2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
4 * Leo Duran <leo.duran@amd.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #define pr_fmt(fmt) "AMD-Vi: " fmt
21 #define dev_fmt(fmt) pr_fmt(fmt)
23 #include <linux/pci.h>
24 #include <linux/acpi.h>
25 #include <linux/list.h>
26 #include <linux/bitmap.h>
27 #include <linux/slab.h>
28 #include <linux/syscore_ops.h>
29 #include <linux/interrupt.h>
30 #include <linux/msi.h>
31 #include <linux/amd-iommu.h>
32 #include <linux/export.h>
33 #include <linux/iommu.h>
34 #include <linux/kmemleak.h>
35 #include <linux/mem_encrypt.h>
36 #include <asm/pci-direct.h>
37 #include <asm/iommu.h>
39 #include <asm/x86_init.h>
40 #include <asm/iommu_table.h>
41 #include <asm/io_apic.h>
42 #include <asm/irq_remapping.h>
44 #include <linux/crash_dump.h>
45 #include "amd_iommu_proto.h"
46 #include "amd_iommu_types.h"
47 #include "irq_remapping.h"
50 * definitions for the ACPI scanning code
52 #define IVRS_HEADER_LENGTH 48
54 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
55 #define ACPI_IVMD_TYPE_ALL 0x20
56 #define ACPI_IVMD_TYPE 0x21
57 #define ACPI_IVMD_TYPE_RANGE 0x22
59 #define IVHD_DEV_ALL 0x01
60 #define IVHD_DEV_SELECT 0x02
61 #define IVHD_DEV_SELECT_RANGE_START 0x03
62 #define IVHD_DEV_RANGE_END 0x04
63 #define IVHD_DEV_ALIAS 0x42
64 #define IVHD_DEV_ALIAS_RANGE 0x43
65 #define IVHD_DEV_EXT_SELECT 0x46
66 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
67 #define IVHD_DEV_SPECIAL 0x48
68 #define IVHD_DEV_ACPI_HID 0xf0
70 #define UID_NOT_PRESENT 0
71 #define UID_IS_INTEGER 1
72 #define UID_IS_CHARACTER 2
74 #define IVHD_SPECIAL_IOAPIC 1
75 #define IVHD_SPECIAL_HPET 2
77 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
78 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
79 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
80 #define IVHD_FLAG_ISOC_EN_MASK 0x08
82 #define IVMD_FLAG_EXCL_RANGE 0x08
83 #define IVMD_FLAG_UNITY_MAP 0x01
85 #define ACPI_DEVFLAG_INITPASS 0x01
86 #define ACPI_DEVFLAG_EXTINT 0x02
87 #define ACPI_DEVFLAG_NMI 0x04
88 #define ACPI_DEVFLAG_SYSMGT1 0x10
89 #define ACPI_DEVFLAG_SYSMGT2 0x20
90 #define ACPI_DEVFLAG_LINT0 0x40
91 #define ACPI_DEVFLAG_LINT1 0x80
92 #define ACPI_DEVFLAG_ATSDIS 0x10000000
94 #define LOOP_TIMEOUT 100000
96 * ACPI table definitions
98 * These data structures are laid over the table to parse the important values
102 extern const struct iommu_ops amd_iommu_ops;
105 * structure describing one IOMMU in the ACPI table. Typically followed by one
106 * or more ivhd_entrys.
119 /* Following only valid on IVHD type 11h and 40h */
120 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
122 } __attribute__((packed));
125 * A device entry describing which devices a specific IOMMU translates and
126 * which requestor ids they use.
138 } __attribute__((packed));
141 * An AMD IOMMU memory definition structure. It defines things like exclusion
142 * ranges for devices and regions that should be unity mapped.
153 } __attribute__((packed));
156 bool amd_iommu_irq_remap __read_mostly;
158 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
159 static int amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
161 static bool amd_iommu_detected;
162 static bool __initdata amd_iommu_disabled;
163 static int amd_iommu_target_ivhd_type;
165 u16 amd_iommu_last_bdf; /* largest PCI device id we have
167 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
169 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
171 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
174 /* Array to assign indices to IOMMUs*/
175 struct amd_iommu *amd_iommus[MAX_IOMMUS];
177 /* Number of IOMMUs present in the system */
178 static int amd_iommus_present;
180 /* IOMMUs have a non-present cache? */
181 bool amd_iommu_np_cache __read_mostly;
182 bool amd_iommu_iotlb_sup __read_mostly = true;
184 u32 amd_iommu_max_pasid __read_mostly = ~0;
186 bool amd_iommu_v2_present __read_mostly;
187 static bool amd_iommu_pc_present __read_mostly;
189 bool amd_iommu_force_isolation __read_mostly;
192 * List of protection domains - used during resume
194 LIST_HEAD(amd_iommu_pd_list);
195 spinlock_t amd_iommu_pd_lock;
198 * Pointer to the device table which is shared by all AMD IOMMUs
199 * it is indexed by the PCI device id or the HT unit id and contains
200 * information about the domain the device belongs to as well as the
201 * page table root pointer.
203 struct dev_table_entry *amd_iommu_dev_table;
205 * Pointer to a device table which the content of old device table
206 * will be copied to. It's only be used in kdump kernel.
208 static struct dev_table_entry *old_dev_tbl_cpy;
211 * The alias table is a driver specific data structure which contains the
212 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
213 * More than one device can share the same requestor id.
215 u16 *amd_iommu_alias_table;
218 * The rlookup table is used to find the IOMMU which is responsible
219 * for a specific device. It is also indexed by the PCI device id.
221 struct amd_iommu **amd_iommu_rlookup_table;
222 EXPORT_SYMBOL(amd_iommu_rlookup_table);
225 * This table is used to find the irq remapping table for a given device id
228 struct irq_remap_table **irq_lookup_table;
231 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
232 * to know which ones are already in use.
234 unsigned long *amd_iommu_pd_alloc_bitmap;
236 static u32 dev_table_size; /* size of the device table */
237 static u32 alias_table_size; /* size of the alias table */
238 static u32 rlookup_table_size; /* size if the rlookup table */
240 enum iommu_init_state {
251 IOMMU_CMDLINE_DISABLED,
254 /* Early ioapic and hpet maps from kernel command line */
255 #define EARLY_MAP_SIZE 4
256 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
257 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
258 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
260 static int __initdata early_ioapic_map_size;
261 static int __initdata early_hpet_map_size;
262 static int __initdata early_acpihid_map_size;
264 static bool __initdata cmdline_maps;
266 static enum iommu_init_state init_state = IOMMU_START_STATE;
268 static int amd_iommu_enable_interrupts(void);
269 static int __init iommu_go_to_state(enum iommu_init_state state);
270 static void init_device_table_dma(void);
272 static bool amd_iommu_pre_enabled = true;
274 bool translation_pre_enabled(struct amd_iommu *iommu)
276 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
278 EXPORT_SYMBOL(translation_pre_enabled);
280 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
282 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
285 static void init_translation_status(struct amd_iommu *iommu)
289 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
290 if (ctrl & (1<<CONTROL_IOMMU_EN))
291 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
294 static inline void update_last_devid(u16 devid)
296 if (devid > amd_iommu_last_bdf)
297 amd_iommu_last_bdf = devid;
300 static inline unsigned long tbl_size(int entry_size)
302 unsigned shift = PAGE_SHIFT +
303 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
308 int amd_iommu_get_num_iommus(void)
310 return amd_iommus_present;
313 /* Access to l1 and l2 indexed register spaces */
315 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
319 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
320 pci_read_config_dword(iommu->dev, 0xfc, &val);
324 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
326 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
327 pci_write_config_dword(iommu->dev, 0xfc, val);
328 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
331 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
335 pci_write_config_dword(iommu->dev, 0xf0, address);
336 pci_read_config_dword(iommu->dev, 0xf4, &val);
340 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
342 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
343 pci_write_config_dword(iommu->dev, 0xf4, val);
346 /****************************************************************************
348 * AMD IOMMU MMIO register space handling functions
350 * These functions are used to program the IOMMU device registers in
351 * MMIO space required for that driver.
353 ****************************************************************************/
356 * This function set the exclusion range in the IOMMU. DMA accesses to the
357 * exclusion range are passed through untranslated
359 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
361 u64 start = iommu->exclusion_start & PAGE_MASK;
362 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
365 if (!iommu->exclusion_start)
368 entry = start | MMIO_EXCL_ENABLE_MASK;
369 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
370 &entry, sizeof(entry));
373 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
374 &entry, sizeof(entry));
377 /* Programs the physical address of the device table into the IOMMU hardware */
378 static void iommu_set_device_table(struct amd_iommu *iommu)
382 BUG_ON(iommu->mmio_base == NULL);
384 entry = iommu_virt_to_phys(amd_iommu_dev_table);
385 entry |= (dev_table_size >> 12) - 1;
386 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
387 &entry, sizeof(entry));
390 /* Generic functions to enable/disable certain features of the IOMMU. */
391 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
395 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
396 ctrl |= (1ULL << bit);
397 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
400 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
404 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
405 ctrl &= ~(1ULL << bit);
406 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
409 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
413 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
414 ctrl &= ~CTRL_INV_TO_MASK;
415 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
416 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
419 /* Function to enable the hardware */
420 static void iommu_enable(struct amd_iommu *iommu)
422 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
425 static void iommu_disable(struct amd_iommu *iommu)
427 /* Disable command buffer */
428 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
430 /* Disable event logging and event interrupts */
431 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
432 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
434 /* Disable IOMMU GA_LOG */
435 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
436 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
438 /* Disable IOMMU hardware itself */
439 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
443 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
444 * the system has one.
446 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
448 if (!request_mem_region(address, end, "amd_iommu")) {
449 pr_err("Can not reserve memory region %llx-%llx for mmio\n",
451 pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
455 return (u8 __iomem *)ioremap_nocache(address, end);
458 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
460 if (iommu->mmio_base)
461 iounmap(iommu->mmio_base);
462 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
465 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
481 /****************************************************************************
483 * The functions below belong to the first pass of AMD IOMMU ACPI table
484 * parsing. In this pass we try to find out the highest device id this
485 * code has to handle. Upon this information the size of the shared data
486 * structures is determined later.
488 ****************************************************************************/
491 * This function calculates the length of a given IVHD entry
493 static inline int ivhd_entry_length(u8 *ivhd)
495 u32 type = ((struct ivhd_entry *)ivhd)->type;
498 return 0x04 << (*ivhd >> 6);
499 } else if (type == IVHD_DEV_ACPI_HID) {
500 /* For ACPI_HID, offset 21 is uid len */
501 return *((u8 *)ivhd + 21) + 22;
507 * After reading the highest device id from the IOMMU PCI capability header
508 * this function looks if there is a higher device id defined in the ACPI table
510 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
512 u8 *p = (void *)h, *end = (void *)h;
513 struct ivhd_entry *dev;
515 u32 ivhd_size = get_ivhd_header_size(h);
518 pr_err("Unsupported IVHD type %#x\n", h->type);
526 dev = (struct ivhd_entry *)p;
529 /* Use maximum BDF value for DEV_ALL */
530 update_last_devid(0xffff);
532 case IVHD_DEV_SELECT:
533 case IVHD_DEV_RANGE_END:
535 case IVHD_DEV_EXT_SELECT:
536 /* all the above subfield types refer to device ids */
537 update_last_devid(dev->devid);
542 p += ivhd_entry_length(p);
550 static int __init check_ivrs_checksum(struct acpi_table_header *table)
553 u8 checksum = 0, *p = (u8 *)table;
555 for (i = 0; i < table->length; ++i)
558 /* ACPI table corrupt */
559 pr_err(FW_BUG "IVRS invalid checksum\n");
567 * Iterate over all IVHD entries in the ACPI table and find the highest device
568 * id which we need to handle. This is the first of three functions which parse
569 * the ACPI table. So we check the checksum here.
571 static int __init find_last_devid_acpi(struct acpi_table_header *table)
573 u8 *p = (u8 *)table, *end = (u8 *)table;
574 struct ivhd_header *h;
576 p += IVRS_HEADER_LENGTH;
578 end += table->length;
580 h = (struct ivhd_header *)p;
581 if (h->type == amd_iommu_target_ivhd_type) {
582 int ret = find_last_devid_from_ivhd(h);
594 /****************************************************************************
596 * The following functions belong to the code path which parses the ACPI table
597 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
598 * data structures, initialize the device/alias/rlookup table and also
599 * basically initialize the hardware.
601 ****************************************************************************/
604 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
605 * write commands to that buffer later and the IOMMU will execute them
608 static int __init alloc_command_buffer(struct amd_iommu *iommu)
610 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
611 get_order(CMD_BUFFER_SIZE));
613 return iommu->cmd_buf ? 0 : -ENOMEM;
617 * This function resets the command buffer if the IOMMU stopped fetching
620 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
622 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
624 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
625 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
626 iommu->cmd_buf_head = 0;
627 iommu->cmd_buf_tail = 0;
629 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
633 * This function writes the command buffer address to the hardware and
636 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
640 BUG_ON(iommu->cmd_buf == NULL);
642 entry = iommu_virt_to_phys(iommu->cmd_buf);
643 entry |= MMIO_CMD_SIZE_512;
645 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
646 &entry, sizeof(entry));
648 amd_iommu_reset_cmd_buffer(iommu);
652 * This function disables the command buffer
654 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
656 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
659 static void __init free_command_buffer(struct amd_iommu *iommu)
661 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
664 /* allocates the memory where the IOMMU will log its events to */
665 static int __init alloc_event_buffer(struct amd_iommu *iommu)
667 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
668 get_order(EVT_BUFFER_SIZE));
670 return iommu->evt_buf ? 0 : -ENOMEM;
673 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
677 BUG_ON(iommu->evt_buf == NULL);
679 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
681 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
682 &entry, sizeof(entry));
684 /* set head and tail to zero manually */
685 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
686 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
688 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
692 * This function disables the event log buffer
694 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
696 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
699 static void __init free_event_buffer(struct amd_iommu *iommu)
701 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
704 /* allocates the memory where the IOMMU will log its events to */
705 static int __init alloc_ppr_log(struct amd_iommu *iommu)
707 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
708 get_order(PPR_LOG_SIZE));
710 return iommu->ppr_log ? 0 : -ENOMEM;
713 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
717 if (iommu->ppr_log == NULL)
720 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
722 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
723 &entry, sizeof(entry));
725 /* set head and tail to zero manually */
726 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
727 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
729 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
730 iommu_feature_enable(iommu, CONTROL_PPR_EN);
733 static void __init free_ppr_log(struct amd_iommu *iommu)
735 if (iommu->ppr_log == NULL)
738 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
741 static void free_ga_log(struct amd_iommu *iommu)
743 #ifdef CONFIG_IRQ_REMAP
745 free_pages((unsigned long)iommu->ga_log,
746 get_order(GA_LOG_SIZE));
747 if (iommu->ga_log_tail)
748 free_pages((unsigned long)iommu->ga_log_tail,
753 static int iommu_ga_log_enable(struct amd_iommu *iommu)
755 #ifdef CONFIG_IRQ_REMAP
761 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
763 /* Check if already running */
764 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
767 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
768 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
770 for (i = 0; i < LOOP_TIMEOUT; ++i) {
771 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
772 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
776 if (i >= LOOP_TIMEOUT)
778 #endif /* CONFIG_IRQ_REMAP */
782 #ifdef CONFIG_IRQ_REMAP
783 static int iommu_init_ga_log(struct amd_iommu *iommu)
787 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
790 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
791 get_order(GA_LOG_SIZE));
795 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
797 if (!iommu->ga_log_tail)
800 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
801 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
802 &entry, sizeof(entry));
803 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
804 (BIT_ULL(52)-1)) & ~7ULL;
805 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
806 &entry, sizeof(entry));
807 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
808 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
815 #endif /* CONFIG_IRQ_REMAP */
817 static int iommu_init_ga(struct amd_iommu *iommu)
821 #ifdef CONFIG_IRQ_REMAP
822 /* Note: We have already checked GASup from IVRS table.
823 * Now, we need to make sure that GAMSup is set.
825 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
826 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
827 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
829 ret = iommu_init_ga_log(iommu);
830 #endif /* CONFIG_IRQ_REMAP */
835 static void iommu_enable_xt(struct amd_iommu *iommu)
837 #ifdef CONFIG_IRQ_REMAP
839 * XT mode (32-bit APIC destination ID) requires
840 * GA mode (128-bit IRTE support) as a prerequisite.
842 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
843 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
844 iommu_feature_enable(iommu, CONTROL_XT_EN);
845 #endif /* CONFIG_IRQ_REMAP */
848 static void iommu_enable_gt(struct amd_iommu *iommu)
850 if (!iommu_feature(iommu, FEATURE_GT))
853 iommu_feature_enable(iommu, CONTROL_GT_EN);
856 /* sets a specific bit in the device table entry. */
857 static void set_dev_entry_bit(u16 devid, u8 bit)
859 int i = (bit >> 6) & 0x03;
860 int _bit = bit & 0x3f;
862 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
865 static int get_dev_entry_bit(u16 devid, u8 bit)
867 int i = (bit >> 6) & 0x03;
868 int _bit = bit & 0x3f;
870 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
874 static bool copy_device_table(void)
876 u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
877 struct dev_table_entry *old_devtb = NULL;
878 u32 lo, hi, devid, old_devtb_size;
879 phys_addr_t old_devtb_phys;
880 struct amd_iommu *iommu;
881 u16 dom_id, dte_v, irq_v;
885 if (!amd_iommu_pre_enabled)
888 pr_warn("Translation is already enabled - trying to copy translation structures\n");
889 for_each_iommu(iommu) {
890 /* All IOMMUs should use the same device table with the same size */
891 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
892 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
893 entry = (((u64) hi) << 32) + lo;
894 if (last_entry && last_entry != entry) {
895 pr_err("IOMMU:%d should use the same dev table as others!\n",
901 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
902 if (old_devtb_size != dev_table_size) {
903 pr_err("The device table size of IOMMU:%d is not expected!\n",
910 * When SME is enabled in the first kernel, the entry includes the
911 * memory encryption mask(sme_me_mask), we must remove the memory
912 * encryption mask to obtain the true physical address in kdump kernel.
914 old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
916 if (old_devtb_phys >= 0x100000000ULL) {
917 pr_err("The address of old device table is above 4G, not trustworthy!\n");
920 old_devtb = (sme_active() && is_kdump_kernel())
921 ? (__force void *)ioremap_encrypted(old_devtb_phys,
923 : memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
928 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
929 old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
930 get_order(dev_table_size));
931 if (old_dev_tbl_cpy == NULL) {
932 pr_err("Failed to allocate memory for copying old device table!\n");
936 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
937 old_dev_tbl_cpy[devid] = old_devtb[devid];
938 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
939 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
941 if (dte_v && dom_id) {
942 old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
943 old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
944 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
945 /* If gcr3 table existed, mask it out */
946 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
947 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
948 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
949 old_dev_tbl_cpy[devid].data[1] &= ~tmp;
950 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
952 old_dev_tbl_cpy[devid].data[0] &= ~tmp;
956 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
957 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
958 int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
959 if (irq_v && (int_ctl || int_tab_len)) {
960 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
961 (int_tab_len != DTE_IRQ_TABLE_LEN)) {
962 pr_err("Wrong old irq remapping flag: %#x\n", devid);
966 old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
974 void amd_iommu_apply_erratum_63(u16 devid)
978 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
979 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
982 set_dev_entry_bit(devid, DEV_ENTRY_IW);
985 /* Writes the specific IOMMU for a device into the rlookup table */
986 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
988 amd_iommu_rlookup_table[devid] = iommu;
992 * This function takes the device specific flags read from the ACPI
993 * table and sets up the device table entry with that information
995 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
996 u16 devid, u32 flags, u32 ext_flags)
998 if (flags & ACPI_DEVFLAG_INITPASS)
999 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
1000 if (flags & ACPI_DEVFLAG_EXTINT)
1001 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
1002 if (flags & ACPI_DEVFLAG_NMI)
1003 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
1004 if (flags & ACPI_DEVFLAG_SYSMGT1)
1005 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
1006 if (flags & ACPI_DEVFLAG_SYSMGT2)
1007 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
1008 if (flags & ACPI_DEVFLAG_LINT0)
1009 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
1010 if (flags & ACPI_DEVFLAG_LINT1)
1011 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
1013 amd_iommu_apply_erratum_63(devid);
1015 set_iommu_for_device(iommu, devid);
1018 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1020 struct devid_map *entry;
1021 struct list_head *list;
1023 if (type == IVHD_SPECIAL_IOAPIC)
1025 else if (type == IVHD_SPECIAL_HPET)
1030 list_for_each_entry(entry, list, list) {
1031 if (!(entry->id == id && entry->cmd_line))
1034 pr_info("Command-line override present for %s id %d - ignoring\n",
1035 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1037 *devid = entry->devid;
1042 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1047 entry->devid = *devid;
1048 entry->cmd_line = cmd_line;
1050 list_add_tail(&entry->list, list);
1055 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1058 struct acpihid_map_entry *entry;
1059 struct list_head *list = &acpihid_map;
1061 list_for_each_entry(entry, list, list) {
1062 if (strcmp(entry->hid, hid) ||
1063 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1067 pr_info("Command-line override for hid:%s uid:%s\n",
1069 *devid = entry->devid;
1073 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1077 memcpy(entry->uid, uid, strlen(uid));
1078 memcpy(entry->hid, hid, strlen(hid));
1079 entry->devid = *devid;
1080 entry->cmd_line = cmd_line;
1081 entry->root_devid = (entry->devid & (~0x7));
1083 pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1084 entry->cmd_line ? "cmd" : "ivrs",
1085 entry->hid, entry->uid, entry->root_devid);
1087 list_add_tail(&entry->list, list);
1091 static int __init add_early_maps(void)
1095 for (i = 0; i < early_ioapic_map_size; ++i) {
1096 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1097 early_ioapic_map[i].id,
1098 &early_ioapic_map[i].devid,
1099 early_ioapic_map[i].cmd_line);
1104 for (i = 0; i < early_hpet_map_size; ++i) {
1105 ret = add_special_device(IVHD_SPECIAL_HPET,
1106 early_hpet_map[i].id,
1107 &early_hpet_map[i].devid,
1108 early_hpet_map[i].cmd_line);
1113 for (i = 0; i < early_acpihid_map_size; ++i) {
1114 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1115 early_acpihid_map[i].uid,
1116 &early_acpihid_map[i].devid,
1117 early_acpihid_map[i].cmd_line);
1126 * Reads the device exclusion range from ACPI and initializes the IOMMU with
1129 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
1131 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
1133 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
1138 * We only can configure exclusion ranges per IOMMU, not
1139 * per device. But we can enable the exclusion range per
1140 * device. This is done here
1142 set_dev_entry_bit(devid, DEV_ENTRY_EX);
1143 iommu->exclusion_start = m->range_start;
1144 iommu->exclusion_length = m->range_length;
1149 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1150 * initializes the hardware and our data structures with it.
1152 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1153 struct ivhd_header *h)
1156 u8 *end = p, flags = 0;
1157 u16 devid = 0, devid_start = 0, devid_to = 0;
1158 u32 dev_i, ext_flags = 0;
1160 struct ivhd_entry *e;
1165 ret = add_early_maps();
1170 * First save the recommended feature enable bits from ACPI
1172 iommu->acpi_flags = h->flags;
1175 * Done. Now parse the device entries
1177 ivhd_size = get_ivhd_header_size(h);
1179 pr_err("Unsupported IVHD type %#x\n", h->type);
1189 e = (struct ivhd_entry *)p;
1193 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1195 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1196 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1198 case IVHD_DEV_SELECT:
1200 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1202 PCI_BUS_NUM(e->devid),
1208 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1210 case IVHD_DEV_SELECT_RANGE_START:
1212 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1213 "devid: %02x:%02x.%x flags: %02x\n",
1214 PCI_BUS_NUM(e->devid),
1219 devid_start = e->devid;
1224 case IVHD_DEV_ALIAS:
1226 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1227 "flags: %02x devid_to: %02x:%02x.%x\n",
1228 PCI_BUS_NUM(e->devid),
1232 PCI_BUS_NUM(e->ext >> 8),
1233 PCI_SLOT(e->ext >> 8),
1234 PCI_FUNC(e->ext >> 8));
1237 devid_to = e->ext >> 8;
1238 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1239 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1240 amd_iommu_alias_table[devid] = devid_to;
1242 case IVHD_DEV_ALIAS_RANGE:
1244 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1245 "devid: %02x:%02x.%x flags: %02x "
1246 "devid_to: %02x:%02x.%x\n",
1247 PCI_BUS_NUM(e->devid),
1251 PCI_BUS_NUM(e->ext >> 8),
1252 PCI_SLOT(e->ext >> 8),
1253 PCI_FUNC(e->ext >> 8));
1255 devid_start = e->devid;
1257 devid_to = e->ext >> 8;
1261 case IVHD_DEV_EXT_SELECT:
1263 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1264 "flags: %02x ext: %08x\n",
1265 PCI_BUS_NUM(e->devid),
1271 set_dev_entry_from_acpi(iommu, devid, e->flags,
1274 case IVHD_DEV_EXT_SELECT_RANGE:
1276 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1277 "%02x:%02x.%x flags: %02x ext: %08x\n",
1278 PCI_BUS_NUM(e->devid),
1283 devid_start = e->devid;
1288 case IVHD_DEV_RANGE_END:
1290 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1291 PCI_BUS_NUM(e->devid),
1293 PCI_FUNC(e->devid));
1296 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1298 amd_iommu_alias_table[dev_i] = devid_to;
1299 set_dev_entry_from_acpi(iommu,
1300 devid_to, flags, ext_flags);
1302 set_dev_entry_from_acpi(iommu, dev_i,
1306 case IVHD_DEV_SPECIAL: {
1312 handle = e->ext & 0xff;
1313 devid = (e->ext >> 8) & 0xffff;
1314 type = (e->ext >> 24) & 0xff;
1316 if (type == IVHD_SPECIAL_IOAPIC)
1318 else if (type == IVHD_SPECIAL_HPET)
1323 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1329 ret = add_special_device(type, handle, &devid, false);
1334 * add_special_device might update the devid in case a
1335 * command-line override is present. So call
1336 * set_dev_entry_from_acpi after add_special_device.
1338 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1342 case IVHD_DEV_ACPI_HID: {
1344 u8 hid[ACPIHID_HID_LEN] = {0};
1345 u8 uid[ACPIHID_UID_LEN] = {0};
1348 if (h->type != 0x40) {
1349 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1354 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1355 hid[ACPIHID_HID_LEN - 1] = '\0';
1358 pr_err(FW_BUG "Invalid HID.\n");
1363 case UID_NOT_PRESENT:
1366 pr_warn(FW_BUG "Invalid UID length.\n");
1369 case UID_IS_INTEGER:
1371 sprintf(uid, "%d", e->uid);
1374 case UID_IS_CHARACTER:
1376 memcpy(uid, (u8 *)(&e->uid), ACPIHID_UID_LEN - 1);
1377 uid[ACPIHID_UID_LEN - 1] = '\0';
1385 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1393 ret = add_acpi_hid_device(hid, uid, &devid, false);
1398 * add_special_device might update the devid in case a
1399 * command-line override is present. So call
1400 * set_dev_entry_from_acpi after add_special_device.
1402 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1410 p += ivhd_entry_length(p);
1416 static void __init free_iommu_one(struct amd_iommu *iommu)
1418 free_command_buffer(iommu);
1419 free_event_buffer(iommu);
1420 free_ppr_log(iommu);
1422 iommu_unmap_mmio_space(iommu);
1425 static void __init free_iommu_all(void)
1427 struct amd_iommu *iommu, *next;
1429 for_each_iommu_safe(iommu, next) {
1430 list_del(&iommu->list);
1431 free_iommu_one(iommu);
1437 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1439 * BIOS should disable L2B micellaneous clock gating by setting
1440 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1442 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1446 if ((boot_cpu_data.x86 != 0x15) ||
1447 (boot_cpu_data.x86_model < 0x10) ||
1448 (boot_cpu_data.x86_model > 0x1f))
1451 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1452 pci_read_config_dword(iommu->dev, 0xf4, &value);
1457 /* Select NB indirect register 0x90 and enable writing */
1458 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1460 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1461 pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1463 /* Clear the enable writing bit */
1464 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1468 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1470 * BIOS should enable ATS write permission check by setting
1471 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1473 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1477 if ((boot_cpu_data.x86 != 0x15) ||
1478 (boot_cpu_data.x86_model < 0x30) ||
1479 (boot_cpu_data.x86_model > 0x3f))
1482 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1483 value = iommu_read_l2(iommu, 0x47);
1488 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1489 iommu_write_l2(iommu, 0x47, value | BIT(0));
1491 pci_info(iommu->dev, "Applying ATS write check workaround\n");
1495 * This function clues the initialization function for one IOMMU
1496 * together and also allocates the command buffer and programs the
1497 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1499 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1503 raw_spin_lock_init(&iommu->lock);
1505 /* Add IOMMU to internal data structures */
1506 list_add_tail(&iommu->list, &amd_iommu_list);
1507 iommu->index = amd_iommus_present++;
1509 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1510 WARN(1, "System has more IOMMUs than supported by this driver\n");
1514 /* Index is fine - add IOMMU to the array */
1515 amd_iommus[iommu->index] = iommu;
1518 * Copy data from ACPI table entry to the iommu struct
1520 iommu->devid = h->devid;
1521 iommu->cap_ptr = h->cap_ptr;
1522 iommu->pci_seg = h->pci_seg;
1523 iommu->mmio_phys = h->mmio_phys;
1527 /* Check if IVHD EFR contains proper max banks/counters */
1528 if ((h->efr_attr != 0) &&
1529 ((h->efr_attr & (0xF << 13)) != 0) &&
1530 ((h->efr_attr & (0x3F << 17)) != 0))
1531 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1533 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1534 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1535 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1536 if (((h->efr_attr & (0x1 << IOMMU_FEAT_XTSUP_SHIFT)) == 0))
1537 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1541 if (h->efr_reg & (1 << 9))
1542 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1544 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1545 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1546 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1547 if (((h->efr_reg & (0x1 << IOMMU_EFR_XTSUP_SHIFT)) == 0))
1548 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1554 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1555 iommu->mmio_phys_end);
1556 if (!iommu->mmio_base)
1559 if (alloc_command_buffer(iommu))
1562 if (alloc_event_buffer(iommu))
1565 iommu->int_enabled = false;
1567 init_translation_status(iommu);
1568 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1569 iommu_disable(iommu);
1570 clear_translation_pre_enabled(iommu);
1571 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1574 if (amd_iommu_pre_enabled)
1575 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1577 ret = init_iommu_from_acpi(iommu, h);
1581 ret = amd_iommu_create_irq_domain(iommu);
1586 * Make sure IOMMU is not considered to translate itself. The IVRS
1587 * table tells us so, but this is a lie!
1589 amd_iommu_rlookup_table[iommu->devid] = NULL;
1595 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1596 * @ivrs Pointer to the IVRS header
1598 * This function search through all IVDB of the maximum supported IVHD
1600 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1602 u8 *base = (u8 *)ivrs;
1603 struct ivhd_header *ivhd = (struct ivhd_header *)
1604 (base + IVRS_HEADER_LENGTH);
1605 u8 last_type = ivhd->type;
1606 u16 devid = ivhd->devid;
1608 while (((u8 *)ivhd - base < ivrs->length) &&
1609 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1610 u8 *p = (u8 *) ivhd;
1612 if (ivhd->devid == devid)
1613 last_type = ivhd->type;
1614 ivhd = (struct ivhd_header *)(p + ivhd->length);
1621 * Iterates over all IOMMU entries in the ACPI table, allocates the
1622 * IOMMU structure and initializes it with init_iommu_one()
1624 static int __init init_iommu_all(struct acpi_table_header *table)
1626 u8 *p = (u8 *)table, *end = (u8 *)table;
1627 struct ivhd_header *h;
1628 struct amd_iommu *iommu;
1631 end += table->length;
1632 p += IVRS_HEADER_LENGTH;
1635 h = (struct ivhd_header *)p;
1636 if (*p == amd_iommu_target_ivhd_type) {
1638 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1639 "seg: %d flags: %01x info %04x\n",
1640 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1641 PCI_FUNC(h->devid), h->cap_ptr,
1642 h->pci_seg, h->flags, h->info);
1643 DUMP_printk(" mmio-addr: %016llx\n",
1646 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1650 ret = init_iommu_one(iommu, h);
1662 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1663 u8 fxn, u64 *value, bool is_write);
1665 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1667 struct pci_dev *pdev = iommu->dev;
1668 u64 val = 0xabcd, val2 = 0;
1670 if (!iommu_feature(iommu, FEATURE_PC))
1673 amd_iommu_pc_present = true;
1675 /* Check if the performance counters can be written to */
1676 if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1677 (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1679 pci_err(pdev, "Unable to write to IOMMU perf counter.\n");
1680 amd_iommu_pc_present = false;
1684 pci_info(pdev, "IOMMU performance counters supported\n");
1686 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1687 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1688 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1691 static ssize_t amd_iommu_show_cap(struct device *dev,
1692 struct device_attribute *attr,
1695 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1696 return sprintf(buf, "%x\n", iommu->cap);
1698 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1700 static ssize_t amd_iommu_show_features(struct device *dev,
1701 struct device_attribute *attr,
1704 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1705 return sprintf(buf, "%llx\n", iommu->features);
1707 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1709 static struct attribute *amd_iommu_attrs[] = {
1711 &dev_attr_features.attr,
1715 static struct attribute_group amd_iommu_group = {
1716 .name = "amd-iommu",
1717 .attrs = amd_iommu_attrs,
1720 static const struct attribute_group *amd_iommu_groups[] = {
1725 static int __init iommu_init_pci(struct amd_iommu *iommu)
1727 int cap_ptr = iommu->cap_ptr;
1728 u32 range, misc, low, high;
1731 iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1732 iommu->devid & 0xff);
1736 /* Prevent binding other PCI device drivers to IOMMU devices */
1737 iommu->dev->match_driver = false;
1739 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1741 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1743 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1746 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1747 amd_iommu_iotlb_sup = false;
1749 /* read extended feature bits */
1750 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1751 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1753 iommu->features = ((u64)high << 32) | low;
1755 if (iommu_feature(iommu, FEATURE_GT)) {
1760 pasmax = iommu->features & FEATURE_PASID_MASK;
1761 pasmax >>= FEATURE_PASID_SHIFT;
1762 max_pasid = (1 << (pasmax + 1)) - 1;
1764 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1766 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1768 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1769 glxval >>= FEATURE_GLXVAL_SHIFT;
1771 if (amd_iommu_max_glx_val == -1)
1772 amd_iommu_max_glx_val = glxval;
1774 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1777 if (iommu_feature(iommu, FEATURE_GT) &&
1778 iommu_feature(iommu, FEATURE_PPR)) {
1779 iommu->is_iommu_v2 = true;
1780 amd_iommu_v2_present = true;
1783 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1786 ret = iommu_init_ga(iommu);
1790 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1791 amd_iommu_np_cache = true;
1793 init_iommu_perf_ctr(iommu);
1795 if (is_rd890_iommu(iommu->dev)) {
1799 pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
1803 * Some rd890 systems may not be fully reconfigured by the
1804 * BIOS, so it's necessary for us to store this information so
1805 * it can be reprogrammed on resume
1807 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1808 &iommu->stored_addr_lo);
1809 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1810 &iommu->stored_addr_hi);
1812 /* Low bit locks writes to configuration space */
1813 iommu->stored_addr_lo &= ~1;
1815 for (i = 0; i < 6; i++)
1816 for (j = 0; j < 0x12; j++)
1817 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1819 for (i = 0; i < 0x83; i++)
1820 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1823 amd_iommu_erratum_746_workaround(iommu);
1824 amd_iommu_ats_write_check_workaround(iommu);
1826 iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1827 amd_iommu_groups, "ivhd%d", iommu->index);
1828 iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1829 iommu_device_register(&iommu->iommu);
1831 return pci_enable_device(iommu->dev);
1834 static void print_iommu_info(void)
1836 static const char * const feat_str[] = {
1837 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1838 "IA", "GA", "HE", "PC"
1840 struct amd_iommu *iommu;
1842 for_each_iommu(iommu) {
1843 struct pci_dev *pdev = iommu->dev;
1846 pci_info(pdev, "Found IOMMU cap 0x%hx\n", iommu->cap_ptr);
1848 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1849 pci_info(pdev, "Extended features (%#llx):\n",
1851 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1852 if (iommu_feature(iommu, (1ULL << i)))
1853 pr_cont(" %s", feat_str[i]);
1856 if (iommu->features & FEATURE_GAM_VAPIC)
1857 pr_cont(" GA_vAPIC");
1862 if (irq_remapping_enabled) {
1863 pr_info("Interrupt remapping enabled\n");
1864 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1865 pr_info("Virtual APIC enabled\n");
1866 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1867 pr_info("X2APIC enabled\n");
1871 static int __init amd_iommu_init_pci(void)
1873 struct amd_iommu *iommu;
1876 for_each_iommu(iommu) {
1877 ret = iommu_init_pci(iommu);
1883 * Order is important here to make sure any unity map requirements are
1884 * fulfilled. The unity mappings are created and written to the device
1885 * table during the amd_iommu_init_api() call.
1887 * After that we call init_device_table_dma() to make sure any
1888 * uninitialized DTE will block DMA, and in the end we flush the caches
1889 * of all IOMMUs to make sure the changes to the device table are
1892 ret = amd_iommu_init_api();
1894 init_device_table_dma();
1896 for_each_iommu(iommu)
1897 iommu_flush_all_caches(iommu);
1905 /****************************************************************************
1907 * The following functions initialize the MSI interrupts for all IOMMUs
1908 * in the system. It's a bit challenging because there could be multiple
1909 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1912 ****************************************************************************/
1914 static int iommu_setup_msi(struct amd_iommu *iommu)
1918 r = pci_enable_msi(iommu->dev);
1922 r = request_threaded_irq(iommu->dev->irq,
1923 amd_iommu_int_handler,
1924 amd_iommu_int_thread,
1929 pci_disable_msi(iommu->dev);
1933 iommu->int_enabled = true;
1938 static int iommu_init_msi(struct amd_iommu *iommu)
1942 if (iommu->int_enabled)
1945 if (iommu->dev->msi_cap)
1946 ret = iommu_setup_msi(iommu);
1954 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1956 if (iommu->ppr_log != NULL)
1957 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1959 iommu_ga_log_enable(iommu);
1964 /****************************************************************************
1966 * The next functions belong to the third pass of parsing the ACPI
1967 * table. In this last pass the memory mapping requirements are
1968 * gathered (like exclusion and unity mapping ranges).
1970 ****************************************************************************/
1972 static void __init free_unity_maps(void)
1974 struct unity_map_entry *entry, *next;
1976 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1977 list_del(&entry->list);
1982 /* called when we find an exclusion range definition in ACPI */
1983 static int __init init_exclusion_range(struct ivmd_header *m)
1988 case ACPI_IVMD_TYPE:
1989 set_device_exclusion_range(m->devid, m);
1991 case ACPI_IVMD_TYPE_ALL:
1992 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1993 set_device_exclusion_range(i, m);
1995 case ACPI_IVMD_TYPE_RANGE:
1996 for (i = m->devid; i <= m->aux; ++i)
1997 set_device_exclusion_range(i, m);
2006 /* called for unity map ACPI definition */
2007 static int __init init_unity_map_range(struct ivmd_header *m)
2009 struct unity_map_entry *e = NULL;
2012 e = kzalloc(sizeof(*e), GFP_KERNEL);
2016 if (m->flags & IVMD_FLAG_EXCL_RANGE)
2017 init_exclusion_range(m);
2023 case ACPI_IVMD_TYPE:
2024 s = "IVMD_TYPEi\t\t\t";
2025 e->devid_start = e->devid_end = m->devid;
2027 case ACPI_IVMD_TYPE_ALL:
2028 s = "IVMD_TYPE_ALL\t\t";
2030 e->devid_end = amd_iommu_last_bdf;
2032 case ACPI_IVMD_TYPE_RANGE:
2033 s = "IVMD_TYPE_RANGE\t\t";
2034 e->devid_start = m->devid;
2035 e->devid_end = m->aux;
2038 e->address_start = PAGE_ALIGN(m->range_start);
2039 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2040 e->prot = m->flags >> 1;
2042 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2043 " range_start: %016llx range_end: %016llx flags: %x\n", s,
2044 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2045 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
2046 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2047 e->address_start, e->address_end, m->flags);
2049 list_add_tail(&e->list, &amd_iommu_unity_map);
2054 /* iterates over all memory definitions we find in the ACPI table */
2055 static int __init init_memory_definitions(struct acpi_table_header *table)
2057 u8 *p = (u8 *)table, *end = (u8 *)table;
2058 struct ivmd_header *m;
2060 end += table->length;
2061 p += IVRS_HEADER_LENGTH;
2064 m = (struct ivmd_header *)p;
2065 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2066 init_unity_map_range(m);
2075 * Init the device table to not allow DMA access for devices
2077 static void init_device_table_dma(void)
2081 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2082 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2083 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2087 static void __init uninit_device_table_dma(void)
2091 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2092 amd_iommu_dev_table[devid].data[0] = 0ULL;
2093 amd_iommu_dev_table[devid].data[1] = 0ULL;
2097 static void init_device_table(void)
2101 if (!amd_iommu_irq_remap)
2104 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2105 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2108 static void iommu_init_flags(struct amd_iommu *iommu)
2110 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2111 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2112 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2114 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2115 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2116 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2118 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2119 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2120 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2122 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2123 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2124 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2127 * make IOMMU memory accesses cache coherent
2129 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2131 /* Set IOTLB invalidation timeout to 1s */
2132 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2135 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2138 u32 ioc_feature_control;
2139 struct pci_dev *pdev = iommu->root_pdev;
2141 /* RD890 BIOSes may not have completely reconfigured the iommu */
2142 if (!is_rd890_iommu(iommu->dev) || !pdev)
2146 * First, we need to ensure that the iommu is enabled. This is
2147 * controlled by a register in the northbridge
2150 /* Select Northbridge indirect register 0x75 and enable writing */
2151 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2152 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2154 /* Enable the iommu */
2155 if (!(ioc_feature_control & 0x1))
2156 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2158 /* Restore the iommu BAR */
2159 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2160 iommu->stored_addr_lo);
2161 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2162 iommu->stored_addr_hi);
2164 /* Restore the l1 indirect regs for each of the 6 l1s */
2165 for (i = 0; i < 6; i++)
2166 for (j = 0; j < 0x12; j++)
2167 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2169 /* Restore the l2 indirect regs */
2170 for (i = 0; i < 0x83; i++)
2171 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2173 /* Lock PCI setup registers */
2174 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2175 iommu->stored_addr_lo | 1);
2178 static void iommu_enable_ga(struct amd_iommu *iommu)
2180 #ifdef CONFIG_IRQ_REMAP
2181 switch (amd_iommu_guest_ir) {
2182 case AMD_IOMMU_GUEST_IR_VAPIC:
2183 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2185 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2186 iommu_feature_enable(iommu, CONTROL_GA_EN);
2187 iommu->irte_ops = &irte_128_ops;
2190 iommu->irte_ops = &irte_32_ops;
2196 static void early_enable_iommu(struct amd_iommu *iommu)
2198 iommu_disable(iommu);
2199 iommu_init_flags(iommu);
2200 iommu_set_device_table(iommu);
2201 iommu_enable_command_buffer(iommu);
2202 iommu_enable_event_buffer(iommu);
2203 iommu_set_exclusion_range(iommu);
2204 iommu_enable_ga(iommu);
2205 iommu_enable_xt(iommu);
2206 iommu_enable(iommu);
2207 iommu_flush_all_caches(iommu);
2211 * This function finally enables all IOMMUs found in the system after
2212 * they have been initialized.
2214 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2215 * the old content of device table entries. Not this case or copy failed,
2216 * just continue as normal kernel does.
2218 static void early_enable_iommus(void)
2220 struct amd_iommu *iommu;
2223 if (!copy_device_table()) {
2225 * If come here because of failure in copying device table from old
2226 * kernel with all IOMMUs enabled, print error message and try to
2227 * free allocated old_dev_tbl_cpy.
2229 if (amd_iommu_pre_enabled)
2230 pr_err("Failed to copy DEV table from previous kernel.\n");
2231 if (old_dev_tbl_cpy != NULL)
2232 free_pages((unsigned long)old_dev_tbl_cpy,
2233 get_order(dev_table_size));
2235 for_each_iommu(iommu) {
2236 clear_translation_pre_enabled(iommu);
2237 early_enable_iommu(iommu);
2240 pr_info("Copied DEV table from previous kernel.\n");
2241 free_pages((unsigned long)amd_iommu_dev_table,
2242 get_order(dev_table_size));
2243 amd_iommu_dev_table = old_dev_tbl_cpy;
2244 for_each_iommu(iommu) {
2245 iommu_disable_command_buffer(iommu);
2246 iommu_disable_event_buffer(iommu);
2247 iommu_enable_command_buffer(iommu);
2248 iommu_enable_event_buffer(iommu);
2249 iommu_enable_ga(iommu);
2250 iommu_enable_xt(iommu);
2251 iommu_set_device_table(iommu);
2252 iommu_flush_all_caches(iommu);
2256 #ifdef CONFIG_IRQ_REMAP
2257 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2258 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2262 static void enable_iommus_v2(void)
2264 struct amd_iommu *iommu;
2266 for_each_iommu(iommu) {
2267 iommu_enable_ppr_log(iommu);
2268 iommu_enable_gt(iommu);
2272 static void enable_iommus(void)
2274 early_enable_iommus();
2279 static void disable_iommus(void)
2281 struct amd_iommu *iommu;
2283 for_each_iommu(iommu)
2284 iommu_disable(iommu);
2286 #ifdef CONFIG_IRQ_REMAP
2287 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2288 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2293 * Suspend/Resume support
2294 * disable suspend until real resume implemented
2297 static void amd_iommu_resume(void)
2299 struct amd_iommu *iommu;
2301 for_each_iommu(iommu)
2302 iommu_apply_resume_quirks(iommu);
2304 /* re-load the hardware */
2307 amd_iommu_enable_interrupts();
2310 static int amd_iommu_suspend(void)
2312 /* disable IOMMUs to go out of the way for BIOS */
2318 static struct syscore_ops amd_iommu_syscore_ops = {
2319 .suspend = amd_iommu_suspend,
2320 .resume = amd_iommu_resume,
2323 static void __init free_iommu_resources(void)
2325 kmemleak_free(irq_lookup_table);
2326 free_pages((unsigned long)irq_lookup_table,
2327 get_order(rlookup_table_size));
2328 irq_lookup_table = NULL;
2330 kmem_cache_destroy(amd_iommu_irq_cache);
2331 amd_iommu_irq_cache = NULL;
2333 free_pages((unsigned long)amd_iommu_rlookup_table,
2334 get_order(rlookup_table_size));
2335 amd_iommu_rlookup_table = NULL;
2337 free_pages((unsigned long)amd_iommu_alias_table,
2338 get_order(alias_table_size));
2339 amd_iommu_alias_table = NULL;
2341 free_pages((unsigned long)amd_iommu_dev_table,
2342 get_order(dev_table_size));
2343 amd_iommu_dev_table = NULL;
2347 #ifdef CONFIG_GART_IOMMU
2349 * We failed to initialize the AMD IOMMU - try fallback to GART
2357 /* SB IOAPIC is always on this device in AMD systems */
2358 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2360 static bool __init check_ioapic_information(void)
2362 const char *fw_bug = FW_BUG;
2363 bool ret, has_sb_ioapic;
2366 has_sb_ioapic = false;
2370 * If we have map overrides on the kernel command line the
2371 * messages in this function might not describe firmware bugs
2372 * anymore - so be careful
2377 for (idx = 0; idx < nr_ioapics; idx++) {
2378 int devid, id = mpc_ioapic_id(idx);
2380 devid = get_ioapic_devid(id);
2382 pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2385 } else if (devid == IOAPIC_SB_DEVID) {
2386 has_sb_ioapic = true;
2391 if (!has_sb_ioapic) {
2393 * We expect the SB IOAPIC to be listed in the IVRS
2394 * table. The system timer is connected to the SB IOAPIC
2395 * and if we don't have it in the list the system will
2396 * panic at boot time. This situation usually happens
2397 * when the BIOS is buggy and provides us the wrong
2398 * device id for the IOAPIC in the system.
2400 pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2404 pr_err("Disabling interrupt remapping\n");
2409 static void __init free_dma_resources(void)
2411 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2412 get_order(MAX_DOMAIN_ID/8));
2413 amd_iommu_pd_alloc_bitmap = NULL;
2419 * This is the hardware init function for AMD IOMMU in the system.
2420 * This function is called either from amd_iommu_init or from the interrupt
2421 * remapping setup code.
2423 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2426 * 1 pass) Discover the most comprehensive IVHD type to use.
2428 * 2 pass) Find the highest PCI device id the driver has to handle.
2429 * Upon this information the size of the data structures is
2430 * determined that needs to be allocated.
2432 * 3 pass) Initialize the data structures just allocated with the
2433 * information in the ACPI table about available AMD IOMMUs
2434 * in the system. It also maps the PCI devices in the
2435 * system to specific IOMMUs
2437 * 4 pass) After the basic data structures are allocated and
2438 * initialized we update them with information about memory
2439 * remapping requirements parsed out of the ACPI table in
2442 * After everything is set up the IOMMUs are enabled and the necessary
2443 * hotplug and suspend notifiers are registered.
2445 static int __init early_amd_iommu_init(void)
2447 struct acpi_table_header *ivrs_base;
2449 int i, remap_cache_sz, ret = 0;
2451 if (!amd_iommu_detected)
2454 status = acpi_get_table("IVRS", 0, &ivrs_base);
2455 if (status == AE_NOT_FOUND)
2457 else if (ACPI_FAILURE(status)) {
2458 const char *err = acpi_format_exception(status);
2459 pr_err("IVRS table error: %s\n", err);
2464 * Validate checksum here so we don't need to do it when
2465 * we actually parse the table
2467 ret = check_ivrs_checksum(ivrs_base);
2471 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2472 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2475 * First parse ACPI tables to find the largest Bus/Dev/Func
2476 * we need to handle. Upon this information the shared data
2477 * structures for the IOMMUs in the system will be allocated
2479 ret = find_last_devid_acpi(ivrs_base);
2483 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2484 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2485 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2487 /* Device table - directly used by all IOMMUs */
2489 amd_iommu_dev_table = (void *)__get_free_pages(
2490 GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2491 get_order(dev_table_size));
2492 if (amd_iommu_dev_table == NULL)
2496 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2497 * IOMMU see for that device
2499 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2500 get_order(alias_table_size));
2501 if (amd_iommu_alias_table == NULL)
2504 /* IOMMU rlookup table - find the IOMMU for a specific device */
2505 amd_iommu_rlookup_table = (void *)__get_free_pages(
2506 GFP_KERNEL | __GFP_ZERO,
2507 get_order(rlookup_table_size));
2508 if (amd_iommu_rlookup_table == NULL)
2511 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2512 GFP_KERNEL | __GFP_ZERO,
2513 get_order(MAX_DOMAIN_ID/8));
2514 if (amd_iommu_pd_alloc_bitmap == NULL)
2518 * let all alias entries point to itself
2520 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2521 amd_iommu_alias_table[i] = i;
2524 * never allocate domain 0 because its used as the non-allocated and
2525 * error value placeholder
2527 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2529 spin_lock_init(&amd_iommu_pd_lock);
2532 * now the data structures are allocated and basically initialized
2533 * start the real acpi table scan
2535 ret = init_iommu_all(ivrs_base);
2539 /* Disable any previously enabled IOMMUs */
2540 if (!is_kdump_kernel() || amd_iommu_disabled)
2543 if (amd_iommu_irq_remap)
2544 amd_iommu_irq_remap = check_ioapic_information();
2546 if (amd_iommu_irq_remap) {
2548 * Interrupt remapping enabled, create kmem_cache for the
2552 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2553 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2555 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2556 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2558 IRQ_TABLE_ALIGNMENT,
2560 if (!amd_iommu_irq_cache)
2563 irq_lookup_table = (void *)__get_free_pages(
2564 GFP_KERNEL | __GFP_ZERO,
2565 get_order(rlookup_table_size));
2566 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2568 if (!irq_lookup_table)
2572 ret = init_memory_definitions(ivrs_base);
2576 /* init the device table */
2577 init_device_table();
2580 /* Don't leak any ACPI memory */
2581 acpi_put_table(ivrs_base);
2587 static int amd_iommu_enable_interrupts(void)
2589 struct amd_iommu *iommu;
2592 for_each_iommu(iommu) {
2593 ret = iommu_init_msi(iommu);
2602 static bool detect_ivrs(void)
2604 struct acpi_table_header *ivrs_base;
2607 status = acpi_get_table("IVRS", 0, &ivrs_base);
2608 if (status == AE_NOT_FOUND)
2610 else if (ACPI_FAILURE(status)) {
2611 const char *err = acpi_format_exception(status);
2612 pr_err("IVRS table error: %s\n", err);
2616 acpi_put_table(ivrs_base);
2618 /* Make sure ACS will be enabled during PCI probe */
2624 /****************************************************************************
2626 * AMD IOMMU Initialization State Machine
2628 ****************************************************************************/
2630 static int __init state_next(void)
2634 switch (init_state) {
2635 case IOMMU_START_STATE:
2636 if (!detect_ivrs()) {
2637 init_state = IOMMU_NOT_FOUND;
2640 init_state = IOMMU_IVRS_DETECTED;
2643 case IOMMU_IVRS_DETECTED:
2644 ret = early_amd_iommu_init();
2645 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2646 if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2647 pr_info("AMD IOMMU disabled on kernel command-line\n");
2648 free_dma_resources();
2649 free_iommu_resources();
2650 init_state = IOMMU_CMDLINE_DISABLED;
2654 case IOMMU_ACPI_FINISHED:
2655 early_enable_iommus();
2656 x86_platform.iommu_shutdown = disable_iommus;
2657 init_state = IOMMU_ENABLED;
2660 register_syscore_ops(&amd_iommu_syscore_ops);
2661 ret = amd_iommu_init_pci();
2662 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2665 case IOMMU_PCI_INIT:
2666 ret = amd_iommu_enable_interrupts();
2667 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2669 case IOMMU_INTERRUPTS_EN:
2670 ret = amd_iommu_init_dma_ops();
2671 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2674 init_state = IOMMU_INITIALIZED;
2676 case IOMMU_INITIALIZED:
2679 case IOMMU_NOT_FOUND:
2680 case IOMMU_INIT_ERROR:
2681 case IOMMU_CMDLINE_DISABLED:
2682 /* Error states => do nothing */
2693 static int __init iommu_go_to_state(enum iommu_init_state state)
2697 while (init_state != state) {
2698 if (init_state == IOMMU_NOT_FOUND ||
2699 init_state == IOMMU_INIT_ERROR ||
2700 init_state == IOMMU_CMDLINE_DISABLED)
2708 #ifdef CONFIG_IRQ_REMAP
2709 int __init amd_iommu_prepare(void)
2713 amd_iommu_irq_remap = true;
2715 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2718 return amd_iommu_irq_remap ? 0 : -ENODEV;
2721 int __init amd_iommu_enable(void)
2725 ret = iommu_go_to_state(IOMMU_ENABLED);
2729 irq_remapping_enabled = 1;
2730 return amd_iommu_xt_mode;
2733 void amd_iommu_disable(void)
2735 amd_iommu_suspend();
2738 int amd_iommu_reenable(int mode)
2745 int __init amd_iommu_enable_faulting(void)
2747 /* We enable MSI later when PCI is initialized */
2753 * This is the core init function for AMD IOMMU hardware in the system.
2754 * This function is called from the generic x86 DMA layer initialization
2757 static int __init amd_iommu_init(void)
2759 struct amd_iommu *iommu;
2762 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2764 free_dma_resources();
2765 if (!irq_remapping_enabled) {
2767 free_iommu_resources();
2769 uninit_device_table_dma();
2770 for_each_iommu(iommu)
2771 iommu_flush_all_caches(iommu);
2775 for_each_iommu(iommu)
2776 amd_iommu_debugfs_setup(iommu);
2781 static bool amd_iommu_sme_check(void)
2783 if (!sme_active() || (boot_cpu_data.x86 != 0x17))
2786 /* For Fam17h, a specific level of support is required */
2787 if (boot_cpu_data.microcode >= 0x08001205)
2790 if ((boot_cpu_data.microcode >= 0x08001126) &&
2791 (boot_cpu_data.microcode <= 0x080011ff))
2794 pr_notice("IOMMU not currently supported when SME is active\n");
2799 /****************************************************************************
2801 * Early detect code. This code runs at IOMMU detection time in the DMA
2802 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2805 ****************************************************************************/
2806 int __init amd_iommu_detect(void)
2810 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2813 if (!amd_iommu_sme_check())
2816 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2820 amd_iommu_detected = true;
2822 x86_init.iommu.iommu_init = amd_iommu_init;
2827 /****************************************************************************
2829 * Parsing functions for the AMD IOMMU specific kernel command line
2832 ****************************************************************************/
2834 static int __init parse_amd_iommu_dump(char *str)
2836 amd_iommu_dump = true;
2841 static int __init parse_amd_iommu_intr(char *str)
2843 for (; *str; ++str) {
2844 if (strncmp(str, "legacy", 6) == 0) {
2845 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
2848 if (strncmp(str, "vapic", 5) == 0) {
2849 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2856 static int __init parse_amd_iommu_options(char *str)
2858 for (; *str; ++str) {
2859 if (strncmp(str, "fullflush", 9) == 0)
2860 amd_iommu_unmap_flush = true;
2861 if (strncmp(str, "off", 3) == 0)
2862 amd_iommu_disabled = true;
2863 if (strncmp(str, "force_isolation", 15) == 0)
2864 amd_iommu_force_isolation = true;
2870 static int __init parse_ivrs_ioapic(char *str)
2872 unsigned int bus, dev, fn;
2876 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2879 pr_err("Invalid command line: ivrs_ioapic%s\n", str);
2883 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2884 pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2889 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2891 cmdline_maps = true;
2892 i = early_ioapic_map_size++;
2893 early_ioapic_map[i].id = id;
2894 early_ioapic_map[i].devid = devid;
2895 early_ioapic_map[i].cmd_line = true;
2900 static int __init parse_ivrs_hpet(char *str)
2902 unsigned int bus, dev, fn;
2906 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2909 pr_err("Invalid command line: ivrs_hpet%s\n", str);
2913 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2914 pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
2919 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2921 cmdline_maps = true;
2922 i = early_hpet_map_size++;
2923 early_hpet_map[i].id = id;
2924 early_hpet_map[i].devid = devid;
2925 early_hpet_map[i].cmd_line = true;
2930 static int __init parse_ivrs_acpihid(char *str)
2933 char *hid, *uid, *p;
2934 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
2937 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
2939 pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
2944 hid = strsep(&p, ":");
2947 if (!hid || !(*hid) || !uid) {
2948 pr_err("Invalid command line: hid or uid\n");
2952 i = early_acpihid_map_size++;
2953 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
2954 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
2955 early_acpihid_map[i].devid =
2956 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2957 early_acpihid_map[i].cmd_line = true;
2962 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2963 __setup("amd_iommu=", parse_amd_iommu_options);
2964 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
2965 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2966 __setup("ivrs_hpet", parse_ivrs_hpet);
2967 __setup("ivrs_acpihid", parse_ivrs_acpihid);
2969 IOMMU_INIT_FINISH(amd_iommu_detect,
2970 gart_iommu_hole_init,
2974 bool amd_iommu_v2_supported(void)
2976 return amd_iommu_v2_present;
2978 EXPORT_SYMBOL(amd_iommu_v2_supported);
2980 struct amd_iommu *get_amd_iommu(unsigned int idx)
2983 struct amd_iommu *iommu;
2985 for_each_iommu(iommu)
2990 EXPORT_SYMBOL(get_amd_iommu);
2992 /****************************************************************************
2994 * IOMMU EFR Performance Counter support functionality. This code allows
2995 * access to the IOMMU PC functionality.
2997 ****************************************************************************/
2999 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3001 struct amd_iommu *iommu = get_amd_iommu(idx);
3004 return iommu->max_banks;
3008 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3010 bool amd_iommu_pc_supported(void)
3012 return amd_iommu_pc_present;
3014 EXPORT_SYMBOL(amd_iommu_pc_supported);
3016 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3018 struct amd_iommu *iommu = get_amd_iommu(idx);
3021 return iommu->max_counters;
3025 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3027 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3028 u8 fxn, u64 *value, bool is_write)
3033 /* Make sure the IOMMU PC resource is available */
3034 if (!amd_iommu_pc_present)
3037 /* Check for valid iommu and pc register indexing */
3038 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3041 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3043 /* Limit the offset to the hw defined mmio region aperture */
3044 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3045 (iommu->max_counters << 8) | 0x28);
3046 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3047 (offset > max_offset_lim))
3051 u64 val = *value & GENMASK_ULL(47, 0);
3053 writel((u32)val, iommu->mmio_base + offset);
3054 writel((val >> 32), iommu->mmio_base + offset + 4);
3056 *value = readl(iommu->mmio_base + offset + 4);
3058 *value |= readl(iommu->mmio_base + offset);
3059 *value &= GENMASK_ULL(47, 0);
3065 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3070 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3072 EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3074 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3079 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3081 EXPORT_SYMBOL(amd_iommu_pc_set_reg);