclusion_range(m->devid, m);
1483 case ACPI_IVMD_TYPE_ALL:
1484 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1485 set_device_exclusion_range(i, m);
1487 case ACPI_IVMD_TYPE_RANGE:
1488 for (i = m->devid; i <= m->aux; ++i)
1489 set_device_exclusion_range(i, m);
1498 /* called for unity map ACPI definition */
1499 static int __init init_unity_map_range(struct ivmd_header *m)
1501 struct unity_map_entry *e = NULL;
1504 e = kzalloc(sizeof(*e), GFP_KERNEL);
1512 case ACPI_IVMD_TYPE:
1513 s = "IVMD_TYPEi\t\t\t";
1514 e->devid_start = e->devid_end = m->devid;
1516 case ACPI_IVMD_TYPE_ALL:
1517 s = "IVMD_TYPE_ALL\t\t";
1519 e->devid_end = amd_iommu_last_bdf;
1521 case ACPI_IVMD_TYPE_RANGE:
1522 s = "IVMD_TYPE_RANGE\t\t";
1523 e->devid_start = m->devid;
1524 e->devid_end = m->aux;
1527 e->address_start = PAGE_ALIGN(m->range_start);
1528 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1529 e->prot = m->flags >> 1;
1531 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1532 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1533 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1534 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1535 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1536 e->address_start, e->address_end, m->flags);
1538 list_add_tail(&e->list, &amd_iommu_unity_map);
1543 /* iterates over all memory definitions we find in the ACPI table */
1544 static int __init init_memory_definitions(struct acpi_table_header *table)
1546 u8 *p = (u8 *)table, *end = (u8 *)table;
1547 struct ivmd_header *m;
1549 end += table->length;
1550 p += IVRS_HEADER_LENGTH;
1553 m = (struct ivmd_header *)p;
1554 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1555 init_exclusion_range(m);
1556 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1557 init_unity_map_range(m);
1566 * Init the device table to not allow DMA access for devices and
1567 * suppress all page faults
1569 static void init_device_table_dma(void)
1573 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1574 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1575 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1579 static void __init uninit_device_table_dma(void)
1583 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1584 amd_iommu_dev_table[devid].data[0] = 0ULL;
1585 amd_iommu_dev_table[devid].data[1] = 0ULL;
1589 static void init_device_table(void)
1593 if (!amd_iommu_irq_remap)
1596 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1597 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1600 static void iommu_init_flags(struct amd_iommu *iommu)
1602 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1603 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1604 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1606 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1607 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1608 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1610 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1611 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1612 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1614 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1615 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1616 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1619 * make IOMMU memory accesses cache coherent
1621 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1623 /* Set IOTLB invalidation timeout to 1s */
1624 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1627 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1630 u32 ioc_feature_control;
1631 struct pci_dev *pdev = iommu->root_pdev;
1633 /* RD890 BIOSes may not have completely reconfigured the iommu */
1634 if (!is_rd890_iommu(iommu->dev) || !pdev)
1638 * First, we need to ensure that the iommu is enabled. This is
1639 * controlled by a register in the northbridge
1642 /* Select Northbridge indirect register 0x75 and enable writing */
1643 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1644 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1646 /* Enable the iommu */
1647 if (!(ioc_feature_control & 0x1))
1648 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1650 /* Restore the iommu BAR */
1651 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1652 iommu->stored_addr_lo);
1653 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1654 iommu->stored_addr_hi);
1656 /* Restore the l1 indirect regs for each of the 6 l1s */
1657 for (i = 0; i < 6; i++)
1658 for (j = 0; j < 0x12; j++)
1659 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1661 /* Restore the l2 indirect regs */
1662 for (i = 0; i < 0x83; i++)
1663 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1665 /* Lock PCI setup registers */
1666 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1667 iommu->stored_addr_lo | 1);
1671 * This function finally enables all IOMMUs found in the system after
1672 * they have been initialized
1674 static void early_enable_iommus(void)
1676 struct amd_iommu *iommu;
1678 for_each_iommu(iommu) {
1679 iommu_disable(iommu);
1680 iommu_init_flags(iommu);
1681 iommu_set_device_table(iommu);
1682 iommu_enable_command_buffer(iommu);
1683 iommu_enable_event_buffer(iommu);
1684 iommu_set_exclusion_range(iommu);
1685 iommu_enable(iommu);
1686 iommu_flush_all_caches(iommu);
1690 static void enable_iommus_v2(void)
1692 struct amd_iommu *iommu;
1694 for_each_iommu(iommu) {
1695 iommu_enable_ppr_log(iommu);
1696 iommu_enable_gt(iommu);
1700 static void enable_iommus(void)
1702 early_enable_iommus();
1707 static void disable_iommus(void)
1709 struct amd_iommu *iommu;
1711 for_each_iommu(iommu)
1712 iommu_disable(iommu);
1716 * Suspend/Resume support
1717 * disable suspend until real resume implemented
1720 static void amd_iommu_resume(void)
1722 struct amd_iommu *iommu;
1724 for_each_iommu(iommu)
1725 iommu_apply_resume_quirks(iommu);
1727 /* re-load the hardware */
1730 amd_iommu_enable_interrupts();
1733 static int amd_iommu_suspend(void)
1735 /* disable IOMMUs to go out of the way for BIOS */
1741 static struct syscore_ops amd_iommu_syscore_ops = {
1742 .suspend = amd_iommu_suspend,
1743 .resume = amd_iommu_resume,
1746 static void __init free_on_init_error(void)
1748 free_pages((unsigned long)irq_lookup_table,
1749 get_order(rlookup_table_size));
1751 if (amd_iommu_irq_cache) {
1752 kmem_cache_destroy(amd_iommu_irq_cache);
1753 amd_iommu_irq_cache = NULL;
1757 free_pages((unsigned long)amd_iommu_rlookup_table,
1758 get_order(rlookup_table_size));
1760 free_pages((unsigned long)amd_iommu_alias_table,
1761 get_order(alias_table_size));
1763 free_pages((unsigned long)amd_iommu_dev_table,
1764 get_order(dev_table_size));
1768 #ifdef CONFIG_GART_IOMMU
1770 * We failed to initialize the AMD IOMMU - try fallback to GART
1778 /* SB IOAPIC is always on this device in AMD systems */
1779 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1781 static bool __init check_ioapic_information(void)
1783 const char *fw_bug = FW_BUG;
1784 bool ret, has_sb_ioapic;
1787 has_sb_ioapic = false;
1791 * If we have map overrides on the kernel command line the
1792 * messages in this function might not describe firmware bugs
1793 * anymore - so be careful
1798 for (idx = 0; idx < nr_ioapics; idx++) {
1799 int devid, id = mpc_ioapic_id(idx);
1801 devid = get_ioapic_devid(id);
1803 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1806 } else if (devid == IOAPIC_SB_DEVID) {
1807 has_sb_ioapic = true;
1812 if (!has_sb_ioapic) {
1814 * We expect the SB IOAPIC to be listed in the IVRS
1815 * table. The system timer is connected to the SB IOAPIC
1816 * and if we don't have it in the list the system will
1817 * panic at boot time. This situation usually happens
1818 * when the BIOS is buggy and provides us the wrong
1819 * device id for the IOAPIC in the system.
1821 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
1825 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1830 static void __init free_dma_resources(void)
1832 amd_iommu_uninit_devices();
1834 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1835 get_order(MAX_DOMAIN_ID/8));
1841 * This is the hardware init function for AMD IOMMU in the system.
1842 * This function is called either from amd_iommu_init or from the interrupt
1843 * remapping setup code.
1845 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1848 * 1 pass) Find the highest PCI device id the driver has to handle.
1849 * Upon this information the size of the data structures is
1850 * determined that needs to be allocated.
1852 * 2 pass) Initialize the data structures just allocated with the
1853 * information in the ACPI table about available AMD IOMMUs
1854 * in the system. It also maps the PCI devices in the
1855 * system to specific IOMMUs
1857 * 3 pass) After the basic data structures are allocated and
1858 * initialized we update them with information about memory
1859 * remapping requirements parsed out of the ACPI table in
1862 * After everything is set up the IOMMUs are enabled and the necessary
1863 * hotplug and suspend notifiers are registered.
1865 static int __init early_amd_iommu_init(void)
1867 struct acpi_table_header *ivrs_base;
1868 acpi_size ivrs_size;
1872 if (!amd_iommu_detected)
1875 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1876 if (status == AE_NOT_FOUND)
1878 else if (ACPI_FAILURE(status)) {
1879 const char *err = acpi_format_exception(status);
1880 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1885 * First parse ACPI tables to find the largest Bus/Dev/Func
1886 * we need to handle. Upon this information the shared data
1887 * structures for the IOMMUs in the system will be allocated
1889 ret = find_last_devid_acpi(ivrs_base);
1893 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1894 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1895 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1897 /* Device table - directly used by all IOMMUs */
1899 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1900 get_order(dev_table_size));
1901 if (amd_iommu_dev_table == NULL)
1905 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1906 * IOMMU see for that device
1908 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1909 get_order(alias_table_size));
1910 if (amd_iommu_alias_table == NULL)
1913 /* IOMMU rlookup table - find the IOMMU for a specific device */
1914 amd_iommu_rlookup_table = (void *)__get_free_pages(
1915 GFP_KERNEL | __GFP_ZERO,
1916 get_order(rlookup_table_size));
1917 if (amd_iommu_rlookup_table == NULL)
1920 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1921 GFP_KERNEL | __GFP_ZERO,
1922 get_order(MAX_DOMAIN_ID/8));
1923 if (amd_iommu_pd_alloc_bitmap == NULL)
1927 * let all alias entries point to itself
1929 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1930 amd_iommu_alias_table[i] = i;
1933 * never allocate domain 0 because its used as the non-allocated and
1934 * error value placeholder
1936 amd_iommu_pd_alloc_bitmap[0] = 1;
1938 spin_lock_init(&amd_iommu_pd_lock);
1941 * now the data structures are allocated and basically initialized
1942 * start the real acpi table scan
1944 ret = init_iommu_all(ivrs_base);
1948 if (amd_iommu_irq_remap)
1949 amd_iommu_irq_remap = check_ioapic_information();
1951 if (amd_iommu_irq_remap) {
1953 * Interrupt remapping enabled, create kmem_cache for the
1957 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
1958 MAX_IRQS_PER_TABLE * sizeof(u32),
1959 IRQ_TABLE_ALIGNMENT,
1961 if (!amd_iommu_irq_cache)
1964 irq_lookup_table = (void *)__get_free_pages(
1965 GFP_KERNEL | __GFP_ZERO,
1966 get_order(rlookup_table_size));
1967 if (!irq_lookup_table)
1971 ret = init_memory_definitions(ivrs_base);
1975 /* init the device table */
1976 init_device_table();
1979 /* Don't leak any ACPI memory */
1980 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1986 static int amd_iommu_enable_interrupts(void)
1988 struct amd_iommu *iommu;
1991 for_each_iommu(iommu) {
1992 ret = iommu_init_msi(iommu);
2001 static bool detect_ivrs(void)
2003 struct acpi_table_header *ivrs_base;
2004 acpi_size ivrs_size;
2007 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
2008 if (status == AE_NOT_FOUND)
2010 else if (ACPI_FAILURE(status)) {
2011 const char *err = acpi_format_exception(status);
2012 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2016 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2018 /* Make sure ACS will be enabled during PCI probe */
2024 static int amd_iommu_init_dma(void)
2026 struct amd_iommu *iommu;
2029 if (iommu_pass_through)
2030 ret = amd_iommu_init_passthrough();
2032 ret = amd_iommu_init_dma_ops();
2037 init_device_table_dma();
2039 for_each_iommu(iommu)
2040 iommu_flush_all_caches(iommu);
2042 amd_iommu_init_api();
2044 amd_iommu_init_notifier();
2049 /****************************************************************************
2051 * AMD IOMMU Initialization State Machine
2053 ****************************************************************************/
2055 static int __init state_next(void)
2059 switch (init_state) {
2060 case IOMMU_START_STATE:
2061 if (!detect_ivrs()) {
2062 init_state = IOMMU_NOT_FOUND;
2065 init_state = IOMMU_IVRS_DETECTED;
2068 case IOMMU_IVRS_DETECTED:
2069 ret = early_amd_iommu_init();
2070 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2072 case IOMMU_ACPI_FINISHED:
2073 early_enable_iommus();
2074 register_syscore_ops(&amd_iommu_syscore_ops);
2075 x86_platform.iommu_shutdown = disable_iommus;
2076 init_state = IOMMU_ENABLED;
2079 ret = amd_iommu_init_pci();
2080 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2083 case IOMMU_PCI_INIT:
2084 ret = amd_iommu_enable_interrupts();
2085 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2087 case IOMMU_INTERRUPTS_EN:
2088 ret = amd_iommu_init_dma();
2089 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2092 init_state = IOMMU_INITIALIZED;
2094 case IOMMU_INITIALIZED:
2097 case IOMMU_NOT_FOUND:
2098 case IOMMU_INIT_ERROR:
2099 /* Error states => do nothing */
2110 static int __init iommu_go_to_state(enum iommu_init_state state)
2114 while (init_state != state) {
2116 if (init_state == IOMMU_NOT_FOUND ||
2117 init_state == IOMMU_INIT_ERROR)
2124 #ifdef CONFIG_IRQ_REMAP
2125 int __init amd_iommu_prepare(void)
2129 amd_iommu_irq_remap = true;
2131 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2134 return amd_iommu_irq_remap ? 0 : -ENODEV;
2137 int __init amd_iommu_enable(void)
2141 ret = iommu_go_to_state(IOMMU_ENABLED);
2145 irq_remapping_enabled = 1;
2150 void amd_iommu_disable(void)
2152 amd_iommu_suspend();
2155 int amd_iommu_reenable(int mode)
2162 int __init amd_iommu_enable_faulting(void)
2164 /* We enable MSI later when PCI is initialized */
2170 * This is the core init function for AMD IOMMU hardware in the system.
2171 * This function is called from the generic x86 DMA layer initialization
2174 static int __init amd_iommu_init(void)
2178 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2180 free_dma_resources();
2181 if (!irq_remapping_enabled) {
2183 free_on_init_error();
2185 struct amd_iommu *iommu;
2187 uninit_device_table_dma();
2188 for_each_iommu(iommu)
2189 iommu_flush_all_caches(iommu);
2196 /****************************************************************************
2198 * Early detect code. This code runs at IOMMU detection time in the DMA
2199 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2202 ****************************************************************************/
2203 int __init amd_iommu_detect(void)
2207 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2210 if (amd_iommu_disabled)
2213 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2217 amd_iommu_detected = true;
2219 x86_init.iommu.iommu_init = amd_iommu_init;
2224 /****************************************************************************
2226 * Parsing functions for the AMD IOMMU specific kernel command line
2229 ****************************************************************************/
2231 static int __init parse_amd_iommu_dump(char *str)
2233 amd_iommu_dump = true;
2238 static int __init parse_amd_iommu_options(char *str)
2240 for (; *str; ++str) {
2241 if (strncmp(str, "fullflush", 9) == 0)
2242 amd_iommu_unmap_flush = true;
2243 if (strncmp(str, "off", 3) == 0)
2244 amd_iommu_disabled = true;
2245 if (strncmp(str, "force_isolation", 15) == 0)
2246 amd_iommu_force_isolation = true;
2252 static int __init parse_ivrs_ioapic(char *str)
2254 unsigned int bus, dev, fn;
2258 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2261 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2265 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2266 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2271 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2273 cmdline_maps = true;
2274 i = early_ioapic_map_size++;
2275 early_ioapic_map[i].id = id;
2276 early_ioapic_map[i].devid = devid;
2277 early_ioapic_map[i].cmd_line = true;
2282 static int __init parse_ivrs_hpet(char *str)
2284 unsigned int bus, dev, fn;
2288 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2291 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2295 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2296 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2301 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2303 cmdline_maps = true;
2304 i = early_hpet_map_size++;
2305 early_hpet_map[i].id = id;
2306 early_hpet_map[i].devid = devid;
2307 early_hpet_map[i].cmd_line = true;
2312 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2313 __setup("amd_iommu=", parse_amd_iommu_options);
2314 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2315 __setup("ivrs_hpet", parse_ivrs_hpet);
2317 IOMMU_INIT_FINISH(amd_iommu_detect,
2318 gart_iommu_hole_init,
2322 bool amd_iommu_v2_supported(void)
2324 return amd_iommu_v2_present;
2326 EXPORT_SYMBOL(amd_iommu_v2_supported);
2328 /****************************************************************************
2330 * IOMMU EFR Performance Counter support functionality. This code allows
2331 * access to the IOMMU PC functionality.
2333 ****************************************************************************/
2335 u8 amd_iommu_pc_get_max_banks(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_banks;
2347 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2349 bool amd_iommu_pc_supported(void)
2351 return amd_iommu_pc_present;
2353 EXPORT_SYMBOL(amd_iommu_pc_supported);
2355 u8 amd_iommu_pc_get_max_counters(u16 devid)
2357 struct amd_iommu *iommu;
2360 /* locate the iommu governing the devid */
2361 iommu = amd_iommu_rlookup_table[devid];
2363 ret = iommu->max_counters;
2367 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2369 int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2370 u64 *value, bool is_write)
2372 struct amd_iommu *iommu;
2376 /* Make sure the IOMMU PC resource is available */
2377 if (!amd_iommu_pc_present)
2380 /* Locate the iommu associated with the device ID */
2381 iommu = amd_iommu_rlookup_table[devid];
2383 /* Check for valid iommu and pc register indexing */
2384 if (WARN_ON((iommu == NULL) || (fxn > 0x28) || (fxn & 7)))
2387 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2389 /* Limit the offset to the hw defined mmio region aperture */
2390 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2391 (iommu->max_counters << 8) | 0x28);
2392 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2393 (offset > max_offset_lim))
2397 writel((u32)*value, iommu->mmio_base + offset);
2398 writel((*value >> 32), iommu->mmio_base + offset + 4);
2400 *value = readl(iommu->mmio_base + offset + 4);
2402 *value = readl(iommu->mmio_base + offset);
2407 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);