1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
4 * Author: Joerg Roedel <jroedel@suse.de>
5 * Leo Duran <leo.duran@amd.com>
8 #define pr_fmt(fmt) "AMD-Vi: " fmt
9 #define dev_fmt(fmt) pr_fmt(fmt)
11 #include <linux/pci.h>
12 #include <linux/acpi.h>
13 #include <linux/list.h>
14 #include <linux/bitmap.h>
15 #include <linux/slab.h>
16 #include <linux/syscore_ops.h>
17 #include <linux/interrupt.h>
18 #include <linux/msi.h>
19 #include <linux/amd-iommu.h>
20 #include <linux/export.h>
21 #include <linux/iommu.h>
22 #include <linux/kmemleak.h>
23 #include <linux/mem_encrypt.h>
24 #include <asm/pci-direct.h>
25 #include <asm/iommu.h>
27 #include <asm/msidef.h>
29 #include <asm/x86_init.h>
30 #include <asm/iommu_table.h>
31 #include <asm/io_apic.h>
32 #include <asm/irq_remapping.h>
34 #include <linux/crash_dump.h>
35 #include "amd_iommu_proto.h"
36 #include "amd_iommu_types.h"
37 #include "irq_remapping.h"
40 * definitions for the ACPI scanning code
42 #define IVRS_HEADER_LENGTH 48
44 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
45 #define ACPI_IVMD_TYPE_ALL 0x20
46 #define ACPI_IVMD_TYPE 0x21
47 #define ACPI_IVMD_TYPE_RANGE 0x22
49 #define IVHD_DEV_ALL 0x01
50 #define IVHD_DEV_SELECT 0x02
51 #define IVHD_DEV_SELECT_RANGE_START 0x03
52 #define IVHD_DEV_RANGE_END 0x04
53 #define IVHD_DEV_ALIAS 0x42
54 #define IVHD_DEV_ALIAS_RANGE 0x43
55 #define IVHD_DEV_EXT_SELECT 0x46
56 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
57 #define IVHD_DEV_SPECIAL 0x48
58 #define IVHD_DEV_ACPI_HID 0xf0
60 #define UID_NOT_PRESENT 0
61 #define UID_IS_INTEGER 1
62 #define UID_IS_CHARACTER 2
64 #define IVHD_SPECIAL_IOAPIC 1
65 #define IVHD_SPECIAL_HPET 2
67 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
68 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
69 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
70 #define IVHD_FLAG_ISOC_EN_MASK 0x08
72 #define IVMD_FLAG_EXCL_RANGE 0x08
73 #define IVMD_FLAG_UNITY_MAP 0x01
75 #define ACPI_DEVFLAG_INITPASS 0x01
76 #define ACPI_DEVFLAG_EXTINT 0x02
77 #define ACPI_DEVFLAG_NMI 0x04
78 #define ACPI_DEVFLAG_SYSMGT1 0x10
79 #define ACPI_DEVFLAG_SYSMGT2 0x20
80 #define ACPI_DEVFLAG_LINT0 0x40
81 #define ACPI_DEVFLAG_LINT1 0x80
82 #define ACPI_DEVFLAG_ATSDIS 0x10000000
84 #define LOOP_TIMEOUT 100000
86 * ACPI table definitions
88 * These data structures are laid over the table to parse the important values
92 extern const struct iommu_ops amd_iommu_ops;
95 * structure describing one IOMMU in the ACPI table. Typically followed by one
96 * or more ivhd_entrys.
109 /* Following only valid on IVHD type 11h and 40h */
110 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
112 } __attribute__((packed));
115 * A device entry describing which devices a specific IOMMU translates and
116 * which requestor ids they use.
128 } __attribute__((packed));
131 * An AMD IOMMU memory definition structure. It defines things like exclusion
132 * ranges for devices and regions that should be unity mapped.
143 } __attribute__((packed));
146 bool amd_iommu_irq_remap __read_mostly;
148 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
149 static int amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
151 static bool amd_iommu_detected;
152 static bool __initdata amd_iommu_disabled;
153 static int amd_iommu_target_ivhd_type;
155 u16 amd_iommu_last_bdf; /* largest PCI device id we have
157 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
159 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
161 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
164 /* Array to assign indices to IOMMUs*/
165 struct amd_iommu *amd_iommus[MAX_IOMMUS];
167 /* Number of IOMMUs present in the system */
168 static int amd_iommus_present;
170 /* IOMMUs have a non-present cache? */
171 bool amd_iommu_np_cache __read_mostly;
172 bool amd_iommu_iotlb_sup __read_mostly = true;
174 u32 amd_iommu_max_pasid __read_mostly = ~0;
176 bool amd_iommu_v2_present __read_mostly;
177 static bool amd_iommu_pc_present __read_mostly;
179 bool amd_iommu_force_isolation __read_mostly;
182 * Pointer to the device table which is shared by all AMD IOMMUs
183 * it is indexed by the PCI device id or the HT unit id and contains
184 * information about the domain the device belongs to as well as the
185 * page table root pointer.
187 struct dev_table_entry *amd_iommu_dev_table;
189 * Pointer to a device table which the content of old device table
190 * will be copied to. It's only be used in kdump kernel.
192 static struct dev_table_entry *old_dev_tbl_cpy;
195 * The alias table is a driver specific data structure which contains the
196 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
197 * More than one device can share the same requestor id.
199 u16 *amd_iommu_alias_table;
202 * The rlookup table is used to find the IOMMU which is responsible
203 * for a specific device. It is also indexed by the PCI device id.
205 struct amd_iommu **amd_iommu_rlookup_table;
206 EXPORT_SYMBOL(amd_iommu_rlookup_table);
209 * This table is used to find the irq remapping table for a given device id
212 struct irq_remap_table **irq_lookup_table;
215 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
216 * to know which ones are already in use.
218 unsigned long *amd_iommu_pd_alloc_bitmap;
220 static u32 dev_table_size; /* size of the device table */
221 static u32 alias_table_size; /* size of the alias table */
222 static u32 rlookup_table_size; /* size if the rlookup table */
224 enum iommu_init_state {
235 IOMMU_CMDLINE_DISABLED,
238 /* Early ioapic and hpet maps from kernel command line */
239 #define EARLY_MAP_SIZE 4
240 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
241 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
242 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
244 static int __initdata early_ioapic_map_size;
245 static int __initdata early_hpet_map_size;
246 static int __initdata early_acpihid_map_size;
248 static bool __initdata cmdline_maps;
250 static enum iommu_init_state init_state = IOMMU_START_STATE;
252 static int amd_iommu_enable_interrupts(void);
253 static int __init iommu_go_to_state(enum iommu_init_state state);
254 static void init_device_table_dma(void);
256 static bool amd_iommu_pre_enabled = true;
258 bool translation_pre_enabled(struct amd_iommu *iommu)
260 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
262 EXPORT_SYMBOL(translation_pre_enabled);
264 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
266 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
269 static void init_translation_status(struct amd_iommu *iommu)
273 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
274 if (ctrl & (1<<CONTROL_IOMMU_EN))
275 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
278 static inline void update_last_devid(u16 devid)
280 if (devid > amd_iommu_last_bdf)
281 amd_iommu_last_bdf = devid;
284 static inline unsigned long tbl_size(int entry_size)
286 unsigned shift = PAGE_SHIFT +
287 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
292 int amd_iommu_get_num_iommus(void)
294 return amd_iommus_present;
297 /* Access to l1 and l2 indexed register spaces */
299 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
303 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
304 pci_read_config_dword(iommu->dev, 0xfc, &val);
308 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
310 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
311 pci_write_config_dword(iommu->dev, 0xfc, val);
312 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
315 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
319 pci_write_config_dword(iommu->dev, 0xf0, address);
320 pci_read_config_dword(iommu->dev, 0xf4, &val);
324 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
326 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
327 pci_write_config_dword(iommu->dev, 0xf4, val);
330 /****************************************************************************
332 * AMD IOMMU MMIO register space handling functions
334 * These functions are used to program the IOMMU device registers in
335 * MMIO space required for that driver.
337 ****************************************************************************/
340 * This function set the exclusion range in the IOMMU. DMA accesses to the
341 * exclusion range are passed through untranslated
343 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
345 u64 start = iommu->exclusion_start & PAGE_MASK;
346 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
349 if (!iommu->exclusion_start)
352 entry = start | MMIO_EXCL_ENABLE_MASK;
353 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
354 &entry, sizeof(entry));
357 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
358 &entry, sizeof(entry));
361 /* Programs the physical address of the device table into the IOMMU hardware */
362 static void iommu_set_device_table(struct amd_iommu *iommu)
366 BUG_ON(iommu->mmio_base == NULL);
368 entry = iommu_virt_to_phys(amd_iommu_dev_table);
369 entry |= (dev_table_size >> 12) - 1;
370 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
371 &entry, sizeof(entry));
374 /* Generic functions to enable/disable certain features of the IOMMU. */
375 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
379 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
380 ctrl |= (1ULL << bit);
381 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
384 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
388 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
389 ctrl &= ~(1ULL << bit);
390 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
393 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
397 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
398 ctrl &= ~CTRL_INV_TO_MASK;
399 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
400 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
403 /* Function to enable the hardware */
404 static void iommu_enable(struct amd_iommu *iommu)
406 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
409 static void iommu_disable(struct amd_iommu *iommu)
411 if (!iommu->mmio_base)
414 /* Disable command buffer */
415 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
417 /* Disable event logging and event interrupts */
418 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
419 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
421 /* Disable IOMMU GA_LOG */
422 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
423 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
425 /* Disable IOMMU hardware itself */
426 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
430 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
431 * the system has one.
433 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
435 if (!request_mem_region(address, end, "amd_iommu")) {
436 pr_err("Can not reserve memory region %llx-%llx for mmio\n",
438 pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
442 return (u8 __iomem *)ioremap_nocache(address, end);
445 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
447 if (iommu->mmio_base)
448 iounmap(iommu->mmio_base);
449 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
452 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
468 /****************************************************************************
470 * The functions below belong to the first pass of AMD IOMMU ACPI table
471 * parsing. In this pass we try to find out the highest device id this
472 * code has to handle. Upon this information the size of the shared data
473 * structures is determined later.
475 ****************************************************************************/
478 * This function calculates the length of a given IVHD entry
480 static inline int ivhd_entry_length(u8 *ivhd)
482 u32 type = ((struct ivhd_entry *)ivhd)->type;
485 return 0x04 << (*ivhd >> 6);
486 } else if (type == IVHD_DEV_ACPI_HID) {
487 /* For ACPI_HID, offset 21 is uid len */
488 return *((u8 *)ivhd + 21) + 22;
494 * After reading the highest device id from the IOMMU PCI capability header
495 * this function looks if there is a higher device id defined in the ACPI table
497 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
499 u8 *p = (void *)h, *end = (void *)h;
500 struct ivhd_entry *dev;
502 u32 ivhd_size = get_ivhd_header_size(h);
505 pr_err("Unsupported IVHD type %#x\n", h->type);
513 dev = (struct ivhd_entry *)p;
516 /* Use maximum BDF value for DEV_ALL */
517 update_last_devid(0xffff);
519 case IVHD_DEV_SELECT:
520 case IVHD_DEV_RANGE_END:
522 case IVHD_DEV_EXT_SELECT:
523 /* all the above subfield types refer to device ids */
524 update_last_devid(dev->devid);
529 p += ivhd_entry_length(p);
537 static int __init check_ivrs_checksum(struct acpi_table_header *table)
540 u8 checksum = 0, *p = (u8 *)table;
542 for (i = 0; i < table->length; ++i)
545 /* ACPI table corrupt */
546 pr_err(FW_BUG "IVRS invalid checksum\n");
554 * Iterate over all IVHD entries in the ACPI table and find the highest device
555 * id which we need to handle. This is the first of three functions which parse
556 * the ACPI table. So we check the checksum here.
558 static int __init find_last_devid_acpi(struct acpi_table_header *table)
560 u8 *p = (u8 *)table, *end = (u8 *)table;
561 struct ivhd_header *h;
563 p += IVRS_HEADER_LENGTH;
565 end += table->length;
567 h = (struct ivhd_header *)p;
568 if (h->type == amd_iommu_target_ivhd_type) {
569 int ret = find_last_devid_from_ivhd(h);
581 /****************************************************************************
583 * The following functions belong to the code path which parses the ACPI table
584 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
585 * data structures, initialize the device/alias/rlookup table and also
586 * basically initialize the hardware.
588 ****************************************************************************/
591 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
592 * write commands to that buffer later and the IOMMU will execute them
595 static int __init alloc_command_buffer(struct amd_iommu *iommu)
597 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
598 get_order(CMD_BUFFER_SIZE));
600 return iommu->cmd_buf ? 0 : -ENOMEM;
604 * This function resets the command buffer if the IOMMU stopped fetching
607 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
609 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
611 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
612 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
613 iommu->cmd_buf_head = 0;
614 iommu->cmd_buf_tail = 0;
616 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
620 * This function writes the command buffer address to the hardware and
623 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
627 BUG_ON(iommu->cmd_buf == NULL);
629 entry = iommu_virt_to_phys(iommu->cmd_buf);
630 entry |= MMIO_CMD_SIZE_512;
632 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
633 &entry, sizeof(entry));
635 amd_iommu_reset_cmd_buffer(iommu);
639 * This function disables the command buffer
641 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
643 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
646 static void __init free_command_buffer(struct amd_iommu *iommu)
648 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
651 /* allocates the memory where the IOMMU will log its events to */
652 static int __init alloc_event_buffer(struct amd_iommu *iommu)
654 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
655 get_order(EVT_BUFFER_SIZE));
657 return iommu->evt_buf ? 0 : -ENOMEM;
660 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
664 BUG_ON(iommu->evt_buf == NULL);
666 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
668 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
669 &entry, sizeof(entry));
671 /* set head and tail to zero manually */
672 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
673 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
675 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
679 * This function disables the event log buffer
681 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
683 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
686 static void __init free_event_buffer(struct amd_iommu *iommu)
688 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
691 /* allocates the memory where the IOMMU will log its events to */
692 static int __init alloc_ppr_log(struct amd_iommu *iommu)
694 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
695 get_order(PPR_LOG_SIZE));
697 return iommu->ppr_log ? 0 : -ENOMEM;
700 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
704 if (iommu->ppr_log == NULL)
707 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
709 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
710 &entry, sizeof(entry));
712 /* set head and tail to zero manually */
713 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
714 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
716 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
717 iommu_feature_enable(iommu, CONTROL_PPR_EN);
720 static void __init free_ppr_log(struct amd_iommu *iommu)
722 if (iommu->ppr_log == NULL)
725 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
728 static void free_ga_log(struct amd_iommu *iommu)
730 #ifdef CONFIG_IRQ_REMAP
732 free_pages((unsigned long)iommu->ga_log,
733 get_order(GA_LOG_SIZE));
734 if (iommu->ga_log_tail)
735 free_pages((unsigned long)iommu->ga_log_tail,
740 static int iommu_ga_log_enable(struct amd_iommu *iommu)
742 #ifdef CONFIG_IRQ_REMAP
748 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
750 /* Check if already running */
751 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
754 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
755 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
757 for (i = 0; i < LOOP_TIMEOUT; ++i) {
758 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
759 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
763 if (i >= LOOP_TIMEOUT)
765 #endif /* CONFIG_IRQ_REMAP */
769 #ifdef CONFIG_IRQ_REMAP
770 static int iommu_init_ga_log(struct amd_iommu *iommu)
774 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
777 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
778 get_order(GA_LOG_SIZE));
782 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
784 if (!iommu->ga_log_tail)
787 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
788 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
789 &entry, sizeof(entry));
790 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
791 (BIT_ULL(52)-1)) & ~7ULL;
792 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
793 &entry, sizeof(entry));
794 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
795 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
802 #endif /* CONFIG_IRQ_REMAP */
804 static int iommu_init_ga(struct amd_iommu *iommu)
808 #ifdef CONFIG_IRQ_REMAP
809 /* Note: We have already checked GASup from IVRS table.
810 * Now, we need to make sure that GAMSup is set.
812 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
813 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
814 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
816 ret = iommu_init_ga_log(iommu);
817 #endif /* CONFIG_IRQ_REMAP */
822 static void iommu_enable_xt(struct amd_iommu *iommu)
824 #ifdef CONFIG_IRQ_REMAP
826 * XT mode (32-bit APIC destination ID) requires
827 * GA mode (128-bit IRTE support) as a prerequisite.
829 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
830 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
831 iommu_feature_enable(iommu, CONTROL_XT_EN);
832 #endif /* CONFIG_IRQ_REMAP */
835 static void iommu_enable_gt(struct amd_iommu *iommu)
837 if (!iommu_feature(iommu, FEATURE_GT))
840 iommu_feature_enable(iommu, CONTROL_GT_EN);
843 /* sets a specific bit in the device table entry. */
844 static void set_dev_entry_bit(u16 devid, u8 bit)
846 int i = (bit >> 6) & 0x03;
847 int _bit = bit & 0x3f;
849 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
852 static int get_dev_entry_bit(u16 devid, u8 bit)
854 int i = (bit >> 6) & 0x03;
855 int _bit = bit & 0x3f;
857 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
861 static bool copy_device_table(void)
863 u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
864 struct dev_table_entry *old_devtb = NULL;
865 u32 lo, hi, devid, old_devtb_size;
866 phys_addr_t old_devtb_phys;
867 struct amd_iommu *iommu;
868 u16 dom_id, dte_v, irq_v;
872 if (!amd_iommu_pre_enabled)
875 pr_warn("Translation is already enabled - trying to copy translation structures\n");
876 for_each_iommu(iommu) {
877 /* All IOMMUs should use the same device table with the same size */
878 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
879 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
880 entry = (((u64) hi) << 32) + lo;
881 if (last_entry && last_entry != entry) {
882 pr_err("IOMMU:%d should use the same dev table as others!\n",
888 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
889 if (old_devtb_size != dev_table_size) {
890 pr_err("The device table size of IOMMU:%d is not expected!\n",
897 * When SME is enabled in the first kernel, the entry includes the
898 * memory encryption mask(sme_me_mask), we must remove the memory
899 * encryption mask to obtain the true physical address in kdump kernel.
901 old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
903 if (old_devtb_phys >= 0x100000000ULL) {
904 pr_err("The address of old device table is above 4G, not trustworthy!\n");
907 old_devtb = (sme_active() && is_kdump_kernel())
908 ? (__force void *)ioremap_encrypted(old_devtb_phys,
910 : memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
915 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
916 old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
917 get_order(dev_table_size));
918 if (old_dev_tbl_cpy == NULL) {
919 pr_err("Failed to allocate memory for copying old device table!\n");
923 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
924 old_dev_tbl_cpy[devid] = old_devtb[devid];
925 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
926 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
928 if (dte_v && dom_id) {
929 old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
930 old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
931 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
932 /* If gcr3 table existed, mask it out */
933 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
934 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
935 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
936 old_dev_tbl_cpy[devid].data[1] &= ~tmp;
937 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
939 old_dev_tbl_cpy[devid].data[0] &= ~tmp;
943 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
944 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
945 int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
946 if (irq_v && (int_ctl || int_tab_len)) {
947 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
948 (int_tab_len != DTE_IRQ_TABLE_LEN)) {
949 pr_err("Wrong old irq remapping flag: %#x\n", devid);
953 old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
961 void amd_iommu_apply_erratum_63(u16 devid)
965 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
966 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
969 set_dev_entry_bit(devid, DEV_ENTRY_IW);
972 /* Writes the specific IOMMU for a device into the rlookup table */
973 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
975 amd_iommu_rlookup_table[devid] = iommu;
979 * This function takes the device specific flags read from the ACPI
980 * table and sets up the device table entry with that information
982 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
983 u16 devid, u32 flags, u32 ext_flags)
985 if (flags & ACPI_DEVFLAG_INITPASS)
986 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
987 if (flags & ACPI_DEVFLAG_EXTINT)
988 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
989 if (flags & ACPI_DEVFLAG_NMI)
990 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
991 if (flags & ACPI_DEVFLAG_SYSMGT1)
992 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
993 if (flags & ACPI_DEVFLAG_SYSMGT2)
994 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
995 if (flags & ACPI_DEVFLAG_LINT0)
996 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
997 if (flags & ACPI_DEVFLAG_LINT1)
998 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
1000 amd_iommu_apply_erratum_63(devid);
1002 set_iommu_for_device(iommu, devid);
1005 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1007 struct devid_map *entry;
1008 struct list_head *list;
1010 if (type == IVHD_SPECIAL_IOAPIC)
1012 else if (type == IVHD_SPECIAL_HPET)
1017 list_for_each_entry(entry, list, list) {
1018 if (!(entry->id == id && entry->cmd_line))
1021 pr_info("Command-line override present for %s id %d - ignoring\n",
1022 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1024 *devid = entry->devid;
1029 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1034 entry->devid = *devid;
1035 entry->cmd_line = cmd_line;
1037 list_add_tail(&entry->list, list);
1042 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1045 struct acpihid_map_entry *entry;
1046 struct list_head *list = &acpihid_map;
1048 list_for_each_entry(entry, list, list) {
1049 if (strcmp(entry->hid, hid) ||
1050 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1054 pr_info("Command-line override for hid:%s uid:%s\n",
1056 *devid = entry->devid;
1060 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1064 memcpy(entry->uid, uid, strlen(uid));
1065 memcpy(entry->hid, hid, strlen(hid));
1066 entry->devid = *devid;
1067 entry->cmd_line = cmd_line;
1068 entry->root_devid = (entry->devid & (~0x7));
1070 pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1071 entry->cmd_line ? "cmd" : "ivrs",
1072 entry->hid, entry->uid, entry->root_devid);
1074 list_add_tail(&entry->list, list);
1078 static int __init add_early_maps(void)
1082 for (i = 0; i < early_ioapic_map_size; ++i) {
1083 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1084 early_ioapic_map[i].id,
1085 &early_ioapic_map[i].devid,
1086 early_ioapic_map[i].cmd_line);
1091 for (i = 0; i < early_hpet_map_size; ++i) {
1092 ret = add_special_device(IVHD_SPECIAL_HPET,
1093 early_hpet_map[i].id,
1094 &early_hpet_map[i].devid,
1095 early_hpet_map[i].cmd_line);
1100 for (i = 0; i < early_acpihid_map_size; ++i) {
1101 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1102 early_acpihid_map[i].uid,
1103 &early_acpihid_map[i].devid,
1104 early_acpihid_map[i].cmd_line);
1113 * Reads the device exclusion range from ACPI and initializes the IOMMU with
1116 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
1118 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
1120 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
1125 * We only can configure exclusion ranges per IOMMU, not
1126 * per device. But we can enable the exclusion range per
1127 * device. This is done here
1129 set_dev_entry_bit(devid, DEV_ENTRY_EX);
1130 iommu->exclusion_start = m->range_start;
1131 iommu->exclusion_length = m->range_length;
1136 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1137 * initializes the hardware and our data structures with it.
1139 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1140 struct ivhd_header *h)
1143 u8 *end = p, flags = 0;
1144 u16 devid = 0, devid_start = 0, devid_to = 0;
1145 u32 dev_i, ext_flags = 0;
1147 struct ivhd_entry *e;
1152 ret = add_early_maps();
1157 * First save the recommended feature enable bits from ACPI
1159 iommu->acpi_flags = h->flags;
1162 * Done. Now parse the device entries
1164 ivhd_size = get_ivhd_header_size(h);
1166 pr_err("Unsupported IVHD type %#x\n", h->type);
1176 e = (struct ivhd_entry *)p;
1180 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1182 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1183 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1185 case IVHD_DEV_SELECT:
1187 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1189 PCI_BUS_NUM(e->devid),
1195 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1197 case IVHD_DEV_SELECT_RANGE_START:
1199 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1200 "devid: %02x:%02x.%x flags: %02x\n",
1201 PCI_BUS_NUM(e->devid),
1206 devid_start = e->devid;
1211 case IVHD_DEV_ALIAS:
1213 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1214 "flags: %02x devid_to: %02x:%02x.%x\n",
1215 PCI_BUS_NUM(e->devid),
1219 PCI_BUS_NUM(e->ext >> 8),
1220 PCI_SLOT(e->ext >> 8),
1221 PCI_FUNC(e->ext >> 8));
1224 devid_to = e->ext >> 8;
1225 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1226 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1227 amd_iommu_alias_table[devid] = devid_to;
1229 case IVHD_DEV_ALIAS_RANGE:
1231 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1232 "devid: %02x:%02x.%x flags: %02x "
1233 "devid_to: %02x:%02x.%x\n",
1234 PCI_BUS_NUM(e->devid),
1238 PCI_BUS_NUM(e->ext >> 8),
1239 PCI_SLOT(e->ext >> 8),
1240 PCI_FUNC(e->ext >> 8));
1242 devid_start = e->devid;
1244 devid_to = e->ext >> 8;
1248 case IVHD_DEV_EXT_SELECT:
1250 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1251 "flags: %02x ext: %08x\n",
1252 PCI_BUS_NUM(e->devid),
1258 set_dev_entry_from_acpi(iommu, devid, e->flags,
1261 case IVHD_DEV_EXT_SELECT_RANGE:
1263 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1264 "%02x:%02x.%x flags: %02x ext: %08x\n",
1265 PCI_BUS_NUM(e->devid),
1270 devid_start = e->devid;
1275 case IVHD_DEV_RANGE_END:
1277 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1278 PCI_BUS_NUM(e->devid),
1280 PCI_FUNC(e->devid));
1283 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1285 amd_iommu_alias_table[dev_i] = devid_to;
1286 set_dev_entry_from_acpi(iommu,
1287 devid_to, flags, ext_flags);
1289 set_dev_entry_from_acpi(iommu, dev_i,
1293 case IVHD_DEV_SPECIAL: {
1299 handle = e->ext & 0xff;
1300 devid = (e->ext >> 8) & 0xffff;
1301 type = (e->ext >> 24) & 0xff;
1303 if (type == IVHD_SPECIAL_IOAPIC)
1305 else if (type == IVHD_SPECIAL_HPET)
1310 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1316 ret = add_special_device(type, handle, &devid, false);
1321 * add_special_device might update the devid in case a
1322 * command-line override is present. So call
1323 * set_dev_entry_from_acpi after add_special_device.
1325 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1329 case IVHD_DEV_ACPI_HID: {
1331 u8 hid[ACPIHID_HID_LEN] = {0};
1332 u8 uid[ACPIHID_UID_LEN] = {0};
1335 if (h->type != 0x40) {
1336 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1341 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1342 hid[ACPIHID_HID_LEN - 1] = '\0';
1345 pr_err(FW_BUG "Invalid HID.\n");
1350 case UID_NOT_PRESENT:
1353 pr_warn(FW_BUG "Invalid UID length.\n");
1356 case UID_IS_INTEGER:
1358 sprintf(uid, "%d", e->uid);
1361 case UID_IS_CHARACTER:
1363 memcpy(uid, (u8 *)(&e->uid), ACPIHID_UID_LEN - 1);
1364 uid[ACPIHID_UID_LEN - 1] = '\0';
1372 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1380 ret = add_acpi_hid_device(hid, uid, &devid, false);
1385 * add_special_device might update the devid in case a
1386 * command-line override is present. So call
1387 * set_dev_entry_from_acpi after add_special_device.
1389 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1397 p += ivhd_entry_length(p);
1403 static void __init free_iommu_one(struct amd_iommu *iommu)
1405 free_command_buffer(iommu);
1406 free_event_buffer(iommu);
1407 free_ppr_log(iommu);
1409 iommu_unmap_mmio_space(iommu);
1412 static void __init free_iommu_all(void)
1414 struct amd_iommu *iommu, *next;
1416 for_each_iommu_safe(iommu, next) {
1417 list_del(&iommu->list);
1418 free_iommu_one(iommu);
1424 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1426 * BIOS should disable L2B micellaneous clock gating by setting
1427 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1429 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1433 if ((boot_cpu_data.x86 != 0x15) ||
1434 (boot_cpu_data.x86_model < 0x10) ||
1435 (boot_cpu_data.x86_model > 0x1f))
1438 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1439 pci_read_config_dword(iommu->dev, 0xf4, &value);
1444 /* Select NB indirect register 0x90 and enable writing */
1445 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1447 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1448 pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1450 /* Clear the enable writing bit */
1451 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1455 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1457 * BIOS should enable ATS write permission check by setting
1458 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1460 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1464 if ((boot_cpu_data.x86 != 0x15) ||
1465 (boot_cpu_data.x86_model < 0x30) ||
1466 (boot_cpu_data.x86_model > 0x3f))
1469 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1470 value = iommu_read_l2(iommu, 0x47);
1475 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1476 iommu_write_l2(iommu, 0x47, value | BIT(0));
1478 pci_info(iommu->dev, "Applying ATS write check workaround\n");
1482 * This function clues the initialization function for one IOMMU
1483 * together and also allocates the command buffer and programs the
1484 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1486 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1490 raw_spin_lock_init(&iommu->lock);
1492 /* Add IOMMU to internal data structures */
1493 list_add_tail(&iommu->list, &amd_iommu_list);
1494 iommu->index = amd_iommus_present++;
1496 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1497 WARN(1, "System has more IOMMUs than supported by this driver\n");
1501 /* Index is fine - add IOMMU to the array */
1502 amd_iommus[iommu->index] = iommu;
1505 * Copy data from ACPI table entry to the iommu struct
1507 iommu->devid = h->devid;
1508 iommu->cap_ptr = h->cap_ptr;
1509 iommu->pci_seg = h->pci_seg;
1510 iommu->mmio_phys = h->mmio_phys;
1514 /* Check if IVHD EFR contains proper max banks/counters */
1515 if ((h->efr_attr != 0) &&
1516 ((h->efr_attr & (0xF << 13)) != 0) &&
1517 ((h->efr_attr & (0x3F << 17)) != 0))
1518 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1520 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1521 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1522 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1523 if (((h->efr_attr & (0x1 << IOMMU_FEAT_XTSUP_SHIFT)) == 0))
1524 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1528 if (h->efr_reg & (1 << 9))
1529 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1531 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1532 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1533 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1534 if (((h->efr_reg & (0x1 << IOMMU_EFR_XTSUP_SHIFT)) == 0))
1535 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1541 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1542 iommu->mmio_phys_end);
1543 if (!iommu->mmio_base)
1546 if (alloc_command_buffer(iommu))
1549 if (alloc_event_buffer(iommu))
1552 iommu->int_enabled = false;
1554 init_translation_status(iommu);
1555 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1556 iommu_disable(iommu);
1557 clear_translation_pre_enabled(iommu);
1558 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1561 if (amd_iommu_pre_enabled)
1562 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1564 ret = init_iommu_from_acpi(iommu, h);
1568 ret = amd_iommu_create_irq_domain(iommu);
1573 * Make sure IOMMU is not considered to translate itself. The IVRS
1574 * table tells us so, but this is a lie!
1576 amd_iommu_rlookup_table[iommu->devid] = NULL;
1582 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1583 * @ivrs Pointer to the IVRS header
1585 * This function search through all IVDB of the maximum supported IVHD
1587 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1589 u8 *base = (u8 *)ivrs;
1590 struct ivhd_header *ivhd = (struct ivhd_header *)
1591 (base + IVRS_HEADER_LENGTH);
1592 u8 last_type = ivhd->type;
1593 u16 devid = ivhd->devid;
1595 while (((u8 *)ivhd - base < ivrs->length) &&
1596 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1597 u8 *p = (u8 *) ivhd;
1599 if (ivhd->devid == devid)
1600 last_type = ivhd->type;
1601 ivhd = (struct ivhd_header *)(p + ivhd->length);
1608 * Iterates over all IOMMU entries in the ACPI table, allocates the
1609 * IOMMU structure and initializes it with init_iommu_one()
1611 static int __init init_iommu_all(struct acpi_table_header *table)
1613 u8 *p = (u8 *)table, *end = (u8 *)table;
1614 struct ivhd_header *h;
1615 struct amd_iommu *iommu;
1618 end += table->length;
1619 p += IVRS_HEADER_LENGTH;
1622 h = (struct ivhd_header *)p;
1623 if (*p == amd_iommu_target_ivhd_type) {
1625 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1626 "seg: %d flags: %01x info %04x\n",
1627 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1628 PCI_FUNC(h->devid), h->cap_ptr,
1629 h->pci_seg, h->flags, h->info);
1630 DUMP_printk(" mmio-addr: %016llx\n",
1633 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1637 ret = init_iommu_one(iommu, h);
1649 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1650 u8 fxn, u64 *value, bool is_write);
1652 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1654 struct pci_dev *pdev = iommu->dev;
1655 u64 val = 0xabcd, val2 = 0;
1657 if (!iommu_feature(iommu, FEATURE_PC))
1660 amd_iommu_pc_present = true;
1662 /* Check if the performance counters can be written to */
1663 if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1664 (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1666 pci_err(pdev, "Unable to write to IOMMU perf counter.\n");
1667 amd_iommu_pc_present = false;
1671 pci_info(pdev, "IOMMU performance counters supported\n");
1673 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1674 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1675 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1678 static ssize_t amd_iommu_show_cap(struct device *dev,
1679 struct device_attribute *attr,
1682 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1683 return sprintf(buf, "%x\n", iommu->cap);
1685 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1687 static ssize_t amd_iommu_show_features(struct device *dev,
1688 struct device_attribute *attr,
1691 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1692 return sprintf(buf, "%llx\n", iommu->features);
1694 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1696 static struct attribute *amd_iommu_attrs[] = {
1698 &dev_attr_features.attr,
1702 static struct attribute_group amd_iommu_group = {
1703 .name = "amd-iommu",
1704 .attrs = amd_iommu_attrs,
1707 static const struct attribute_group *amd_iommu_groups[] = {
1712 static int __init iommu_init_pci(struct amd_iommu *iommu)
1714 int cap_ptr = iommu->cap_ptr;
1715 u32 range, misc, low, high;
1718 iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1719 iommu->devid & 0xff);
1723 /* Prevent binding other PCI device drivers to IOMMU devices */
1724 iommu->dev->match_driver = false;
1726 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1728 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1730 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1733 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1734 amd_iommu_iotlb_sup = false;
1736 /* read extended feature bits */
1737 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1738 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1740 iommu->features = ((u64)high << 32) | low;
1742 if (iommu_feature(iommu, FEATURE_GT)) {
1747 pasmax = iommu->features & FEATURE_PASID_MASK;
1748 pasmax >>= FEATURE_PASID_SHIFT;
1749 max_pasid = (1 << (pasmax + 1)) - 1;
1751 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1753 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1755 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1756 glxval >>= FEATURE_GLXVAL_SHIFT;
1758 if (amd_iommu_max_glx_val == -1)
1759 amd_iommu_max_glx_val = glxval;
1761 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1764 if (iommu_feature(iommu, FEATURE_GT) &&
1765 iommu_feature(iommu, FEATURE_PPR)) {
1766 iommu->is_iommu_v2 = true;
1767 amd_iommu_v2_present = true;
1770 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1773 ret = iommu_init_ga(iommu);
1777 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1778 amd_iommu_np_cache = true;
1780 init_iommu_perf_ctr(iommu);
1782 if (is_rd890_iommu(iommu->dev)) {
1786 pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
1790 * Some rd890 systems may not be fully reconfigured by the
1791 * BIOS, so it's necessary for us to store this information so
1792 * it can be reprogrammed on resume
1794 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1795 &iommu->stored_addr_lo);
1796 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1797 &iommu->stored_addr_hi);
1799 /* Low bit locks writes to configuration space */
1800 iommu->stored_addr_lo &= ~1;
1802 for (i = 0; i < 6; i++)
1803 for (j = 0; j < 0x12; j++)
1804 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1806 for (i = 0; i < 0x83; i++)
1807 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1810 amd_iommu_erratum_746_workaround(iommu);
1811 amd_iommu_ats_write_check_workaround(iommu);
1813 iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1814 amd_iommu_groups, "ivhd%d", iommu->index);
1815 iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1816 iommu_device_register(&iommu->iommu);
1818 return pci_enable_device(iommu->dev);
1821 static void print_iommu_info(void)
1823 static const char * const feat_str[] = {
1824 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1825 "IA", "GA", "HE", "PC"
1827 struct amd_iommu *iommu;
1829 for_each_iommu(iommu) {
1830 struct pci_dev *pdev = iommu->dev;
1833 pci_info(pdev, "Found IOMMU cap 0x%hx\n", iommu->cap_ptr);
1835 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1836 pci_info(pdev, "Extended features (%#llx):\n",
1838 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1839 if (iommu_feature(iommu, (1ULL << i)))
1840 pr_cont(" %s", feat_str[i]);
1843 if (iommu->features & FEATURE_GAM_VAPIC)
1844 pr_cont(" GA_vAPIC");
1849 if (irq_remapping_enabled) {
1850 pr_info("Interrupt remapping enabled\n");
1851 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1852 pr_info("Virtual APIC enabled\n");
1853 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1854 pr_info("X2APIC enabled\n");
1858 static int __init amd_iommu_init_pci(void)
1860 struct amd_iommu *iommu;
1863 for_each_iommu(iommu) {
1864 ret = iommu_init_pci(iommu);
1870 * Order is important here to make sure any unity map requirements are
1871 * fulfilled. The unity mappings are created and written to the device
1872 * table during the amd_iommu_init_api() call.
1874 * After that we call init_device_table_dma() to make sure any
1875 * uninitialized DTE will block DMA, and in the end we flush the caches
1876 * of all IOMMUs to make sure the changes to the device table are
1879 ret = amd_iommu_init_api();
1881 init_device_table_dma();
1883 for_each_iommu(iommu)
1884 iommu_flush_all_caches(iommu);
1892 /****************************************************************************
1894 * The following functions initialize the MSI interrupts for all IOMMUs
1895 * in the system. It's a bit challenging because there could be multiple
1896 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1899 ****************************************************************************/
1901 static int iommu_setup_msi(struct amd_iommu *iommu)
1905 r = pci_enable_msi(iommu->dev);
1909 r = request_threaded_irq(iommu->dev->irq,
1910 amd_iommu_int_handler,
1911 amd_iommu_int_thread,
1916 pci_disable_msi(iommu->dev);
1920 iommu->int_enabled = true;
1925 #define XT_INT_DEST_MODE(x) (((x) & 0x1ULL) << 2)
1926 #define XT_INT_DEST_LO(x) (((x) & 0xFFFFFFULL) << 8)
1927 #define XT_INT_VEC(x) (((x) & 0xFFULL) << 32)
1928 #define XT_INT_DEST_HI(x) ((((x) >> 24) & 0xFFULL) << 56)
1931 * Setup the IntCapXT registers with interrupt routing information
1932 * based on the PCI MSI capability block registers, accessed via
1933 * MMIO MSI address low/hi and MSI data registers.
1935 static void iommu_update_intcapxt(struct amd_iommu *iommu)
1938 u32 addr_lo = readl(iommu->mmio_base + MMIO_MSI_ADDR_LO_OFFSET);
1939 u32 addr_hi = readl(iommu->mmio_base + MMIO_MSI_ADDR_HI_OFFSET);
1940 u32 data = readl(iommu->mmio_base + MMIO_MSI_DATA_OFFSET);
1941 bool dm = (addr_lo >> MSI_ADDR_DEST_MODE_SHIFT) & 0x1;
1942 u32 dest = ((addr_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xFF);
1944 if (x2apic_enabled())
1945 dest |= MSI_ADDR_EXT_DEST_ID(addr_hi);
1947 val = XT_INT_VEC(data & 0xFF) |
1948 XT_INT_DEST_MODE(dm) |
1949 XT_INT_DEST_LO(dest) |
1950 XT_INT_DEST_HI(dest);
1953 * Current IOMMU implemtation uses the same IRQ for all
1954 * 3 IOMMU interrupts.
1956 writeq(val, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
1957 writeq(val, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
1958 writeq(val, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
1961 static void _irq_notifier_notify(struct irq_affinity_notify *notify,
1962 const cpumask_t *mask)
1964 struct amd_iommu *iommu;
1966 for_each_iommu(iommu) {
1967 if (iommu->dev->irq == notify->irq) {
1968 iommu_update_intcapxt(iommu);
1974 static void _irq_notifier_release(struct kref *ref)
1978 static int iommu_init_intcapxt(struct amd_iommu *iommu)
1981 struct irq_affinity_notify *notify = &iommu->intcapxt_notify;
1984 * IntCapXT requires XTSup=1, which can be inferred
1985 * amd_iommu_xt_mode.
1987 if (amd_iommu_xt_mode != IRQ_REMAP_X2APIC_MODE)
1991 * Also, we need to setup notifier to update the IntCapXT registers
1992 * whenever the irq affinity is changed from user-space.
1994 notify->irq = iommu->dev->irq;
1995 notify->notify = _irq_notifier_notify,
1996 notify->release = _irq_notifier_release,
1997 ret = irq_set_affinity_notifier(iommu->dev->irq, notify);
1999 pr_err("Failed to register irq affinity notifier (devid=%#x, irq %d)\n",
2000 iommu->devid, iommu->dev->irq);
2004 iommu_update_intcapxt(iommu);
2005 iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2009 static int iommu_init_msi(struct amd_iommu *iommu)
2013 if (iommu->int_enabled)
2016 if (iommu->dev->msi_cap)
2017 ret = iommu_setup_msi(iommu);
2025 ret = iommu_init_intcapxt(iommu);
2029 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2031 if (iommu->ppr_log != NULL)
2032 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
2034 iommu_ga_log_enable(iommu);
2039 /****************************************************************************
2041 * The next functions belong to the third pass of parsing the ACPI
2042 * table. In this last pass the memory mapping requirements are
2043 * gathered (like exclusion and unity mapping ranges).
2045 ****************************************************************************/
2047 static void __init free_unity_maps(void)
2049 struct unity_map_entry *entry, *next;
2051 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
2052 list_del(&entry->list);
2057 /* called when we find an exclusion range definition in ACPI */
2058 static int __init init_exclusion_range(struct ivmd_header *m)
2063 case ACPI_IVMD_TYPE:
2064 set_device_exclusion_range(m->devid, m);
2066 case ACPI_IVMD_TYPE_ALL:
2067 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2068 set_device_exclusion_range(i, m);
2070 case ACPI_IVMD_TYPE_RANGE:
2071 for (i = m->devid; i <= m->aux; ++i)
2072 set_device_exclusion_range(i, m);
2081 /* called for unity map ACPI definition */
2082 static int __init init_unity_map_range(struct ivmd_header *m)
2084 struct unity_map_entry *e = NULL;
2087 e = kzalloc(sizeof(*e), GFP_KERNEL);
2091 if (m->flags & IVMD_FLAG_EXCL_RANGE)
2092 init_exclusion_range(m);
2098 case ACPI_IVMD_TYPE:
2099 s = "IVMD_TYPEi\t\t\t";
2100 e->devid_start = e->devid_end = m->devid;
2102 case ACPI_IVMD_TYPE_ALL:
2103 s = "IVMD_TYPE_ALL\t\t";
2105 e->devid_end = amd_iommu_last_bdf;
2107 case ACPI_IVMD_TYPE_RANGE:
2108 s = "IVMD_TYPE_RANGE\t\t";
2109 e->devid_start = m->devid;
2110 e->devid_end = m->aux;
2113 e->address_start = PAGE_ALIGN(m->range_start);
2114 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2115 e->prot = m->flags >> 1;
2117 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2118 " range_start: %016llx range_end: %016llx flags: %x\n", s,
2119 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2120 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
2121 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2122 e->address_start, e->address_end, m->flags);
2124 list_add_tail(&e->list, &amd_iommu_unity_map);
2129 /* iterates over all memory definitions we find in the ACPI table */
2130 static int __init init_memory_definitions(struct acpi_table_header *table)
2132 u8 *p = (u8 *)table, *end = (u8 *)table;
2133 struct ivmd_header *m;
2135 end += table->length;
2136 p += IVRS_HEADER_LENGTH;
2139 m = (struct ivmd_header *)p;
2140 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2141 init_unity_map_range(m);
2150 * Init the device table to not allow DMA access for devices
2152 static void init_device_table_dma(void)
2156 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2157 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2158 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2162 static void __init uninit_device_table_dma(void)
2166 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2167 amd_iommu_dev_table[devid].data[0] = 0ULL;
2168 amd_iommu_dev_table[devid].data[1] = 0ULL;
2172 static void init_device_table(void)
2176 if (!amd_iommu_irq_remap)
2179 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2180 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2183 static void iommu_init_flags(struct amd_iommu *iommu)
2185 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2186 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2187 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2189 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2190 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2191 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2193 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2194 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2195 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2197 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2198 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2199 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2202 * make IOMMU memory accesses cache coherent
2204 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2206 /* Set IOTLB invalidation timeout to 1s */
2207 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2210 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2213 u32 ioc_feature_control;
2214 struct pci_dev *pdev = iommu->root_pdev;
2216 /* RD890 BIOSes may not have completely reconfigured the iommu */
2217 if (!is_rd890_iommu(iommu->dev) || !pdev)
2221 * First, we need to ensure that the iommu is enabled. This is
2222 * controlled by a register in the northbridge
2225 /* Select Northbridge indirect register 0x75 and enable writing */
2226 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2227 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2229 /* Enable the iommu */
2230 if (!(ioc_feature_control & 0x1))
2231 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2233 /* Restore the iommu BAR */
2234 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2235 iommu->stored_addr_lo);
2236 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2237 iommu->stored_addr_hi);
2239 /* Restore the l1 indirect regs for each of the 6 l1s */
2240 for (i = 0; i < 6; i++)
2241 for (j = 0; j < 0x12; j++)
2242 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2244 /* Restore the l2 indirect regs */
2245 for (i = 0; i < 0x83; i++)
2246 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2248 /* Lock PCI setup registers */
2249 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2250 iommu->stored_addr_lo | 1);
2253 static void iommu_enable_ga(struct amd_iommu *iommu)
2255 #ifdef CONFIG_IRQ_REMAP
2256 switch (amd_iommu_guest_ir) {
2257 case AMD_IOMMU_GUEST_IR_VAPIC:
2258 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2260 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2261 iommu_feature_enable(iommu, CONTROL_GA_EN);
2262 iommu->irte_ops = &irte_128_ops;
2265 iommu->irte_ops = &irte_32_ops;
2271 static void early_enable_iommu(struct amd_iommu *iommu)
2273 iommu_disable(iommu);
2274 iommu_init_flags(iommu);
2275 iommu_set_device_table(iommu);
2276 iommu_enable_command_buffer(iommu);
2277 iommu_enable_event_buffer(iommu);
2278 iommu_set_exclusion_range(iommu);
2279 iommu_enable_ga(iommu);
2280 iommu_enable_xt(iommu);
2281 iommu_enable(iommu);
2282 iommu_flush_all_caches(iommu);
2286 * This function finally enables all IOMMUs found in the system after
2287 * they have been initialized.
2289 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2290 * the old content of device table entries. Not this case or copy failed,
2291 * just continue as normal kernel does.
2293 static void early_enable_iommus(void)
2295 struct amd_iommu *iommu;
2298 if (!copy_device_table()) {
2300 * If come here because of failure in copying device table from old
2301 * kernel with all IOMMUs enabled, print error message and try to
2302 * free allocated old_dev_tbl_cpy.
2304 if (amd_iommu_pre_enabled)
2305 pr_err("Failed to copy DEV table from previous kernel.\n");
2306 if (old_dev_tbl_cpy != NULL)
2307 free_pages((unsigned long)old_dev_tbl_cpy,
2308 get_order(dev_table_size));
2310 for_each_iommu(iommu) {
2311 clear_translation_pre_enabled(iommu);
2312 early_enable_iommu(iommu);
2315 pr_info("Copied DEV table from previous kernel.\n");
2316 free_pages((unsigned long)amd_iommu_dev_table,
2317 get_order(dev_table_size));
2318 amd_iommu_dev_table = old_dev_tbl_cpy;
2319 for_each_iommu(iommu) {
2320 iommu_disable_command_buffer(iommu);
2321 iommu_disable_event_buffer(iommu);
2322 iommu_enable_command_buffer(iommu);
2323 iommu_enable_event_buffer(iommu);
2324 iommu_enable_ga(iommu);
2325 iommu_enable_xt(iommu);
2326 iommu_set_device_table(iommu);
2327 iommu_flush_all_caches(iommu);
2331 #ifdef CONFIG_IRQ_REMAP
2332 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2333 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2337 static void enable_iommus_v2(void)
2339 struct amd_iommu *iommu;
2341 for_each_iommu(iommu) {
2342 iommu_enable_ppr_log(iommu);
2343 iommu_enable_gt(iommu);
2347 static void enable_iommus(void)
2349 early_enable_iommus();
2354 static void disable_iommus(void)
2356 struct amd_iommu *iommu;
2358 for_each_iommu(iommu)
2359 iommu_disable(iommu);
2361 #ifdef CONFIG_IRQ_REMAP
2362 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2363 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2368 * Suspend/Resume support
2369 * disable suspend until real resume implemented
2372 static void amd_iommu_resume(void)
2374 struct amd_iommu *iommu;
2376 for_each_iommu(iommu)
2377 iommu_apply_resume_quirks(iommu);
2379 /* re-load the hardware */
2382 amd_iommu_enable_interrupts();
2385 static int amd_iommu_suspend(void)
2387 /* disable IOMMUs to go out of the way for BIOS */
2393 static struct syscore_ops amd_iommu_syscore_ops = {
2394 .suspend = amd_iommu_suspend,
2395 .resume = amd_iommu_resume,
2398 static void __init free_iommu_resources(void)
2400 kmemleak_free(irq_lookup_table);
2401 free_pages((unsigned long)irq_lookup_table,
2402 get_order(rlookup_table_size));
2403 irq_lookup_table = NULL;
2405 kmem_cache_destroy(amd_iommu_irq_cache);
2406 amd_iommu_irq_cache = NULL;
2408 free_pages((unsigned long)amd_iommu_rlookup_table,
2409 get_order(rlookup_table_size));
2410 amd_iommu_rlookup_table = NULL;
2412 free_pages((unsigned long)amd_iommu_alias_table,
2413 get_order(alias_table_size));
2414 amd_iommu_alias_table = NULL;
2416 free_pages((unsigned long)amd_iommu_dev_table,
2417 get_order(dev_table_size));
2418 amd_iommu_dev_table = NULL;
2423 /* SB IOAPIC is always on this device in AMD systems */
2424 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2426 static bool __init check_ioapic_information(void)
2428 const char *fw_bug = FW_BUG;
2429 bool ret, has_sb_ioapic;
2432 has_sb_ioapic = false;
2436 * If we have map overrides on the kernel command line the
2437 * messages in this function might not describe firmware bugs
2438 * anymore - so be careful
2443 for (idx = 0; idx < nr_ioapics; idx++) {
2444 int devid, id = mpc_ioapic_id(idx);
2446 devid = get_ioapic_devid(id);
2448 pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2451 } else if (devid == IOAPIC_SB_DEVID) {
2452 has_sb_ioapic = true;
2457 if (!has_sb_ioapic) {
2459 * We expect the SB IOAPIC to be listed in the IVRS
2460 * table. The system timer is connected to the SB IOAPIC
2461 * and if we don't have it in the list the system will
2462 * panic at boot time. This situation usually happens
2463 * when the BIOS is buggy and provides us the wrong
2464 * device id for the IOAPIC in the system.
2466 pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2470 pr_err("Disabling interrupt remapping\n");
2475 static void __init free_dma_resources(void)
2477 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2478 get_order(MAX_DOMAIN_ID/8));
2479 amd_iommu_pd_alloc_bitmap = NULL;
2485 * This is the hardware init function for AMD IOMMU in the system.
2486 * This function is called either from amd_iommu_init or from the interrupt
2487 * remapping setup code.
2489 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2492 * 1 pass) Discover the most comprehensive IVHD type to use.
2494 * 2 pass) Find the highest PCI device id the driver has to handle.
2495 * Upon this information the size of the data structures is
2496 * determined that needs to be allocated.
2498 * 3 pass) Initialize the data structures just allocated with the
2499 * information in the ACPI table about available AMD IOMMUs
2500 * in the system. It also maps the PCI devices in the
2501 * system to specific IOMMUs
2503 * 4 pass) After the basic data structures are allocated and
2504 * initialized we update them with information about memory
2505 * remapping requirements parsed out of the ACPI table in
2508 * After everything is set up the IOMMUs are enabled and the necessary
2509 * hotplug and suspend notifiers are registered.
2511 static int __init early_amd_iommu_init(void)
2513 struct acpi_table_header *ivrs_base;
2515 int i, remap_cache_sz, ret = 0;
2517 if (!amd_iommu_detected)
2520 status = acpi_get_table("IVRS", 0, &ivrs_base);
2521 if (status == AE_NOT_FOUND)
2523 else if (ACPI_FAILURE(status)) {
2524 const char *err = acpi_format_exception(status);
2525 pr_err("IVRS table error: %s\n", err);
2530 * Validate checksum here so we don't need to do it when
2531 * we actually parse the table
2533 ret = check_ivrs_checksum(ivrs_base);
2537 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2538 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2541 * First parse ACPI tables to find the largest Bus/Dev/Func
2542 * we need to handle. Upon this information the shared data
2543 * structures for the IOMMUs in the system will be allocated
2545 ret = find_last_devid_acpi(ivrs_base);
2549 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2550 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2551 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2553 /* Device table - directly used by all IOMMUs */
2555 amd_iommu_dev_table = (void *)__get_free_pages(
2556 GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2557 get_order(dev_table_size));
2558 if (amd_iommu_dev_table == NULL)
2562 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2563 * IOMMU see for that device
2565 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2566 get_order(alias_table_size));
2567 if (amd_iommu_alias_table == NULL)
2570 /* IOMMU rlookup table - find the IOMMU for a specific device */
2571 amd_iommu_rlookup_table = (void *)__get_free_pages(
2572 GFP_KERNEL | __GFP_ZERO,
2573 get_order(rlookup_table_size));
2574 if (amd_iommu_rlookup_table == NULL)
2577 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2578 GFP_KERNEL | __GFP_ZERO,
2579 get_order(MAX_DOMAIN_ID/8));
2580 if (amd_iommu_pd_alloc_bitmap == NULL)
2584 * let all alias entries point to itself
2586 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2587 amd_iommu_alias_table[i] = i;
2590 * never allocate domain 0 because its used as the non-allocated and
2591 * error value placeholder
2593 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2596 * now the data structures are allocated and basically initialized
2597 * start the real acpi table scan
2599 ret = init_iommu_all(ivrs_base);
2603 /* Disable any previously enabled IOMMUs */
2604 if (!is_kdump_kernel() || amd_iommu_disabled)
2607 if (amd_iommu_irq_remap)
2608 amd_iommu_irq_remap = check_ioapic_information();
2610 if (amd_iommu_irq_remap) {
2612 * Interrupt remapping enabled, create kmem_cache for the
2616 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2617 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2619 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2620 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2622 IRQ_TABLE_ALIGNMENT,
2624 if (!amd_iommu_irq_cache)
2627 irq_lookup_table = (void *)__get_free_pages(
2628 GFP_KERNEL | __GFP_ZERO,
2629 get_order(rlookup_table_size));
2630 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2632 if (!irq_lookup_table)
2636 ret = init_memory_definitions(ivrs_base);
2640 /* init the device table */
2641 init_device_table();
2644 /* Don't leak any ACPI memory */
2645 acpi_put_table(ivrs_base);
2651 static int amd_iommu_enable_interrupts(void)
2653 struct amd_iommu *iommu;
2656 for_each_iommu(iommu) {
2657 ret = iommu_init_msi(iommu);
2666 static bool detect_ivrs(void)
2668 struct acpi_table_header *ivrs_base;
2671 status = acpi_get_table("IVRS", 0, &ivrs_base);
2672 if (status == AE_NOT_FOUND)
2674 else if (ACPI_FAILURE(status)) {
2675 const char *err = acpi_format_exception(status);
2676 pr_err("IVRS table error: %s\n", err);
2680 acpi_put_table(ivrs_base);
2682 /* Make sure ACS will be enabled during PCI probe */
2688 /****************************************************************************
2690 * AMD IOMMU Initialization State Machine
2692 ****************************************************************************/
2694 static int __init state_next(void)
2698 switch (init_state) {
2699 case IOMMU_START_STATE:
2700 if (!detect_ivrs()) {
2701 init_state = IOMMU_NOT_FOUND;
2704 init_state = IOMMU_IVRS_DETECTED;
2707 case IOMMU_IVRS_DETECTED:
2708 ret = early_amd_iommu_init();
2709 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2710 if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2711 pr_info("AMD IOMMU disabled on kernel command-line\n");
2712 init_state = IOMMU_CMDLINE_DISABLED;
2716 case IOMMU_ACPI_FINISHED:
2717 early_enable_iommus();
2718 x86_platform.iommu_shutdown = disable_iommus;
2719 init_state = IOMMU_ENABLED;
2722 register_syscore_ops(&amd_iommu_syscore_ops);
2723 ret = amd_iommu_init_pci();
2724 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2727 case IOMMU_PCI_INIT:
2728 ret = amd_iommu_enable_interrupts();
2729 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2731 case IOMMU_INTERRUPTS_EN:
2732 ret = amd_iommu_init_dma_ops();
2733 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2736 init_state = IOMMU_INITIALIZED;
2738 case IOMMU_INITIALIZED:
2741 case IOMMU_NOT_FOUND:
2742 case IOMMU_INIT_ERROR:
2743 case IOMMU_CMDLINE_DISABLED:
2744 /* Error states => do nothing */
2753 free_dma_resources();
2754 if (!irq_remapping_enabled) {
2756 free_iommu_resources();
2758 struct amd_iommu *iommu;
2760 uninit_device_table_dma();
2761 for_each_iommu(iommu)
2762 iommu_flush_all_caches(iommu);
2768 static int __init iommu_go_to_state(enum iommu_init_state state)
2772 while (init_state != state) {
2773 if (init_state == IOMMU_NOT_FOUND ||
2774 init_state == IOMMU_INIT_ERROR ||
2775 init_state == IOMMU_CMDLINE_DISABLED)
2783 #ifdef CONFIG_IRQ_REMAP
2784 int __init amd_iommu_prepare(void)
2788 amd_iommu_irq_remap = true;
2790 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2793 return amd_iommu_irq_remap ? 0 : -ENODEV;
2796 int __init amd_iommu_enable(void)
2800 ret = iommu_go_to_state(IOMMU_ENABLED);
2804 irq_remapping_enabled = 1;
2805 return amd_iommu_xt_mode;
2808 void amd_iommu_disable(void)
2810 amd_iommu_suspend();
2813 int amd_iommu_reenable(int mode)
2820 int __init amd_iommu_enable_faulting(void)
2822 /* We enable MSI later when PCI is initialized */
2828 * This is the core init function for AMD IOMMU hardware in the system.
2829 * This function is called from the generic x86 DMA layer initialization
2832 static int __init amd_iommu_init(void)
2834 struct amd_iommu *iommu;
2837 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2838 #ifdef CONFIG_GART_IOMMU
2839 if (ret && list_empty(&amd_iommu_list)) {
2841 * We failed to initialize the AMD IOMMU - try fallback
2842 * to GART if possible.
2848 for_each_iommu(iommu)
2849 amd_iommu_debugfs_setup(iommu);
2854 static bool amd_iommu_sme_check(void)
2856 if (!sme_active() || (boot_cpu_data.x86 != 0x17))
2859 /* For Fam17h, a specific level of support is required */
2860 if (boot_cpu_data.microcode >= 0x08001205)
2863 if ((boot_cpu_data.microcode >= 0x08001126) &&
2864 (boot_cpu_data.microcode <= 0x080011ff))
2867 pr_notice("IOMMU not currently supported when SME is active\n");
2872 /****************************************************************************
2874 * Early detect code. This code runs at IOMMU detection time in the DMA
2875 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2878 ****************************************************************************/
2879 int __init amd_iommu_detect(void)
2883 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2886 if (!amd_iommu_sme_check())
2889 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2893 amd_iommu_detected = true;
2895 x86_init.iommu.iommu_init = amd_iommu_init;
2900 /****************************************************************************
2902 * Parsing functions for the AMD IOMMU specific kernel command line
2905 ****************************************************************************/
2907 static int __init parse_amd_iommu_dump(char *str)
2909 amd_iommu_dump = true;
2914 static int __init parse_amd_iommu_intr(char *str)
2916 for (; *str; ++str) {
2917 if (strncmp(str, "legacy", 6) == 0) {
2918 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
2921 if (strncmp(str, "vapic", 5) == 0) {
2922 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2929 static int __init parse_amd_iommu_options(char *str)
2931 for (; *str; ++str) {
2932 if (strncmp(str, "fullflush", 9) == 0)
2933 amd_iommu_unmap_flush = true;
2934 if (strncmp(str, "off", 3) == 0)
2935 amd_iommu_disabled = true;
2936 if (strncmp(str, "force_isolation", 15) == 0)
2937 amd_iommu_force_isolation = true;
2943 static int __init parse_ivrs_ioapic(char *str)
2945 unsigned int bus, dev, fn;
2949 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2952 pr_err("Invalid command line: ivrs_ioapic%s\n", str);
2956 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2957 pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2962 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2964 cmdline_maps = true;
2965 i = early_ioapic_map_size++;
2966 early_ioapic_map[i].id = id;
2967 early_ioapic_map[i].devid = devid;
2968 early_ioapic_map[i].cmd_line = true;
2973 static int __init parse_ivrs_hpet(char *str)
2975 unsigned int bus, dev, fn;
2979 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2982 pr_err("Invalid command line: ivrs_hpet%s\n", str);
2986 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2987 pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
2992 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2994 cmdline_maps = true;
2995 i = early_hpet_map_size++;
2996 early_hpet_map[i].id = id;
2997 early_hpet_map[i].devid = devid;
2998 early_hpet_map[i].cmd_line = true;
3003 static int __init parse_ivrs_acpihid(char *str)
3006 char *hid, *uid, *p;
3007 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3010 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
3012 pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
3017 hid = strsep(&p, ":");
3020 if (!hid || !(*hid) || !uid) {
3021 pr_err("Invalid command line: hid or uid\n");
3025 i = early_acpihid_map_size++;
3026 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3027 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3028 early_acpihid_map[i].devid =
3029 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3030 early_acpihid_map[i].cmd_line = true;
3035 __setup("amd_iommu_dump", parse_amd_iommu_dump);
3036 __setup("amd_iommu=", parse_amd_iommu_options);
3037 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
3038 __setup("ivrs_ioapic", parse_ivrs_ioapic);
3039 __setup("ivrs_hpet", parse_ivrs_hpet);
3040 __setup("ivrs_acpihid", parse_ivrs_acpihid);
3042 IOMMU_INIT_FINISH(amd_iommu_detect,
3043 gart_iommu_hole_init,
3047 bool amd_iommu_v2_supported(void)
3049 return amd_iommu_v2_present;
3051 EXPORT_SYMBOL(amd_iommu_v2_supported);
3053 struct amd_iommu *get_amd_iommu(unsigned int idx)
3056 struct amd_iommu *iommu;
3058 for_each_iommu(iommu)
3063 EXPORT_SYMBOL(get_amd_iommu);
3065 /****************************************************************************
3067 * IOMMU EFR Performance Counter support functionality. This code allows
3068 * access to the IOMMU PC functionality.
3070 ****************************************************************************/
3072 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3074 struct amd_iommu *iommu = get_amd_iommu(idx);
3077 return iommu->max_banks;
3081 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3083 bool amd_iommu_pc_supported(void)
3085 return amd_iommu_pc_present;
3087 EXPORT_SYMBOL(amd_iommu_pc_supported);
3089 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3091 struct amd_iommu *iommu = get_amd_iommu(idx);
3094 return iommu->max_counters;
3098 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3100 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3101 u8 fxn, u64 *value, bool is_write)
3106 /* Make sure the IOMMU PC resource is available */
3107 if (!amd_iommu_pc_present)
3110 /* Check for valid iommu and pc register indexing */
3111 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3114 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3116 /* Limit the offset to the hw defined mmio region aperture */
3117 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3118 (iommu->max_counters << 8) | 0x28);
3119 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3120 (offset > max_offset_lim))
3124 u64 val = *value & GENMASK_ULL(47, 0);
3126 writel((u32)val, iommu->mmio_base + offset);
3127 writel((val >> 32), iommu->mmio_base + offset + 4);
3129 *value = readl(iommu->mmio_base + offset + 4);
3131 *value |= readl(iommu->mmio_base + offset);
3132 *value &= GENMASK_ULL(47, 0);
3138 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3143 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3145 EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3147 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3152 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3154 EXPORT_SYMBOL(amd_iommu_pc_set_reg);