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/kmemleak.h>
22 #include <linux/mem_encrypt.h>
23 #include <asm/pci-direct.h>
24 #include <asm/iommu.h>
26 #include <asm/msidef.h>
28 #include <asm/x86_init.h>
29 #include <asm/iommu_table.h>
30 #include <asm/io_apic.h>
31 #include <asm/irq_remapping.h>
33 #include <linux/crash_dump.h>
35 #include "amd_iommu.h"
36 #include "../irq_remapping.h"
39 * definitions for the ACPI scanning code
41 #define IVRS_HEADER_LENGTH 48
43 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
44 #define ACPI_IVMD_TYPE_ALL 0x20
45 #define ACPI_IVMD_TYPE 0x21
46 #define ACPI_IVMD_TYPE_RANGE 0x22
48 #define IVHD_DEV_ALL 0x01
49 #define IVHD_DEV_SELECT 0x02
50 #define IVHD_DEV_SELECT_RANGE_START 0x03
51 #define IVHD_DEV_RANGE_END 0x04
52 #define IVHD_DEV_ALIAS 0x42
53 #define IVHD_DEV_ALIAS_RANGE 0x43
54 #define IVHD_DEV_EXT_SELECT 0x46
55 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
56 #define IVHD_DEV_SPECIAL 0x48
57 #define IVHD_DEV_ACPI_HID 0xf0
59 #define UID_NOT_PRESENT 0
60 #define UID_IS_INTEGER 1
61 #define UID_IS_CHARACTER 2
63 #define IVHD_SPECIAL_IOAPIC 1
64 #define IVHD_SPECIAL_HPET 2
66 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
67 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
68 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
69 #define IVHD_FLAG_ISOC_EN_MASK 0x08
71 #define IVMD_FLAG_EXCL_RANGE 0x08
72 #define IVMD_FLAG_IW 0x04
73 #define IVMD_FLAG_IR 0x02
74 #define IVMD_FLAG_UNITY_MAP 0x01
76 #define ACPI_DEVFLAG_INITPASS 0x01
77 #define ACPI_DEVFLAG_EXTINT 0x02
78 #define ACPI_DEVFLAG_NMI 0x04
79 #define ACPI_DEVFLAG_SYSMGT1 0x10
80 #define ACPI_DEVFLAG_SYSMGT2 0x20
81 #define ACPI_DEVFLAG_LINT0 0x40
82 #define ACPI_DEVFLAG_LINT1 0x80
83 #define ACPI_DEVFLAG_ATSDIS 0x10000000
85 #define LOOP_TIMEOUT 100000
87 * ACPI table definitions
89 * These data structures are laid over the table to parse the important values
93 extern const struct iommu_ops amd_iommu_ops;
96 * structure describing one IOMMU in the ACPI table. Typically followed by one
97 * or more ivhd_entrys.
110 /* Following only valid on IVHD type 11h and 40h */
111 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
113 } __attribute__((packed));
116 * A device entry describing which devices a specific IOMMU translates and
117 * which requestor ids they use.
129 } __attribute__((packed));
132 * An AMD IOMMU memory definition structure. It defines things like exclusion
133 * ranges for devices and regions that should be unity mapped.
144 } __attribute__((packed));
147 bool amd_iommu_irq_remap __read_mostly;
149 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
150 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
152 static bool amd_iommu_detected;
153 static bool __initdata amd_iommu_disabled;
154 static int amd_iommu_target_ivhd_type;
156 u16 amd_iommu_last_bdf; /* largest PCI device id we have
158 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
160 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
162 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
165 /* Array to assign indices to IOMMUs*/
166 struct amd_iommu *amd_iommus[MAX_IOMMUS];
168 /* Number of IOMMUs present in the system */
169 static int amd_iommus_present;
171 /* IOMMUs have a non-present cache? */
172 bool amd_iommu_np_cache __read_mostly;
173 bool amd_iommu_iotlb_sup __read_mostly = true;
175 u32 amd_iommu_max_pasid __read_mostly = ~0;
177 bool amd_iommu_v2_present __read_mostly;
178 static bool amd_iommu_pc_present __read_mostly;
180 bool amd_iommu_force_isolation __read_mostly;
183 * Pointer to the device table which is shared by all AMD IOMMUs
184 * it is indexed by the PCI device id or the HT unit id and contains
185 * information about the domain the device belongs to as well as the
186 * page table root pointer.
188 struct dev_table_entry *amd_iommu_dev_table;
190 * Pointer to a device table which the content of old device table
191 * will be copied to. It's only be used in kdump kernel.
193 static struct dev_table_entry *old_dev_tbl_cpy;
196 * The alias table is a driver specific data structure which contains the
197 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
198 * More than one device can share the same requestor id.
200 u16 *amd_iommu_alias_table;
203 * The rlookup table is used to find the IOMMU which is responsible
204 * for a specific device. It is also indexed by the PCI device id.
206 struct amd_iommu **amd_iommu_rlookup_table;
207 EXPORT_SYMBOL(amd_iommu_rlookup_table);
210 * This table is used to find the irq remapping table for a given device id
213 struct irq_remap_table **irq_lookup_table;
216 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
217 * to know which ones are already in use.
219 unsigned long *amd_iommu_pd_alloc_bitmap;
221 static u32 dev_table_size; /* size of the device table */
222 static u32 alias_table_size; /* size of the alias table */
223 static u32 rlookup_table_size; /* size if the rlookup table */
225 enum iommu_init_state {
236 IOMMU_CMDLINE_DISABLED,
239 /* Early ioapic and hpet maps from kernel command line */
240 #define EARLY_MAP_SIZE 4
241 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
242 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
243 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
245 static int __initdata early_ioapic_map_size;
246 static int __initdata early_hpet_map_size;
247 static int __initdata early_acpihid_map_size;
249 static bool __initdata cmdline_maps;
251 static enum iommu_init_state init_state = IOMMU_START_STATE;
253 static int amd_iommu_enable_interrupts(void);
254 static int __init iommu_go_to_state(enum iommu_init_state state);
255 static void init_device_table_dma(void);
257 static bool amd_iommu_pre_enabled = true;
259 bool translation_pre_enabled(struct amd_iommu *iommu)
261 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
263 EXPORT_SYMBOL(translation_pre_enabled);
265 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
267 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
270 static void init_translation_status(struct amd_iommu *iommu)
274 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
275 if (ctrl & (1<<CONTROL_IOMMU_EN))
276 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
279 static inline void update_last_devid(u16 devid)
281 if (devid > amd_iommu_last_bdf)
282 amd_iommu_last_bdf = devid;
285 static inline unsigned long tbl_size(int entry_size)
287 unsigned shift = PAGE_SHIFT +
288 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
293 int amd_iommu_get_num_iommus(void)
295 return amd_iommus_present;
298 /* Access to l1 and l2 indexed register spaces */
300 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
304 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
305 pci_read_config_dword(iommu->dev, 0xfc, &val);
309 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
311 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
312 pci_write_config_dword(iommu->dev, 0xfc, val);
313 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
316 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
320 pci_write_config_dword(iommu->dev, 0xf0, address);
321 pci_read_config_dword(iommu->dev, 0xf4, &val);
325 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
327 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
328 pci_write_config_dword(iommu->dev, 0xf4, val);
331 /****************************************************************************
333 * AMD IOMMU MMIO register space handling functions
335 * These functions are used to program the IOMMU device registers in
336 * MMIO space required for that driver.
338 ****************************************************************************/
341 * This function set the exclusion range in the IOMMU. DMA accesses to the
342 * exclusion range are passed through untranslated
344 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
346 u64 start = iommu->exclusion_start & PAGE_MASK;
347 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
350 if (!iommu->exclusion_start)
353 entry = start | MMIO_EXCL_ENABLE_MASK;
354 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
355 &entry, sizeof(entry));
358 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
359 &entry, sizeof(entry));
362 /* Programs the physical address of the device table into the IOMMU hardware */
363 static void iommu_set_device_table(struct amd_iommu *iommu)
367 BUG_ON(iommu->mmio_base == NULL);
369 entry = iommu_virt_to_phys(amd_iommu_dev_table);
370 entry |= (dev_table_size >> 12) - 1;
371 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
372 &entry, sizeof(entry));
375 /* Generic functions to enable/disable certain features of the IOMMU. */
376 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
380 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
381 ctrl |= (1ULL << bit);
382 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
385 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
389 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
390 ctrl &= ~(1ULL << bit);
391 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
394 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
398 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
399 ctrl &= ~CTRL_INV_TO_MASK;
400 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
401 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
404 /* Function to enable the hardware */
405 static void iommu_enable(struct amd_iommu *iommu)
407 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
410 static void iommu_disable(struct amd_iommu *iommu)
412 if (!iommu->mmio_base)
415 /* Disable command buffer */
416 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
418 /* Disable event logging and event interrupts */
419 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
420 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
422 /* Disable IOMMU GA_LOG */
423 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
424 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
426 /* Disable IOMMU hardware itself */
427 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
431 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
432 * the system has one.
434 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
436 if (!request_mem_region(address, end, "amd_iommu")) {
437 pr_err("Can not reserve memory region %llx-%llx for mmio\n",
439 pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
443 return (u8 __iomem *)ioremap(address, end);
446 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
448 if (iommu->mmio_base)
449 iounmap(iommu->mmio_base);
450 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
453 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
469 /****************************************************************************
471 * The functions below belong to the first pass of AMD IOMMU ACPI table
472 * parsing. In this pass we try to find out the highest device id this
473 * code has to handle. Upon this information the size of the shared data
474 * structures is determined later.
476 ****************************************************************************/
479 * This function calculates the length of a given IVHD entry
481 static inline int ivhd_entry_length(u8 *ivhd)
483 u32 type = ((struct ivhd_entry *)ivhd)->type;
486 return 0x04 << (*ivhd >> 6);
487 } else if (type == IVHD_DEV_ACPI_HID) {
488 /* For ACPI_HID, offset 21 is uid len */
489 return *((u8 *)ivhd + 21) + 22;
495 * After reading the highest device id from the IOMMU PCI capability header
496 * this function looks if there is a higher device id defined in the ACPI table
498 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
500 u8 *p = (void *)h, *end = (void *)h;
501 struct ivhd_entry *dev;
503 u32 ivhd_size = get_ivhd_header_size(h);
506 pr_err("Unsupported IVHD type %#x\n", h->type);
514 dev = (struct ivhd_entry *)p;
517 /* Use maximum BDF value for DEV_ALL */
518 update_last_devid(0xffff);
520 case IVHD_DEV_SELECT:
521 case IVHD_DEV_RANGE_END:
523 case IVHD_DEV_EXT_SELECT:
524 /* all the above subfield types refer to device ids */
525 update_last_devid(dev->devid);
530 p += ivhd_entry_length(p);
538 static int __init check_ivrs_checksum(struct acpi_table_header *table)
541 u8 checksum = 0, *p = (u8 *)table;
543 for (i = 0; i < table->length; ++i)
546 /* ACPI table corrupt */
547 pr_err(FW_BUG "IVRS invalid checksum\n");
555 * Iterate over all IVHD entries in the ACPI table and find the highest device
556 * id which we need to handle. This is the first of three functions which parse
557 * the ACPI table. So we check the checksum here.
559 static int __init find_last_devid_acpi(struct acpi_table_header *table)
561 u8 *p = (u8 *)table, *end = (u8 *)table;
562 struct ivhd_header *h;
564 p += IVRS_HEADER_LENGTH;
566 end += table->length;
568 h = (struct ivhd_header *)p;
569 if (h->type == amd_iommu_target_ivhd_type) {
570 int ret = find_last_devid_from_ivhd(h);
582 /****************************************************************************
584 * The following functions belong to the code path which parses the ACPI table
585 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
586 * data structures, initialize the device/alias/rlookup table and also
587 * basically initialize the hardware.
589 ****************************************************************************/
592 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
593 * write commands to that buffer later and the IOMMU will execute them
596 static int __init alloc_command_buffer(struct amd_iommu *iommu)
598 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
599 get_order(CMD_BUFFER_SIZE));
601 return iommu->cmd_buf ? 0 : -ENOMEM;
605 * This function resets the command buffer if the IOMMU stopped fetching
608 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
610 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
612 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
613 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
614 iommu->cmd_buf_head = 0;
615 iommu->cmd_buf_tail = 0;
617 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
621 * This function writes the command buffer address to the hardware and
624 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
628 BUG_ON(iommu->cmd_buf == NULL);
630 entry = iommu_virt_to_phys(iommu->cmd_buf);
631 entry |= MMIO_CMD_SIZE_512;
633 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
634 &entry, sizeof(entry));
636 amd_iommu_reset_cmd_buffer(iommu);
640 * This function disables the command buffer
642 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
644 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
647 static void __init free_command_buffer(struct amd_iommu *iommu)
649 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
652 /* allocates the memory where the IOMMU will log its events to */
653 static int __init alloc_event_buffer(struct amd_iommu *iommu)
655 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
656 get_order(EVT_BUFFER_SIZE));
658 return iommu->evt_buf ? 0 : -ENOMEM;
661 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
665 BUG_ON(iommu->evt_buf == NULL);
667 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
669 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
670 &entry, sizeof(entry));
672 /* set head and tail to zero manually */
673 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
674 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
676 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
680 * This function disables the event log buffer
682 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
684 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
687 static void __init free_event_buffer(struct amd_iommu *iommu)
689 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
692 /* allocates the memory where the IOMMU will log its events to */
693 static int __init alloc_ppr_log(struct amd_iommu *iommu)
695 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
696 get_order(PPR_LOG_SIZE));
698 return iommu->ppr_log ? 0 : -ENOMEM;
701 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
705 if (iommu->ppr_log == NULL)
708 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
710 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
711 &entry, sizeof(entry));
713 /* set head and tail to zero manually */
714 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
715 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
717 iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
718 iommu_feature_enable(iommu, CONTROL_PPR_EN);
721 static void __init free_ppr_log(struct amd_iommu *iommu)
723 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
726 static void free_ga_log(struct amd_iommu *iommu)
728 #ifdef CONFIG_IRQ_REMAP
729 free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE));
730 free_pages((unsigned long)iommu->ga_log_tail, get_order(8));
734 static int iommu_ga_log_enable(struct amd_iommu *iommu)
736 #ifdef CONFIG_IRQ_REMAP
742 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
744 /* Check if already running */
745 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
748 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
749 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
751 for (i = 0; i < LOOP_TIMEOUT; ++i) {
752 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
753 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
757 if (i >= LOOP_TIMEOUT)
759 #endif /* CONFIG_IRQ_REMAP */
763 #ifdef CONFIG_IRQ_REMAP
764 static int iommu_init_ga_log(struct amd_iommu *iommu)
768 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
771 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
772 get_order(GA_LOG_SIZE));
776 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
778 if (!iommu->ga_log_tail)
781 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
782 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
783 &entry, sizeof(entry));
784 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
785 (BIT_ULL(52)-1)) & ~7ULL;
786 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
787 &entry, sizeof(entry));
788 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
789 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
796 #endif /* CONFIG_IRQ_REMAP */
798 static int iommu_init_ga(struct amd_iommu *iommu)
802 #ifdef CONFIG_IRQ_REMAP
803 /* Note: We have already checked GASup from IVRS table.
804 * Now, we need to make sure that GAMSup is set.
806 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
807 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
808 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
810 ret = iommu_init_ga_log(iommu);
811 #endif /* CONFIG_IRQ_REMAP */
816 static void iommu_enable_xt(struct amd_iommu *iommu)
818 #ifdef CONFIG_IRQ_REMAP
820 * XT mode (32-bit APIC destination ID) requires
821 * GA mode (128-bit IRTE support) as a prerequisite.
823 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
824 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
825 iommu_feature_enable(iommu, CONTROL_XT_EN);
826 #endif /* CONFIG_IRQ_REMAP */
829 static void iommu_enable_gt(struct amd_iommu *iommu)
831 if (!iommu_feature(iommu, FEATURE_GT))
834 iommu_feature_enable(iommu, CONTROL_GT_EN);
837 /* sets a specific bit in the device table entry. */
838 static void set_dev_entry_bit(u16 devid, u8 bit)
840 int i = (bit >> 6) & 0x03;
841 int _bit = bit & 0x3f;
843 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
846 static int get_dev_entry_bit(u16 devid, u8 bit)
848 int i = (bit >> 6) & 0x03;
849 int _bit = bit & 0x3f;
851 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
855 static bool copy_device_table(void)
857 u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
858 struct dev_table_entry *old_devtb = NULL;
859 u32 lo, hi, devid, old_devtb_size;
860 phys_addr_t old_devtb_phys;
861 struct amd_iommu *iommu;
862 u16 dom_id, dte_v, irq_v;
866 if (!amd_iommu_pre_enabled)
869 pr_warn("Translation is already enabled - trying to copy translation structures\n");
870 for_each_iommu(iommu) {
871 /* All IOMMUs should use the same device table with the same size */
872 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
873 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
874 entry = (((u64) hi) << 32) + lo;
875 if (last_entry && last_entry != entry) {
876 pr_err("IOMMU:%d should use the same dev table as others!\n",
882 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
883 if (old_devtb_size != dev_table_size) {
884 pr_err("The device table size of IOMMU:%d is not expected!\n",
891 * When SME is enabled in the first kernel, the entry includes the
892 * memory encryption mask(sme_me_mask), we must remove the memory
893 * encryption mask to obtain the true physical address in kdump kernel.
895 old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
897 if (old_devtb_phys >= 0x100000000ULL) {
898 pr_err("The address of old device table is above 4G, not trustworthy!\n");
901 old_devtb = (sme_active() && is_kdump_kernel())
902 ? (__force void *)ioremap_encrypted(old_devtb_phys,
904 : memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
909 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
910 old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
911 get_order(dev_table_size));
912 if (old_dev_tbl_cpy == NULL) {
913 pr_err("Failed to allocate memory for copying old device table!\n");
917 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
918 old_dev_tbl_cpy[devid] = old_devtb[devid];
919 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
920 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
922 if (dte_v && dom_id) {
923 old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
924 old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
925 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
926 /* If gcr3 table existed, mask it out */
927 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
928 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
929 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
930 old_dev_tbl_cpy[devid].data[1] &= ~tmp;
931 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
933 old_dev_tbl_cpy[devid].data[0] &= ~tmp;
937 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
938 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
939 int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
940 if (irq_v && (int_ctl || int_tab_len)) {
941 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
942 (int_tab_len != DTE_IRQ_TABLE_LEN)) {
943 pr_err("Wrong old irq remapping flag: %#x\n", devid);
947 old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
955 void amd_iommu_apply_erratum_63(u16 devid)
959 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
960 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
963 set_dev_entry_bit(devid, DEV_ENTRY_IW);
966 /* Writes the specific IOMMU for a device into the rlookup table */
967 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
969 amd_iommu_rlookup_table[devid] = iommu;
973 * This function takes the device specific flags read from the ACPI
974 * table and sets up the device table entry with that information
976 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
977 u16 devid, u32 flags, u32 ext_flags)
979 if (flags & ACPI_DEVFLAG_INITPASS)
980 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
981 if (flags & ACPI_DEVFLAG_EXTINT)
982 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
983 if (flags & ACPI_DEVFLAG_NMI)
984 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
985 if (flags & ACPI_DEVFLAG_SYSMGT1)
986 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
987 if (flags & ACPI_DEVFLAG_SYSMGT2)
988 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
989 if (flags & ACPI_DEVFLAG_LINT0)
990 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
991 if (flags & ACPI_DEVFLAG_LINT1)
992 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
994 amd_iommu_apply_erratum_63(devid);
996 set_iommu_for_device(iommu, devid);
999 int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1001 struct devid_map *entry;
1002 struct list_head *list;
1004 if (type == IVHD_SPECIAL_IOAPIC)
1006 else if (type == IVHD_SPECIAL_HPET)
1011 list_for_each_entry(entry, list, list) {
1012 if (!(entry->id == id && entry->cmd_line))
1015 pr_info("Command-line override present for %s id %d - ignoring\n",
1016 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1018 *devid = entry->devid;
1023 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1028 entry->devid = *devid;
1029 entry->cmd_line = cmd_line;
1031 list_add_tail(&entry->list, list);
1036 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1039 struct acpihid_map_entry *entry;
1040 struct list_head *list = &acpihid_map;
1042 list_for_each_entry(entry, list, list) {
1043 if (strcmp(entry->hid, hid) ||
1044 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1048 pr_info("Command-line override for hid:%s uid:%s\n",
1050 *devid = entry->devid;
1054 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1058 memcpy(entry->uid, uid, strlen(uid));
1059 memcpy(entry->hid, hid, strlen(hid));
1060 entry->devid = *devid;
1061 entry->cmd_line = cmd_line;
1062 entry->root_devid = (entry->devid & (~0x7));
1064 pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1065 entry->cmd_line ? "cmd" : "ivrs",
1066 entry->hid, entry->uid, entry->root_devid);
1068 list_add_tail(&entry->list, list);
1072 static int __init add_early_maps(void)
1076 for (i = 0; i < early_ioapic_map_size; ++i) {
1077 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1078 early_ioapic_map[i].id,
1079 &early_ioapic_map[i].devid,
1080 early_ioapic_map[i].cmd_line);
1085 for (i = 0; i < early_hpet_map_size; ++i) {
1086 ret = add_special_device(IVHD_SPECIAL_HPET,
1087 early_hpet_map[i].id,
1088 &early_hpet_map[i].devid,
1089 early_hpet_map[i].cmd_line);
1094 for (i = 0; i < early_acpihid_map_size; ++i) {
1095 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1096 early_acpihid_map[i].uid,
1097 &early_acpihid_map[i].devid,
1098 early_acpihid_map[i].cmd_line);
1107 * Reads the device exclusion range from ACPI and initializes the IOMMU with
1110 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
1112 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
1116 * Treat per-device exclusion ranges as r/w unity-mapped regions
1117 * since some buggy BIOSes might lead to the overwritten exclusion
1118 * range (exclusion_start and exclusion_length members). This
1119 * happens when there are multiple exclusion ranges (IVMD entries)
1120 * defined in ACPI table.
1122 m->flags = (IVMD_FLAG_IW | IVMD_FLAG_IR | IVMD_FLAG_UNITY_MAP);
1126 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1127 * initializes the hardware and our data structures with it.
1129 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1130 struct ivhd_header *h)
1133 u8 *end = p, flags = 0;
1134 u16 devid = 0, devid_start = 0, devid_to = 0;
1135 u32 dev_i, ext_flags = 0;
1137 struct ivhd_entry *e;
1142 ret = add_early_maps();
1146 amd_iommu_apply_ivrs_quirks();
1149 * First save the recommended feature enable bits from ACPI
1151 iommu->acpi_flags = h->flags;
1154 * Done. Now parse the device entries
1156 ivhd_size = get_ivhd_header_size(h);
1158 pr_err("Unsupported IVHD type %#x\n", h->type);
1168 e = (struct ivhd_entry *)p;
1172 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1174 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1175 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1177 case IVHD_DEV_SELECT:
1179 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1181 PCI_BUS_NUM(e->devid),
1187 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1189 case IVHD_DEV_SELECT_RANGE_START:
1191 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1192 "devid: %02x:%02x.%x flags: %02x\n",
1193 PCI_BUS_NUM(e->devid),
1198 devid_start = e->devid;
1203 case IVHD_DEV_ALIAS:
1205 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1206 "flags: %02x devid_to: %02x:%02x.%x\n",
1207 PCI_BUS_NUM(e->devid),
1211 PCI_BUS_NUM(e->ext >> 8),
1212 PCI_SLOT(e->ext >> 8),
1213 PCI_FUNC(e->ext >> 8));
1216 devid_to = e->ext >> 8;
1217 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1218 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1219 amd_iommu_alias_table[devid] = devid_to;
1221 case IVHD_DEV_ALIAS_RANGE:
1223 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1224 "devid: %02x:%02x.%x flags: %02x "
1225 "devid_to: %02x:%02x.%x\n",
1226 PCI_BUS_NUM(e->devid),
1230 PCI_BUS_NUM(e->ext >> 8),
1231 PCI_SLOT(e->ext >> 8),
1232 PCI_FUNC(e->ext >> 8));
1234 devid_start = e->devid;
1236 devid_to = e->ext >> 8;
1240 case IVHD_DEV_EXT_SELECT:
1242 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1243 "flags: %02x ext: %08x\n",
1244 PCI_BUS_NUM(e->devid),
1250 set_dev_entry_from_acpi(iommu, devid, e->flags,
1253 case IVHD_DEV_EXT_SELECT_RANGE:
1255 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1256 "%02x:%02x.%x flags: %02x ext: %08x\n",
1257 PCI_BUS_NUM(e->devid),
1262 devid_start = e->devid;
1267 case IVHD_DEV_RANGE_END:
1269 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1270 PCI_BUS_NUM(e->devid),
1272 PCI_FUNC(e->devid));
1275 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1277 amd_iommu_alias_table[dev_i] = devid_to;
1278 set_dev_entry_from_acpi(iommu,
1279 devid_to, flags, ext_flags);
1281 set_dev_entry_from_acpi(iommu, dev_i,
1285 case IVHD_DEV_SPECIAL: {
1291 handle = e->ext & 0xff;
1292 devid = (e->ext >> 8) & 0xffff;
1293 type = (e->ext >> 24) & 0xff;
1295 if (type == IVHD_SPECIAL_IOAPIC)
1297 else if (type == IVHD_SPECIAL_HPET)
1302 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1308 ret = add_special_device(type, handle, &devid, false);
1313 * add_special_device might update the devid in case a
1314 * command-line override is present. So call
1315 * set_dev_entry_from_acpi after add_special_device.
1317 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1321 case IVHD_DEV_ACPI_HID: {
1323 u8 hid[ACPIHID_HID_LEN];
1324 u8 uid[ACPIHID_UID_LEN];
1327 if (h->type != 0x40) {
1328 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1333 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1334 hid[ACPIHID_HID_LEN - 1] = '\0';
1337 pr_err(FW_BUG "Invalid HID.\n");
1343 case UID_NOT_PRESENT:
1346 pr_warn(FW_BUG "Invalid UID length.\n");
1349 case UID_IS_INTEGER:
1351 sprintf(uid, "%d", e->uid);
1354 case UID_IS_CHARACTER:
1356 memcpy(uid, &e->uid, e->uidl);
1357 uid[e->uidl] = '\0';
1365 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1373 ret = add_acpi_hid_device(hid, uid, &devid, false);
1378 * add_special_device might update the devid in case a
1379 * command-line override is present. So call
1380 * set_dev_entry_from_acpi after add_special_device.
1382 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1390 p += ivhd_entry_length(p);
1396 static void __init free_iommu_one(struct amd_iommu *iommu)
1398 free_command_buffer(iommu);
1399 free_event_buffer(iommu);
1400 free_ppr_log(iommu);
1402 iommu_unmap_mmio_space(iommu);
1405 static void __init free_iommu_all(void)
1407 struct amd_iommu *iommu, *next;
1409 for_each_iommu_safe(iommu, next) {
1410 list_del(&iommu->list);
1411 free_iommu_one(iommu);
1417 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1419 * BIOS should disable L2B micellaneous clock gating by setting
1420 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1422 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1426 if ((boot_cpu_data.x86 != 0x15) ||
1427 (boot_cpu_data.x86_model < 0x10) ||
1428 (boot_cpu_data.x86_model > 0x1f))
1431 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1432 pci_read_config_dword(iommu->dev, 0xf4, &value);
1437 /* Select NB indirect register 0x90 and enable writing */
1438 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1440 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1441 pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1443 /* Clear the enable writing bit */
1444 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1448 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1450 * BIOS should enable ATS write permission check by setting
1451 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1453 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1457 if ((boot_cpu_data.x86 != 0x15) ||
1458 (boot_cpu_data.x86_model < 0x30) ||
1459 (boot_cpu_data.x86_model > 0x3f))
1462 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1463 value = iommu_read_l2(iommu, 0x47);
1468 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1469 iommu_write_l2(iommu, 0x47, value | BIT(0));
1471 pci_info(iommu->dev, "Applying ATS write check workaround\n");
1475 * This function clues the initialization function for one IOMMU
1476 * together and also allocates the command buffer and programs the
1477 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1479 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1483 raw_spin_lock_init(&iommu->lock);
1485 /* Add IOMMU to internal data structures */
1486 list_add_tail(&iommu->list, &amd_iommu_list);
1487 iommu->index = amd_iommus_present++;
1489 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1490 WARN(1, "System has more IOMMUs than supported by this driver\n");
1494 /* Index is fine - add IOMMU to the array */
1495 amd_iommus[iommu->index] = iommu;
1498 * Copy data from ACPI table entry to the iommu struct
1500 iommu->devid = h->devid;
1501 iommu->cap_ptr = h->cap_ptr;
1502 iommu->pci_seg = h->pci_seg;
1503 iommu->mmio_phys = h->mmio_phys;
1507 /* Check if IVHD EFR contains proper max banks/counters */
1508 if ((h->efr_attr != 0) &&
1509 ((h->efr_attr & (0xF << 13)) != 0) &&
1510 ((h->efr_attr & (0x3F << 17)) != 0))
1511 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1513 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1514 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1515 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1519 if (h->efr_reg & (1 << 9))
1520 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1522 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1523 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1524 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1526 * Note: Since iommu_update_intcapxt() leverages
1527 * the IOMMU MMIO access to MSI capability block registers
1528 * for MSI address lo/hi/data, we need to check both
1529 * EFR[XtSup] and EFR[MsiCapMmioSup] for x2APIC support.
1531 if ((h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT)) &&
1532 (h->efr_reg & BIT(IOMMU_EFR_MSICAPMMIOSUP_SHIFT)))
1533 amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1539 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1540 iommu->mmio_phys_end);
1541 if (!iommu->mmio_base)
1544 if (alloc_command_buffer(iommu))
1547 if (alloc_event_buffer(iommu))
1550 iommu->int_enabled = false;
1552 init_translation_status(iommu);
1553 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1554 iommu_disable(iommu);
1555 clear_translation_pre_enabled(iommu);
1556 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1559 if (amd_iommu_pre_enabled)
1560 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1562 ret = init_iommu_from_acpi(iommu, h);
1566 ret = amd_iommu_create_irq_domain(iommu);
1571 * Make sure IOMMU is not considered to translate itself. The IVRS
1572 * table tells us so, but this is a lie!
1574 amd_iommu_rlookup_table[iommu->devid] = NULL;
1580 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1581 * @ivrs Pointer to the IVRS header
1583 * This function search through all IVDB of the maximum supported IVHD
1585 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1587 u8 *base = (u8 *)ivrs;
1588 struct ivhd_header *ivhd = (struct ivhd_header *)
1589 (base + IVRS_HEADER_LENGTH);
1590 u8 last_type = ivhd->type;
1591 u16 devid = ivhd->devid;
1593 while (((u8 *)ivhd - base < ivrs->length) &&
1594 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1595 u8 *p = (u8 *) ivhd;
1597 if (ivhd->devid == devid)
1598 last_type = ivhd->type;
1599 ivhd = (struct ivhd_header *)(p + ivhd->length);
1606 * Iterates over all IOMMU entries in the ACPI table, allocates the
1607 * IOMMU structure and initializes it with init_iommu_one()
1609 static int __init init_iommu_all(struct acpi_table_header *table)
1611 u8 *p = (u8 *)table, *end = (u8 *)table;
1612 struct ivhd_header *h;
1613 struct amd_iommu *iommu;
1616 end += table->length;
1617 p += IVRS_HEADER_LENGTH;
1620 h = (struct ivhd_header *)p;
1621 if (*p == amd_iommu_target_ivhd_type) {
1623 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1624 "seg: %d flags: %01x info %04x\n",
1625 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1626 PCI_FUNC(h->devid), h->cap_ptr,
1627 h->pci_seg, h->flags, h->info);
1628 DUMP_printk(" mmio-addr: %016llx\n",
1631 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1635 ret = init_iommu_one(iommu, h);
1647 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1648 u8 fxn, u64 *value, bool is_write);
1650 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1652 struct pci_dev *pdev = iommu->dev;
1653 u64 val = 0xabcd, val2 = 0, save_reg = 0;
1655 if (!iommu_feature(iommu, FEATURE_PC))
1658 amd_iommu_pc_present = true;
1660 /* save the value to restore, if writable */
1661 if (iommu_pc_get_set_reg(iommu, 0, 0, 0, &save_reg, false))
1664 /* Check if the performance counters can be written to */
1665 if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1666 (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1671 if (iommu_pc_get_set_reg(iommu, 0, 0, 0, &save_reg, true))
1674 pci_info(pdev, "IOMMU performance counters supported\n");
1676 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1677 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1678 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1683 pci_err(pdev, "Unable to read/write to IOMMU perf counter.\n");
1684 amd_iommu_pc_present = false;
1688 static ssize_t amd_iommu_show_cap(struct device *dev,
1689 struct device_attribute *attr,
1692 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1693 return sprintf(buf, "%x\n", iommu->cap);
1695 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1697 static ssize_t amd_iommu_show_features(struct device *dev,
1698 struct device_attribute *attr,
1701 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1702 return sprintf(buf, "%llx\n", iommu->features);
1704 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1706 static struct attribute *amd_iommu_attrs[] = {
1708 &dev_attr_features.attr,
1712 static struct attribute_group amd_iommu_group = {
1713 .name = "amd-iommu",
1714 .attrs = amd_iommu_attrs,
1717 static const struct attribute_group *amd_iommu_groups[] = {
1722 static int __init iommu_init_pci(struct amd_iommu *iommu)
1724 int cap_ptr = iommu->cap_ptr;
1727 iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1728 iommu->devid & 0xff);
1732 /* Prevent binding other PCI device drivers to IOMMU devices */
1733 iommu->dev->match_driver = false;
1735 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1738 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1739 amd_iommu_iotlb_sup = false;
1741 /* read extended feature bits */
1742 iommu->features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1744 if (iommu_feature(iommu, FEATURE_GT)) {
1749 pasmax = iommu->features & FEATURE_PASID_MASK;
1750 pasmax >>= FEATURE_PASID_SHIFT;
1751 max_pasid = (1 << (pasmax + 1)) - 1;
1753 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1755 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1757 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1758 glxval >>= FEATURE_GLXVAL_SHIFT;
1760 if (amd_iommu_max_glx_val == -1)
1761 amd_iommu_max_glx_val = glxval;
1763 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1766 if (iommu_feature(iommu, FEATURE_GT) &&
1767 iommu_feature(iommu, FEATURE_PPR)) {
1768 iommu->is_iommu_v2 = true;
1769 amd_iommu_v2_present = true;
1772 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1775 ret = iommu_init_ga(iommu);
1779 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1780 amd_iommu_np_cache = true;
1782 init_iommu_perf_ctr(iommu);
1784 if (is_rd890_iommu(iommu->dev)) {
1788 pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
1792 * Some rd890 systems may not be fully reconfigured by the
1793 * BIOS, so it's necessary for us to store this information so
1794 * it can be reprogrammed on resume
1796 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1797 &iommu->stored_addr_lo);
1798 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1799 &iommu->stored_addr_hi);
1801 /* Low bit locks writes to configuration space */
1802 iommu->stored_addr_lo &= ~1;
1804 for (i = 0; i < 6; i++)
1805 for (j = 0; j < 0x12; j++)
1806 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1808 for (i = 0; i < 0x83; i++)
1809 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1812 amd_iommu_erratum_746_workaround(iommu);
1813 amd_iommu_ats_write_check_workaround(iommu);
1815 iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1816 amd_iommu_groups, "ivhd%d", iommu->index);
1817 iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1818 iommu_device_register(&iommu->iommu);
1820 return pci_enable_device(iommu->dev);
1823 static void print_iommu_info(void)
1825 static const char * const feat_str[] = {
1826 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1827 "IA", "GA", "HE", "PC"
1829 struct amd_iommu *iommu;
1831 for_each_iommu(iommu) {
1832 struct pci_dev *pdev = iommu->dev;
1835 pci_info(pdev, "Found IOMMU cap 0x%hx\n", iommu->cap_ptr);
1837 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1838 pci_info(pdev, "Extended features (%#llx):",
1840 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1841 if (iommu_feature(iommu, (1ULL << i)))
1842 pr_cont(" %s", feat_str[i]);
1845 if (iommu->features & FEATURE_GAM_VAPIC)
1846 pr_cont(" GA_vAPIC");
1851 if (irq_remapping_enabled) {
1852 pr_info("Interrupt remapping enabled\n");
1853 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1854 pr_info("Virtual APIC enabled\n");
1855 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1856 pr_info("X2APIC enabled\n");
1860 static int __init amd_iommu_init_pci(void)
1862 struct amd_iommu *iommu;
1865 for_each_iommu(iommu) {
1866 ret = iommu_init_pci(iommu);
1872 * Order is important here to make sure any unity map requirements are
1873 * fulfilled. The unity mappings are created and written to the device
1874 * table during the amd_iommu_init_api() call.
1876 * After that we call init_device_table_dma() to make sure any
1877 * uninitialized DTE will block DMA, and in the end we flush the caches
1878 * of all IOMMUs to make sure the changes to the device table are
1881 ret = amd_iommu_init_api();
1883 init_device_table_dma();
1885 for_each_iommu(iommu)
1886 iommu_flush_all_caches(iommu);
1894 /****************************************************************************
1896 * The following functions initialize the MSI interrupts for all IOMMUs
1897 * in the system. It's a bit challenging because there could be multiple
1898 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1901 ****************************************************************************/
1903 static int iommu_setup_msi(struct amd_iommu *iommu)
1907 r = pci_enable_msi(iommu->dev);
1911 r = request_threaded_irq(iommu->dev->irq,
1912 amd_iommu_int_handler,
1913 amd_iommu_int_thread,
1918 pci_disable_msi(iommu->dev);
1922 iommu->int_enabled = true;
1927 #define XT_INT_DEST_MODE(x) (((x) & 0x1ULL) << 2)
1928 #define XT_INT_DEST_LO(x) (((x) & 0xFFFFFFULL) << 8)
1929 #define XT_INT_VEC(x) (((x) & 0xFFULL) << 32)
1930 #define XT_INT_DEST_HI(x) ((((x) >> 24) & 0xFFULL) << 56)
1933 * Setup the IntCapXT registers with interrupt routing information
1934 * based on the PCI MSI capability block registers, accessed via
1935 * MMIO MSI address low/hi and MSI data registers.
1937 static void iommu_update_intcapxt(struct amd_iommu *iommu)
1940 u32 addr_lo = readl(iommu->mmio_base + MMIO_MSI_ADDR_LO_OFFSET);
1941 u32 addr_hi = readl(iommu->mmio_base + MMIO_MSI_ADDR_HI_OFFSET);
1942 u32 data = readl(iommu->mmio_base + MMIO_MSI_DATA_OFFSET);
1943 bool dm = (addr_lo >> MSI_ADDR_DEST_MODE_SHIFT) & 0x1;
1944 u32 dest = ((addr_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xFF);
1946 if (x2apic_enabled())
1947 dest |= MSI_ADDR_EXT_DEST_ID(addr_hi);
1949 val = XT_INT_VEC(data & 0xFF) |
1950 XT_INT_DEST_MODE(dm) |
1951 XT_INT_DEST_LO(dest) |
1952 XT_INT_DEST_HI(dest);
1955 * Current IOMMU implemtation uses the same IRQ for all
1956 * 3 IOMMU interrupts.
1958 writeq(val, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
1959 writeq(val, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
1960 writeq(val, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
1963 static void _irq_notifier_notify(struct irq_affinity_notify *notify,
1964 const cpumask_t *mask)
1966 struct amd_iommu *iommu;
1968 for_each_iommu(iommu) {
1969 if (iommu->dev->irq == notify->irq) {
1970 iommu_update_intcapxt(iommu);
1976 static void _irq_notifier_release(struct kref *ref)
1980 static int iommu_init_intcapxt(struct amd_iommu *iommu)
1983 struct irq_affinity_notify *notify = &iommu->intcapxt_notify;
1986 * IntCapXT requires XTSup=1 and MsiCapMmioSup=1,
1987 * which can be inferred from amd_iommu_xt_mode.
1989 if (amd_iommu_xt_mode != IRQ_REMAP_X2APIC_MODE)
1993 * Also, we need to setup notifier to update the IntCapXT registers
1994 * whenever the irq affinity is changed from user-space.
1996 notify->irq = iommu->dev->irq;
1997 notify->notify = _irq_notifier_notify,
1998 notify->release = _irq_notifier_release,
1999 ret = irq_set_affinity_notifier(iommu->dev->irq, notify);
2001 pr_err("Failed to register irq affinity notifier (devid=%#x, irq %d)\n",
2002 iommu->devid, iommu->dev->irq);
2006 iommu_update_intcapxt(iommu);
2007 iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2011 static int iommu_init_msi(struct amd_iommu *iommu)
2015 if (iommu->int_enabled)
2018 if (iommu->dev->msi_cap)
2019 ret = iommu_setup_msi(iommu);
2027 ret = iommu_init_intcapxt(iommu);
2031 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2033 if (iommu->ppr_log != NULL)
2034 iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
2036 iommu_ga_log_enable(iommu);
2041 /****************************************************************************
2043 * The next functions belong to the third pass of parsing the ACPI
2044 * table. In this last pass the memory mapping requirements are
2045 * gathered (like exclusion and unity mapping ranges).
2047 ****************************************************************************/
2049 static void __init free_unity_maps(void)
2051 struct unity_map_entry *entry, *next;
2053 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
2054 list_del(&entry->list);
2059 /* called when we find an exclusion range definition in ACPI */
2060 static int __init init_exclusion_range(struct ivmd_header *m)
2065 case ACPI_IVMD_TYPE:
2066 set_device_exclusion_range(m->devid, m);
2068 case ACPI_IVMD_TYPE_ALL:
2069 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2070 set_device_exclusion_range(i, m);
2072 case ACPI_IVMD_TYPE_RANGE:
2073 for (i = m->devid; i <= m->aux; ++i)
2074 set_device_exclusion_range(i, m);
2083 /* called for unity map ACPI definition */
2084 static int __init init_unity_map_range(struct ivmd_header *m)
2086 struct unity_map_entry *e = NULL;
2089 e = kzalloc(sizeof(*e), GFP_KERNEL);
2093 if (m->flags & IVMD_FLAG_EXCL_RANGE)
2094 init_exclusion_range(m);
2100 case ACPI_IVMD_TYPE:
2101 s = "IVMD_TYPEi\t\t\t";
2102 e->devid_start = e->devid_end = m->devid;
2104 case ACPI_IVMD_TYPE_ALL:
2105 s = "IVMD_TYPE_ALL\t\t";
2107 e->devid_end = amd_iommu_last_bdf;
2109 case ACPI_IVMD_TYPE_RANGE:
2110 s = "IVMD_TYPE_RANGE\t\t";
2111 e->devid_start = m->devid;
2112 e->devid_end = m->aux;
2115 e->address_start = PAGE_ALIGN(m->range_start);
2116 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2117 e->prot = m->flags >> 1;
2119 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2120 " range_start: %016llx range_end: %016llx flags: %x\n", s,
2121 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2122 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
2123 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2124 e->address_start, e->address_end, m->flags);
2126 list_add_tail(&e->list, &amd_iommu_unity_map);
2131 /* iterates over all memory definitions we find in the ACPI table */
2132 static int __init init_memory_definitions(struct acpi_table_header *table)
2134 u8 *p = (u8 *)table, *end = (u8 *)table;
2135 struct ivmd_header *m;
2137 end += table->length;
2138 p += IVRS_HEADER_LENGTH;
2141 m = (struct ivmd_header *)p;
2142 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2143 init_unity_map_range(m);
2152 * Init the device table to not allow DMA access for devices
2154 static void init_device_table_dma(void)
2158 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2159 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2160 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2164 static void __init uninit_device_table_dma(void)
2168 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2169 amd_iommu_dev_table[devid].data[0] = 0ULL;
2170 amd_iommu_dev_table[devid].data[1] = 0ULL;
2174 static void init_device_table(void)
2178 if (!amd_iommu_irq_remap)
2181 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2182 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2185 static void iommu_init_flags(struct amd_iommu *iommu)
2187 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2188 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2189 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2191 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2192 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2193 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2195 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2196 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2197 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2199 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2200 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2201 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2204 * make IOMMU memory accesses cache coherent
2206 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2208 /* Set IOTLB invalidation timeout to 1s */
2209 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2212 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2215 u32 ioc_feature_control;
2216 struct pci_dev *pdev = iommu->root_pdev;
2218 /* RD890 BIOSes may not have completely reconfigured the iommu */
2219 if (!is_rd890_iommu(iommu->dev) || !pdev)
2223 * First, we need to ensure that the iommu is enabled. This is
2224 * controlled by a register in the northbridge
2227 /* Select Northbridge indirect register 0x75 and enable writing */
2228 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2229 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2231 /* Enable the iommu */
2232 if (!(ioc_feature_control & 0x1))
2233 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2235 /* Restore the iommu BAR */
2236 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2237 iommu->stored_addr_lo);
2238 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2239 iommu->stored_addr_hi);
2241 /* Restore the l1 indirect regs for each of the 6 l1s */
2242 for (i = 0; i < 6; i++)
2243 for (j = 0; j < 0x12; j++)
2244 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2246 /* Restore the l2 indirect regs */
2247 for (i = 0; i < 0x83; i++)
2248 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2250 /* Lock PCI setup registers */
2251 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2252 iommu->stored_addr_lo | 1);
2255 static void iommu_enable_ga(struct amd_iommu *iommu)
2257 #ifdef CONFIG_IRQ_REMAP
2258 switch (amd_iommu_guest_ir) {
2259 case AMD_IOMMU_GUEST_IR_VAPIC:
2260 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2262 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2263 iommu_feature_enable(iommu, CONTROL_GA_EN);
2264 iommu->irte_ops = &irte_128_ops;
2267 iommu->irte_ops = &irte_32_ops;
2273 static void early_enable_iommu(struct amd_iommu *iommu)
2275 iommu_disable(iommu);
2276 iommu_init_flags(iommu);
2277 iommu_set_device_table(iommu);
2278 iommu_enable_command_buffer(iommu);
2279 iommu_enable_event_buffer(iommu);
2280 iommu_set_exclusion_range(iommu);
2281 iommu_enable_ga(iommu);
2282 iommu_enable_xt(iommu);
2283 iommu_enable(iommu);
2284 iommu_flush_all_caches(iommu);
2288 * This function finally enables all IOMMUs found in the system after
2289 * they have been initialized.
2291 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2292 * the old content of device table entries. Not this case or copy failed,
2293 * just continue as normal kernel does.
2295 static void early_enable_iommus(void)
2297 struct amd_iommu *iommu;
2300 if (!copy_device_table()) {
2302 * If come here because of failure in copying device table from old
2303 * kernel with all IOMMUs enabled, print error message and try to
2304 * free allocated old_dev_tbl_cpy.
2306 if (amd_iommu_pre_enabled)
2307 pr_err("Failed to copy DEV table from previous kernel.\n");
2308 if (old_dev_tbl_cpy != NULL)
2309 free_pages((unsigned long)old_dev_tbl_cpy,
2310 get_order(dev_table_size));
2312 for_each_iommu(iommu) {
2313 clear_translation_pre_enabled(iommu);
2314 early_enable_iommu(iommu);
2317 pr_info("Copied DEV table from previous kernel.\n");
2318 free_pages((unsigned long)amd_iommu_dev_table,
2319 get_order(dev_table_size));
2320 amd_iommu_dev_table = old_dev_tbl_cpy;
2321 for_each_iommu(iommu) {
2322 iommu_disable_command_buffer(iommu);
2323 iommu_disable_event_buffer(iommu);
2324 iommu_enable_command_buffer(iommu);
2325 iommu_enable_event_buffer(iommu);
2326 iommu_enable_ga(iommu);
2327 iommu_enable_xt(iommu);
2328 iommu_set_device_table(iommu);
2329 iommu_flush_all_caches(iommu);
2333 #ifdef CONFIG_IRQ_REMAP
2334 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2335 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2339 static void enable_iommus_v2(void)
2341 struct amd_iommu *iommu;
2343 for_each_iommu(iommu) {
2344 iommu_enable_ppr_log(iommu);
2345 iommu_enable_gt(iommu);
2349 static void enable_iommus(void)
2351 early_enable_iommus();
2356 static void disable_iommus(void)
2358 struct amd_iommu *iommu;
2360 for_each_iommu(iommu)
2361 iommu_disable(iommu);
2363 #ifdef CONFIG_IRQ_REMAP
2364 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2365 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2370 * Suspend/Resume support
2371 * disable suspend until real resume implemented
2374 static void amd_iommu_resume(void)
2376 struct amd_iommu *iommu;
2378 for_each_iommu(iommu)
2379 iommu_apply_resume_quirks(iommu);
2381 /* re-load the hardware */
2384 amd_iommu_enable_interrupts();
2387 static int amd_iommu_suspend(void)
2389 /* disable IOMMUs to go out of the way for BIOS */
2395 static struct syscore_ops amd_iommu_syscore_ops = {
2396 .suspend = amd_iommu_suspend,
2397 .resume = amd_iommu_resume,
2400 static void __init free_iommu_resources(void)
2402 kmemleak_free(irq_lookup_table);
2403 free_pages((unsigned long)irq_lookup_table,
2404 get_order(rlookup_table_size));
2405 irq_lookup_table = NULL;
2407 kmem_cache_destroy(amd_iommu_irq_cache);
2408 amd_iommu_irq_cache = NULL;
2410 free_pages((unsigned long)amd_iommu_rlookup_table,
2411 get_order(rlookup_table_size));
2412 amd_iommu_rlookup_table = NULL;
2414 free_pages((unsigned long)amd_iommu_alias_table,
2415 get_order(alias_table_size));
2416 amd_iommu_alias_table = NULL;
2418 free_pages((unsigned long)amd_iommu_dev_table,
2419 get_order(dev_table_size));
2420 amd_iommu_dev_table = NULL;
2425 /* SB IOAPIC is always on this device in AMD systems */
2426 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2428 static bool __init check_ioapic_information(void)
2430 const char *fw_bug = FW_BUG;
2431 bool ret, has_sb_ioapic;
2434 has_sb_ioapic = false;
2438 * If we have map overrides on the kernel command line the
2439 * messages in this function might not describe firmware bugs
2440 * anymore - so be careful
2445 for (idx = 0; idx < nr_ioapics; idx++) {
2446 int devid, id = mpc_ioapic_id(idx);
2448 devid = get_ioapic_devid(id);
2450 pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2453 } else if (devid == IOAPIC_SB_DEVID) {
2454 has_sb_ioapic = true;
2459 if (!has_sb_ioapic) {
2461 * We expect the SB IOAPIC to be listed in the IVRS
2462 * table. The system timer is connected to the SB IOAPIC
2463 * and if we don't have it in the list the system will
2464 * panic at boot time. This situation usually happens
2465 * when the BIOS is buggy and provides us the wrong
2466 * device id for the IOAPIC in the system.
2468 pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2472 pr_err("Disabling interrupt remapping\n");
2477 static void __init free_dma_resources(void)
2479 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2480 get_order(MAX_DOMAIN_ID/8));
2481 amd_iommu_pd_alloc_bitmap = NULL;
2487 * This is the hardware init function for AMD IOMMU in the system.
2488 * This function is called either from amd_iommu_init or from the interrupt
2489 * remapping setup code.
2491 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2494 * 1 pass) Discover the most comprehensive IVHD type to use.
2496 * 2 pass) Find the highest PCI device id the driver has to handle.
2497 * Upon this information the size of the data structures is
2498 * determined that needs to be allocated.
2500 * 3 pass) Initialize the data structures just allocated with the
2501 * information in the ACPI table about available AMD IOMMUs
2502 * in the system. It also maps the PCI devices in the
2503 * system to specific IOMMUs
2505 * 4 pass) After the basic data structures are allocated and
2506 * initialized we update them with information about memory
2507 * remapping requirements parsed out of the ACPI table in
2510 * After everything is set up the IOMMUs are enabled and the necessary
2511 * hotplug and suspend notifiers are registered.
2513 static int __init early_amd_iommu_init(void)
2515 struct acpi_table_header *ivrs_base;
2517 int i, remap_cache_sz, ret = 0;
2520 if (!amd_iommu_detected)
2523 status = acpi_get_table("IVRS", 0, &ivrs_base);
2524 if (status == AE_NOT_FOUND)
2526 else if (ACPI_FAILURE(status)) {
2527 const char *err = acpi_format_exception(status);
2528 pr_err("IVRS table error: %s\n", err);
2533 * Validate checksum here so we don't need to do it when
2534 * we actually parse the table
2536 ret = check_ivrs_checksum(ivrs_base);
2540 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2541 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2544 * First parse ACPI tables to find the largest Bus/Dev/Func
2545 * we need to handle. Upon this information the shared data
2546 * structures for the IOMMUs in the system will be allocated
2548 ret = find_last_devid_acpi(ivrs_base);
2552 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2553 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2554 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2556 /* Device table - directly used by all IOMMUs */
2558 amd_iommu_dev_table = (void *)__get_free_pages(
2559 GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2560 get_order(dev_table_size));
2561 if (amd_iommu_dev_table == NULL)
2565 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2566 * IOMMU see for that device
2568 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2569 get_order(alias_table_size));
2570 if (amd_iommu_alias_table == NULL)
2573 /* IOMMU rlookup table - find the IOMMU for a specific device */
2574 amd_iommu_rlookup_table = (void *)__get_free_pages(
2575 GFP_KERNEL | __GFP_ZERO,
2576 get_order(rlookup_table_size));
2577 if (amd_iommu_rlookup_table == NULL)
2580 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2581 GFP_KERNEL | __GFP_ZERO,
2582 get_order(MAX_DOMAIN_ID/8));
2583 if (amd_iommu_pd_alloc_bitmap == NULL)
2587 * let all alias entries point to itself
2589 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2590 amd_iommu_alias_table[i] = i;
2593 * never allocate domain 0 because its used as the non-allocated and
2594 * error value placeholder
2596 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2599 * now the data structures are allocated and basically initialized
2600 * start the real acpi table scan
2602 ret = init_iommu_all(ivrs_base);
2606 /* Disable IOMMU if there's Stoney Ridge graphics */
2607 for (i = 0; i < 32; i++) {
2608 pci_id = read_pci_config(0, i, 0, 0);
2609 if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
2610 pr_info("Disable IOMMU on Stoney Ridge\n");
2611 amd_iommu_disabled = true;
2616 /* Disable any previously enabled IOMMUs */
2617 if (!is_kdump_kernel() || amd_iommu_disabled)
2620 if (amd_iommu_irq_remap)
2621 amd_iommu_irq_remap = check_ioapic_information();
2623 if (amd_iommu_irq_remap) {
2625 * Interrupt remapping enabled, create kmem_cache for the
2629 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2630 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2632 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2633 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2635 IRQ_TABLE_ALIGNMENT,
2637 if (!amd_iommu_irq_cache)
2640 irq_lookup_table = (void *)__get_free_pages(
2641 GFP_KERNEL | __GFP_ZERO,
2642 get_order(rlookup_table_size));
2643 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2645 if (!irq_lookup_table)
2649 ret = init_memory_definitions(ivrs_base);
2653 /* init the device table */
2654 init_device_table();
2657 /* Don't leak any ACPI memory */
2658 acpi_put_table(ivrs_base);
2664 static int amd_iommu_enable_interrupts(void)
2666 struct amd_iommu *iommu;
2669 for_each_iommu(iommu) {
2670 ret = iommu_init_msi(iommu);
2679 static bool detect_ivrs(void)
2681 struct acpi_table_header *ivrs_base;
2684 status = acpi_get_table("IVRS", 0, &ivrs_base);
2685 if (status == AE_NOT_FOUND)
2687 else if (ACPI_FAILURE(status)) {
2688 const char *err = acpi_format_exception(status);
2689 pr_err("IVRS table error: %s\n", err);
2693 acpi_put_table(ivrs_base);
2695 /* Make sure ACS will be enabled during PCI probe */
2701 /****************************************************************************
2703 * AMD IOMMU Initialization State Machine
2705 ****************************************************************************/
2707 static int __init state_next(void)
2711 switch (init_state) {
2712 case IOMMU_START_STATE:
2713 if (!detect_ivrs()) {
2714 init_state = IOMMU_NOT_FOUND;
2717 init_state = IOMMU_IVRS_DETECTED;
2720 case IOMMU_IVRS_DETECTED:
2721 ret = early_amd_iommu_init();
2722 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2723 if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2724 pr_info("AMD IOMMU disabled\n");
2725 init_state = IOMMU_CMDLINE_DISABLED;
2729 case IOMMU_ACPI_FINISHED:
2730 early_enable_iommus();
2731 x86_platform.iommu_shutdown = disable_iommus;
2732 init_state = IOMMU_ENABLED;
2735 register_syscore_ops(&amd_iommu_syscore_ops);
2736 ret = amd_iommu_init_pci();
2737 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2740 case IOMMU_PCI_INIT:
2741 ret = amd_iommu_enable_interrupts();
2742 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2744 case IOMMU_INTERRUPTS_EN:
2745 ret = amd_iommu_init_dma_ops();
2746 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2749 init_state = IOMMU_INITIALIZED;
2751 case IOMMU_INITIALIZED:
2754 case IOMMU_NOT_FOUND:
2755 case IOMMU_INIT_ERROR:
2756 case IOMMU_CMDLINE_DISABLED:
2757 /* Error states => do nothing */
2766 free_dma_resources();
2767 if (!irq_remapping_enabled) {
2769 free_iommu_resources();
2771 struct amd_iommu *iommu;
2773 uninit_device_table_dma();
2774 for_each_iommu(iommu)
2775 iommu_flush_all_caches(iommu);
2781 static int __init iommu_go_to_state(enum iommu_init_state state)
2785 while (init_state != state) {
2786 if (init_state == IOMMU_NOT_FOUND ||
2787 init_state == IOMMU_INIT_ERROR ||
2788 init_state == IOMMU_CMDLINE_DISABLED)
2796 #ifdef CONFIG_IRQ_REMAP
2797 int __init amd_iommu_prepare(void)
2801 amd_iommu_irq_remap = true;
2803 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2806 return amd_iommu_irq_remap ? 0 : -ENODEV;
2809 int __init amd_iommu_enable(void)
2813 ret = iommu_go_to_state(IOMMU_ENABLED);
2817 irq_remapping_enabled = 1;
2818 return amd_iommu_xt_mode;
2821 void amd_iommu_disable(void)
2823 amd_iommu_suspend();
2826 int amd_iommu_reenable(int mode)
2833 int __init amd_iommu_enable_faulting(void)
2835 /* We enable MSI later when PCI is initialized */
2841 * This is the core init function for AMD IOMMU hardware in the system.
2842 * This function is called from the generic x86 DMA layer initialization
2845 static int __init amd_iommu_init(void)
2847 struct amd_iommu *iommu;
2850 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2851 #ifdef CONFIG_GART_IOMMU
2852 if (ret && list_empty(&amd_iommu_list)) {
2854 * We failed to initialize the AMD IOMMU - try fallback
2855 * to GART if possible.
2861 for_each_iommu(iommu)
2862 amd_iommu_debugfs_setup(iommu);
2867 static bool amd_iommu_sme_check(void)
2869 if (!sme_active() || (boot_cpu_data.x86 != 0x17))
2872 /* For Fam17h, a specific level of support is required */
2873 if (boot_cpu_data.microcode >= 0x08001205)
2876 if ((boot_cpu_data.microcode >= 0x08001126) &&
2877 (boot_cpu_data.microcode <= 0x080011ff))
2880 pr_notice("IOMMU not currently supported when SME is active\n");
2885 /****************************************************************************
2887 * Early detect code. This code runs at IOMMU detection time in the DMA
2888 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2891 ****************************************************************************/
2892 int __init amd_iommu_detect(void)
2896 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2899 if (!amd_iommu_sme_check())
2902 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2906 amd_iommu_detected = true;
2908 x86_init.iommu.iommu_init = amd_iommu_init;
2913 /****************************************************************************
2915 * Parsing functions for the AMD IOMMU specific kernel command line
2918 ****************************************************************************/
2920 static int __init parse_amd_iommu_dump(char *str)
2922 amd_iommu_dump = true;
2927 static int __init parse_amd_iommu_intr(char *str)
2929 for (; *str; ++str) {
2930 if (strncmp(str, "legacy", 6) == 0) {
2931 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2934 if (strncmp(str, "vapic", 5) == 0) {
2935 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2942 static int __init parse_amd_iommu_options(char *str)
2944 for (; *str; ++str) {
2945 if (strncmp(str, "fullflush", 9) == 0)
2946 amd_iommu_unmap_flush = true;
2947 if (strncmp(str, "off", 3) == 0)
2948 amd_iommu_disabled = true;
2949 if (strncmp(str, "force_isolation", 15) == 0)
2950 amd_iommu_force_isolation = true;
2956 static int __init parse_ivrs_ioapic(char *str)
2958 unsigned int bus, dev, fn;
2962 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2965 pr_err("Invalid command line: ivrs_ioapic%s\n", str);
2969 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2970 pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2975 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2977 cmdline_maps = true;
2978 i = early_ioapic_map_size++;
2979 early_ioapic_map[i].id = id;
2980 early_ioapic_map[i].devid = devid;
2981 early_ioapic_map[i].cmd_line = true;
2986 static int __init parse_ivrs_hpet(char *str)
2988 unsigned int bus, dev, fn;
2992 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2995 pr_err("Invalid command line: ivrs_hpet%s\n", str);
2999 if (early_hpet_map_size == EARLY_MAP_SIZE) {
3000 pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3005 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3007 cmdline_maps = true;
3008 i = early_hpet_map_size++;
3009 early_hpet_map[i].id = id;
3010 early_hpet_map[i].devid = devid;
3011 early_hpet_map[i].cmd_line = true;
3016 static int __init parse_ivrs_acpihid(char *str)
3019 char *hid, *uid, *p;
3020 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3023 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
3025 pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
3030 hid = strsep(&p, ":");
3033 if (!hid || !(*hid) || !uid) {
3034 pr_err("Invalid command line: hid or uid\n");
3038 i = early_acpihid_map_size++;
3039 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3040 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3041 early_acpihid_map[i].devid =
3042 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3043 early_acpihid_map[i].cmd_line = true;
3048 __setup("amd_iommu_dump", parse_amd_iommu_dump);
3049 __setup("amd_iommu=", parse_amd_iommu_options);
3050 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
3051 __setup("ivrs_ioapic", parse_ivrs_ioapic);
3052 __setup("ivrs_hpet", parse_ivrs_hpet);
3053 __setup("ivrs_acpihid", parse_ivrs_acpihid);
3055 IOMMU_INIT_FINISH(amd_iommu_detect,
3056 gart_iommu_hole_init,
3060 bool amd_iommu_v2_supported(void)
3062 return amd_iommu_v2_present;
3064 EXPORT_SYMBOL(amd_iommu_v2_supported);
3066 struct amd_iommu *get_amd_iommu(unsigned int idx)
3069 struct amd_iommu *iommu;
3071 for_each_iommu(iommu)
3076 EXPORT_SYMBOL(get_amd_iommu);
3078 /****************************************************************************
3080 * IOMMU EFR Performance Counter support functionality. This code allows
3081 * access to the IOMMU PC functionality.
3083 ****************************************************************************/
3085 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3087 struct amd_iommu *iommu = get_amd_iommu(idx);
3090 return iommu->max_banks;
3094 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3096 bool amd_iommu_pc_supported(void)
3098 return amd_iommu_pc_present;
3100 EXPORT_SYMBOL(amd_iommu_pc_supported);
3102 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3104 struct amd_iommu *iommu = get_amd_iommu(idx);
3107 return iommu->max_counters;
3111 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3113 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3114 u8 fxn, u64 *value, bool is_write)
3119 /* Make sure the IOMMU PC resource is available */
3120 if (!amd_iommu_pc_present)
3123 /* Check for valid iommu and pc register indexing */
3124 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3127 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3129 /* Limit the offset to the hw defined mmio region aperture */
3130 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3131 (iommu->max_counters << 8) | 0x28);
3132 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3133 (offset > max_offset_lim))
3137 u64 val = *value & GENMASK_ULL(47, 0);
3139 writel((u32)val, iommu->mmio_base + offset);
3140 writel((val >> 32), iommu->mmio_base + offset + 4);
3142 *value = readl(iommu->mmio_base + offset + 4);
3144 *value |= readl(iommu->mmio_base + offset);
3145 *value &= GENMASK_ULL(47, 0);
3151 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3156 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3158 EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3160 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3165 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3167 EXPORT_SYMBOL(amd_iommu_pc_set_reg);