2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
4 * Leo Duran <leo.duran@amd.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/pci.h>
21 #include <linux/acpi.h>
22 #include <linux/list.h>
23 #include <linux/bitmap.h>
24 #include <linux/slab.h>
25 #include <linux/syscore_ops.h>
26 #include <linux/interrupt.h>
27 #include <linux/msi.h>
28 #include <linux/amd-iommu.h>
29 #include <linux/export.h>
30 #include <linux/iommu.h>
31 #include <linux/kmemleak.h>
32 #include <linux/mem_encrypt.h>
33 #include <asm/pci-direct.h>
34 #include <asm/iommu.h>
36 #include <asm/x86_init.h>
37 #include <asm/iommu_table.h>
38 #include <asm/io_apic.h>
39 #include <asm/irq_remapping.h>
41 #include <linux/crash_dump.h>
42 #include "amd_iommu_proto.h"
43 #include "amd_iommu_types.h"
44 #include "irq_remapping.h"
47 * definitions for the ACPI scanning code
49 #define IVRS_HEADER_LENGTH 48
51 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
52 #define ACPI_IVMD_TYPE_ALL 0x20
53 #define ACPI_IVMD_TYPE 0x21
54 #define ACPI_IVMD_TYPE_RANGE 0x22
56 #define IVHD_DEV_ALL 0x01
57 #define IVHD_DEV_SELECT 0x02
58 #define IVHD_DEV_SELECT_RANGE_START 0x03
59 #define IVHD_DEV_RANGE_END 0x04
60 #define IVHD_DEV_ALIAS 0x42
61 #define IVHD_DEV_ALIAS_RANGE 0x43
62 #define IVHD_DEV_EXT_SELECT 0x46
63 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
64 #define IVHD_DEV_SPECIAL 0x48
65 #define IVHD_DEV_ACPI_HID 0xf0
67 #define UID_NOT_PRESENT 0
68 #define UID_IS_INTEGER 1
69 #define UID_IS_CHARACTER 2
71 #define IVHD_SPECIAL_IOAPIC 1
72 #define IVHD_SPECIAL_HPET 2
74 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
75 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
76 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
77 #define IVHD_FLAG_ISOC_EN_MASK 0x08
79 #define IVMD_FLAG_EXCL_RANGE 0x08
80 #define IVMD_FLAG_UNITY_MAP 0x01
82 #define ACPI_DEVFLAG_INITPASS 0x01
83 #define ACPI_DEVFLAG_EXTINT 0x02
84 #define ACPI_DEVFLAG_NMI 0x04
85 #define ACPI_DEVFLAG_SYSMGT1 0x10
86 #define ACPI_DEVFLAG_SYSMGT2 0x20
87 #define ACPI_DEVFLAG_LINT0 0x40
88 #define ACPI_DEVFLAG_LINT1 0x80
89 #define ACPI_DEVFLAG_ATSDIS 0x10000000
91 #define LOOP_TIMEOUT 100000
93 * ACPI table definitions
95 * These data structures are laid over the table to parse the important values
99 extern const struct iommu_ops amd_iommu_ops;
102 * structure describing one IOMMU in the ACPI table. Typically followed by one
103 * or more ivhd_entrys.
116 /* Following only valid on IVHD type 11h and 40h */
117 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
119 } __attribute__((packed));
122 * A device entry describing which devices a specific IOMMU translates and
123 * which requestor ids they use.
135 } __attribute__((packed));
138 * An AMD IOMMU memory definition structure. It defines things like exclusion
139 * ranges for devices and regions that should be unity mapped.
150 } __attribute__((packed));
153 bool amd_iommu_irq_remap __read_mostly;
155 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
157 static bool amd_iommu_detected;
158 static bool __initdata amd_iommu_disabled;
159 static int amd_iommu_target_ivhd_type;
161 u16 amd_iommu_last_bdf; /* largest PCI device id we have
163 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
165 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
167 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
170 /* Array to assign indices to IOMMUs*/
171 struct amd_iommu *amd_iommus[MAX_IOMMUS];
173 /* Number of IOMMUs present in the system */
174 static int amd_iommus_present;
176 /* IOMMUs have a non-present cache? */
177 bool amd_iommu_np_cache __read_mostly;
178 bool amd_iommu_iotlb_sup __read_mostly = true;
180 u32 amd_iommu_max_pasid __read_mostly = ~0;
182 bool amd_iommu_v2_present __read_mostly;
183 static bool amd_iommu_pc_present __read_mostly;
185 bool amd_iommu_force_isolation __read_mostly;
188 * List of protection domains - used during resume
190 LIST_HEAD(amd_iommu_pd_list);
191 spinlock_t amd_iommu_pd_lock;
194 * Pointer to the device table which is shared by all AMD IOMMUs
195 * it is indexed by the PCI device id or the HT unit id and contains
196 * information about the domain the device belongs to as well as the
197 * page table root pointer.
199 struct dev_table_entry *amd_iommu_dev_table;
201 * Pointer to a device table which the content of old device table
202 * will be copied to. It's only be used in kdump kernel.
204 static struct dev_table_entry *old_dev_tbl_cpy;
207 * The alias table is a driver specific data structure which contains the
208 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
209 * More than one device can share the same requestor id.
211 u16 *amd_iommu_alias_table;
214 * The rlookup table is used to find the IOMMU which is responsible
215 * for a specific device. It is also indexed by the PCI device id.
217 struct amd_iommu **amd_iommu_rlookup_table;
218 EXPORT_SYMBOL(amd_iommu_rlookup_table);
221 * This table is used to find the irq remapping table for a given device id
224 struct irq_remap_table **irq_lookup_table;
227 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
228 * to know which ones are already in use.
230 unsigned long *amd_iommu_pd_alloc_bitmap;
232 static u32 dev_table_size; /* size of the device table */
233 static u32 alias_table_size; /* size of the alias table */
234 static u32 rlookup_table_size; /* size if the rlookup table */
236 enum iommu_init_state {
247 IOMMU_CMDLINE_DISABLED,
250 /* Early ioapic and hpet maps from kernel command line */
251 #define EARLY_MAP_SIZE 4
252 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
253 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
254 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
256 static int __initdata early_ioapic_map_size;
257 static int __initdata early_hpet_map_size;
258 static int __initdata early_acpihid_map_size;
260 static bool __initdata cmdline_maps;
262 static enum iommu_init_state init_state = IOMMU_START_STATE;
264 static int amd_iommu_enable_interrupts(void);
265 static int __init iommu_go_to_state(enum iommu_init_state state);
266 static void init_device_table_dma(void);
268 static bool amd_iommu_pre_enabled = true;
270 bool translation_pre_enabled(struct amd_iommu *iommu)
272 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
274 EXPORT_SYMBOL(translation_pre_enabled);
276 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
278 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
281 static void init_translation_status(struct amd_iommu *iommu)
285 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
286 if (ctrl & (1<<CONTROL_IOMMU_EN))
287 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
290 static inline void update_last_devid(u16 devid)
292 if (devid > amd_iommu_last_bdf)
293 amd_iommu_last_bdf = devid;
296 static inline unsigned long tbl_size(int entry_size)
298 unsigned shift = PAGE_SHIFT +
299 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
304 int amd_iommu_get_num_iommus(void)
306 return amd_iommus_present;
309 /* Access to l1 and l2 indexed register spaces */
311 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
315 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
316 pci_read_config_dword(iommu->dev, 0xfc, &val);
320 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
322 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
323 pci_write_config_dword(iommu->dev, 0xfc, val);
324 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
327 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
331 pci_write_config_dword(iommu->dev, 0xf0, address);
332 pci_read_config_dword(iommu->dev, 0xf4, &val);
336 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
338 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
339 pci_write_config_dword(iommu->dev, 0xf4, val);
342 /****************************************************************************
344 * AMD IOMMU MMIO register space handling functions
346 * These functions are used to program the IOMMU device registers in
347 * MMIO space required for that driver.
349 ****************************************************************************/
352 * This function set the exclusion range in the IOMMU. DMA accesses to the
353 * exclusion range are passed through untranslated
355 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
357 u64 start = iommu->exclusion_start & PAGE_MASK;
358 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
361 if (!iommu->exclusion_start)
364 entry = start | MMIO_EXCL_ENABLE_MASK;
365 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
366 &entry, sizeof(entry));
369 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
370 &entry, sizeof(entry));
373 /* Programs the physical address of the device table into the IOMMU hardware */
374 static void iommu_set_device_table(struct amd_iommu *iommu)
378 BUG_ON(iommu->mmio_base == NULL);
380 entry = iommu_virt_to_phys(amd_iommu_dev_table);
381 entry |= (dev_table_size >> 12) - 1;
382 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
383 &entry, sizeof(entry));
386 /* Generic functions to enable/disable certain features of the IOMMU. */
387 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
391 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
393 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
396 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
400 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
402 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
405 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
409 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
410 ctrl &= ~CTRL_INV_TO_MASK;
411 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
412 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
415 /* Function to enable the hardware */
416 static void iommu_enable(struct amd_iommu *iommu)
418 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
421 static void iommu_disable(struct amd_iommu *iommu)
423 if (!iommu->mmio_base)
426 /* Disable command buffer */
427 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
429 /* Disable event logging and event interrupts */
430 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
431 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
433 /* Disable IOMMU GA_LOG */
434 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
435 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
437 /* Disable IOMMU hardware itself */
438 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
442 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
443 * the system has one.
445 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
447 if (!request_mem_region(address, end, "amd_iommu")) {
448 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
450 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
454 return (u8 __iomem *)ioremap_nocache(address, end);
457 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
459 if (iommu->mmio_base)
460 iounmap(iommu->mmio_base);
461 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
464 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
480 /****************************************************************************
482 * The functions below belong to the first pass of AMD IOMMU ACPI table
483 * parsing. In this pass we try to find out the highest device id this
484 * code has to handle. Upon this information the size of the shared data
485 * structures is determined later.
487 ****************************************************************************/
490 * This function calculates the length of a given IVHD entry
492 static inline int ivhd_entry_length(u8 *ivhd)
494 u32 type = ((struct ivhd_entry *)ivhd)->type;
497 return 0x04 << (*ivhd >> 6);
498 } else if (type == IVHD_DEV_ACPI_HID) {
499 /* For ACPI_HID, offset 21 is uid len */
500 return *((u8 *)ivhd + 21) + 22;
506 * After reading the highest device id from the IOMMU PCI capability header
507 * this function looks if there is a higher device id defined in the ACPI table
509 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
511 u8 *p = (void *)h, *end = (void *)h;
512 struct ivhd_entry *dev;
514 u32 ivhd_size = get_ivhd_header_size(h);
517 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
525 dev = (struct ivhd_entry *)p;
528 /* Use maximum BDF value for DEV_ALL */
529 update_last_devid(0xffff);
531 case IVHD_DEV_SELECT:
532 case IVHD_DEV_RANGE_END:
534 case IVHD_DEV_EXT_SELECT:
535 /* all the above subfield types refer to device ids */
536 update_last_devid(dev->devid);
541 p += ivhd_entry_length(p);
549 static int __init check_ivrs_checksum(struct acpi_table_header *table)
552 u8 checksum = 0, *p = (u8 *)table;
554 for (i = 0; i < table->length; ++i)
557 /* ACPI table corrupt */
558 pr_err(FW_BUG "AMD-Vi: IVRS invalid checksum\n");
566 * Iterate over all IVHD entries in the ACPI table and find the highest device
567 * id which we need to handle. This is the first of three functions which parse
568 * the ACPI table. So we check the checksum here.
570 static int __init find_last_devid_acpi(struct acpi_table_header *table)
572 u8 *p = (u8 *)table, *end = (u8 *)table;
573 struct ivhd_header *h;
575 p += IVRS_HEADER_LENGTH;
577 end += table->length;
579 h = (struct ivhd_header *)p;
580 if (h->type == amd_iommu_target_ivhd_type) {
581 int ret = find_last_devid_from_ivhd(h);
593 /****************************************************************************
595 * The following functions belong to the code path which parses the ACPI table
596 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
597 * data structures, initialize the device/alias/rlookup table and also
598 * basically initialize the hardware.
600 ****************************************************************************/
603 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
604 * write commands to that buffer later and the IOMMU will execute them
607 static int __init alloc_command_buffer(struct amd_iommu *iommu)
609 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
610 get_order(CMD_BUFFER_SIZE));
612 return iommu->cmd_buf ? 0 : -ENOMEM;
616 * This function resets the command buffer if the IOMMU stopped fetching
619 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
621 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
623 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
624 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
625 iommu->cmd_buf_head = 0;
626 iommu->cmd_buf_tail = 0;
628 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
632 * This function writes the command buffer address to the hardware and
635 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
639 BUG_ON(iommu->cmd_buf == NULL);
641 entry = iommu_virt_to_phys(iommu->cmd_buf);
642 entry |= MMIO_CMD_SIZE_512;
644 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
645 &entry, sizeof(entry));
647 amd_iommu_reset_cmd_buffer(iommu);
651 * This function disables the command buffer
653 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
655 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
658 static void __init free_command_buffer(struct amd_iommu *iommu)
660 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
663 /* allocates the memory where the IOMMU will log its events to */
664 static int __init alloc_event_buffer(struct amd_iommu *iommu)
666 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
667 get_order(EVT_BUFFER_SIZE));
669 return iommu->evt_buf ? 0 : -ENOMEM;
672 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
676 BUG_ON(iommu->evt_buf == NULL);
678 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
680 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
681 &entry, sizeof(entry));
683 /* set head and tail to zero manually */
684 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
685 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
687 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
691 * This function disables the event log buffer
693 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
695 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
698 static void __init free_event_buffer(struct amd_iommu *iommu)
700 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
703 /* allocates the memory where the IOMMU will log its events to */
704 static int __init alloc_ppr_log(struct amd_iommu *iommu)
706 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
707 get_order(PPR_LOG_SIZE));
709 return iommu->ppr_log ? 0 : -ENOMEM;
712 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
716 if (iommu->ppr_log == NULL)
719 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
721 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
722 &entry, sizeof(entry));
724 /* set head and tail to zero manually */
725 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
726 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
728 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
729 iommu_feature_enable(iommu, CONTROL_PPR_EN);
732 static void __init free_ppr_log(struct amd_iommu *iommu)
734 if (iommu->ppr_log == NULL)
737 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
740 static void free_ga_log(struct amd_iommu *iommu)
742 #ifdef CONFIG_IRQ_REMAP
744 free_pages((unsigned long)iommu->ga_log,
745 get_order(GA_LOG_SIZE));
746 if (iommu->ga_log_tail)
747 free_pages((unsigned long)iommu->ga_log_tail,
752 static int iommu_ga_log_enable(struct amd_iommu *iommu)
754 #ifdef CONFIG_IRQ_REMAP
760 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
762 /* Check if already running */
763 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
766 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
767 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
769 for (i = 0; i < LOOP_TIMEOUT; ++i) {
770 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
771 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
775 if (i >= LOOP_TIMEOUT)
777 #endif /* CONFIG_IRQ_REMAP */
781 #ifdef CONFIG_IRQ_REMAP
782 static int iommu_init_ga_log(struct amd_iommu *iommu)
786 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
789 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
790 get_order(GA_LOG_SIZE));
794 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
796 if (!iommu->ga_log_tail)
799 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
800 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
801 &entry, sizeof(entry));
802 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
803 (BIT_ULL(52)-1)) & ~7ULL;
804 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
805 &entry, sizeof(entry));
806 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
807 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
814 #endif /* CONFIG_IRQ_REMAP */
816 static int iommu_init_ga(struct amd_iommu *iommu)
820 #ifdef CONFIG_IRQ_REMAP
821 /* Note: We have already checked GASup from IVRS table.
822 * Now, we need to make sure that GAMSup is set.
824 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
825 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
826 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
828 ret = iommu_init_ga_log(iommu);
829 #endif /* CONFIG_IRQ_REMAP */
834 static void iommu_enable_gt(struct amd_iommu *iommu)
836 if (!iommu_feature(iommu, FEATURE_GT))
839 iommu_feature_enable(iommu, CONTROL_GT_EN);
842 /* sets a specific bit in the device table entry. */
843 static void set_dev_entry_bit(u16 devid, u8 bit)
845 int i = (bit >> 6) & 0x03;
846 int _bit = bit & 0x3f;
848 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
851 static int get_dev_entry_bit(u16 devid, u8 bit)
853 int i = (bit >> 6) & 0x03;
854 int _bit = bit & 0x3f;
856 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
860 static bool copy_device_table(void)
862 u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
863 struct dev_table_entry *old_devtb = NULL;
864 u32 lo, hi, devid, old_devtb_size;
865 phys_addr_t old_devtb_phys;
866 struct amd_iommu *iommu;
867 u16 dom_id, dte_v, irq_v;
871 if (!amd_iommu_pre_enabled)
874 pr_warn("Translation is already enabled - trying to copy translation structures\n");
875 for_each_iommu(iommu) {
876 /* All IOMMUs should use the same device table with the same size */
877 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
878 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
879 entry = (((u64) hi) << 32) + lo;
880 if (last_entry && last_entry != entry) {
881 pr_err("IOMMU:%d should use the same dev table as others!\n",
887 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
888 if (old_devtb_size != dev_table_size) {
889 pr_err("The device table size of IOMMU:%d is not expected!\n",
895 old_devtb_phys = entry & PAGE_MASK;
896 if (old_devtb_phys >= 0x100000000ULL) {
897 pr_err("The address of old device table is above 4G, not trustworthy!\n");
900 old_devtb = memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
904 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
905 old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
906 get_order(dev_table_size));
907 if (old_dev_tbl_cpy == NULL) {
908 pr_err("Failed to allocate memory for copying old device table!\n");
912 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
913 old_dev_tbl_cpy[devid] = old_devtb[devid];
914 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
915 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
917 if (dte_v && dom_id) {
918 old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
919 old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
920 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
921 /* If gcr3 table existed, mask it out */
922 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
923 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
924 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
925 old_dev_tbl_cpy[devid].data[1] &= ~tmp;
926 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
928 old_dev_tbl_cpy[devid].data[0] &= ~tmp;
932 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
933 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
934 int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
935 if (irq_v && (int_ctl || int_tab_len)) {
936 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
937 (int_tab_len != DTE_IRQ_TABLE_LEN)) {
938 pr_err("Wrong old irq remapping flag: %#x\n", devid);
942 old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
950 void amd_iommu_apply_erratum_63(u16 devid)
954 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
955 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
958 set_dev_entry_bit(devid, DEV_ENTRY_IW);
961 /* Writes the specific IOMMU for a device into the rlookup table */
962 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
964 amd_iommu_rlookup_table[devid] = iommu;
968 * This function takes the device specific flags read from the ACPI
969 * table and sets up the device table entry with that information
971 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
972 u16 devid, u32 flags, u32 ext_flags)
974 if (flags & ACPI_DEVFLAG_INITPASS)
975 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
976 if (flags & ACPI_DEVFLAG_EXTINT)
977 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
978 if (flags & ACPI_DEVFLAG_NMI)
979 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
980 if (flags & ACPI_DEVFLAG_SYSMGT1)
981 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
982 if (flags & ACPI_DEVFLAG_SYSMGT2)
983 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
984 if (flags & ACPI_DEVFLAG_LINT0)
985 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
986 if (flags & ACPI_DEVFLAG_LINT1)
987 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
989 amd_iommu_apply_erratum_63(devid);
991 set_iommu_for_device(iommu, devid);
994 static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
996 struct devid_map *entry;
997 struct list_head *list;
999 if (type == IVHD_SPECIAL_IOAPIC)
1001 else if (type == IVHD_SPECIAL_HPET)
1006 list_for_each_entry(entry, list, list) {
1007 if (!(entry->id == id && entry->cmd_line))
1010 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
1011 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1013 *devid = entry->devid;
1018 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1023 entry->devid = *devid;
1024 entry->cmd_line = cmd_line;
1026 list_add_tail(&entry->list, list);
1031 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1034 struct acpihid_map_entry *entry;
1035 struct list_head *list = &acpihid_map;
1037 list_for_each_entry(entry, list, list) {
1038 if (strcmp(entry->hid, hid) ||
1039 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1043 pr_info("AMD-Vi: Command-line override for hid:%s uid:%s\n",
1045 *devid = entry->devid;
1049 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1053 memcpy(entry->uid, uid, strlen(uid));
1054 memcpy(entry->hid, hid, strlen(hid));
1055 entry->devid = *devid;
1056 entry->cmd_line = cmd_line;
1057 entry->root_devid = (entry->devid & (~0x7));
1059 pr_info("AMD-Vi:%s, add hid:%s, uid:%s, rdevid:%d\n",
1060 entry->cmd_line ? "cmd" : "ivrs",
1061 entry->hid, entry->uid, entry->root_devid);
1063 list_add_tail(&entry->list, list);
1067 static int __init add_early_maps(void)
1071 for (i = 0; i < early_ioapic_map_size; ++i) {
1072 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1073 early_ioapic_map[i].id,
1074 &early_ioapic_map[i].devid,
1075 early_ioapic_map[i].cmd_line);
1080 for (i = 0; i < early_hpet_map_size; ++i) {
1081 ret = add_special_device(IVHD_SPECIAL_HPET,
1082 early_hpet_map[i].id,
1083 &early_hpet_map[i].devid,
1084 early_hpet_map[i].cmd_line);
1089 for (i = 0; i < early_acpihid_map_size; ++i) {
1090 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1091 early_acpihid_map[i].uid,
1092 &early_acpihid_map[i].devid,
1093 early_acpihid_map[i].cmd_line);
1102 * Reads the device exclusion range from ACPI and initializes the IOMMU with
1105 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
1107 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
1109 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
1114 * We only can configure exclusion ranges per IOMMU, not
1115 * per device. But we can enable the exclusion range per
1116 * device. This is done here
1118 set_dev_entry_bit(devid, DEV_ENTRY_EX);
1119 iommu->exclusion_start = m->range_start;
1120 iommu->exclusion_length = m->range_length;
1125 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1126 * initializes the hardware and our data structures with it.
1128 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1129 struct ivhd_header *h)
1132 u8 *end = p, flags = 0;
1133 u16 devid = 0, devid_start = 0, devid_to = 0;
1134 u32 dev_i, ext_flags = 0;
1136 struct ivhd_entry *e;
1141 ret = add_early_maps();
1146 * First save the recommended feature enable bits from ACPI
1148 iommu->acpi_flags = h->flags;
1151 * Done. Now parse the device entries
1153 ivhd_size = get_ivhd_header_size(h);
1155 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
1165 e = (struct ivhd_entry *)p;
1169 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1171 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1172 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1174 case IVHD_DEV_SELECT:
1176 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1178 PCI_BUS_NUM(e->devid),
1184 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1186 case IVHD_DEV_SELECT_RANGE_START:
1188 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1189 "devid: %02x:%02x.%x flags: %02x\n",
1190 PCI_BUS_NUM(e->devid),
1195 devid_start = e->devid;
1200 case IVHD_DEV_ALIAS:
1202 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1203 "flags: %02x devid_to: %02x:%02x.%x\n",
1204 PCI_BUS_NUM(e->devid),
1208 PCI_BUS_NUM(e->ext >> 8),
1209 PCI_SLOT(e->ext >> 8),
1210 PCI_FUNC(e->ext >> 8));
1213 devid_to = e->ext >> 8;
1214 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1215 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1216 amd_iommu_alias_table[devid] = devid_to;
1218 case IVHD_DEV_ALIAS_RANGE:
1220 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1221 "devid: %02x:%02x.%x flags: %02x "
1222 "devid_to: %02x:%02x.%x\n",
1223 PCI_BUS_NUM(e->devid),
1227 PCI_BUS_NUM(e->ext >> 8),
1228 PCI_SLOT(e->ext >> 8),
1229 PCI_FUNC(e->ext >> 8));
1231 devid_start = e->devid;
1233 devid_to = e->ext >> 8;
1237 case IVHD_DEV_EXT_SELECT:
1239 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1240 "flags: %02x ext: %08x\n",
1241 PCI_BUS_NUM(e->devid),
1247 set_dev_entry_from_acpi(iommu, devid, e->flags,
1250 case IVHD_DEV_EXT_SELECT_RANGE:
1252 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1253 "%02x:%02x.%x flags: %02x ext: %08x\n",
1254 PCI_BUS_NUM(e->devid),
1259 devid_start = e->devid;
1264 case IVHD_DEV_RANGE_END:
1266 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1267 PCI_BUS_NUM(e->devid),
1269 PCI_FUNC(e->devid));
1272 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1274 amd_iommu_alias_table[dev_i] = devid_to;
1275 set_dev_entry_from_acpi(iommu,
1276 devid_to, flags, ext_flags);
1278 set_dev_entry_from_acpi(iommu, dev_i,
1282 case IVHD_DEV_SPECIAL: {
1288 handle = e->ext & 0xff;
1289 devid = (e->ext >> 8) & 0xffff;
1290 type = (e->ext >> 24) & 0xff;
1292 if (type == IVHD_SPECIAL_IOAPIC)
1294 else if (type == IVHD_SPECIAL_HPET)
1299 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1305 ret = add_special_device(type, handle, &devid, false);
1310 * add_special_device might update the devid in case a
1311 * command-line override is present. So call
1312 * set_dev_entry_from_acpi after add_special_device.
1314 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1318 case IVHD_DEV_ACPI_HID: {
1320 u8 hid[ACPIHID_HID_LEN];
1321 u8 uid[ACPIHID_UID_LEN];
1324 if (h->type != 0x40) {
1325 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1330 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1331 hid[ACPIHID_HID_LEN - 1] = '\0';
1334 pr_err(FW_BUG "Invalid HID.\n");
1340 case UID_NOT_PRESENT:
1343 pr_warn(FW_BUG "Invalid UID length.\n");
1346 case UID_IS_INTEGER:
1348 sprintf(uid, "%d", e->uid);
1351 case UID_IS_CHARACTER:
1353 memcpy(uid, &e->uid, e->uidl);
1354 uid[e->uidl] = '\0';
1362 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1370 ret = add_acpi_hid_device(hid, uid, &devid, false);
1375 * add_special_device might update the devid in case a
1376 * command-line override is present. So call
1377 * set_dev_entry_from_acpi after add_special_device.
1379 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1387 p += ivhd_entry_length(p);
1393 static void __init free_iommu_one(struct amd_iommu *iommu)
1395 free_command_buffer(iommu);
1396 free_event_buffer(iommu);
1397 free_ppr_log(iommu);
1399 iommu_unmap_mmio_space(iommu);
1402 static void __init free_iommu_all(void)
1404 struct amd_iommu *iommu, *next;
1406 for_each_iommu_safe(iommu, next) {
1407 list_del(&iommu->list);
1408 free_iommu_one(iommu);
1414 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1416 * BIOS should disable L2B micellaneous clock gating by setting
1417 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1419 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1423 if ((boot_cpu_data.x86 != 0x15) ||
1424 (boot_cpu_data.x86_model < 0x10) ||
1425 (boot_cpu_data.x86_model > 0x1f))
1428 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1429 pci_read_config_dword(iommu->dev, 0xf4, &value);
1434 /* Select NB indirect register 0x90 and enable writing */
1435 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1437 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1438 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1439 dev_name(&iommu->dev->dev));
1441 /* Clear the enable writing bit */
1442 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1446 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1448 * BIOS should enable ATS write permission check by setting
1449 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1451 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1455 if ((boot_cpu_data.x86 != 0x15) ||
1456 (boot_cpu_data.x86_model < 0x30) ||
1457 (boot_cpu_data.x86_model > 0x3f))
1460 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1461 value = iommu_read_l2(iommu, 0x47);
1466 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1467 iommu_write_l2(iommu, 0x47, value | BIT(0));
1469 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1470 dev_name(&iommu->dev->dev));
1474 * This function clues the initialization function for one IOMMU
1475 * together and also allocates the command buffer and programs the
1476 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1478 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1482 spin_lock_init(&iommu->lock);
1484 /* Add IOMMU to internal data structures */
1485 list_add_tail(&iommu->list, &amd_iommu_list);
1486 iommu->index = amd_iommus_present++;
1488 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1489 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1493 /* Index is fine - add IOMMU to the array */
1494 amd_iommus[iommu->index] = iommu;
1497 * Copy data from ACPI table entry to the iommu struct
1499 iommu->devid = h->devid;
1500 iommu->cap_ptr = h->cap_ptr;
1501 iommu->pci_seg = h->pci_seg;
1502 iommu->mmio_phys = h->mmio_phys;
1506 /* Check if IVHD EFR contains proper max banks/counters */
1507 if ((h->efr_attr != 0) &&
1508 ((h->efr_attr & (0xF << 13)) != 0) &&
1509 ((h->efr_attr & (0x3F << 17)) != 0))
1510 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1512 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1513 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1514 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1518 if (h->efr_reg & (1 << 9))
1519 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1521 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1522 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1523 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1529 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1530 iommu->mmio_phys_end);
1531 if (!iommu->mmio_base)
1534 if (alloc_command_buffer(iommu))
1537 if (alloc_event_buffer(iommu))
1540 iommu->int_enabled = false;
1542 init_translation_status(iommu);
1543 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1544 iommu_disable(iommu);
1545 clear_translation_pre_enabled(iommu);
1546 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1549 if (amd_iommu_pre_enabled)
1550 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1552 ret = init_iommu_from_acpi(iommu, h);
1556 ret = amd_iommu_create_irq_domain(iommu);
1561 * Make sure IOMMU is not considered to translate itself. The IVRS
1562 * table tells us so, but this is a lie!
1564 amd_iommu_rlookup_table[iommu->devid] = NULL;
1570 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1571 * @ivrs Pointer to the IVRS header
1573 * This function search through all IVDB of the maximum supported IVHD
1575 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1577 u8 *base = (u8 *)ivrs;
1578 struct ivhd_header *ivhd = (struct ivhd_header *)
1579 (base + IVRS_HEADER_LENGTH);
1580 u8 last_type = ivhd->type;
1581 u16 devid = ivhd->devid;
1583 while (((u8 *)ivhd - base < ivrs->length) &&
1584 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1585 u8 *p = (u8 *) ivhd;
1587 if (ivhd->devid == devid)
1588 last_type = ivhd->type;
1589 ivhd = (struct ivhd_header *)(p + ivhd->length);
1596 * Iterates over all IOMMU entries in the ACPI table, allocates the
1597 * IOMMU structure and initializes it with init_iommu_one()
1599 static int __init init_iommu_all(struct acpi_table_header *table)
1601 u8 *p = (u8 *)table, *end = (u8 *)table;
1602 struct ivhd_header *h;
1603 struct amd_iommu *iommu;
1606 end += table->length;
1607 p += IVRS_HEADER_LENGTH;
1610 h = (struct ivhd_header *)p;
1611 if (*p == amd_iommu_target_ivhd_type) {
1613 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1614 "seg: %d flags: %01x info %04x\n",
1615 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1616 PCI_FUNC(h->devid), h->cap_ptr,
1617 h->pci_seg, h->flags, h->info);
1618 DUMP_printk(" mmio-addr: %016llx\n",
1621 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1625 ret = init_iommu_one(iommu, h);
1637 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1638 u8 fxn, u64 *value, bool is_write);
1640 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1642 u64 val = 0xabcd, val2 = 0;
1644 if (!iommu_feature(iommu, FEATURE_PC))
1647 amd_iommu_pc_present = true;
1649 /* Check if the performance counters can be written to */
1650 if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1651 (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1653 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1654 amd_iommu_pc_present = false;
1658 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1660 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1661 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1662 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1665 static ssize_t amd_iommu_show_cap(struct device *dev,
1666 struct device_attribute *attr,
1669 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1670 return sprintf(buf, "%x\n", iommu->cap);
1672 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1674 static ssize_t amd_iommu_show_features(struct device *dev,
1675 struct device_attribute *attr,
1678 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1679 return sprintf(buf, "%llx\n", iommu->features);
1681 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1683 static struct attribute *amd_iommu_attrs[] = {
1685 &dev_attr_features.attr,
1689 static struct attribute_group amd_iommu_group = {
1690 .name = "amd-iommu",
1691 .attrs = amd_iommu_attrs,
1694 static const struct attribute_group *amd_iommu_groups[] = {
1699 static int __init iommu_init_pci(struct amd_iommu *iommu)
1701 int cap_ptr = iommu->cap_ptr;
1702 u32 range, misc, low, high;
1705 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1706 iommu->devid & 0xff);
1710 /* Prevent binding other PCI device drivers to IOMMU devices */
1711 iommu->dev->match_driver = false;
1713 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1715 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1717 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1720 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1721 amd_iommu_iotlb_sup = false;
1723 /* read extended feature bits */
1724 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1725 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1727 iommu->features = ((u64)high << 32) | low;
1729 if (iommu_feature(iommu, FEATURE_GT)) {
1734 pasmax = iommu->features & FEATURE_PASID_MASK;
1735 pasmax >>= FEATURE_PASID_SHIFT;
1736 max_pasid = (1 << (pasmax + 1)) - 1;
1738 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1740 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1742 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1743 glxval >>= FEATURE_GLXVAL_SHIFT;
1745 if (amd_iommu_max_glx_val == -1)
1746 amd_iommu_max_glx_val = glxval;
1748 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1751 if (iommu_feature(iommu, FEATURE_GT) &&
1752 iommu_feature(iommu, FEATURE_PPR)) {
1753 iommu->is_iommu_v2 = true;
1754 amd_iommu_v2_present = true;
1757 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1760 ret = iommu_init_ga(iommu);
1764 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1765 amd_iommu_np_cache = true;
1767 init_iommu_perf_ctr(iommu);
1769 if (is_rd890_iommu(iommu->dev)) {
1772 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1776 * Some rd890 systems may not be fully reconfigured by the
1777 * BIOS, so it's necessary for us to store this information so
1778 * it can be reprogrammed on resume
1780 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1781 &iommu->stored_addr_lo);
1782 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1783 &iommu->stored_addr_hi);
1785 /* Low bit locks writes to configuration space */
1786 iommu->stored_addr_lo &= ~1;
1788 for (i = 0; i < 6; i++)
1789 for (j = 0; j < 0x12; j++)
1790 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1792 for (i = 0; i < 0x83; i++)
1793 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1796 amd_iommu_erratum_746_workaround(iommu);
1797 amd_iommu_ats_write_check_workaround(iommu);
1799 iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1800 amd_iommu_groups, "ivhd%d", iommu->index);
1801 iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1802 iommu_device_register(&iommu->iommu);
1804 return pci_enable_device(iommu->dev);
1807 static void print_iommu_info(void)
1809 static const char * const feat_str[] = {
1810 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1811 "IA", "GA", "HE", "PC"
1813 struct amd_iommu *iommu;
1815 for_each_iommu(iommu) {
1818 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1819 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1821 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1822 pr_info("AMD-Vi: Extended features (%#llx):\n",
1824 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1825 if (iommu_feature(iommu, (1ULL << i)))
1826 pr_cont(" %s", feat_str[i]);
1829 if (iommu->features & FEATURE_GAM_VAPIC)
1830 pr_cont(" GA_vAPIC");
1835 if (irq_remapping_enabled) {
1836 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1837 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1838 pr_info("AMD-Vi: virtual APIC enabled\n");
1842 static int __init amd_iommu_init_pci(void)
1844 struct amd_iommu *iommu;
1847 for_each_iommu(iommu) {
1848 ret = iommu_init_pci(iommu);
1854 * Order is important here to make sure any unity map requirements are
1855 * fulfilled. The unity mappings are created and written to the device
1856 * table during the amd_iommu_init_api() call.
1858 * After that we call init_device_table_dma() to make sure any
1859 * uninitialized DTE will block DMA, and in the end we flush the caches
1860 * of all IOMMUs to make sure the changes to the device table are
1863 ret = amd_iommu_init_api();
1865 init_device_table_dma();
1867 for_each_iommu(iommu)
1868 iommu_flush_all_caches(iommu);
1876 /****************************************************************************
1878 * The following functions initialize the MSI interrupts for all IOMMUs
1879 * in the system. It's a bit challenging because there could be multiple
1880 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1883 ****************************************************************************/
1885 static int iommu_setup_msi(struct amd_iommu *iommu)
1889 r = pci_enable_msi(iommu->dev);
1893 r = request_threaded_irq(iommu->dev->irq,
1894 amd_iommu_int_handler,
1895 amd_iommu_int_thread,
1900 pci_disable_msi(iommu->dev);
1904 iommu->int_enabled = true;
1909 static int iommu_init_msi(struct amd_iommu *iommu)
1913 if (iommu->int_enabled)
1916 if (iommu->dev->msi_cap)
1917 ret = iommu_setup_msi(iommu);
1925 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1927 if (iommu->ppr_log != NULL)
1928 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1930 iommu_ga_log_enable(iommu);
1935 /****************************************************************************
1937 * The next functions belong to the third pass of parsing the ACPI
1938 * table. In this last pass the memory mapping requirements are
1939 * gathered (like exclusion and unity mapping ranges).
1941 ****************************************************************************/
1943 static void __init free_unity_maps(void)
1945 struct unity_map_entry *entry, *next;
1947 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1948 list_del(&entry->list);
1953 /* called when we find an exclusion range definition in ACPI */
1954 static int __init init_exclusion_range(struct ivmd_header *m)
1959 case ACPI_IVMD_TYPE:
1960 set_device_exclusion_range(m->devid, m);
1962 case ACPI_IVMD_TYPE_ALL:
1963 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1964 set_device_exclusion_range(i, m);
1966 case ACPI_IVMD_TYPE_RANGE:
1967 for (i = m->devid; i <= m->aux; ++i)
1968 set_device_exclusion_range(i, m);
1977 /* called for unity map ACPI definition */
1978 static int __init init_unity_map_range(struct ivmd_header *m)
1980 struct unity_map_entry *e = NULL;
1983 e = kzalloc(sizeof(*e), GFP_KERNEL);
1987 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1988 init_exclusion_range(m);
1994 case ACPI_IVMD_TYPE:
1995 s = "IVMD_TYPEi\t\t\t";
1996 e->devid_start = e->devid_end = m->devid;
1998 case ACPI_IVMD_TYPE_ALL:
1999 s = "IVMD_TYPE_ALL\t\t";
2001 e->devid_end = amd_iommu_last_bdf;
2003 case ACPI_IVMD_TYPE_RANGE:
2004 s = "IVMD_TYPE_RANGE\t\t";
2005 e->devid_start = m->devid;
2006 e->devid_end = m->aux;
2009 e->address_start = PAGE_ALIGN(m->range_start);
2010 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2011 e->prot = m->flags >> 1;
2013 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2014 " range_start: %016llx range_end: %016llx flags: %x\n", s,
2015 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2016 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
2017 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2018 e->address_start, e->address_end, m->flags);
2020 list_add_tail(&e->list, &amd_iommu_unity_map);
2025 /* iterates over all memory definitions we find in the ACPI table */
2026 static int __init init_memory_definitions(struct acpi_table_header *table)
2028 u8 *p = (u8 *)table, *end = (u8 *)table;
2029 struct ivmd_header *m;
2031 end += table->length;
2032 p += IVRS_HEADER_LENGTH;
2035 m = (struct ivmd_header *)p;
2036 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2037 init_unity_map_range(m);
2046 * Init the device table to not allow DMA access for devices
2048 static void init_device_table_dma(void)
2052 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2053 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2054 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2058 static void __init uninit_device_table_dma(void)
2062 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2063 amd_iommu_dev_table[devid].data[0] = 0ULL;
2064 amd_iommu_dev_table[devid].data[1] = 0ULL;
2068 static void init_device_table(void)
2072 if (!amd_iommu_irq_remap)
2075 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2076 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2079 static void iommu_init_flags(struct amd_iommu *iommu)
2081 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2082 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2083 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2085 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2086 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2087 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2089 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2090 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2091 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2093 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2094 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2095 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2098 * make IOMMU memory accesses cache coherent
2100 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2102 /* Set IOTLB invalidation timeout to 1s */
2103 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2106 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2109 u32 ioc_feature_control;
2110 struct pci_dev *pdev = iommu->root_pdev;
2112 /* RD890 BIOSes may not have completely reconfigured the iommu */
2113 if (!is_rd890_iommu(iommu->dev) || !pdev)
2117 * First, we need to ensure that the iommu is enabled. This is
2118 * controlled by a register in the northbridge
2121 /* Select Northbridge indirect register 0x75 and enable writing */
2122 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2123 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2125 /* Enable the iommu */
2126 if (!(ioc_feature_control & 0x1))
2127 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2129 /* Restore the iommu BAR */
2130 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2131 iommu->stored_addr_lo);
2132 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2133 iommu->stored_addr_hi);
2135 /* Restore the l1 indirect regs for each of the 6 l1s */
2136 for (i = 0; i < 6; i++)
2137 for (j = 0; j < 0x12; j++)
2138 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2140 /* Restore the l2 indirect regs */
2141 for (i = 0; i < 0x83; i++)
2142 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2144 /* Lock PCI setup registers */
2145 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2146 iommu->stored_addr_lo | 1);
2149 static void iommu_enable_ga(struct amd_iommu *iommu)
2151 #ifdef CONFIG_IRQ_REMAP
2152 switch (amd_iommu_guest_ir) {
2153 case AMD_IOMMU_GUEST_IR_VAPIC:
2154 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2156 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2157 iommu_feature_enable(iommu, CONTROL_GA_EN);
2158 iommu->irte_ops = &irte_128_ops;
2161 iommu->irte_ops = &irte_32_ops;
2167 static void early_enable_iommu(struct amd_iommu *iommu)
2169 iommu_disable(iommu);
2170 iommu_init_flags(iommu);
2171 iommu_set_device_table(iommu);
2172 iommu_enable_command_buffer(iommu);
2173 iommu_enable_event_buffer(iommu);
2174 iommu_set_exclusion_range(iommu);
2175 iommu_enable_ga(iommu);
2176 iommu_enable(iommu);
2177 iommu_flush_all_caches(iommu);
2181 * This function finally enables all IOMMUs found in the system after
2182 * they have been initialized.
2184 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2185 * the old content of device table entries. Not this case or copy failed,
2186 * just continue as normal kernel does.
2188 static void early_enable_iommus(void)
2190 struct amd_iommu *iommu;
2193 if (!copy_device_table()) {
2195 * If come here because of failure in copying device table from old
2196 * kernel with all IOMMUs enabled, print error message and try to
2197 * free allocated old_dev_tbl_cpy.
2199 if (amd_iommu_pre_enabled)
2200 pr_err("Failed to copy DEV table from previous kernel.\n");
2201 if (old_dev_tbl_cpy != NULL)
2202 free_pages((unsigned long)old_dev_tbl_cpy,
2203 get_order(dev_table_size));
2205 for_each_iommu(iommu) {
2206 clear_translation_pre_enabled(iommu);
2207 early_enable_iommu(iommu);
2210 pr_info("Copied DEV table from previous kernel.\n");
2211 free_pages((unsigned long)amd_iommu_dev_table,
2212 get_order(dev_table_size));
2213 amd_iommu_dev_table = old_dev_tbl_cpy;
2214 for_each_iommu(iommu) {
2215 iommu_disable_command_buffer(iommu);
2216 iommu_disable_event_buffer(iommu);
2217 iommu_enable_command_buffer(iommu);
2218 iommu_enable_event_buffer(iommu);
2219 iommu_enable_ga(iommu);
2220 iommu_set_device_table(iommu);
2221 iommu_flush_all_caches(iommu);
2225 #ifdef CONFIG_IRQ_REMAP
2226 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2227 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2231 static void enable_iommus_v2(void)
2233 struct amd_iommu *iommu;
2235 for_each_iommu(iommu) {
2236 iommu_enable_ppr_log(iommu);
2237 iommu_enable_gt(iommu);
2241 static void enable_iommus(void)
2243 early_enable_iommus();
2248 static void disable_iommus(void)
2250 struct amd_iommu *iommu;
2252 for_each_iommu(iommu)
2253 iommu_disable(iommu);
2255 #ifdef CONFIG_IRQ_REMAP
2256 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2257 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2262 * Suspend/Resume support
2263 * disable suspend until real resume implemented
2266 static void amd_iommu_resume(void)
2268 struct amd_iommu *iommu;
2270 for_each_iommu(iommu)
2271 iommu_apply_resume_quirks(iommu);
2273 /* re-load the hardware */
2276 amd_iommu_enable_interrupts();
2279 static int amd_iommu_suspend(void)
2281 /* disable IOMMUs to go out of the way for BIOS */
2287 static struct syscore_ops amd_iommu_syscore_ops = {
2288 .suspend = amd_iommu_suspend,
2289 .resume = amd_iommu_resume,
2292 static void __init free_iommu_resources(void)
2294 kmemleak_free(irq_lookup_table);
2295 free_pages((unsigned long)irq_lookup_table,
2296 get_order(rlookup_table_size));
2297 irq_lookup_table = NULL;
2299 kmem_cache_destroy(amd_iommu_irq_cache);
2300 amd_iommu_irq_cache = NULL;
2302 free_pages((unsigned long)amd_iommu_rlookup_table,
2303 get_order(rlookup_table_size));
2304 amd_iommu_rlookup_table = NULL;
2306 free_pages((unsigned long)amd_iommu_alias_table,
2307 get_order(alias_table_size));
2308 amd_iommu_alias_table = NULL;
2310 free_pages((unsigned long)amd_iommu_dev_table,
2311 get_order(dev_table_size));
2312 amd_iommu_dev_table = NULL;
2316 #ifdef CONFIG_GART_IOMMU
2318 * We failed to initialize the AMD IOMMU - try fallback to GART
2326 /* SB IOAPIC is always on this device in AMD systems */
2327 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2329 static bool __init check_ioapic_information(void)
2331 const char *fw_bug = FW_BUG;
2332 bool ret, has_sb_ioapic;
2335 has_sb_ioapic = false;
2339 * If we have map overrides on the kernel command line the
2340 * messages in this function might not describe firmware bugs
2341 * anymore - so be careful
2346 for (idx = 0; idx < nr_ioapics; idx++) {
2347 int devid, id = mpc_ioapic_id(idx);
2349 devid = get_ioapic_devid(id);
2351 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
2354 } else if (devid == IOAPIC_SB_DEVID) {
2355 has_sb_ioapic = true;
2360 if (!has_sb_ioapic) {
2362 * We expect the SB IOAPIC to be listed in the IVRS
2363 * table. The system timer is connected to the SB IOAPIC
2364 * and if we don't have it in the list the system will
2365 * panic at boot time. This situation usually happens
2366 * when the BIOS is buggy and provides us the wrong
2367 * device id for the IOAPIC in the system.
2369 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
2373 pr_err("AMD-Vi: Disabling interrupt remapping\n");
2378 static void __init free_dma_resources(void)
2380 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2381 get_order(MAX_DOMAIN_ID/8));
2382 amd_iommu_pd_alloc_bitmap = NULL;
2388 * This is the hardware init function for AMD IOMMU in the system.
2389 * This function is called either from amd_iommu_init or from the interrupt
2390 * remapping setup code.
2392 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2395 * 1 pass) Discover the most comprehensive IVHD type to use.
2397 * 2 pass) Find the highest PCI device id the driver has to handle.
2398 * Upon this information the size of the data structures is
2399 * determined that needs to be allocated.
2401 * 3 pass) Initialize the data structures just allocated with the
2402 * information in the ACPI table about available AMD IOMMUs
2403 * in the system. It also maps the PCI devices in the
2404 * system to specific IOMMUs
2406 * 4 pass) After the basic data structures are allocated and
2407 * initialized we update them with information about memory
2408 * remapping requirements parsed out of the ACPI table in
2411 * After everything is set up the IOMMUs are enabled and the necessary
2412 * hotplug and suspend notifiers are registered.
2414 static int __init early_amd_iommu_init(void)
2416 struct acpi_table_header *ivrs_base;
2418 int i, remap_cache_sz, ret = 0;
2420 if (!amd_iommu_detected)
2423 status = acpi_get_table("IVRS", 0, &ivrs_base);
2424 if (status == AE_NOT_FOUND)
2426 else if (ACPI_FAILURE(status)) {
2427 const char *err = acpi_format_exception(status);
2428 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2433 * Validate checksum here so we don't need to do it when
2434 * we actually parse the table
2436 ret = check_ivrs_checksum(ivrs_base);
2440 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2441 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2444 * First parse ACPI tables to find the largest Bus/Dev/Func
2445 * we need to handle. Upon this information the shared data
2446 * structures for the IOMMUs in the system will be allocated
2448 ret = find_last_devid_acpi(ivrs_base);
2452 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2453 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2454 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2456 /* Device table - directly used by all IOMMUs */
2458 amd_iommu_dev_table = (void *)__get_free_pages(
2459 GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2460 get_order(dev_table_size));
2461 if (amd_iommu_dev_table == NULL)
2465 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2466 * IOMMU see for that device
2468 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2469 get_order(alias_table_size));
2470 if (amd_iommu_alias_table == NULL)
2473 /* IOMMU rlookup table - find the IOMMU for a specific device */
2474 amd_iommu_rlookup_table = (void *)__get_free_pages(
2475 GFP_KERNEL | __GFP_ZERO,
2476 get_order(rlookup_table_size));
2477 if (amd_iommu_rlookup_table == NULL)
2480 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2481 GFP_KERNEL | __GFP_ZERO,
2482 get_order(MAX_DOMAIN_ID/8));
2483 if (amd_iommu_pd_alloc_bitmap == NULL)
2487 * let all alias entries point to itself
2489 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2490 amd_iommu_alias_table[i] = i;
2493 * never allocate domain 0 because its used as the non-allocated and
2494 * error value placeholder
2496 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2498 spin_lock_init(&amd_iommu_pd_lock);
2501 * now the data structures are allocated and basically initialized
2502 * start the real acpi table scan
2504 ret = init_iommu_all(ivrs_base);
2508 /* Disable any previously enabled IOMMUs */
2509 if (!is_kdump_kernel() || amd_iommu_disabled)
2512 if (amd_iommu_irq_remap)
2513 amd_iommu_irq_remap = check_ioapic_information();
2515 if (amd_iommu_irq_remap) {
2517 * Interrupt remapping enabled, create kmem_cache for the
2521 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2522 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2524 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2525 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2527 IRQ_TABLE_ALIGNMENT,
2529 if (!amd_iommu_irq_cache)
2532 irq_lookup_table = (void *)__get_free_pages(
2533 GFP_KERNEL | __GFP_ZERO,
2534 get_order(rlookup_table_size));
2535 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2537 if (!irq_lookup_table)
2541 ret = init_memory_definitions(ivrs_base);
2545 /* init the device table */
2546 init_device_table();
2549 /* Don't leak any ACPI memory */
2550 acpi_put_table(ivrs_base);
2556 static int amd_iommu_enable_interrupts(void)
2558 struct amd_iommu *iommu;
2561 for_each_iommu(iommu) {
2562 ret = iommu_init_msi(iommu);
2571 static bool detect_ivrs(void)
2573 struct acpi_table_header *ivrs_base;
2576 status = acpi_get_table("IVRS", 0, &ivrs_base);
2577 if (status == AE_NOT_FOUND)
2579 else if (ACPI_FAILURE(status)) {
2580 const char *err = acpi_format_exception(status);
2581 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2585 acpi_put_table(ivrs_base);
2587 /* Make sure ACS will be enabled during PCI probe */
2593 /****************************************************************************
2595 * AMD IOMMU Initialization State Machine
2597 ****************************************************************************/
2599 static int __init state_next(void)
2603 switch (init_state) {
2604 case IOMMU_START_STATE:
2605 if (!detect_ivrs()) {
2606 init_state = IOMMU_NOT_FOUND;
2609 init_state = IOMMU_IVRS_DETECTED;
2612 case IOMMU_IVRS_DETECTED:
2613 ret = early_amd_iommu_init();
2614 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2615 if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2616 pr_info("AMD-Vi: AMD IOMMU disabled on kernel command-line\n");
2617 free_dma_resources();
2618 free_iommu_resources();
2619 init_state = IOMMU_CMDLINE_DISABLED;
2623 case IOMMU_ACPI_FINISHED:
2624 early_enable_iommus();
2625 x86_platform.iommu_shutdown = disable_iommus;
2626 init_state = IOMMU_ENABLED;
2629 register_syscore_ops(&amd_iommu_syscore_ops);
2630 ret = amd_iommu_init_pci();
2631 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2634 case IOMMU_PCI_INIT:
2635 ret = amd_iommu_enable_interrupts();
2636 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2638 case IOMMU_INTERRUPTS_EN:
2639 ret = amd_iommu_init_dma_ops();
2640 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2643 init_state = IOMMU_INITIALIZED;
2645 case IOMMU_INITIALIZED:
2648 case IOMMU_NOT_FOUND:
2649 case IOMMU_INIT_ERROR:
2650 case IOMMU_CMDLINE_DISABLED:
2651 /* Error states => do nothing */
2662 static int __init iommu_go_to_state(enum iommu_init_state state)
2666 while (init_state != state) {
2667 if (init_state == IOMMU_NOT_FOUND ||
2668 init_state == IOMMU_INIT_ERROR ||
2669 init_state == IOMMU_CMDLINE_DISABLED)
2677 #ifdef CONFIG_IRQ_REMAP
2678 int __init amd_iommu_prepare(void)
2682 amd_iommu_irq_remap = true;
2684 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2687 return amd_iommu_irq_remap ? 0 : -ENODEV;
2690 int __init amd_iommu_enable(void)
2694 ret = iommu_go_to_state(IOMMU_ENABLED);
2698 irq_remapping_enabled = 1;
2703 void amd_iommu_disable(void)
2705 amd_iommu_suspend();
2708 int amd_iommu_reenable(int mode)
2715 int __init amd_iommu_enable_faulting(void)
2717 /* We enable MSI later when PCI is initialized */
2723 * This is the core init function for AMD IOMMU hardware in the system.
2724 * This function is called from the generic x86 DMA layer initialization
2727 static int __init amd_iommu_init(void)
2731 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2733 free_dma_resources();
2734 if (!irq_remapping_enabled) {
2736 free_iommu_resources();
2738 struct amd_iommu *iommu;
2740 uninit_device_table_dma();
2741 for_each_iommu(iommu)
2742 iommu_flush_all_caches(iommu);
2749 static bool amd_iommu_sme_check(void)
2751 if (!sme_active() || (boot_cpu_data.x86 != 0x17))
2754 /* For Fam17h, a specific level of support is required */
2755 if (boot_cpu_data.microcode >= 0x08001205)
2758 if ((boot_cpu_data.microcode >= 0x08001126) &&
2759 (boot_cpu_data.microcode <= 0x080011ff))
2762 pr_notice("AMD-Vi: IOMMU not currently supported when SME is active\n");
2767 /****************************************************************************
2769 * Early detect code. This code runs at IOMMU detection time in the DMA
2770 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2773 ****************************************************************************/
2774 int __init amd_iommu_detect(void)
2778 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2781 if (!amd_iommu_sme_check())
2784 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2788 amd_iommu_detected = true;
2790 x86_init.iommu.iommu_init = amd_iommu_init;
2795 /****************************************************************************
2797 * Parsing functions for the AMD IOMMU specific kernel command line
2800 ****************************************************************************/
2802 static int __init parse_amd_iommu_dump(char *str)
2804 amd_iommu_dump = true;
2809 static int __init parse_amd_iommu_intr(char *str)
2811 for (; *str; ++str) {
2812 if (strncmp(str, "legacy", 6) == 0) {
2813 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2816 if (strncmp(str, "vapic", 5) == 0) {
2817 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2824 static int __init parse_amd_iommu_options(char *str)
2826 for (; *str; ++str) {
2827 if (strncmp(str, "fullflush", 9) == 0)
2828 amd_iommu_unmap_flush = true;
2829 if (strncmp(str, "off", 3) == 0)
2830 amd_iommu_disabled = true;
2831 if (strncmp(str, "force_isolation", 15) == 0)
2832 amd_iommu_force_isolation = true;
2838 static int __init parse_ivrs_ioapic(char *str)
2840 unsigned int bus, dev, fn;
2844 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2847 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2851 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2852 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2857 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2859 cmdline_maps = true;
2860 i = early_ioapic_map_size++;
2861 early_ioapic_map[i].id = id;
2862 early_ioapic_map[i].devid = devid;
2863 early_ioapic_map[i].cmd_line = true;
2868 static int __init parse_ivrs_hpet(char *str)
2870 unsigned int bus, dev, fn;
2874 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2877 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2881 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2882 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2887 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2889 cmdline_maps = true;
2890 i = early_hpet_map_size++;
2891 early_hpet_map[i].id = id;
2892 early_hpet_map[i].devid = devid;
2893 early_hpet_map[i].cmd_line = true;
2898 static int __init parse_ivrs_acpihid(char *str)
2901 char *hid, *uid, *p;
2902 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
2905 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
2907 pr_err("AMD-Vi: Invalid command line: ivrs_acpihid(%s)\n", str);
2912 hid = strsep(&p, ":");
2915 if (!hid || !(*hid) || !uid) {
2916 pr_err("AMD-Vi: Invalid command line: hid or uid\n");
2920 i = early_acpihid_map_size++;
2921 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
2922 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
2923 early_acpihid_map[i].devid =
2924 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2925 early_acpihid_map[i].cmd_line = true;
2930 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2931 __setup("amd_iommu=", parse_amd_iommu_options);
2932 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
2933 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2934 __setup("ivrs_hpet", parse_ivrs_hpet);
2935 __setup("ivrs_acpihid", parse_ivrs_acpihid);
2937 IOMMU_INIT_FINISH(amd_iommu_detect,
2938 gart_iommu_hole_init,
2942 bool amd_iommu_v2_supported(void)
2944 return amd_iommu_v2_present;
2946 EXPORT_SYMBOL(amd_iommu_v2_supported);
2948 struct amd_iommu *get_amd_iommu(unsigned int idx)
2951 struct amd_iommu *iommu;
2953 for_each_iommu(iommu)
2958 EXPORT_SYMBOL(get_amd_iommu);
2960 /****************************************************************************
2962 * IOMMU EFR Performance Counter support functionality. This code allows
2963 * access to the IOMMU PC functionality.
2965 ****************************************************************************/
2967 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
2969 struct amd_iommu *iommu = get_amd_iommu(idx);
2972 return iommu->max_banks;
2976 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2978 bool amd_iommu_pc_supported(void)
2980 return amd_iommu_pc_present;
2982 EXPORT_SYMBOL(amd_iommu_pc_supported);
2984 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
2986 struct amd_iommu *iommu = get_amd_iommu(idx);
2989 return iommu->max_counters;
2993 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2995 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
2996 u8 fxn, u64 *value, bool is_write)
3001 /* Make sure the IOMMU PC resource is available */
3002 if (!amd_iommu_pc_present)
3005 /* Check for valid iommu and pc register indexing */
3006 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3009 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3011 /* Limit the offset to the hw defined mmio region aperture */
3012 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3013 (iommu->max_counters << 8) | 0x28);
3014 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3015 (offset > max_offset_lim))
3019 u64 val = *value & GENMASK_ULL(47, 0);
3021 writel((u32)val, iommu->mmio_base + offset);
3022 writel((val >> 32), iommu->mmio_base + offset + 4);
3024 *value = readl(iommu->mmio_base + offset + 4);
3026 *value |= readl(iommu->mmio_base + offset);
3027 *value &= GENMASK_ULL(47, 0);
3033 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3038 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3040 EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3042 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3047 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3049 EXPORT_SYMBOL(amd_iommu_pc_set_reg);