2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
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/bitmap.h>
24 #include <linux/slab.h>
25 #include <linux/syscore_ops.h>
26 #include <linux/interrupt.h>
27 #include <linux/msi.h>
28 #include <linux/amd-iommu.h>
29 #include <linux/export.h>
30 #include <linux/iommu.h>
31 #include <linux/kmemleak.h>
32 #include <linux/mem_encrypt.h>
33 #include <asm/pci-direct.h>
34 #include <asm/iommu.h>
36 #include <asm/x86_init.h>
37 #include <asm/iommu_table.h>
38 #include <asm/io_apic.h>
39 #include <asm/irq_remapping.h>
41 #include <linux/crash_dump.h>
42 #include "amd_iommu_proto.h"
43 #include "amd_iommu_types.h"
44 #include "irq_remapping.h"
47 * definitions for the ACPI scanning code
49 #define IVRS_HEADER_LENGTH 48
51 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
52 #define ACPI_IVMD_TYPE_ALL 0x20
53 #define ACPI_IVMD_TYPE 0x21
54 #define ACPI_IVMD_TYPE_RANGE 0x22
56 #define IVHD_DEV_ALL 0x01
57 #define IVHD_DEV_SELECT 0x02
58 #define IVHD_DEV_SELECT_RANGE_START 0x03
59 #define IVHD_DEV_RANGE_END 0x04
60 #define IVHD_DEV_ALIAS 0x42
61 #define IVHD_DEV_ALIAS_RANGE 0x43
62 #define IVHD_DEV_EXT_SELECT 0x46
63 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
64 #define IVHD_DEV_SPECIAL 0x48
65 #define IVHD_DEV_ACPI_HID 0xf0
67 #define UID_NOT_PRESENT 0
68 #define UID_IS_INTEGER 1
69 #define UID_IS_CHARACTER 2
71 #define IVHD_SPECIAL_IOAPIC 1
72 #define IVHD_SPECIAL_HPET 2
74 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
75 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
76 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
77 #define IVHD_FLAG_ISOC_EN_MASK 0x08
79 #define IVMD_FLAG_EXCL_RANGE 0x08
80 #define IVMD_FLAG_UNITY_MAP 0x01
82 #define ACPI_DEVFLAG_INITPASS 0x01
83 #define ACPI_DEVFLAG_EXTINT 0x02
84 #define ACPI_DEVFLAG_NMI 0x04
85 #define ACPI_DEVFLAG_SYSMGT1 0x10
86 #define ACPI_DEVFLAG_SYSMGT2 0x20
87 #define ACPI_DEVFLAG_LINT0 0x40
88 #define ACPI_DEVFLAG_LINT1 0x80
89 #define ACPI_DEVFLAG_ATSDIS 0x10000000
91 #define LOOP_TIMEOUT 100000
93 * ACPI table definitions
95 * These data structures are laid over the table to parse the important values
99 extern const struct iommu_ops amd_iommu_ops;
102 * structure describing one IOMMU in the ACPI table. Typically followed by one
103 * or more ivhd_entrys.
116 /* Following only valid on IVHD type 11h and 40h */
117 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
119 } __attribute__((packed));
122 * A device entry describing which devices a specific IOMMU translates and
123 * which requestor ids they use.
135 } __attribute__((packed));
138 * An AMD IOMMU memory definition structure. It defines things like exclusion
139 * ranges for devices and regions that should be unity mapped.
150 } __attribute__((packed));
153 bool amd_iommu_irq_remap __read_mostly;
155 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
156 static int amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
158 static bool amd_iommu_detected;
159 static bool __initdata amd_iommu_disabled;
160 static int amd_iommu_target_ivhd_type;
162 u16 amd_iommu_last_bdf; /* largest PCI device id we have
164 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
166 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
168 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
171 /* Array to assign indices to IOMMUs*/
172 struct amd_iommu *amd_iommus[MAX_IOMMUS];
174 /* Number of IOMMUs present in the system */
175 static int amd_iommus_present;
177 /* IOMMUs have a non-present cache? */
178 bool amd_iommu_np_cache __read_mostly;
179 bool amd_iommu_iotlb_sup __read_mostly = true;
181 u32 amd_iommu_max_pasid __read_mostly = ~0;
183 bool amd_iommu_v2_present __read_mostly;
184 static bool amd_iommu_pc_present __read_mostly;
186 bool amd_iommu_force_isolation __read_mostly;
189 * List of protection domains - used during resume
191 LIST_HEAD(amd_iommu_pd_list);
192 spinlock_t amd_iommu_pd_lock;
195 * Pointer to the device table which is shared by all AMD IOMMUs
196 * it is indexed by the PCI device id or the HT unit id and contains
197 * information about the domain the device belongs to as well as the
198 * page table root pointer.
200 struct dev_table_entry *amd_iommu_dev_table;
202 * Pointer to a device table which the content of old device table
203 * will be copied to. It's only be used in kdump kernel.
205 static struct dev_table_entry *old_dev_tbl_cpy;
208 * The alias table is a driver specific data structure which contains the
209 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
210 * More than one device can share the same requestor id.
212 u16 *amd_iommu_alias_table;
215 * The rlookup table is used to find the IOMMU which is responsible
216 * for a specific device. It is also indexed by the PCI device id.
218 struct amd_iommu **amd_iommu_rlookup_table;
219 EXPORT_SYMBOL(amd_iommu_rlookup_table);
222 * This table is used to find the irq remapping table for a given device id
225 struct irq_remap_table **irq_lookup_table;
228 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
229 * to know which ones are already in use.
231 unsigned long *amd_iommu_pd_alloc_bitmap;
233 static u32 dev_table_size; /* size of the device table */
234 static u32 alias_table_size; /* size of the alias table */
235 static u32 rlookup_table_size; /* size if the rlookup table */
237 enum iommu_init_state {
248 IOMMU_CMDLINE_DISABLED,
251 /* Early ioapic and hpet maps from kernel command line */
252 #define EARLY_MAP_SIZE 4
253 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
254 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
255 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
257 static int __initdata early_ioapic_map_size;
258 static int __initdata early_hpet_map_size;
259 static int __initdata early_acpihid_map_size;
261 static bool __initdata cmdline_maps;
263 static enum iommu_init_state init_state = IOMMU_START_STATE;
265 static int amd_iommu_enable_interrupts(void);
266 static int __init iommu_go_to_state(enum iommu_init_state state);
267 static void init_device_table_dma(void);
269 static bool amd_iommu_pre_enabled = true;
271 bool translation_pre_enabled(struct amd_iommu *iommu)
273 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
275 EXPORT_SYMBOL(translation_pre_enabled);
277 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
279 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
282 static void init_translation_status(struct amd_iommu *iommu)
286 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
287 if (ctrl & (1<<CONTROL_IOMMU_EN))
288 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
291 static inline void update_last_devid(u16 devid)
293 if (devid > amd_iommu_last_bdf)
294 amd_iommu_last_bdf = devid;
297 static inline unsigned long tbl_size(int entry_size)
299 unsigned shift = PAGE_SHIFT +
300 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
305 int amd_iommu_get_num_iommus(void)
307 return amd_iommus_present;
310 /* Access to l1 and l2 indexed register spaces */
312 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
316 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
317 pci_read_config_dword(iommu->dev, 0xfc, &val);
321 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
323 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
324 pci_write_config_dword(iommu->dev, 0xfc, val);
325 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
328 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
332 pci_write_config_dword(iommu->dev, 0xf0, address);
333 pci_read_config_dword(iommu->dev, 0xf4, &val);
337 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
339 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
340 pci_write_config_dword(iommu->dev, 0xf4, val);
343 /****************************************************************************
345 * AMD IOMMU MMIO register space handling functions
347 * These functions are used to program the IOMMU device registers in
348 * MMIO space required for that driver.
350 ****************************************************************************/
353 * This function set the exclusion range in the IOMMU. DMA accesses to the
354 * exclusion range are passed through untranslated
356 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
358 u64 start = iommu->exclusion_start & PAGE_MASK;
359 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
362 if (!iommu->exclusion_start)
365 entry = start | MMIO_EXCL_ENABLE_MASK;
366 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
367 &entry, sizeof(entry));
370 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
371 &entry, sizeof(entry));
374 /* Programs the physical address of the device table into the IOMMU hardware */
375 static void iommu_set_device_table(struct amd_iommu *iommu)
379 BUG_ON(iommu->mmio_base == NULL);
381 entry = iommu_virt_to_phys(amd_iommu_dev_table);
382 entry |= (dev_table_size >> 12) - 1;
383 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
384 &entry, sizeof(entry));
387 /* Generic functions to enable/disable certain features of the IOMMU. */
388 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
392 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
393 ctrl |= (1ULL << bit);
394 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
397 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
401 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
402 ctrl &= ~(1ULL << bit);
403 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
406 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
410 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
411 ctrl &= ~CTRL_INV_TO_MASK;
412 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
413 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
416 /* Function to enable the hardware */
417 static void iommu_enable(struct amd_iommu *iommu)
419 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
422 static void iommu_disable(struct amd_iommu *iommu)
424 /* Disable command buffer */
425 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
427 /* Disable event logging and event interrupts */
428 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
429 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
431 /* Disable IOMMU GA_LOG */
432 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
433 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
435 /* Disable IOMMU hardware itself */
436 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
440 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
441 * the system has one.
443 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
445 if (!request_mem_region(address, end, "amd_iommu")) {
446 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
448 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
452 return (u8 __iomem *)ioremap_nocache(address, end);
455 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
457 if (iommu->mmio_base)
458 iounmap(iommu->mmio_base);
459 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
462 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
478 /****************************************************************************
480 * The functions below belong to the first pass of AMD IOMMU ACPI table
481 * parsing. In this pass we try to find out the highest device id this
482 * code has to handle. Upon this information the size of the shared data
483 * structures is determined later.
485 ****************************************************************************/
488 * This function calculates the length of a given IVHD entry
490 static inline int ivhd_entry_length(u8 *ivhd)
492 u32 type = ((struct ivhd_entry *)ivhd)->type;
495 return 0x04 << (*ivhd >> 6);
496 } else if (type == IVHD_DEV_ACPI_HID) {
497 /* For ACPI_HID, offset 21 is uid len */
498 return *((u8 *)ivhd + 21) + 22;
504 * After reading the highest device id from the IOMMU PCI capability header
505 * this function looks if there is a higher device id defined in the ACPI table
507 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
509 u8 *p = (void *)h, *end = (void *)h;
510 struct ivhd_entry *dev;
512 u32 ivhd_size = get_ivhd_header_size(h);
515 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
523 dev = (struct ivhd_entry *)p;
526 /* Use maximum BDF value for DEV_ALL */
527 update_last_devid(0xffff);
529 case IVHD_DEV_SELECT:
530 case IVHD_DEV_RANGE_END:
532 case IVHD_DEV_EXT_SELECT:
533 /* all the above subfield types refer to device ids */
534 update_last_devid(dev->devid);
539 p += ivhd_entry_length(p);
547 static int __init check_ivrs_checksum(struct acpi_table_header *table)
550 u8 checksum = 0, *p = (u8 *)table;
552 for (i = 0; i < table->length; ++i)
555 /* ACPI table corrupt */
556 pr_err(FW_BUG "AMD-Vi: IVRS invalid checksum\n");
564 * Iterate over all IVHD entries in the ACPI table and find the highest device
565 * id which we need to handle. This is the first of three functions which parse
566 * the ACPI table. So we check the checksum here.
568 static int __init find_last_devid_acpi(struct acpi_table_header *table)
570 u8 *p = (u8 *)table, *end = (u8 *)table;
571 struct ivhd_header *h;
573 p += IVRS_HEADER_LENGTH;
575 end += table->length;
577 h = (struct ivhd_header *)p;
578 if (h->type == amd_iommu_target_ivhd_type) {
579 int ret = find_last_devid_from_ivhd(h);
591 /****************************************************************************
593 * The following functions belong to the code path which parses the ACPI table
594 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
595 * data structures, initialize the device/alias/rlookup table and also
596 * basically initialize the hardware.
598 ****************************************************************************/
601 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
602 * write commands to that buffer later and the IOMMU will execute them
605 static int __init alloc_command_buffer(struct amd_iommu *iommu)
607 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
608 get_order(CMD_BUFFER_SIZE));
610 return iommu->cmd_buf ? 0 : -ENOMEM;
614 * This function resets the command buffer if the IOMMU stopped fetching
617 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
619 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
621 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
622 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
623 iommu->cmd_buf_head = 0;
624 iommu->cmd_buf_tail = 0;
626 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
630 * This function writes the command buffer address to the hardware and
633 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
637 BUG_ON(iommu->cmd_buf == NULL);
639 entry = iommu_virt_to_phys(iommu->cmd_buf);
640 entry |= MMIO_CMD_SIZE_512;
642 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
643 &entry, sizeof(entry));
645 amd_iommu_reset_cmd_buffer(iommu);
649 * This function disables the command buffer
651 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
653 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
656 static void __init free_command_buffer(struct amd_iommu *iommu)
658 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
661 /* allocates the memory where the IOMMU will log its events to */
662 static int __init alloc_event_buffer(struct amd_iommu *iommu)
664 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
665 get_order(EVT_BUFFER_SIZE));
667 return iommu->evt_buf ? 0 : -ENOMEM;
670 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
674 BUG_ON(iommu->evt_buf == NULL);
676 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
678 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
679 &entry, sizeof(entry));
681 /* set head and tail to zero manually */
682 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
683 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
685 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
689 * This function disables the event log buffer
691 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
693 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
696 static void __init free_event_buffer(struct amd_iommu *iommu)
698 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
701 /* allocates the memory where the IOMMU will log its events to */
702 static int __init alloc_ppr_log(struct amd_iommu *iommu)
704 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
705 get_order(PPR_LOG_SIZE));
707 return iommu->ppr_log ? 0 : -ENOMEM;
710 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
714 if (iommu->ppr_log == NULL)
717 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
719 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
720 &entry, sizeof(entry));
722 /* set head and tail to zero manually */
723 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
724 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
726 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
727 iommu_feature_enable(iommu, CONTROL_PPR_EN);
730 static void __init free_ppr_log(struct amd_iommu *iommu)
732 if (iommu->ppr_log == NULL)
735 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
738 static void free_ga_log(struct amd_iommu *iommu)
740 #ifdef CONFIG_IRQ_REMAP
742 free_pages((unsigned long)iommu->ga_log,
743 get_order(GA_LOG_SIZE));
744 if (iommu->ga_log_tail)
745 free_pages((unsigned long)iommu->ga_log_tail,
750 static int iommu_ga_log_enable(struct amd_iommu *iommu)
752 #ifdef CONFIG_IRQ_REMAP
758 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
760 /* Check if already running */
761 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
764 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
765 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
767 for (i = 0; i < LOOP_TIMEOUT; ++i) {
768 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
769 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
773 if (i >= LOOP_TIMEOUT)
775 #endif /* CONFIG_IRQ_REMAP */
779 #ifdef CONFIG_IRQ_REMAP
780 static int iommu_init_ga_log(struct amd_iommu *iommu)
784 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
787 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
788 get_order(GA_LOG_SIZE));
792 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
794 if (!iommu->ga_log_tail)
797 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
798 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
799 &entry, sizeof(entry));
800 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
801 (BIT_ULL(52)-1)) & ~7ULL;
802 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
803 &entry, sizeof(entry));
804 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
805 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
812 #endif /* CONFIG_IRQ_REMAP */
814 static int iommu_init_ga(struct amd_iommu *iommu)
818 #ifdef CONFIG_IRQ_REMAP
819 /* Note: We have already checked GASup from IVRS table.
820 * Now, we need to make sure that GAMSup is set.
822 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
823 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
824 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
826 ret = iommu_init_ga_log(iommu);
827 #endif /* CONFIG_IRQ_REMAP */
832 static void iommu_enable_xt(struct amd_iommu *iommu)
834 #ifdef CONFIG_IRQ_REMAP
836 * XT mode (32-bit APIC destination ID) requires
837 * GA mode (128-bit IRTE support) as a prerequisite.
839 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
840 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
841 iommu_feature_enable(iommu, CONTROL_XT_EN);
842 #endif /* CONFIG_IRQ_REMAP */
845 static void iommu_enable_gt(struct amd_iommu *iommu)
847 if (!iommu_feature(iommu, FEATURE_GT))
850 iommu_feature_enable(iommu, CONTROL_GT_EN);
853 /* sets a specific bit in the device table entry. */
854 static void set_dev_entry_bit(u16 devid, u8 bit)
856 int i = (bit >> 6) & 0x03;
857 int _bit = bit & 0x3f;
859 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
862 static int get_dev_entry_bit(u16 devid, u8 bit)
864 int i = (bit >> 6) & 0x03;
865 int _bit = bit & 0x3f;
867 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
871 static bool copy_device_table(void)
873 u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
874 struct dev_table_entry *old_devtb = NULL;
875 u32 lo, hi, devid, old_devtb_size;
876 phys_addr_t old_devtb_phys;
877 struct amd_iommu *iommu;
878 u16 dom_id, dte_v, irq_v;
882 if (!amd_iommu_pre_enabled)
885 pr_warn("Translation is already enabled - trying to copy translation structures\n");
886 for_each_iommu(iommu) {
887 /* All IOMMUs should use the same device table with the same size */
888 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
889 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
890 entry = (((u64) hi) << 32) + lo;
891 if (last_entry && last_entry != entry) {
892 pr_err("IOMMU:%d should use the same dev table as others!\n",
898 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
899 if (old_devtb_size != dev_table_size) {
900 pr_err("The device table size of IOMMU:%d is not expected!\n",
907 * When SME is enabled in the first kernel, the entry includes the
908 * memory encryption mask(sme_me_mask), we must remove the memory
909 * encryption mask to obtain the true physical address in kdump kernel.
911 old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
913 if (old_devtb_phys >= 0x100000000ULL) {
914 pr_err("The address of old device table is above 4G, not trustworthy!\n");
917 old_devtb = (sme_active() && is_kdump_kernel())
918 ? (__force void *)ioremap_encrypted(old_devtb_phys,
920 : memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
925 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
926 old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
927 get_order(dev_table_size));
928 if (old_dev_tbl_cpy == NULL) {
929 pr_err("Failed to allocate memory for copying old device table!\n");
933 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
934 old_dev_tbl_cpy[devid] = old_devtb[devid];
935 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
936 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
938 if (dte_v && dom_id) {
939 old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
940 old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
941 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
942 /* If gcr3 table existed, mask it out */
943 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
944 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
945 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
946 old_dev_tbl_cpy[devid].data[1] &= ~tmp;
947 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
949 old_dev_tbl_cpy[devid].data[0] &= ~tmp;
953 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
954 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
955 int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
956 if (irq_v && (int_ctl || int_tab_len)) {
957 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
958 (int_tab_len != DTE_IRQ_TABLE_LEN)) {
959 pr_err("Wrong old irq remapping flag: %#x\n", devid);
963 old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
971 void amd_iommu_apply_erratum_63(u16 devid)
975 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
976 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
979 set_dev_entry_bit(devid, DEV_ENTRY_IW);
982 /* Writes the specific IOMMU for a device into the rlookup table */
983 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
985 amd_iommu_rlookup_table[devid] = iommu;
989 * This function takes the device specific flags read from the ACPI
990 * table and sets up the device table entry with that information
992 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
993 u16 devid, u32 flags, u32 ext_flags)
995 if (flags & ACPI_DEVFLAG_INITPASS)
996 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
997 if (flags & ACPI_DEVFLAG_EXTINT)
998 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
999 if (flags & ACPI_DEVFLAG_NMI)
1000 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
1001 if (flags & ACPI_DEVFLAG_SYSMGT1)
1002 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
1003 if (flags & ACPI_DEVFLAG_SYSMGT2)
1004 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
1005 if (flags & ACPI_DEVFLAG_LINT0)
1006 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
1007 if (flags & ACPI_DEVFLAG_LINT1)
1008 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
1010 amd_iommu_apply_erratum_63(devid);
1012 set_iommu_for_device(iommu, devid);
1015 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1017 struct devid_map *entry;
1018 struct list_head *list;
1020 if (type == IVHD_SPECIAL_IOAPIC)
1022 else if (type == IVHD_SPECIAL_HPET)
1027 list_for_each_entry(entry, list, list) {
1028 if (!(entry->id == id && entry->cmd_line))
1031 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
1032 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1034 *devid = entry->devid;
1039 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1044 entry->devid = *devid;
1045 entry->cmd_line = cmd_line;
1047 list_add_tail(&entry->list, list);
1052 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1055 struct acpihid_map_entry *entry;
1056 struct list_head *list = &acpihid_map;
1058 list_for_each_entry(entry, list, list) {
1059 if (strcmp(entry->hid, hid) ||
1060 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1064 pr_info("AMD-Vi: Command-line override for hid:%s uid:%s\n",
1066 *devid = entry->devid;
1070 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1074 memcpy(entry->uid, uid, strlen(uid));
1075 memcpy(entry->hid, hid, strlen(hid));
1076 entry->devid = *devid;
1077 entry->cmd_line = cmd_line;
1078 entry->root_devid = (entry->devid & (~0x7));
1080 pr_info("AMD-Vi:%s, add hid:%s, uid:%s, rdevid:%d\n",
1081 entry->cmd_line ? "cmd" : "ivrs",
1082 entry->hid, entry->uid, entry->root_devid);
1084 list_add_tail(&entry->list, list);
1088 static int __init add_early_maps(void)
1092 for (i = 0; i < early_ioapic_map_size; ++i) {
1093 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1094 early_ioapic_map[i].id,
1095 &early_ioapic_map[i].devid,
1096 early_ioapic_map[i].cmd_line);
1101 for (i = 0; i < early_hpet_map_size; ++i) {
1102 ret = add_special_device(IVHD_SPECIAL_HPET,
1103 early_hpet_map[i].id,
1104 &early_hpet_map[i].devid,
1105 early_hpet_map[i].cmd_line);
1110 for (i = 0; i < early_acpihid_map_size; ++i) {
1111 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1112 early_acpihid_map[i].uid,
1113 &early_acpihid_map[i].devid,
1114 early_acpihid_map[i].cmd_line);
1123 * Reads the device exclusion range from ACPI and initializes the IOMMU with
1126 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
1128 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
1130 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
1135 * We only can configure exclusion ranges per IOMMU, not
1136 * per device. But we can enable the exclusion range per
1137 * device. This is done here
1139 set_dev_entry_bit(devid, DEV_ENTRY_EX);
1140 iommu->exclusion_start = m->range_start;
1141 iommu->exclusion_length = m->range_length;
1146 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1147 * initializes the hardware and our data structures with it.
1149 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1150 struct ivhd_header *h)
1153 u8 *end = p, flags = 0;
1154 u16 devid = 0, devid_start = 0, devid_to = 0;
1155 u32 dev_i, ext_flags = 0;
1157 struct ivhd_entry *e;
1162 ret = add_early_maps();
1167 * First save the recommended feature enable bits from ACPI
1169 iommu->acpi_flags = h->flags;
1172 * Done. Now parse the device entries
1174 ivhd_size = get_ivhd_header_size(h);
1176 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
1186 e = (struct ivhd_entry *)p;
1190 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1192 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1193 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1195 case IVHD_DEV_SELECT:
1197 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1199 PCI_BUS_NUM(e->devid),
1205 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1207 case IVHD_DEV_SELECT_RANGE_START:
1209 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1210 "devid: %02x:%02x.%x flags: %02x\n",
1211 PCI_BUS_NUM(e->devid),
1216 devid_start = e->devid;
1221 case IVHD_DEV_ALIAS:
1223 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1224 "flags: %02x devid_to: %02x:%02x.%x\n",
1225 PCI_BUS_NUM(e->devid),
1229 PCI_BUS_NUM(e->ext >> 8),
1230 PCI_SLOT(e->ext >> 8),
1231 PCI_FUNC(e->ext >> 8));
1234 devid_to = e->ext >> 8;
1235 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1236 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1237 amd_iommu_alias_table[devid] = devid_to;
1239 case IVHD_DEV_ALIAS_RANGE:
1241 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1242 "devid: %02x:%02x.%x flags: %02x "
1243 "devid_to: %02x:%02x.%x\n",
1244 PCI_BUS_NUM(e->devid),
1248 PCI_BUS_NUM(e->ext >> 8),
1249 PCI_SLOT(e->ext >> 8),
1250 PCI_FUNC(e->ext >> 8));
1252 devid_start = e->devid;
1254 devid_to = e->ext >> 8;
1258 case IVHD_DEV_EXT_SELECT:
1260 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1261 "flags: %02x ext: %08x\n",
1262 PCI_BUS_NUM(e->devid),
1268 set_dev_entry_from_acpi(iommu, devid, e->flags,
1271 case IVHD_DEV_EXT_SELECT_RANGE:
1273 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1274 "%02x:%02x.%x flags: %02x ext: %08x\n",
1275 PCI_BUS_NUM(e->devid),
1280 devid_start = e->devid;
1285 case IVHD_DEV_RANGE_END:
1287 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1288 PCI_BUS_NUM(e->devid),
1290 PCI_FUNC(e->devid));
1293 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1295 amd_iommu_alias_table[dev_i] = devid_to;
1296 set_dev_entry_from_acpi(iommu,
1297 devid_to, flags, ext_flags);
1299 set_dev_entry_from_acpi(iommu, dev_i,
1303 case IVHD_DEV_SPECIAL: {
1309 handle = e->ext & 0xff;
1310 devid = (e->ext >> 8) & 0xffff;
1311 type = (e->ext >> 24) & 0xff;
1313 if (type == IVHD_SPECIAL_IOAPIC)
1315 else if (type == IVHD_SPECIAL_HPET)
1320 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1326 ret = add_special_device(type, handle, &devid, false);
1331 * add_special_device might update the devid in case a
1332 * command-line override is present. So call
1333 * set_dev_entry_from_acpi after add_special_device.
1335 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1339 case IVHD_DEV_ACPI_HID: {
1341 u8 hid[ACPIHID_HID_LEN] = {0};
1342 u8 uid[ACPIHID_UID_LEN] = {0};
1345 if (h->type != 0x40) {
1346 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1351 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1352 hid[ACPIHID_HID_LEN - 1] = '\0';
1355 pr_err(FW_BUG "Invalid HID.\n");
1360 case UID_NOT_PRESENT:
1363 pr_warn(FW_BUG "Invalid UID length.\n");
1366 case UID_IS_INTEGER:
1368 sprintf(uid, "%d", e->uid);
1371 case UID_IS_CHARACTER:
1373 memcpy(uid, (u8 *)(&e->uid), ACPIHID_UID_LEN - 1);
1374 uid[ACPIHID_UID_LEN - 1] = '\0';
1382 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1390 ret = add_acpi_hid_device(hid, uid, &devid, false);
1395 * add_special_device might update the devid in case a
1396 * command-line override is present. So call
1397 * set_dev_entry_from_acpi after add_special_device.
1399 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1407 p += ivhd_entry_length(p);
1413 static void __init free_iommu_one(struct amd_iommu *iommu)
1415 free_command_buffer(iommu);
1416 free_event_buffer(iommu);
1417 free_ppr_log(iommu);
1419 iommu_unmap_mmio_space(iommu);
1422 static void __init free_iommu_all(void)
1424 struct amd_iommu *iommu, *next;
1426 for_each_iommu_safe(iommu, next) {
1427 list_del(&iommu->list);
1428 free_iommu_one(iommu);
1434 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1436 * BIOS should disable L2B micellaneous clock gating by setting
1437 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1439 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1443 if ((boot_cpu_data.x86 != 0x15) ||
1444 (boot_cpu_data.x86_model < 0x10) ||
1445 (boot_cpu_data.x86_model > 0x1f))
1448 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1449 pci_read_config_dword(iommu->dev, 0xf4, &value);
1454 /* Select NB indirect register 0x90 and enable writing */
1455 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1457 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1458 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1459 dev_name(&iommu->dev->dev));
1461 /* Clear the enable writing bit */
1462 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1466 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1468 * BIOS should enable ATS write permission check by setting
1469 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1471 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1475 if ((boot_cpu_data.x86 != 0x15) ||
1476 (boot_cpu_data.x86_model < 0x30) ||
1477 (boot_cpu_data.x86_model > 0x3f))
1480 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1481 value = iommu_read_l2(iommu, 0x47);
1486 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1487 iommu_write_l2(iommu, 0x47, value | BIT(0));
1489 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1490 dev_name(&iommu->dev->dev));
1494 * This function clues the initialization function for one IOMMU
1495 * together and also allocates the command buffer and programs the
1496 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1498 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1502 raw_spin_lock_init(&iommu->lock);
1504 /* Add IOMMU to internal data structures */
1505 list_add_tail(&iommu->list, &amd_iommu_list);
1506 iommu->index = amd_iommus_present++;
1508 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1509 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1513 /* Index is fine - add IOMMU to the array */
1514 amd_iommus[iommu->index] = iommu;
1517 * Copy data from ACPI table entry to the iommu struct
1519 iommu->devid = h->devid;
1520 iommu->cap_ptr = h->cap_ptr;
1521 iommu->pci_seg = h->pci_seg;
1522 iommu->mmio_phys = h->mmio_phys;
1526 /* Check if IVHD EFR contains proper max banks/counters */
1527 if ((h->efr_attr != 0) &&
1528 ((h->efr_attr & (0xF << 13)) != 0) &&
1529 ((h->efr_attr & (0x3F << 17)) != 0))
1530 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1532 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1533 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1534 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1535 if (((h->efr_attr & (0x1 << IOMMU_FEAT_XTSUP_SHIFT)) == 0))
1536 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1540 if (h->efr_reg & (1 << 9))
1541 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1543 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1544 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1545 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1546 if (((h->efr_reg & (0x1 << IOMMU_EFR_XTSUP_SHIFT)) == 0))
1547 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1553 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1554 iommu->mmio_phys_end);
1555 if (!iommu->mmio_base)
1558 if (alloc_command_buffer(iommu))
1561 if (alloc_event_buffer(iommu))
1564 iommu->int_enabled = false;
1566 init_translation_status(iommu);
1567 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1568 iommu_disable(iommu);
1569 clear_translation_pre_enabled(iommu);
1570 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1573 if (amd_iommu_pre_enabled)
1574 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1576 ret = init_iommu_from_acpi(iommu, h);
1580 ret = amd_iommu_create_irq_domain(iommu);
1585 * Make sure IOMMU is not considered to translate itself. The IVRS
1586 * table tells us so, but this is a lie!
1588 amd_iommu_rlookup_table[iommu->devid] = NULL;
1594 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1595 * @ivrs Pointer to the IVRS header
1597 * This function search through all IVDB of the maximum supported IVHD
1599 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1601 u8 *base = (u8 *)ivrs;
1602 struct ivhd_header *ivhd = (struct ivhd_header *)
1603 (base + IVRS_HEADER_LENGTH);
1604 u8 last_type = ivhd->type;
1605 u16 devid = ivhd->devid;
1607 while (((u8 *)ivhd - base < ivrs->length) &&
1608 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1609 u8 *p = (u8 *) ivhd;
1611 if (ivhd->devid == devid)
1612 last_type = ivhd->type;
1613 ivhd = (struct ivhd_header *)(p + ivhd->length);
1620 * Iterates over all IOMMU entries in the ACPI table, allocates the
1621 * IOMMU structure and initializes it with init_iommu_one()
1623 static int __init init_iommu_all(struct acpi_table_header *table)
1625 u8 *p = (u8 *)table, *end = (u8 *)table;
1626 struct ivhd_header *h;
1627 struct amd_iommu *iommu;
1630 end += table->length;
1631 p += IVRS_HEADER_LENGTH;
1634 h = (struct ivhd_header *)p;
1635 if (*p == amd_iommu_target_ivhd_type) {
1637 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1638 "seg: %d flags: %01x info %04x\n",
1639 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1640 PCI_FUNC(h->devid), h->cap_ptr,
1641 h->pci_seg, h->flags, h->info);
1642 DUMP_printk(" mmio-addr: %016llx\n",
1645 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1649 ret = init_iommu_one(iommu, h);
1661 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1662 u8 fxn, u64 *value, bool is_write);
1664 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1666 u64 val = 0xabcd, val2 = 0;
1668 if (!iommu_feature(iommu, FEATURE_PC))
1671 amd_iommu_pc_present = true;
1673 /* Check if the performance counters can be written to */
1674 if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1675 (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1677 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1678 amd_iommu_pc_present = false;
1682 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1684 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1685 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1686 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1689 static ssize_t amd_iommu_show_cap(struct device *dev,
1690 struct device_attribute *attr,
1693 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1694 return sprintf(buf, "%x\n", iommu->cap);
1696 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1698 static ssize_t amd_iommu_show_features(struct device *dev,
1699 struct device_attribute *attr,
1702 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1703 return sprintf(buf, "%llx\n", iommu->features);
1705 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1707 static struct attribute *amd_iommu_attrs[] = {
1709 &dev_attr_features.attr,
1713 static struct attribute_group amd_iommu_group = {
1714 .name = "amd-iommu",
1715 .attrs = amd_iommu_attrs,
1718 static const struct attribute_group *amd_iommu_groups[] = {
1723 static int __init iommu_init_pci(struct amd_iommu *iommu)
1725 int cap_ptr = iommu->cap_ptr;
1726 u32 range, misc, low, high;
1729 iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1730 iommu->devid & 0xff);
1734 /* Prevent binding other PCI device drivers to IOMMU devices */
1735 iommu->dev->match_driver = false;
1737 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1739 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1741 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1744 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1745 amd_iommu_iotlb_sup = false;
1747 /* read extended feature bits */
1748 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1749 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1751 iommu->features = ((u64)high << 32) | low;
1753 if (iommu_feature(iommu, FEATURE_GT)) {
1758 pasmax = iommu->features & FEATURE_PASID_MASK;
1759 pasmax >>= FEATURE_PASID_SHIFT;
1760 max_pasid = (1 << (pasmax + 1)) - 1;
1762 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1764 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1766 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1767 glxval >>= FEATURE_GLXVAL_SHIFT;
1769 if (amd_iommu_max_glx_val == -1)
1770 amd_iommu_max_glx_val = glxval;
1772 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1775 if (iommu_feature(iommu, FEATURE_GT) &&
1776 iommu_feature(iommu, FEATURE_PPR)) {
1777 iommu->is_iommu_v2 = true;
1778 amd_iommu_v2_present = true;
1781 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1784 ret = iommu_init_ga(iommu);
1788 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1789 amd_iommu_np_cache = true;
1791 init_iommu_perf_ctr(iommu);
1793 if (is_rd890_iommu(iommu->dev)) {
1797 pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
1801 * Some rd890 systems may not be fully reconfigured by the
1802 * BIOS, so it's necessary for us to store this information so
1803 * it can be reprogrammed on resume
1805 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1806 &iommu->stored_addr_lo);
1807 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1808 &iommu->stored_addr_hi);
1810 /* Low bit locks writes to configuration space */
1811 iommu->stored_addr_lo &= ~1;
1813 for (i = 0; i < 6; i++)
1814 for (j = 0; j < 0x12; j++)
1815 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1817 for (i = 0; i < 0x83; i++)
1818 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1821 amd_iommu_erratum_746_workaround(iommu);
1822 amd_iommu_ats_write_check_workaround(iommu);
1824 iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1825 amd_iommu_groups, "ivhd%d", iommu->index);
1826 iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1827 iommu_device_register(&iommu->iommu);
1829 return pci_enable_device(iommu->dev);
1832 static void print_iommu_info(void)
1834 static const char * const feat_str[] = {
1835 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1836 "IA", "GA", "HE", "PC"
1838 struct amd_iommu *iommu;
1840 for_each_iommu(iommu) {
1843 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1844 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1846 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1847 pr_info("AMD-Vi: Extended features (%#llx):\n",
1849 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1850 if (iommu_feature(iommu, (1ULL << i)))
1851 pr_cont(" %s", feat_str[i]);
1854 if (iommu->features & FEATURE_GAM_VAPIC)
1855 pr_cont(" GA_vAPIC");
1860 if (irq_remapping_enabled) {
1861 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1862 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1863 pr_info("AMD-Vi: virtual APIC enabled\n");
1864 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1865 pr_info("AMD-Vi: X2APIC enabled\n");
1869 static int __init amd_iommu_init_pci(void)
1871 struct amd_iommu *iommu;
1874 for_each_iommu(iommu) {
1875 ret = iommu_init_pci(iommu);
1881 * Order is important here to make sure any unity map requirements are
1882 * fulfilled. The unity mappings are created and written to the device
1883 * table during the amd_iommu_init_api() call.
1885 * After that we call init_device_table_dma() to make sure any
1886 * uninitialized DTE will block DMA, and in the end we flush the caches
1887 * of all IOMMUs to make sure the changes to the device table are
1890 ret = amd_iommu_init_api();
1892 init_device_table_dma();
1894 for_each_iommu(iommu)
1895 iommu_flush_all_caches(iommu);
1903 /****************************************************************************
1905 * The following functions initialize the MSI interrupts for all IOMMUs
1906 * in the system. It's a bit challenging because there could be multiple
1907 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1910 ****************************************************************************/
1912 static int iommu_setup_msi(struct amd_iommu *iommu)
1916 r = pci_enable_msi(iommu->dev);
1920 r = request_threaded_irq(iommu->dev->irq,
1921 amd_iommu_int_handler,
1922 amd_iommu_int_thread,
1927 pci_disable_msi(iommu->dev);
1931 iommu->int_enabled = true;
1936 static int iommu_init_msi(struct amd_iommu *iommu)
1940 if (iommu->int_enabled)
1943 if (iommu->dev->msi_cap)
1944 ret = iommu_setup_msi(iommu);
1952 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1954 if (iommu->ppr_log != NULL)
1955 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1957 iommu_ga_log_enable(iommu);
1962 /****************************************************************************
1964 * The next functions belong to the third pass of parsing the ACPI
1965 * table. In this last pass the memory mapping requirements are
1966 * gathered (like exclusion and unity mapping ranges).
1968 ****************************************************************************/
1970 static void __init free_unity_maps(void)
1972 struct unity_map_entry *entry, *next;
1974 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1975 list_del(&entry->list);
1980 /* called when we find an exclusion range definition in ACPI */
1981 static int __init init_exclusion_range(struct ivmd_header *m)
1986 case ACPI_IVMD_TYPE:
1987 set_device_exclusion_range(m->devid, m);
1989 case ACPI_IVMD_TYPE_ALL:
1990 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1991 set_device_exclusion_range(i, m);
1993 case ACPI_IVMD_TYPE_RANGE:
1994 for (i = m->devid; i <= m->aux; ++i)
1995 set_device_exclusion_range(i, m);
2004 /* called for unity map ACPI definition */
2005 static int __init init_unity_map_range(struct ivmd_header *m)
2007 struct unity_map_entry *e = NULL;
2010 e = kzalloc(sizeof(*e), GFP_KERNEL);
2018 case ACPI_IVMD_TYPE:
2019 s = "IVMD_TYPEi\t\t\t";
2020 e->devid_start = e->devid_end = m->devid;
2022 case ACPI_IVMD_TYPE_ALL:
2023 s = "IVMD_TYPE_ALL\t\t";
2025 e->devid_end = amd_iommu_last_bdf;
2027 case ACPI_IVMD_TYPE_RANGE:
2028 s = "IVMD_TYPE_RANGE\t\t";
2029 e->devid_start = m->devid;
2030 e->devid_end = m->aux;
2033 e->address_start = PAGE_ALIGN(m->range_start);
2034 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2035 e->prot = m->flags >> 1;
2037 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2038 " range_start: %016llx range_end: %016llx flags: %x\n", s,
2039 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2040 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
2041 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2042 e->address_start, e->address_end, m->flags);
2044 list_add_tail(&e->list, &amd_iommu_unity_map);
2049 /* iterates over all memory definitions we find in the ACPI table */
2050 static int __init init_memory_definitions(struct acpi_table_header *table)
2052 u8 *p = (u8 *)table, *end = (u8 *)table;
2053 struct ivmd_header *m;
2055 end += table->length;
2056 p += IVRS_HEADER_LENGTH;
2059 m = (struct ivmd_header *)p;
2060 if (m->flags & IVMD_FLAG_EXCL_RANGE)
2061 init_exclusion_range(m);
2062 else if (m->flags & IVMD_FLAG_UNITY_MAP)
2063 init_unity_map_range(m);
2072 * Init the device table to not allow DMA access for devices
2074 static void init_device_table_dma(void)
2078 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2079 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2080 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2084 static void __init uninit_device_table_dma(void)
2088 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2089 amd_iommu_dev_table[devid].data[0] = 0ULL;
2090 amd_iommu_dev_table[devid].data[1] = 0ULL;
2094 static void init_device_table(void)
2098 if (!amd_iommu_irq_remap)
2101 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2102 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2105 static void iommu_init_flags(struct amd_iommu *iommu)
2107 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2108 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2109 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2111 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2112 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2113 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2115 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2116 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2117 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2119 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2120 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2121 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2124 * make IOMMU memory accesses cache coherent
2126 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2128 /* Set IOTLB invalidation timeout to 1s */
2129 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2132 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2135 u32 ioc_feature_control;
2136 struct pci_dev *pdev = iommu->root_pdev;
2138 /* RD890 BIOSes may not have completely reconfigured the iommu */
2139 if (!is_rd890_iommu(iommu->dev) || !pdev)
2143 * First, we need to ensure that the iommu is enabled. This is
2144 * controlled by a register in the northbridge
2147 /* Select Northbridge indirect register 0x75 and enable writing */
2148 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2149 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2151 /* Enable the iommu */
2152 if (!(ioc_feature_control & 0x1))
2153 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2155 /* Restore the iommu BAR */
2156 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2157 iommu->stored_addr_lo);
2158 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2159 iommu->stored_addr_hi);
2161 /* Restore the l1 indirect regs for each of the 6 l1s */
2162 for (i = 0; i < 6; i++)
2163 for (j = 0; j < 0x12; j++)
2164 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2166 /* Restore the l2 indirect regs */
2167 for (i = 0; i < 0x83; i++)
2168 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2170 /* Lock PCI setup registers */
2171 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2172 iommu->stored_addr_lo | 1);
2175 static void iommu_enable_ga(struct amd_iommu *iommu)
2177 #ifdef CONFIG_IRQ_REMAP
2178 switch (amd_iommu_guest_ir) {
2179 case AMD_IOMMU_GUEST_IR_VAPIC:
2180 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2182 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2183 iommu_feature_enable(iommu, CONTROL_GA_EN);
2184 iommu->irte_ops = &irte_128_ops;
2187 iommu->irte_ops = &irte_32_ops;
2193 static void early_enable_iommu(struct amd_iommu *iommu)
2195 iommu_disable(iommu);
2196 iommu_init_flags(iommu);
2197 iommu_set_device_table(iommu);
2198 iommu_enable_command_buffer(iommu);
2199 iommu_enable_event_buffer(iommu);
2200 iommu_set_exclusion_range(iommu);
2201 iommu_enable_ga(iommu);
2202 iommu_enable_xt(iommu);
2203 iommu_enable(iommu);
2204 iommu_flush_all_caches(iommu);
2208 * This function finally enables all IOMMUs found in the system after
2209 * they have been initialized.
2211 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2212 * the old content of device table entries. Not this case or copy failed,
2213 * just continue as normal kernel does.
2215 static void early_enable_iommus(void)
2217 struct amd_iommu *iommu;
2220 if (!copy_device_table()) {
2222 * If come here because of failure in copying device table from old
2223 * kernel with all IOMMUs enabled, print error message and try to
2224 * free allocated old_dev_tbl_cpy.
2226 if (amd_iommu_pre_enabled)
2227 pr_err("Failed to copy DEV table from previous kernel.\n");
2228 if (old_dev_tbl_cpy != NULL)
2229 free_pages((unsigned long)old_dev_tbl_cpy,
2230 get_order(dev_table_size));
2232 for_each_iommu(iommu) {
2233 clear_translation_pre_enabled(iommu);
2234 early_enable_iommu(iommu);
2237 pr_info("Copied DEV table from previous kernel.\n");
2238 free_pages((unsigned long)amd_iommu_dev_table,
2239 get_order(dev_table_size));
2240 amd_iommu_dev_table = old_dev_tbl_cpy;
2241 for_each_iommu(iommu) {
2242 iommu_disable_command_buffer(iommu);
2243 iommu_disable_event_buffer(iommu);
2244 iommu_enable_command_buffer(iommu);
2245 iommu_enable_event_buffer(iommu);
2246 iommu_enable_ga(iommu);
2247 iommu_enable_xt(iommu);
2248 iommu_set_device_table(iommu);
2249 iommu_flush_all_caches(iommu);
2253 #ifdef CONFIG_IRQ_REMAP
2254 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2255 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2259 static void enable_iommus_v2(void)
2261 struct amd_iommu *iommu;
2263 for_each_iommu(iommu) {
2264 iommu_enable_ppr_log(iommu);
2265 iommu_enable_gt(iommu);
2269 static void enable_iommus(void)
2271 early_enable_iommus();
2276 static void disable_iommus(void)
2278 struct amd_iommu *iommu;
2280 for_each_iommu(iommu)
2281 iommu_disable(iommu);
2283 #ifdef CONFIG_IRQ_REMAP
2284 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2285 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2290 * Suspend/Resume support
2291 * disable suspend until real resume implemented
2294 static void amd_iommu_resume(void)
2296 struct amd_iommu *iommu;
2298 for_each_iommu(iommu)
2299 iommu_apply_resume_quirks(iommu);
2301 /* re-load the hardware */
2304 amd_iommu_enable_interrupts();
2307 static int amd_iommu_suspend(void)
2309 /* disable IOMMUs to go out of the way for BIOS */
2315 static struct syscore_ops amd_iommu_syscore_ops = {
2316 .suspend = amd_iommu_suspend,
2317 .resume = amd_iommu_resume,
2320 static void __init free_iommu_resources(void)
2322 kmemleak_free(irq_lookup_table);
2323 free_pages((unsigned long)irq_lookup_table,
2324 get_order(rlookup_table_size));
2325 irq_lookup_table = NULL;
2327 kmem_cache_destroy(amd_iommu_irq_cache);
2328 amd_iommu_irq_cache = NULL;
2330 free_pages((unsigned long)amd_iommu_rlookup_table,
2331 get_order(rlookup_table_size));
2332 amd_iommu_rlookup_table = NULL;
2334 free_pages((unsigned long)amd_iommu_alias_table,
2335 get_order(alias_table_size));
2336 amd_iommu_alias_table = NULL;
2338 free_pages((unsigned long)amd_iommu_dev_table,
2339 get_order(dev_table_size));
2340 amd_iommu_dev_table = NULL;
2344 #ifdef CONFIG_GART_IOMMU
2346 * We failed to initialize the AMD IOMMU - try fallback to GART
2354 /* SB IOAPIC is always on this device in AMD systems */
2355 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2357 static bool __init check_ioapic_information(void)
2359 const char *fw_bug = FW_BUG;
2360 bool ret, has_sb_ioapic;
2363 has_sb_ioapic = false;
2367 * If we have map overrides on the kernel command line the
2368 * messages in this function might not describe firmware bugs
2369 * anymore - so be careful
2374 for (idx = 0; idx < nr_ioapics; idx++) {
2375 int devid, id = mpc_ioapic_id(idx);
2377 devid = get_ioapic_devid(id);
2379 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
2382 } else if (devid == IOAPIC_SB_DEVID) {
2383 has_sb_ioapic = true;
2388 if (!has_sb_ioapic) {
2390 * We expect the SB IOAPIC to be listed in the IVRS
2391 * table. The system timer is connected to the SB IOAPIC
2392 * and if we don't have it in the list the system will
2393 * panic at boot time. This situation usually happens
2394 * when the BIOS is buggy and provides us the wrong
2395 * device id for the IOAPIC in the system.
2397 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
2401 pr_err("AMD-Vi: Disabling interrupt remapping\n");
2406 static void __init free_dma_resources(void)
2408 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2409 get_order(MAX_DOMAIN_ID/8));
2410 amd_iommu_pd_alloc_bitmap = NULL;
2416 * This is the hardware init function for AMD IOMMU in the system.
2417 * This function is called either from amd_iommu_init or from the interrupt
2418 * remapping setup code.
2420 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2423 * 1 pass) Discover the most comprehensive IVHD type to use.
2425 * 2 pass) Find the highest PCI device id the driver has to handle.
2426 * Upon this information the size of the data structures is
2427 * determined that needs to be allocated.
2429 * 3 pass) Initialize the data structures just allocated with the
2430 * information in the ACPI table about available AMD IOMMUs
2431 * in the system. It also maps the PCI devices in the
2432 * system to specific IOMMUs
2434 * 4 pass) After the basic data structures are allocated and
2435 * initialized we update them with information about memory
2436 * remapping requirements parsed out of the ACPI table in
2439 * After everything is set up the IOMMUs are enabled and the necessary
2440 * hotplug and suspend notifiers are registered.
2442 static int __init early_amd_iommu_init(void)
2444 struct acpi_table_header *ivrs_base;
2446 int i, remap_cache_sz, ret = 0;
2448 if (!amd_iommu_detected)
2451 status = acpi_get_table("IVRS", 0, &ivrs_base);
2452 if (status == AE_NOT_FOUND)
2454 else if (ACPI_FAILURE(status)) {
2455 const char *err = acpi_format_exception(status);
2456 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2461 * Validate checksum here so we don't need to do it when
2462 * we actually parse the table
2464 ret = check_ivrs_checksum(ivrs_base);
2468 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2469 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2472 * First parse ACPI tables to find the largest Bus/Dev/Func
2473 * we need to handle. Upon this information the shared data
2474 * structures for the IOMMUs in the system will be allocated
2476 ret = find_last_devid_acpi(ivrs_base);
2480 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2481 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2482 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2484 /* Device table - directly used by all IOMMUs */
2486 amd_iommu_dev_table = (void *)__get_free_pages(
2487 GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2488 get_order(dev_table_size));
2489 if (amd_iommu_dev_table == NULL)
2493 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2494 * IOMMU see for that device
2496 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2497 get_order(alias_table_size));
2498 if (amd_iommu_alias_table == NULL)
2501 /* IOMMU rlookup table - find the IOMMU for a specific device */
2502 amd_iommu_rlookup_table = (void *)__get_free_pages(
2503 GFP_KERNEL | __GFP_ZERO,
2504 get_order(rlookup_table_size));
2505 if (amd_iommu_rlookup_table == NULL)
2508 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2509 GFP_KERNEL | __GFP_ZERO,
2510 get_order(MAX_DOMAIN_ID/8));
2511 if (amd_iommu_pd_alloc_bitmap == NULL)
2515 * let all alias entries point to itself
2517 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2518 amd_iommu_alias_table[i] = i;
2521 * never allocate domain 0 because its used as the non-allocated and
2522 * error value placeholder
2524 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2526 spin_lock_init(&amd_iommu_pd_lock);
2529 * now the data structures are allocated and basically initialized
2530 * start the real acpi table scan
2532 ret = init_iommu_all(ivrs_base);
2536 /* Disable any previously enabled IOMMUs */
2537 if (!is_kdump_kernel() || amd_iommu_disabled)
2540 if (amd_iommu_irq_remap)
2541 amd_iommu_irq_remap = check_ioapic_information();
2543 if (amd_iommu_irq_remap) {
2545 * Interrupt remapping enabled, create kmem_cache for the
2549 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2550 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2552 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2553 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2555 IRQ_TABLE_ALIGNMENT,
2557 if (!amd_iommu_irq_cache)
2560 irq_lookup_table = (void *)__get_free_pages(
2561 GFP_KERNEL | __GFP_ZERO,
2562 get_order(rlookup_table_size));
2563 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2565 if (!irq_lookup_table)
2569 ret = init_memory_definitions(ivrs_base);
2573 /* init the device table */
2574 init_device_table();
2577 /* Don't leak any ACPI memory */
2578 acpi_put_table(ivrs_base);
2584 static int amd_iommu_enable_interrupts(void)
2586 struct amd_iommu *iommu;
2589 for_each_iommu(iommu) {
2590 ret = iommu_init_msi(iommu);
2599 static bool detect_ivrs(void)
2601 struct acpi_table_header *ivrs_base;
2604 status = acpi_get_table("IVRS", 0, &ivrs_base);
2605 if (status == AE_NOT_FOUND)
2607 else if (ACPI_FAILURE(status)) {
2608 const char *err = acpi_format_exception(status);
2609 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2613 acpi_put_table(ivrs_base);
2615 /* Make sure ACS will be enabled during PCI probe */
2621 /****************************************************************************
2623 * AMD IOMMU Initialization State Machine
2625 ****************************************************************************/
2627 static int __init state_next(void)
2631 switch (init_state) {
2632 case IOMMU_START_STATE:
2633 if (!detect_ivrs()) {
2634 init_state = IOMMU_NOT_FOUND;
2637 init_state = IOMMU_IVRS_DETECTED;
2640 case IOMMU_IVRS_DETECTED:
2641 ret = early_amd_iommu_init();
2642 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2643 if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2644 pr_info("AMD-Vi: AMD IOMMU disabled on kernel command-line\n");
2645 free_dma_resources();
2646 free_iommu_resources();
2647 init_state = IOMMU_CMDLINE_DISABLED;
2651 case IOMMU_ACPI_FINISHED:
2652 early_enable_iommus();
2653 x86_platform.iommu_shutdown = disable_iommus;
2654 init_state = IOMMU_ENABLED;
2657 register_syscore_ops(&amd_iommu_syscore_ops);
2658 ret = amd_iommu_init_pci();
2659 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2662 case IOMMU_PCI_INIT:
2663 ret = amd_iommu_enable_interrupts();
2664 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2666 case IOMMU_INTERRUPTS_EN:
2667 ret = amd_iommu_init_dma_ops();
2668 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2671 init_state = IOMMU_INITIALIZED;
2673 case IOMMU_INITIALIZED:
2676 case IOMMU_NOT_FOUND:
2677 case IOMMU_INIT_ERROR:
2678 case IOMMU_CMDLINE_DISABLED:
2679 /* Error states => do nothing */
2690 static int __init iommu_go_to_state(enum iommu_init_state state)
2694 while (init_state != state) {
2695 if (init_state == IOMMU_NOT_FOUND ||
2696 init_state == IOMMU_INIT_ERROR ||
2697 init_state == IOMMU_CMDLINE_DISABLED)
2705 #ifdef CONFIG_IRQ_REMAP
2706 int __init amd_iommu_prepare(void)
2710 amd_iommu_irq_remap = true;
2712 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2715 return amd_iommu_irq_remap ? 0 : -ENODEV;
2718 int __init amd_iommu_enable(void)
2722 ret = iommu_go_to_state(IOMMU_ENABLED);
2726 irq_remapping_enabled = 1;
2727 return amd_iommu_xt_mode;
2730 void amd_iommu_disable(void)
2732 amd_iommu_suspend();
2735 int amd_iommu_reenable(int mode)
2742 int __init amd_iommu_enable_faulting(void)
2744 /* We enable MSI later when PCI is initialized */
2750 * This is the core init function for AMD IOMMU hardware in the system.
2751 * This function is called from the generic x86 DMA layer initialization
2754 static int __init amd_iommu_init(void)
2756 struct amd_iommu *iommu;
2759 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2761 free_dma_resources();
2762 if (!irq_remapping_enabled) {
2764 free_iommu_resources();
2766 uninit_device_table_dma();
2767 for_each_iommu(iommu)
2768 iommu_flush_all_caches(iommu);
2772 for_each_iommu(iommu)
2773 amd_iommu_debugfs_setup(iommu);
2778 static bool amd_iommu_sme_check(void)
2780 if (!sme_active() || (boot_cpu_data.x86 != 0x17))
2783 /* For Fam17h, a specific level of support is required */
2784 if (boot_cpu_data.microcode >= 0x08001205)
2787 if ((boot_cpu_data.microcode >= 0x08001126) &&
2788 (boot_cpu_data.microcode <= 0x080011ff))
2791 pr_notice("AMD-Vi: IOMMU not currently supported when SME is active\n");
2796 /****************************************************************************
2798 * Early detect code. This code runs at IOMMU detection time in the DMA
2799 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2802 ****************************************************************************/
2803 int __init amd_iommu_detect(void)
2807 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2810 if (!amd_iommu_sme_check())
2813 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2817 amd_iommu_detected = true;
2819 x86_init.iommu.iommu_init = amd_iommu_init;
2824 /****************************************************************************
2826 * Parsing functions for the AMD IOMMU specific kernel command line
2829 ****************************************************************************/
2831 static int __init parse_amd_iommu_dump(char *str)
2833 amd_iommu_dump = true;
2838 static int __init parse_amd_iommu_intr(char *str)
2840 for (; *str; ++str) {
2841 if (strncmp(str, "legacy", 6) == 0) {
2842 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
2845 if (strncmp(str, "vapic", 5) == 0) {
2846 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2853 static int __init parse_amd_iommu_options(char *str)
2855 for (; *str; ++str) {
2856 if (strncmp(str, "fullflush", 9) == 0)
2857 amd_iommu_unmap_flush = true;
2858 if (strncmp(str, "off", 3) == 0)
2859 amd_iommu_disabled = true;
2860 if (strncmp(str, "force_isolation", 15) == 0)
2861 amd_iommu_force_isolation = true;
2867 static int __init parse_ivrs_ioapic(char *str)
2869 unsigned int bus, dev, fn;
2873 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2876 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2880 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2881 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2886 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2888 cmdline_maps = true;
2889 i = early_ioapic_map_size++;
2890 early_ioapic_map[i].id = id;
2891 early_ioapic_map[i].devid = devid;
2892 early_ioapic_map[i].cmd_line = true;
2897 static int __init parse_ivrs_hpet(char *str)
2899 unsigned int bus, dev, fn;
2903 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2906 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2910 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2911 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2916 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2918 cmdline_maps = true;
2919 i = early_hpet_map_size++;
2920 early_hpet_map[i].id = id;
2921 early_hpet_map[i].devid = devid;
2922 early_hpet_map[i].cmd_line = true;
2927 static int __init parse_ivrs_acpihid(char *str)
2930 char *hid, *uid, *p;
2931 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
2934 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
2936 pr_err("AMD-Vi: Invalid command line: ivrs_acpihid(%s)\n", str);
2941 hid = strsep(&p, ":");
2944 if (!hid || !(*hid) || !uid) {
2945 pr_err("AMD-Vi: Invalid command line: hid or uid\n");
2949 i = early_acpihid_map_size++;
2950 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
2951 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
2952 early_acpihid_map[i].devid =
2953 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2954 early_acpihid_map[i].cmd_line = true;
2959 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2960 __setup("amd_iommu=", parse_amd_iommu_options);
2961 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
2962 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2963 __setup("ivrs_hpet", parse_ivrs_hpet);
2964 __setup("ivrs_acpihid", parse_ivrs_acpihid);
2966 IOMMU_INIT_FINISH(amd_iommu_detect,
2967 gart_iommu_hole_init,
2971 bool amd_iommu_v2_supported(void)
2973 return amd_iommu_v2_present;
2975 EXPORT_SYMBOL(amd_iommu_v2_supported);
2977 struct amd_iommu *get_amd_iommu(unsigned int idx)
2980 struct amd_iommu *iommu;
2982 for_each_iommu(iommu)
2987 EXPORT_SYMBOL(get_amd_iommu);
2989 /****************************************************************************
2991 * IOMMU EFR Performance Counter support functionality. This code allows
2992 * access to the IOMMU PC functionality.
2994 ****************************************************************************/
2996 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
2998 struct amd_iommu *iommu = get_amd_iommu(idx);
3001 return iommu->max_banks;
3005 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3007 bool amd_iommu_pc_supported(void)
3009 return amd_iommu_pc_present;
3011 EXPORT_SYMBOL(amd_iommu_pc_supported);
3013 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3015 struct amd_iommu *iommu = get_amd_iommu(idx);
3018 return iommu->max_counters;
3022 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3024 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3025 u8 fxn, u64 *value, bool is_write)
3030 /* Make sure the IOMMU PC resource is available */
3031 if (!amd_iommu_pc_present)
3034 /* Check for valid iommu and pc register indexing */
3035 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3038 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3040 /* Limit the offset to the hw defined mmio region aperture */
3041 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3042 (iommu->max_counters << 8) | 0x28);
3043 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3044 (offset > max_offset_lim))
3048 u64 val = *value & GENMASK_ULL(47, 0);
3050 writel((u32)val, iommu->mmio_base + offset);
3051 writel((val >> 32), iommu->mmio_base + offset + 4);
3053 *value = readl(iommu->mmio_base + offset + 4);
3055 *value |= readl(iommu->mmio_base + offset);
3056 *value &= GENMASK_ULL(47, 0);
3062 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3067 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3069 EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3071 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3076 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3078 EXPORT_SYMBOL(amd_iommu_pc_set_reg);