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_MAX_SUPPORTED 0x40
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
61 #define IVHD_DEV_ACPI_HID 0xf0
63 #define IVHD_SPECIAL_IOAPIC 1
64 #define IVHD_SPECIAL_HPET 2
66 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
67 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
68 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
69 #define IVHD_FLAG_ISOC_EN_MASK 0x08
71 #define IVMD_FLAG_EXCL_RANGE 0x08
72 #define IVMD_FLAG_UNITY_MAP 0x01
74 #define ACPI_DEVFLAG_INITPASS 0x01
75 #define ACPI_DEVFLAG_EXTINT 0x02
76 #define ACPI_DEVFLAG_NMI 0x04
77 #define ACPI_DEVFLAG_SYSMGT1 0x10
78 #define ACPI_DEVFLAG_SYSMGT2 0x20
79 #define ACPI_DEVFLAG_LINT0 0x40
80 #define ACPI_DEVFLAG_LINT1 0x80
81 #define ACPI_DEVFLAG_ATSDIS 0x10000000
84 * ACPI table definitions
86 * These data structures are laid over the table to parse the important values
91 * structure describing one IOMMU in the ACPI table. Typically followed by one
92 * or more ivhd_entrys.
105 /* Following only valid on IVHD type 11h and 40h */
106 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
108 } __attribute__((packed));
111 * A device entry describing which devices a specific IOMMU translates and
112 * which requestor ids they use.
119 } __attribute__((packed));
122 * An AMD IOMMU memory definition structure. It defines things like exclusion
123 * ranges for devices and regions that should be unity mapped.
134 } __attribute__((packed));
137 bool amd_iommu_irq_remap __read_mostly;
139 static bool amd_iommu_detected;
140 static bool __initdata amd_iommu_disabled;
141 static int amd_iommu_target_ivhd_type;
143 u16 amd_iommu_last_bdf; /* largest PCI device id we have
145 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
147 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
149 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
152 /* Array to assign indices to IOMMUs*/
153 struct amd_iommu *amd_iommus[MAX_IOMMUS];
154 int amd_iommus_present;
156 /* IOMMUs have a non-present cache? */
157 bool amd_iommu_np_cache __read_mostly;
158 bool amd_iommu_iotlb_sup __read_mostly = true;
160 u32 amd_iommu_max_pasid __read_mostly = ~0;
162 bool amd_iommu_v2_present __read_mostly;
163 static bool amd_iommu_pc_present __read_mostly;
165 bool amd_iommu_force_isolation __read_mostly;
168 * List of protection domains - used during resume
170 LIST_HEAD(amd_iommu_pd_list);
171 spinlock_t amd_iommu_pd_lock;
174 * Pointer to the device table which is shared by all AMD IOMMUs
175 * it is indexed by the PCI device id or the HT unit id and contains
176 * information about the domain the device belongs to as well as the
177 * page table root pointer.
179 struct dev_table_entry *amd_iommu_dev_table;
182 * The alias table is a driver specific data structure which contains the
183 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
184 * More than one device can share the same requestor id.
186 u16 *amd_iommu_alias_table;
189 * The rlookup table is used to find the IOMMU which is responsible
190 * for a specific device. It is also indexed by the PCI device id.
192 struct amd_iommu **amd_iommu_rlookup_table;
195 * This table is used to find the irq remapping table for a given device id
198 struct irq_remap_table **irq_lookup_table;
201 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
202 * to know which ones are already in use.
204 unsigned long *amd_iommu_pd_alloc_bitmap;
206 static u32 dev_table_size; /* size of the device table */
207 static u32 alias_table_size; /* size of the alias table */
208 static u32 rlookup_table_size; /* size if the rlookup table */
210 enum iommu_init_state {
223 /* Early ioapic and hpet maps from kernel command line */
224 #define EARLY_MAP_SIZE 4
225 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
226 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
227 static int __initdata early_ioapic_map_size;
228 static int __initdata early_hpet_map_size;
229 static bool __initdata cmdline_maps;
231 static enum iommu_init_state init_state = IOMMU_START_STATE;
233 static int amd_iommu_enable_interrupts(void);
234 static int __init iommu_go_to_state(enum iommu_init_state state);
235 static void init_device_table_dma(void);
237 static int iommu_pc_get_set_reg_val(struct amd_iommu *iommu,
238 u8 bank, u8 cntr, u8 fxn,
239 u64 *value, bool is_write);
241 static inline void update_last_devid(u16 devid)
243 if (devid > amd_iommu_last_bdf)
244 amd_iommu_last_bdf = devid;
247 static inline unsigned long tbl_size(int entry_size)
249 unsigned shift = PAGE_SHIFT +
250 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
255 /* Access to l1 and l2 indexed register spaces */
257 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
261 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
262 pci_read_config_dword(iommu->dev, 0xfc, &val);
266 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
268 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
269 pci_write_config_dword(iommu->dev, 0xfc, val);
270 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
273 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
277 pci_write_config_dword(iommu->dev, 0xf0, address);
278 pci_read_config_dword(iommu->dev, 0xf4, &val);
282 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
284 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
285 pci_write_config_dword(iommu->dev, 0xf4, val);
288 /****************************************************************************
290 * AMD IOMMU MMIO register space handling functions
292 * These functions are used to program the IOMMU device registers in
293 * MMIO space required for that driver.
295 ****************************************************************************/
298 * This function set the exclusion range in the IOMMU. DMA accesses to the
299 * exclusion range are passed through untranslated
301 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
303 u64 start = iommu->exclusion_start & PAGE_MASK;
304 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
307 if (!iommu->exclusion_start)
310 entry = start | MMIO_EXCL_ENABLE_MASK;
311 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
312 &entry, sizeof(entry));
315 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
316 &entry, sizeof(entry));
319 /* Programs the physical address of the device table into the IOMMU hardware */
320 static void iommu_set_device_table(struct amd_iommu *iommu)
324 BUG_ON(iommu->mmio_base == NULL);
326 entry = virt_to_phys(amd_iommu_dev_table);
327 entry |= (dev_table_size >> 12) - 1;
328 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
329 &entry, sizeof(entry));
332 /* Generic functions to enable/disable certain features of the IOMMU. */
333 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
337 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
339 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
342 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
346 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
348 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
351 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
355 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
356 ctrl &= ~CTRL_INV_TO_MASK;
357 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
358 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
361 /* Function to enable the hardware */
362 static void iommu_enable(struct amd_iommu *iommu)
364 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
367 static void iommu_disable(struct amd_iommu *iommu)
369 /* Disable command buffer */
370 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
372 /* Disable event logging and event interrupts */
373 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
374 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
376 /* Disable IOMMU hardware itself */
377 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
381 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
382 * the system has one.
384 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
386 if (!request_mem_region(address, end, "amd_iommu")) {
387 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
389 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
393 return (u8 __iomem *)ioremap_nocache(address, end);
396 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
398 if (iommu->mmio_base)
399 iounmap(iommu->mmio_base);
400 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
403 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
419 /****************************************************************************
421 * The functions below belong to the first pass of AMD IOMMU ACPI table
422 * parsing. In this pass we try to find out the highest device id this
423 * code has to handle. Upon this information the size of the shared data
424 * structures is determined later.
426 ****************************************************************************/
429 * This function calculates the length of a given IVHD entry
431 static inline int ivhd_entry_length(u8 *ivhd)
433 u32 type = ((struct ivhd_entry *)ivhd)->type;
436 return 0x04 << (*ivhd >> 6);
437 } else if (type == IVHD_DEV_ACPI_HID) {
438 /* For ACPI_HID, offset 21 is uid len */
439 return *((u8 *)ivhd + 21) + 22;
445 * After reading the highest device id from the IOMMU PCI capability header
446 * this function looks if there is a higher device id defined in the ACPI table
448 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
450 u8 *p = (void *)h, *end = (void *)h;
451 struct ivhd_entry *dev;
453 u32 ivhd_size = get_ivhd_header_size(h);
456 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
464 dev = (struct ivhd_entry *)p;
467 /* Use maximum BDF value for DEV_ALL */
468 update_last_devid(0xffff);
470 case IVHD_DEV_SELECT:
471 case IVHD_DEV_RANGE_END:
473 case IVHD_DEV_EXT_SELECT:
474 /* all the above subfield types refer to device ids */
475 update_last_devid(dev->devid);
480 p += ivhd_entry_length(p);
488 static int __init check_ivrs_checksum(struct acpi_table_header *table)
491 u8 checksum = 0, *p = (u8 *)table;
493 for (i = 0; i < table->length; ++i)
496 /* ACPI table corrupt */
497 pr_err(FW_BUG "AMD-Vi: IVRS invalid checksum\n");
505 * Iterate over all IVHD entries in the ACPI table and find the highest device
506 * id which we need to handle. This is the first of three functions which parse
507 * the ACPI table. So we check the checksum here.
509 static int __init find_last_devid_acpi(struct acpi_table_header *table)
511 u8 *p = (u8 *)table, *end = (u8 *)table;
512 struct ivhd_header *h;
514 p += IVRS_HEADER_LENGTH;
516 end += table->length;
518 h = (struct ivhd_header *)p;
519 if (h->type == amd_iommu_target_ivhd_type) {
520 int ret = find_last_devid_from_ivhd(h);
532 /****************************************************************************
534 * The following functions belong to the code path which parses the ACPI table
535 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
536 * data structures, initialize the device/alias/rlookup table and also
537 * basically initialize the hardware.
539 ****************************************************************************/
542 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
543 * write commands to that buffer later and the IOMMU will execute them
546 static int __init alloc_command_buffer(struct amd_iommu *iommu)
548 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
549 get_order(CMD_BUFFER_SIZE));
551 return iommu->cmd_buf ? 0 : -ENOMEM;
555 * This function resets the command buffer if the IOMMU stopped fetching
558 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
560 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
562 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
563 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
565 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
569 * This function writes the command buffer address to the hardware and
572 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
576 BUG_ON(iommu->cmd_buf == NULL);
578 entry = (u64)virt_to_phys(iommu->cmd_buf);
579 entry |= MMIO_CMD_SIZE_512;
581 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
582 &entry, sizeof(entry));
584 amd_iommu_reset_cmd_buffer(iommu);
587 static void __init free_command_buffer(struct amd_iommu *iommu)
589 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
592 /* allocates the memory where the IOMMU will log its events to */
593 static int __init alloc_event_buffer(struct amd_iommu *iommu)
595 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
596 get_order(EVT_BUFFER_SIZE));
598 return iommu->evt_buf ? 0 : -ENOMEM;
601 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
605 BUG_ON(iommu->evt_buf == NULL);
607 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
609 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
610 &entry, sizeof(entry));
612 /* set head and tail to zero manually */
613 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
614 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
616 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
619 static void __init free_event_buffer(struct amd_iommu *iommu)
621 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
624 /* allocates the memory where the IOMMU will log its events to */
625 static int __init alloc_ppr_log(struct amd_iommu *iommu)
627 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
628 get_order(PPR_LOG_SIZE));
630 return iommu->ppr_log ? 0 : -ENOMEM;
633 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
637 if (iommu->ppr_log == NULL)
640 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
642 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
643 &entry, sizeof(entry));
645 /* set head and tail to zero manually */
646 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
647 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
649 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
650 iommu_feature_enable(iommu, CONTROL_PPR_EN);
653 static void __init free_ppr_log(struct amd_iommu *iommu)
655 if (iommu->ppr_log == NULL)
658 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
661 static void iommu_enable_gt(struct amd_iommu *iommu)
663 if (!iommu_feature(iommu, FEATURE_GT))
666 iommu_feature_enable(iommu, CONTROL_GT_EN);
669 /* sets a specific bit in the device table entry. */
670 static void set_dev_entry_bit(u16 devid, u8 bit)
672 int i = (bit >> 6) & 0x03;
673 int _bit = bit & 0x3f;
675 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
678 static int get_dev_entry_bit(u16 devid, u8 bit)
680 int i = (bit >> 6) & 0x03;
681 int _bit = bit & 0x3f;
683 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
687 void amd_iommu_apply_erratum_63(u16 devid)
691 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
692 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
695 set_dev_entry_bit(devid, DEV_ENTRY_IW);
698 /* Writes the specific IOMMU for a device into the rlookup table */
699 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
701 amd_iommu_rlookup_table[devid] = iommu;
705 * This function takes the device specific flags read from the ACPI
706 * table and sets up the device table entry with that information
708 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
709 u16 devid, u32 flags, u32 ext_flags)
711 if (flags & ACPI_DEVFLAG_INITPASS)
712 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
713 if (flags & ACPI_DEVFLAG_EXTINT)
714 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
715 if (flags & ACPI_DEVFLAG_NMI)
716 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
717 if (flags & ACPI_DEVFLAG_SYSMGT1)
718 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
719 if (flags & ACPI_DEVFLAG_SYSMGT2)
720 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
721 if (flags & ACPI_DEVFLAG_LINT0)
722 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
723 if (flags & ACPI_DEVFLAG_LINT1)
724 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
726 amd_iommu_apply_erratum_63(devid);
728 set_iommu_for_device(iommu, devid);
731 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
733 struct devid_map *entry;
734 struct list_head *list;
736 if (type == IVHD_SPECIAL_IOAPIC)
738 else if (type == IVHD_SPECIAL_HPET)
743 list_for_each_entry(entry, list, list) {
744 if (!(entry->id == id && entry->cmd_line))
747 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
748 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
750 *devid = entry->devid;
755 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
760 entry->devid = *devid;
761 entry->cmd_line = cmd_line;
763 list_add_tail(&entry->list, list);
768 static int __init add_early_maps(void)
772 for (i = 0; i < early_ioapic_map_size; ++i) {
773 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
774 early_ioapic_map[i].id,
775 &early_ioapic_map[i].devid,
776 early_ioapic_map[i].cmd_line);
781 for (i = 0; i < early_hpet_map_size; ++i) {
782 ret = add_special_device(IVHD_SPECIAL_HPET,
783 early_hpet_map[i].id,
784 &early_hpet_map[i].devid,
785 early_hpet_map[i].cmd_line);
794 * Reads the device exclusion range from ACPI and initializes the IOMMU with
797 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
799 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
801 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
806 * We only can configure exclusion ranges per IOMMU, not
807 * per device. But we can enable the exclusion range per
808 * device. This is done here
810 set_dev_entry_bit(devid, DEV_ENTRY_EX);
811 iommu->exclusion_start = m->range_start;
812 iommu->exclusion_length = m->range_length;
817 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
818 * initializes the hardware and our data structures with it.
820 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
821 struct ivhd_header *h)
824 u8 *end = p, flags = 0;
825 u16 devid = 0, devid_start = 0, devid_to = 0;
826 u32 dev_i, ext_flags = 0;
828 struct ivhd_entry *e;
833 ret = add_early_maps();
838 * First save the recommended feature enable bits from ACPI
840 iommu->acpi_flags = h->flags;
843 * Done. Now parse the device entries
845 ivhd_size = get_ivhd_header_size(h);
847 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
857 e = (struct ivhd_entry *)p;
861 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
863 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
864 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
866 case IVHD_DEV_SELECT:
868 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
870 PCI_BUS_NUM(e->devid),
876 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
878 case IVHD_DEV_SELECT_RANGE_START:
880 DUMP_printk(" DEV_SELECT_RANGE_START\t "
881 "devid: %02x:%02x.%x flags: %02x\n",
882 PCI_BUS_NUM(e->devid),
887 devid_start = e->devid;
894 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
895 "flags: %02x devid_to: %02x:%02x.%x\n",
896 PCI_BUS_NUM(e->devid),
900 PCI_BUS_NUM(e->ext >> 8),
901 PCI_SLOT(e->ext >> 8),
902 PCI_FUNC(e->ext >> 8));
905 devid_to = e->ext >> 8;
906 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
907 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
908 amd_iommu_alias_table[devid] = devid_to;
910 case IVHD_DEV_ALIAS_RANGE:
912 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
913 "devid: %02x:%02x.%x flags: %02x "
914 "devid_to: %02x:%02x.%x\n",
915 PCI_BUS_NUM(e->devid),
919 PCI_BUS_NUM(e->ext >> 8),
920 PCI_SLOT(e->ext >> 8),
921 PCI_FUNC(e->ext >> 8));
923 devid_start = e->devid;
925 devid_to = e->ext >> 8;
929 case IVHD_DEV_EXT_SELECT:
931 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
932 "flags: %02x ext: %08x\n",
933 PCI_BUS_NUM(e->devid),
939 set_dev_entry_from_acpi(iommu, devid, e->flags,
942 case IVHD_DEV_EXT_SELECT_RANGE:
944 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
945 "%02x:%02x.%x flags: %02x ext: %08x\n",
946 PCI_BUS_NUM(e->devid),
951 devid_start = e->devid;
956 case IVHD_DEV_RANGE_END:
958 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
959 PCI_BUS_NUM(e->devid),
964 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
966 amd_iommu_alias_table[dev_i] = devid_to;
967 set_dev_entry_from_acpi(iommu,
968 devid_to, flags, ext_flags);
970 set_dev_entry_from_acpi(iommu, dev_i,
974 case IVHD_DEV_SPECIAL: {
980 handle = e->ext & 0xff;
981 devid = (e->ext >> 8) & 0xffff;
982 type = (e->ext >> 24) & 0xff;
984 if (type == IVHD_SPECIAL_IOAPIC)
986 else if (type == IVHD_SPECIAL_HPET)
991 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
997 ret = add_special_device(type, handle, &devid, false);
1002 * add_special_device might update the devid in case a
1003 * command-line override is present. So call
1004 * set_dev_entry_from_acpi after add_special_device.
1006 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1014 p += ivhd_entry_length(p);
1020 static void __init free_iommu_one(struct amd_iommu *iommu)
1022 free_command_buffer(iommu);
1023 free_event_buffer(iommu);
1024 free_ppr_log(iommu);
1025 iommu_unmap_mmio_space(iommu);
1028 static void __init free_iommu_all(void)
1030 struct amd_iommu *iommu, *next;
1032 for_each_iommu_safe(iommu, next) {
1033 list_del(&iommu->list);
1034 free_iommu_one(iommu);
1040 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1042 * BIOS should disable L2B micellaneous clock gating by setting
1043 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1045 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1049 if ((boot_cpu_data.x86 != 0x15) ||
1050 (boot_cpu_data.x86_model < 0x10) ||
1051 (boot_cpu_data.x86_model > 0x1f))
1054 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1055 pci_read_config_dword(iommu->dev, 0xf4, &value);
1060 /* Select NB indirect register 0x90 and enable writing */
1061 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1063 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1064 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1065 dev_name(&iommu->dev->dev));
1067 /* Clear the enable writing bit */
1068 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1072 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1074 * BIOS should enable ATS write permission check by setting
1075 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1077 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1081 if ((boot_cpu_data.x86 != 0x15) ||
1082 (boot_cpu_data.x86_model < 0x30) ||
1083 (boot_cpu_data.x86_model > 0x3f))
1086 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1087 value = iommu_read_l2(iommu, 0x47);
1092 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1093 iommu_write_l2(iommu, 0x47, value | BIT(0));
1095 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1096 dev_name(&iommu->dev->dev));
1100 * This function clues the initialization function for one IOMMU
1101 * together and also allocates the command buffer and programs the
1102 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1104 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1108 spin_lock_init(&iommu->lock);
1110 /* Add IOMMU to internal data structures */
1111 list_add_tail(&iommu->list, &amd_iommu_list);
1112 iommu->index = amd_iommus_present++;
1114 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1115 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1119 /* Index is fine - add IOMMU to the array */
1120 amd_iommus[iommu->index] = iommu;
1123 * Copy data from ACPI table entry to the iommu struct
1125 iommu->devid = h->devid;
1126 iommu->cap_ptr = h->cap_ptr;
1127 iommu->pci_seg = h->pci_seg;
1128 iommu->mmio_phys = h->mmio_phys;
1132 /* Check if IVHD EFR contains proper max banks/counters */
1133 if ((h->efr_attr != 0) &&
1134 ((h->efr_attr & (0xF << 13)) != 0) &&
1135 ((h->efr_attr & (0x3F << 17)) != 0))
1136 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1138 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1142 if (h->efr_reg & (1 << 9))
1143 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1145 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1151 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1152 iommu->mmio_phys_end);
1153 if (!iommu->mmio_base)
1156 if (alloc_command_buffer(iommu))
1159 if (alloc_event_buffer(iommu))
1162 iommu->int_enabled = false;
1164 ret = init_iommu_from_acpi(iommu, h);
1168 ret = amd_iommu_create_irq_domain(iommu);
1173 * Make sure IOMMU is not considered to translate itself. The IVRS
1174 * table tells us so, but this is a lie!
1176 amd_iommu_rlookup_table[iommu->devid] = NULL;
1182 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1183 * @ivrs Pointer to the IVRS header
1185 * This function search through all IVDB of the maximum supported IVHD
1187 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1189 u8 *base = (u8 *)ivrs;
1190 struct ivhd_header *ivhd = (struct ivhd_header *)
1191 (base + IVRS_HEADER_LENGTH);
1192 u8 last_type = ivhd->type;
1193 u16 devid = ivhd->devid;
1195 while (((u8 *)ivhd - base < ivrs->length) &&
1196 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1197 u8 *p = (u8 *) ivhd;
1199 if (ivhd->devid == devid)
1200 last_type = ivhd->type;
1201 ivhd = (struct ivhd_header *)(p + ivhd->length);
1208 * Iterates over all IOMMU entries in the ACPI table, allocates the
1209 * IOMMU structure and initializes it with init_iommu_one()
1211 static int __init init_iommu_all(struct acpi_table_header *table)
1213 u8 *p = (u8 *)table, *end = (u8 *)table;
1214 struct ivhd_header *h;
1215 struct amd_iommu *iommu;
1218 end += table->length;
1219 p += IVRS_HEADER_LENGTH;
1222 h = (struct ivhd_header *)p;
1223 if (*p == amd_iommu_target_ivhd_type) {
1225 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1226 "seg: %d flags: %01x info %04x\n",
1227 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1228 PCI_FUNC(h->devid), h->cap_ptr,
1229 h->pci_seg, h->flags, h->info);
1230 DUMP_printk(" mmio-addr: %016llx\n",
1233 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1237 ret = init_iommu_one(iommu, h);
1250 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1252 u64 val = 0xabcd, val2 = 0;
1254 if (!iommu_feature(iommu, FEATURE_PC))
1257 amd_iommu_pc_present = true;
1259 /* Check if the performance counters can be written to */
1260 if ((0 != iommu_pc_get_set_reg_val(iommu, 0, 0, 0, &val, true)) ||
1261 (0 != iommu_pc_get_set_reg_val(iommu, 0, 0, 0, &val2, false)) ||
1263 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1264 amd_iommu_pc_present = false;
1268 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1270 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1271 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1272 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1275 static ssize_t amd_iommu_show_cap(struct device *dev,
1276 struct device_attribute *attr,
1279 struct amd_iommu *iommu = dev_get_drvdata(dev);
1280 return sprintf(buf, "%x\n", iommu->cap);
1282 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1284 static ssize_t amd_iommu_show_features(struct device *dev,
1285 struct device_attribute *attr,
1288 struct amd_iommu *iommu = dev_get_drvdata(dev);
1289 return sprintf(buf, "%llx\n", iommu->features);
1291 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1293 static struct attribute *amd_iommu_attrs[] = {
1295 &dev_attr_features.attr,
1299 static struct attribute_group amd_iommu_group = {
1300 .name = "amd-iommu",
1301 .attrs = amd_iommu_attrs,
1304 static const struct attribute_group *amd_iommu_groups[] = {
1309 static int iommu_init_pci(struct amd_iommu *iommu)
1311 int cap_ptr = iommu->cap_ptr;
1312 u32 range, misc, low, high;
1314 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1315 iommu->devid & 0xff);
1319 /* Prevent binding other PCI device drivers to IOMMU devices */
1320 iommu->dev->match_driver = false;
1322 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1324 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1326 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1329 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1330 amd_iommu_iotlb_sup = false;
1332 /* read extended feature bits */
1333 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1334 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1336 iommu->features = ((u64)high << 32) | low;
1338 if (iommu_feature(iommu, FEATURE_GT)) {
1343 pasmax = iommu->features & FEATURE_PASID_MASK;
1344 pasmax >>= FEATURE_PASID_SHIFT;
1345 max_pasid = (1 << (pasmax + 1)) - 1;
1347 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1349 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1351 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1352 glxval >>= FEATURE_GLXVAL_SHIFT;
1354 if (amd_iommu_max_glx_val == -1)
1355 amd_iommu_max_glx_val = glxval;
1357 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1360 if (iommu_feature(iommu, FEATURE_GT) &&
1361 iommu_feature(iommu, FEATURE_PPR)) {
1362 iommu->is_iommu_v2 = true;
1363 amd_iommu_v2_present = true;
1366 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1369 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1370 amd_iommu_np_cache = true;
1372 init_iommu_perf_ctr(iommu);
1374 if (is_rd890_iommu(iommu->dev)) {
1377 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1381 * Some rd890 systems may not be fully reconfigured by the
1382 * BIOS, so it's necessary for us to store this information so
1383 * it can be reprogrammed on resume
1385 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1386 &iommu->stored_addr_lo);
1387 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1388 &iommu->stored_addr_hi);
1390 /* Low bit locks writes to configuration space */
1391 iommu->stored_addr_lo &= ~1;
1393 for (i = 0; i < 6; i++)
1394 for (j = 0; j < 0x12; j++)
1395 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1397 for (i = 0; i < 0x83; i++)
1398 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1401 amd_iommu_erratum_746_workaround(iommu);
1402 amd_iommu_ats_write_check_workaround(iommu);
1404 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1405 amd_iommu_groups, "ivhd%d",
1408 return pci_enable_device(iommu->dev);
1411 static void print_iommu_info(void)
1413 static const char * const feat_str[] = {
1414 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1415 "IA", "GA", "HE", "PC"
1417 struct amd_iommu *iommu;
1419 for_each_iommu(iommu) {
1422 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1423 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1425 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1426 pr_info("AMD-Vi: Extended features: ");
1427 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1428 if (iommu_feature(iommu, (1ULL << i)))
1429 pr_cont(" %s", feat_str[i]);
1434 if (irq_remapping_enabled)
1435 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1438 static int __init amd_iommu_init_pci(void)
1440 struct amd_iommu *iommu;
1443 for_each_iommu(iommu) {
1444 ret = iommu_init_pci(iommu);
1449 init_device_table_dma();
1451 for_each_iommu(iommu)
1452 iommu_flush_all_caches(iommu);
1454 ret = amd_iommu_init_api();
1462 /****************************************************************************
1464 * The following functions initialize the MSI interrupts for all IOMMUs
1465 * in the system. It's a bit challenging because there could be multiple
1466 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1469 ****************************************************************************/
1471 static int iommu_setup_msi(struct amd_iommu *iommu)
1475 r = pci_enable_msi(iommu->dev);
1479 r = request_threaded_irq(iommu->dev->irq,
1480 amd_iommu_int_handler,
1481 amd_iommu_int_thread,
1486 pci_disable_msi(iommu->dev);
1490 iommu->int_enabled = true;
1495 static int iommu_init_msi(struct amd_iommu *iommu)
1499 if (iommu->int_enabled)
1502 if (iommu->dev->msi_cap)
1503 ret = iommu_setup_msi(iommu);
1511 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1513 if (iommu->ppr_log != NULL)
1514 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1519 /****************************************************************************
1521 * The next functions belong to the third pass of parsing the ACPI
1522 * table. In this last pass the memory mapping requirements are
1523 * gathered (like exclusion and unity mapping ranges).
1525 ****************************************************************************/
1527 static void __init free_unity_maps(void)
1529 struct unity_map_entry *entry, *next;
1531 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1532 list_del(&entry->list);
1537 /* called when we find an exclusion range definition in ACPI */
1538 static int __init init_exclusion_range(struct ivmd_header *m)
1543 case ACPI_IVMD_TYPE:
1544 set_device_exclusion_range(m->devid, m);
1546 case ACPI_IVMD_TYPE_ALL:
1547 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1548 set_device_exclusion_range(i, m);
1550 case ACPI_IVMD_TYPE_RANGE:
1551 for (i = m->devid; i <= m->aux; ++i)
1552 set_device_exclusion_range(i, m);
1561 /* called for unity map ACPI definition */
1562 static int __init init_unity_map_range(struct ivmd_header *m)
1564 struct unity_map_entry *e = NULL;
1567 e = kzalloc(sizeof(*e), GFP_KERNEL);
1575 case ACPI_IVMD_TYPE:
1576 s = "IVMD_TYPEi\t\t\t";
1577 e->devid_start = e->devid_end = m->devid;
1579 case ACPI_IVMD_TYPE_ALL:
1580 s = "IVMD_TYPE_ALL\t\t";
1582 e->devid_end = amd_iommu_last_bdf;
1584 case ACPI_IVMD_TYPE_RANGE:
1585 s = "IVMD_TYPE_RANGE\t\t";
1586 e->devid_start = m->devid;
1587 e->devid_end = m->aux;
1590 e->address_start = PAGE_ALIGN(m->range_start);
1591 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1592 e->prot = m->flags >> 1;
1594 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1595 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1596 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1597 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1598 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1599 e->address_start, e->address_end, m->flags);
1601 list_add_tail(&e->list, &amd_iommu_unity_map);
1606 /* iterates over all memory definitions we find in the ACPI table */
1607 static int __init init_memory_definitions(struct acpi_table_header *table)
1609 u8 *p = (u8 *)table, *end = (u8 *)table;
1610 struct ivmd_header *m;
1612 end += table->length;
1613 p += IVRS_HEADER_LENGTH;
1616 m = (struct ivmd_header *)p;
1617 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1618 init_exclusion_range(m);
1619 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1620 init_unity_map_range(m);
1629 * Init the device table to not allow DMA access for devices and
1630 * suppress all page faults
1632 static void init_device_table_dma(void)
1636 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1637 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1638 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1642 static void __init uninit_device_table_dma(void)
1646 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1647 amd_iommu_dev_table[devid].data[0] = 0ULL;
1648 amd_iommu_dev_table[devid].data[1] = 0ULL;
1652 static void init_device_table(void)
1656 if (!amd_iommu_irq_remap)
1659 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1660 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1663 static void iommu_init_flags(struct amd_iommu *iommu)
1665 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1666 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1667 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1669 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1670 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1671 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1673 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1674 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1675 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1677 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1678 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1679 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1682 * make IOMMU memory accesses cache coherent
1684 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1686 /* Set IOTLB invalidation timeout to 1s */
1687 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1690 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1693 u32 ioc_feature_control;
1694 struct pci_dev *pdev = iommu->root_pdev;
1696 /* RD890 BIOSes may not have completely reconfigured the iommu */
1697 if (!is_rd890_iommu(iommu->dev) || !pdev)
1701 * First, we need to ensure that the iommu is enabled. This is
1702 * controlled by a register in the northbridge
1705 /* Select Northbridge indirect register 0x75 and enable writing */
1706 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1707 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1709 /* Enable the iommu */
1710 if (!(ioc_feature_control & 0x1))
1711 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1713 /* Restore the iommu BAR */
1714 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1715 iommu->stored_addr_lo);
1716 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1717 iommu->stored_addr_hi);
1719 /* Restore the l1 indirect regs for each of the 6 l1s */
1720 for (i = 0; i < 6; i++)
1721 for (j = 0; j < 0x12; j++)
1722 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1724 /* Restore the l2 indirect regs */
1725 for (i = 0; i < 0x83; i++)
1726 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1728 /* Lock PCI setup registers */
1729 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1730 iommu->stored_addr_lo | 1);
1734 * This function finally enables all IOMMUs found in the system after
1735 * they have been initialized
1737 static void early_enable_iommus(void)
1739 struct amd_iommu *iommu;
1741 for_each_iommu(iommu) {
1742 iommu_disable(iommu);
1743 iommu_init_flags(iommu);
1744 iommu_set_device_table(iommu);
1745 iommu_enable_command_buffer(iommu);
1746 iommu_enable_event_buffer(iommu);
1747 iommu_set_exclusion_range(iommu);
1748 iommu_enable(iommu);
1749 iommu_flush_all_caches(iommu);
1753 static void enable_iommus_v2(void)
1755 struct amd_iommu *iommu;
1757 for_each_iommu(iommu) {
1758 iommu_enable_ppr_log(iommu);
1759 iommu_enable_gt(iommu);
1763 static void enable_iommus(void)
1765 early_enable_iommus();
1770 static void disable_iommus(void)
1772 struct amd_iommu *iommu;
1774 for_each_iommu(iommu)
1775 iommu_disable(iommu);
1779 * Suspend/Resume support
1780 * disable suspend until real resume implemented
1783 static void amd_iommu_resume(void)
1785 struct amd_iommu *iommu;
1787 for_each_iommu(iommu)
1788 iommu_apply_resume_quirks(iommu);
1790 /* re-load the hardware */
1793 amd_iommu_enable_interrupts();
1796 static int amd_iommu_suspend(void)
1798 /* disable IOMMUs to go out of the way for BIOS */
1804 static struct syscore_ops amd_iommu_syscore_ops = {
1805 .suspend = amd_iommu_suspend,
1806 .resume = amd_iommu_resume,
1809 static void __init free_on_init_error(void)
1811 free_pages((unsigned long)irq_lookup_table,
1812 get_order(rlookup_table_size));
1814 kmem_cache_destroy(amd_iommu_irq_cache);
1815 amd_iommu_irq_cache = NULL;
1817 free_pages((unsigned long)amd_iommu_rlookup_table,
1818 get_order(rlookup_table_size));
1820 free_pages((unsigned long)amd_iommu_alias_table,
1821 get_order(alias_table_size));
1823 free_pages((unsigned long)amd_iommu_dev_table,
1824 get_order(dev_table_size));
1828 #ifdef CONFIG_GART_IOMMU
1830 * We failed to initialize the AMD IOMMU - try fallback to GART
1838 /* SB IOAPIC is always on this device in AMD systems */
1839 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1841 static bool __init check_ioapic_information(void)
1843 const char *fw_bug = FW_BUG;
1844 bool ret, has_sb_ioapic;
1847 has_sb_ioapic = false;
1851 * If we have map overrides on the kernel command line the
1852 * messages in this function might not describe firmware bugs
1853 * anymore - so be careful
1858 for (idx = 0; idx < nr_ioapics; idx++) {
1859 int devid, id = mpc_ioapic_id(idx);
1861 devid = get_ioapic_devid(id);
1863 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1866 } else if (devid == IOAPIC_SB_DEVID) {
1867 has_sb_ioapic = true;
1872 if (!has_sb_ioapic) {
1874 * We expect the SB IOAPIC to be listed in the IVRS
1875 * table. The system timer is connected to the SB IOAPIC
1876 * and if we don't have it in the list the system will
1877 * panic at boot time. This situation usually happens
1878 * when the BIOS is buggy and provides us the wrong
1879 * device id for the IOAPIC in the system.
1881 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
1885 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1890 static void __init free_dma_resources(void)
1892 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1893 get_order(MAX_DOMAIN_ID/8));
1899 * This is the hardware init function for AMD IOMMU in the system.
1900 * This function is called either from amd_iommu_init or from the interrupt
1901 * remapping setup code.
1903 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1906 * 1 pass) Discover the most comprehensive IVHD type to use.
1908 * 2 pass) Find the highest PCI device id the driver has to handle.
1909 * Upon this information the size of the data structures is
1910 * determined that needs to be allocated.
1912 * 3 pass) Initialize the data structures just allocated with the
1913 * information in the ACPI table about available AMD IOMMUs
1914 * in the system. It also maps the PCI devices in the
1915 * system to specific IOMMUs
1917 * 4 pass) After the basic data structures are allocated and
1918 * initialized we update them with information about memory
1919 * remapping requirements parsed out of the ACPI table in
1922 * After everything is set up the IOMMUs are enabled and the necessary
1923 * hotplug and suspend notifiers are registered.
1925 static int __init early_amd_iommu_init(void)
1927 struct acpi_table_header *ivrs_base;
1928 acpi_size ivrs_size;
1932 if (!amd_iommu_detected)
1935 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1936 if (status == AE_NOT_FOUND)
1938 else if (ACPI_FAILURE(status)) {
1939 const char *err = acpi_format_exception(status);
1940 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1945 * Validate checksum here so we don't need to do it when
1946 * we actually parse the table
1948 ret = check_ivrs_checksum(ivrs_base);
1952 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
1953 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
1956 * First parse ACPI tables to find the largest Bus/Dev/Func
1957 * we need to handle. Upon this information the shared data
1958 * structures for the IOMMUs in the system will be allocated
1960 ret = find_last_devid_acpi(ivrs_base);
1964 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1965 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1966 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1968 /* Device table - directly used by all IOMMUs */
1970 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1971 get_order(dev_table_size));
1972 if (amd_iommu_dev_table == NULL)
1976 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1977 * IOMMU see for that device
1979 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1980 get_order(alias_table_size));
1981 if (amd_iommu_alias_table == NULL)
1984 /* IOMMU rlookup table - find the IOMMU for a specific device */
1985 amd_iommu_rlookup_table = (void *)__get_free_pages(
1986 GFP_KERNEL | __GFP_ZERO,
1987 get_order(rlookup_table_size));
1988 if (amd_iommu_rlookup_table == NULL)
1991 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1992 GFP_KERNEL | __GFP_ZERO,
1993 get_order(MAX_DOMAIN_ID/8));
1994 if (amd_iommu_pd_alloc_bitmap == NULL)
1998 * let all alias entries point to itself
2000 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2001 amd_iommu_alias_table[i] = i;
2004 * never allocate domain 0 because its used as the non-allocated and
2005 * error value placeholder
2007 amd_iommu_pd_alloc_bitmap[0] = 1;
2009 spin_lock_init(&amd_iommu_pd_lock);
2012 * now the data structures are allocated and basically initialized
2013 * start the real acpi table scan
2015 ret = init_iommu_all(ivrs_base);
2019 if (amd_iommu_irq_remap)
2020 amd_iommu_irq_remap = check_ioapic_information();
2022 if (amd_iommu_irq_remap) {
2024 * Interrupt remapping enabled, create kmem_cache for the
2028 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2029 MAX_IRQS_PER_TABLE * sizeof(u32),
2030 IRQ_TABLE_ALIGNMENT,
2032 if (!amd_iommu_irq_cache)
2035 irq_lookup_table = (void *)__get_free_pages(
2036 GFP_KERNEL | __GFP_ZERO,
2037 get_order(rlookup_table_size));
2038 if (!irq_lookup_table)
2042 ret = init_memory_definitions(ivrs_base);
2046 /* init the device table */
2047 init_device_table();
2050 /* Don't leak any ACPI memory */
2051 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2057 static int amd_iommu_enable_interrupts(void)
2059 struct amd_iommu *iommu;
2062 for_each_iommu(iommu) {
2063 ret = iommu_init_msi(iommu);
2072 static bool detect_ivrs(void)
2074 struct acpi_table_header *ivrs_base;
2075 acpi_size ivrs_size;
2078 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
2079 if (status == AE_NOT_FOUND)
2081 else if (ACPI_FAILURE(status)) {
2082 const char *err = acpi_format_exception(status);
2083 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2087 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2089 /* Make sure ACS will be enabled during PCI probe */
2095 /****************************************************************************
2097 * AMD IOMMU Initialization State Machine
2099 ****************************************************************************/
2101 static int __init state_next(void)
2105 switch (init_state) {
2106 case IOMMU_START_STATE:
2107 if (!detect_ivrs()) {
2108 init_state = IOMMU_NOT_FOUND;
2111 init_state = IOMMU_IVRS_DETECTED;
2114 case IOMMU_IVRS_DETECTED:
2115 ret = early_amd_iommu_init();
2116 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2118 case IOMMU_ACPI_FINISHED:
2119 early_enable_iommus();
2120 register_syscore_ops(&amd_iommu_syscore_ops);
2121 x86_platform.iommu_shutdown = disable_iommus;
2122 init_state = IOMMU_ENABLED;
2125 ret = amd_iommu_init_pci();
2126 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2129 case IOMMU_PCI_INIT:
2130 ret = amd_iommu_enable_interrupts();
2131 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2133 case IOMMU_INTERRUPTS_EN:
2134 ret = amd_iommu_init_dma_ops();
2135 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2138 init_state = IOMMU_INITIALIZED;
2140 case IOMMU_INITIALIZED:
2143 case IOMMU_NOT_FOUND:
2144 case IOMMU_INIT_ERROR:
2145 /* Error states => do nothing */
2156 static int __init iommu_go_to_state(enum iommu_init_state state)
2160 while (init_state != state) {
2162 if (init_state == IOMMU_NOT_FOUND ||
2163 init_state == IOMMU_INIT_ERROR)
2170 #ifdef CONFIG_IRQ_REMAP
2171 int __init amd_iommu_prepare(void)
2175 amd_iommu_irq_remap = true;
2177 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2180 return amd_iommu_irq_remap ? 0 : -ENODEV;
2183 int __init amd_iommu_enable(void)
2187 ret = iommu_go_to_state(IOMMU_ENABLED);
2191 irq_remapping_enabled = 1;
2196 void amd_iommu_disable(void)
2198 amd_iommu_suspend();
2201 int amd_iommu_reenable(int mode)
2208 int __init amd_iommu_enable_faulting(void)
2210 /* We enable MSI later when PCI is initialized */
2216 * This is the core init function for AMD IOMMU hardware in the system.
2217 * This function is called from the generic x86 DMA layer initialization
2220 static int __init amd_iommu_init(void)
2224 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2226 free_dma_resources();
2227 if (!irq_remapping_enabled) {
2229 free_on_init_error();
2231 struct amd_iommu *iommu;
2233 uninit_device_table_dma();
2234 for_each_iommu(iommu)
2235 iommu_flush_all_caches(iommu);
2242 /****************************************************************************
2244 * Early detect code. This code runs at IOMMU detection time in the DMA
2245 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2248 ****************************************************************************/
2249 int __init amd_iommu_detect(void)
2253 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2256 if (amd_iommu_disabled)
2259 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2263 amd_iommu_detected = true;
2265 x86_init.iommu.iommu_init = amd_iommu_init;
2270 /****************************************************************************
2272 * Parsing functions for the AMD IOMMU specific kernel command line
2275 ****************************************************************************/
2277 static int __init parse_amd_iommu_dump(char *str)
2279 amd_iommu_dump = true;
2284 static int __init parse_amd_iommu_options(char *str)
2286 for (; *str; ++str) {
2287 if (strncmp(str, "fullflush", 9) == 0)
2288 amd_iommu_unmap_flush = true;
2289 if (strncmp(str, "off", 3) == 0)
2290 amd_iommu_disabled = true;
2291 if (strncmp(str, "force_isolation", 15) == 0)
2292 amd_iommu_force_isolation = true;
2298 static int __init parse_ivrs_ioapic(char *str)
2300 unsigned int bus, dev, fn;
2304 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2307 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2311 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2312 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2317 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2319 cmdline_maps = true;
2320 i = early_ioapic_map_size++;
2321 early_ioapic_map[i].id = id;
2322 early_ioapic_map[i].devid = devid;
2323 early_ioapic_map[i].cmd_line = true;
2328 static int __init parse_ivrs_hpet(char *str)
2330 unsigned int bus, dev, fn;
2334 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2337 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2341 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2342 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2347 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2349 cmdline_maps = true;
2350 i = early_hpet_map_size++;
2351 early_hpet_map[i].id = id;
2352 early_hpet_map[i].devid = devid;
2353 early_hpet_map[i].cmd_line = true;
2358 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2359 __setup("amd_iommu=", parse_amd_iommu_options);
2360 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2361 __setup("ivrs_hpet", parse_ivrs_hpet);
2363 IOMMU_INIT_FINISH(amd_iommu_detect,
2364 gart_iommu_hole_init,
2368 bool amd_iommu_v2_supported(void)
2370 return amd_iommu_v2_present;
2372 EXPORT_SYMBOL(amd_iommu_v2_supported);
2374 /****************************************************************************
2376 * IOMMU EFR Performance Counter support functionality. This code allows
2377 * access to the IOMMU PC functionality.
2379 ****************************************************************************/
2381 u8 amd_iommu_pc_get_max_banks(u16 devid)
2383 struct amd_iommu *iommu;
2386 /* locate the iommu governing the devid */
2387 iommu = amd_iommu_rlookup_table[devid];
2389 ret = iommu->max_banks;
2393 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2395 bool amd_iommu_pc_supported(void)
2397 return amd_iommu_pc_present;
2399 EXPORT_SYMBOL(amd_iommu_pc_supported);
2401 u8 amd_iommu_pc_get_max_counters(u16 devid)
2403 struct amd_iommu *iommu;
2406 /* locate the iommu governing the devid */
2407 iommu = amd_iommu_rlookup_table[devid];
2409 ret = iommu->max_counters;
2413 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2415 static int iommu_pc_get_set_reg_val(struct amd_iommu *iommu,
2416 u8 bank, u8 cntr, u8 fxn,
2417 u64 *value, bool is_write)
2422 /* Check for valid iommu and pc register indexing */
2423 if (WARN_ON((fxn > 0x28) || (fxn & 7)))
2426 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2428 /* Limit the offset to the hw defined mmio region aperture */
2429 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2430 (iommu->max_counters << 8) | 0x28);
2431 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2432 (offset > max_offset_lim))
2436 writel((u32)*value, iommu->mmio_base + offset);
2437 writel((*value >> 32), iommu->mmio_base + offset + 4);
2439 *value = readl(iommu->mmio_base + offset + 4);
2441 *value = readl(iommu->mmio_base + offset);
2446 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);
2448 int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2449 u64 *value, bool is_write)
2451 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
2453 /* Make sure the IOMMU PC resource is available */
2454 if (!amd_iommu_pc_present || iommu == NULL)
2457 return iommu_pc_get_set_reg_val(iommu, bank, cntr, fxn,