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/crash_dump.h>
33 #include <linux/mem_encrypt.h>
34 #include <asm/pci-direct.h>
35 #include <asm/iommu.h>
37 #include <asm/x86_init.h>
38 #include <asm/iommu_table.h>
39 #include <asm/io_apic.h>
40 #include <asm/irq_remapping.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
;
202 * The alias table is a driver specific data structure which contains the
203 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
204 * More than one device can share the same requestor id.
206 u16
*amd_iommu_alias_table
;
209 * The rlookup table is used to find the IOMMU which is responsible
210 * for a specific device. It is also indexed by the PCI device id.
212 struct amd_iommu
**amd_iommu_rlookup_table
;
215 * This table is used to find the irq remapping table for a given device id
218 struct irq_remap_table
**irq_lookup_table
;
221 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
222 * to know which ones are already in use.
224 unsigned long *amd_iommu_pd_alloc_bitmap
;
226 static u32 dev_table_size
; /* size of the device table */
227 static u32 alias_table_size
; /* size of the alias table */
228 static u32 rlookup_table_size
; /* size if the rlookup table */
230 enum iommu_init_state
{
241 IOMMU_CMDLINE_DISABLED
,
244 /* Early ioapic and hpet maps from kernel command line */
245 #define EARLY_MAP_SIZE 4
246 static struct devid_map __initdata early_ioapic_map
[EARLY_MAP_SIZE
];
247 static struct devid_map __initdata early_hpet_map
[EARLY_MAP_SIZE
];
248 static struct acpihid_map_entry __initdata early_acpihid_map
[EARLY_MAP_SIZE
];
250 static int __initdata early_ioapic_map_size
;
251 static int __initdata early_hpet_map_size
;
252 static int __initdata early_acpihid_map_size
;
254 static bool __initdata cmdline_maps
;
256 static enum iommu_init_state init_state
= IOMMU_START_STATE
;
258 static int amd_iommu_enable_interrupts(void);
259 static int __init
iommu_go_to_state(enum iommu_init_state state
);
260 static void init_device_table_dma(void);
262 static inline void update_last_devid(u16 devid
)
264 if (devid
> amd_iommu_last_bdf
)
265 amd_iommu_last_bdf
= devid
;
268 static inline unsigned long tbl_size(int entry_size
)
270 unsigned shift
= PAGE_SHIFT
+
271 get_order(((int)amd_iommu_last_bdf
+ 1) * entry_size
);
276 int amd_iommu_get_num_iommus(void)
278 return amd_iommus_present
;
281 /* Access to l1 and l2 indexed register spaces */
283 static u32
iommu_read_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
)
287 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
288 pci_read_config_dword(iommu
->dev
, 0xfc, &val
);
292 static void iommu_write_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
, u32 val
)
294 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16 | 1 << 31));
295 pci_write_config_dword(iommu
->dev
, 0xfc, val
);
296 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
299 static u32
iommu_read_l2(struct amd_iommu
*iommu
, u8 address
)
303 pci_write_config_dword(iommu
->dev
, 0xf0, address
);
304 pci_read_config_dword(iommu
->dev
, 0xf4, &val
);
308 static void iommu_write_l2(struct amd_iommu
*iommu
, u8 address
, u32 val
)
310 pci_write_config_dword(iommu
->dev
, 0xf0, (address
| 1 << 8));
311 pci_write_config_dword(iommu
->dev
, 0xf4, val
);
314 /****************************************************************************
316 * AMD IOMMU MMIO register space handling functions
318 * These functions are used to program the IOMMU device registers in
319 * MMIO space required for that driver.
321 ****************************************************************************/
324 * This function set the exclusion range in the IOMMU. DMA accesses to the
325 * exclusion range are passed through untranslated
327 static void iommu_set_exclusion_range(struct amd_iommu
*iommu
)
329 u64 start
= iommu
->exclusion_start
& PAGE_MASK
;
330 u64 limit
= (start
+ iommu
->exclusion_length
) & PAGE_MASK
;
333 if (!iommu
->exclusion_start
)
336 entry
= start
| MMIO_EXCL_ENABLE_MASK
;
337 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_BASE_OFFSET
,
338 &entry
, sizeof(entry
));
341 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_LIMIT_OFFSET
,
342 &entry
, sizeof(entry
));
345 /* Programs the physical address of the device table into the IOMMU hardware */
346 static void iommu_set_device_table(struct amd_iommu
*iommu
)
350 BUG_ON(iommu
->mmio_base
== NULL
);
352 entry
= iommu_virt_to_phys(amd_iommu_dev_table
);
353 entry
|= (dev_table_size
>> 12) - 1;
354 memcpy_toio(iommu
->mmio_base
+ MMIO_DEV_TABLE_OFFSET
,
355 &entry
, sizeof(entry
));
358 /* Generic functions to enable/disable certain features of the IOMMU. */
359 static void iommu_feature_enable(struct amd_iommu
*iommu
, u8 bit
)
363 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
365 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
368 static void iommu_feature_disable(struct amd_iommu
*iommu
, u8 bit
)
372 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
374 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
377 static void iommu_set_inv_tlb_timeout(struct amd_iommu
*iommu
, int timeout
)
381 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
382 ctrl
&= ~CTRL_INV_TO_MASK
;
383 ctrl
|= (timeout
<< CONTROL_INV_TIMEOUT
) & CTRL_INV_TO_MASK
;
384 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
387 /* Function to enable the hardware */
388 static void iommu_enable(struct amd_iommu
*iommu
)
390 iommu_feature_enable(iommu
, CONTROL_IOMMU_EN
);
393 static void iommu_disable(struct amd_iommu
*iommu
)
395 /* Disable command buffer */
396 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
398 /* Disable event logging and event interrupts */
399 iommu_feature_disable(iommu
, CONTROL_EVT_INT_EN
);
400 iommu_feature_disable(iommu
, CONTROL_EVT_LOG_EN
);
402 /* Disable IOMMU GA_LOG */
403 iommu_feature_disable(iommu
, CONTROL_GALOG_EN
);
404 iommu_feature_disable(iommu
, CONTROL_GAINT_EN
);
406 /* Disable IOMMU hardware itself */
407 iommu_feature_disable(iommu
, CONTROL_IOMMU_EN
);
411 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
412 * the system has one.
414 static u8 __iomem
* __init
iommu_map_mmio_space(u64 address
, u64 end
)
416 if (!request_mem_region(address
, end
, "amd_iommu")) {
417 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
419 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
423 return (u8 __iomem
*)ioremap_nocache(address
, end
);
426 static void __init
iommu_unmap_mmio_space(struct amd_iommu
*iommu
)
428 if (iommu
->mmio_base
)
429 iounmap(iommu
->mmio_base
);
430 release_mem_region(iommu
->mmio_phys
, iommu
->mmio_phys_end
);
433 static inline u32
get_ivhd_header_size(struct ivhd_header
*h
)
449 /****************************************************************************
451 * The functions below belong to the first pass of AMD IOMMU ACPI table
452 * parsing. In this pass we try to find out the highest device id this
453 * code has to handle. Upon this information the size of the shared data
454 * structures is determined later.
456 ****************************************************************************/
459 * This function calculates the length of a given IVHD entry
461 static inline int ivhd_entry_length(u8
*ivhd
)
463 u32 type
= ((struct ivhd_entry
*)ivhd
)->type
;
466 return 0x04 << (*ivhd
>> 6);
467 } else if (type
== IVHD_DEV_ACPI_HID
) {
468 /* For ACPI_HID, offset 21 is uid len */
469 return *((u8
*)ivhd
+ 21) + 22;
475 * After reading the highest device id from the IOMMU PCI capability header
476 * this function looks if there is a higher device id defined in the ACPI table
478 static int __init
find_last_devid_from_ivhd(struct ivhd_header
*h
)
480 u8
*p
= (void *)h
, *end
= (void *)h
;
481 struct ivhd_entry
*dev
;
483 u32 ivhd_size
= get_ivhd_header_size(h
);
486 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h
->type
);
494 dev
= (struct ivhd_entry
*)p
;
497 /* Use maximum BDF value for DEV_ALL */
498 update_last_devid(0xffff);
500 case IVHD_DEV_SELECT
:
501 case IVHD_DEV_RANGE_END
:
503 case IVHD_DEV_EXT_SELECT
:
504 /* all the above subfield types refer to device ids */
505 update_last_devid(dev
->devid
);
510 p
+= ivhd_entry_length(p
);
518 static int __init
check_ivrs_checksum(struct acpi_table_header
*table
)
521 u8 checksum
= 0, *p
= (u8
*)table
;
523 for (i
= 0; i
< table
->length
; ++i
)
526 /* ACPI table corrupt */
527 pr_err(FW_BUG
"AMD-Vi: IVRS invalid checksum\n");
535 * Iterate over all IVHD entries in the ACPI table and find the highest device
536 * id which we need to handle. This is the first of three functions which parse
537 * the ACPI table. So we check the checksum here.
539 static int __init
find_last_devid_acpi(struct acpi_table_header
*table
)
541 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
542 struct ivhd_header
*h
;
544 p
+= IVRS_HEADER_LENGTH
;
546 end
+= table
->length
;
548 h
= (struct ivhd_header
*)p
;
549 if (h
->type
== amd_iommu_target_ivhd_type
) {
550 int ret
= find_last_devid_from_ivhd(h
);
562 /****************************************************************************
564 * The following functions belong to the code path which parses the ACPI table
565 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
566 * data structures, initialize the device/alias/rlookup table and also
567 * basically initialize the hardware.
569 ****************************************************************************/
572 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
573 * write commands to that buffer later and the IOMMU will execute them
576 static int __init
alloc_command_buffer(struct amd_iommu
*iommu
)
578 iommu
->cmd_buf
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
579 get_order(CMD_BUFFER_SIZE
));
581 return iommu
->cmd_buf
? 0 : -ENOMEM
;
585 * This function resets the command buffer if the IOMMU stopped fetching
588 void amd_iommu_reset_cmd_buffer(struct amd_iommu
*iommu
)
590 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
592 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_HEAD_OFFSET
);
593 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_TAIL_OFFSET
);
594 iommu
->cmd_buf_head
= 0;
595 iommu
->cmd_buf_tail
= 0;
597 iommu_feature_enable(iommu
, CONTROL_CMDBUF_EN
);
601 * This function writes the command buffer address to the hardware and
604 static void iommu_enable_command_buffer(struct amd_iommu
*iommu
)
608 BUG_ON(iommu
->cmd_buf
== NULL
);
610 entry
= iommu_virt_to_phys(iommu
->cmd_buf
);
611 entry
|= MMIO_CMD_SIZE_512
;
613 memcpy_toio(iommu
->mmio_base
+ MMIO_CMD_BUF_OFFSET
,
614 &entry
, sizeof(entry
));
616 amd_iommu_reset_cmd_buffer(iommu
);
619 static void __init
free_command_buffer(struct amd_iommu
*iommu
)
621 free_pages((unsigned long)iommu
->cmd_buf
, get_order(CMD_BUFFER_SIZE
));
624 /* allocates the memory where the IOMMU will log its events to */
625 static int __init
alloc_event_buffer(struct amd_iommu
*iommu
)
627 iommu
->evt_buf
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
628 get_order(EVT_BUFFER_SIZE
));
630 return iommu
->evt_buf
? 0 : -ENOMEM
;
633 static void iommu_enable_event_buffer(struct amd_iommu
*iommu
)
637 BUG_ON(iommu
->evt_buf
== NULL
);
639 entry
= iommu_virt_to_phys(iommu
->evt_buf
) | EVT_LEN_MASK
;
641 memcpy_toio(iommu
->mmio_base
+ MMIO_EVT_BUF_OFFSET
,
642 &entry
, sizeof(entry
));
644 /* set head and tail to zero manually */
645 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_HEAD_OFFSET
);
646 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_TAIL_OFFSET
);
648 iommu_feature_enable(iommu
, CONTROL_EVT_LOG_EN
);
651 static void __init
free_event_buffer(struct amd_iommu
*iommu
)
653 free_pages((unsigned long)iommu
->evt_buf
, get_order(EVT_BUFFER_SIZE
));
656 /* allocates the memory where the IOMMU will log its events to */
657 static int __init
alloc_ppr_log(struct amd_iommu
*iommu
)
659 iommu
->ppr_log
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
660 get_order(PPR_LOG_SIZE
));
662 return iommu
->ppr_log
? 0 : -ENOMEM
;
665 static void iommu_enable_ppr_log(struct amd_iommu
*iommu
)
669 if (iommu
->ppr_log
== NULL
)
672 entry
= iommu_virt_to_phys(iommu
->ppr_log
) | PPR_LOG_SIZE_512
;
674 memcpy_toio(iommu
->mmio_base
+ MMIO_PPR_LOG_OFFSET
,
675 &entry
, sizeof(entry
));
677 /* set head and tail to zero manually */
678 writel(0x00, iommu
->mmio_base
+ MMIO_PPR_HEAD_OFFSET
);
679 writel(0x00, iommu
->mmio_base
+ MMIO_PPR_TAIL_OFFSET
);
681 iommu_feature_enable(iommu
, CONTROL_PPFLOG_EN
);
682 iommu_feature_enable(iommu
, CONTROL_PPR_EN
);
685 static void __init
free_ppr_log(struct amd_iommu
*iommu
)
687 if (iommu
->ppr_log
== NULL
)
690 free_pages((unsigned long)iommu
->ppr_log
, get_order(PPR_LOG_SIZE
));
693 static void free_ga_log(struct amd_iommu
*iommu
)
695 #ifdef CONFIG_IRQ_REMAP
697 free_pages((unsigned long)iommu
->ga_log
,
698 get_order(GA_LOG_SIZE
));
699 if (iommu
->ga_log_tail
)
700 free_pages((unsigned long)iommu
->ga_log_tail
,
705 static int iommu_ga_log_enable(struct amd_iommu
*iommu
)
707 #ifdef CONFIG_IRQ_REMAP
713 status
= readl(iommu
->mmio_base
+ MMIO_STATUS_OFFSET
);
715 /* Check if already running */
716 if (status
& (MMIO_STATUS_GALOG_RUN_MASK
))
719 iommu_feature_enable(iommu
, CONTROL_GAINT_EN
);
720 iommu_feature_enable(iommu
, CONTROL_GALOG_EN
);
722 for (i
= 0; i
< LOOP_TIMEOUT
; ++i
) {
723 status
= readl(iommu
->mmio_base
+ MMIO_STATUS_OFFSET
);
724 if (status
& (MMIO_STATUS_GALOG_RUN_MASK
))
728 if (i
>= LOOP_TIMEOUT
)
730 #endif /* CONFIG_IRQ_REMAP */
734 #ifdef CONFIG_IRQ_REMAP
735 static int iommu_init_ga_log(struct amd_iommu
*iommu
)
739 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
742 iommu
->ga_log
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
743 get_order(GA_LOG_SIZE
));
747 iommu
->ga_log_tail
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
749 if (!iommu
->ga_log_tail
)
752 entry
= iommu_virt_to_phys(iommu
->ga_log
) | GA_LOG_SIZE_512
;
753 memcpy_toio(iommu
->mmio_base
+ MMIO_GA_LOG_BASE_OFFSET
,
754 &entry
, sizeof(entry
));
755 entry
= (iommu_virt_to_phys(iommu
->ga_log
) & 0xFFFFFFFFFFFFFULL
) & ~7ULL;
756 memcpy_toio(iommu
->mmio_base
+ MMIO_GA_LOG_TAIL_OFFSET
,
757 &entry
, sizeof(entry
));
758 writel(0x00, iommu
->mmio_base
+ MMIO_GA_HEAD_OFFSET
);
759 writel(0x00, iommu
->mmio_base
+ MMIO_GA_TAIL_OFFSET
);
766 #endif /* CONFIG_IRQ_REMAP */
768 static int iommu_init_ga(struct amd_iommu
*iommu
)
772 #ifdef CONFIG_IRQ_REMAP
773 /* Note: We have already checked GASup from IVRS table.
774 * Now, we need to make sure that GAMSup is set.
776 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
) &&
777 !iommu_feature(iommu
, FEATURE_GAM_VAPIC
))
778 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY_GA
;
780 ret
= iommu_init_ga_log(iommu
);
781 #endif /* CONFIG_IRQ_REMAP */
786 static void iommu_enable_gt(struct amd_iommu
*iommu
)
788 if (!iommu_feature(iommu
, FEATURE_GT
))
791 iommu_feature_enable(iommu
, CONTROL_GT_EN
);
794 /* sets a specific bit in the device table entry. */
795 static void set_dev_entry_bit(u16 devid
, u8 bit
)
797 int i
= (bit
>> 6) & 0x03;
798 int _bit
= bit
& 0x3f;
800 amd_iommu_dev_table
[devid
].data
[i
] |= (1UL << _bit
);
803 static int get_dev_entry_bit(u16 devid
, u8 bit
)
805 int i
= (bit
>> 6) & 0x03;
806 int _bit
= bit
& 0x3f;
808 return (amd_iommu_dev_table
[devid
].data
[i
] & (1UL << _bit
)) >> _bit
;
812 void amd_iommu_apply_erratum_63(u16 devid
)
816 sysmgt
= get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
) |
817 (get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
) << 1);
820 set_dev_entry_bit(devid
, DEV_ENTRY_IW
);
823 /* Writes the specific IOMMU for a device into the rlookup table */
824 static void __init
set_iommu_for_device(struct amd_iommu
*iommu
, u16 devid
)
826 amd_iommu_rlookup_table
[devid
] = iommu
;
830 * This function takes the device specific flags read from the ACPI
831 * table and sets up the device table entry with that information
833 static void __init
set_dev_entry_from_acpi(struct amd_iommu
*iommu
,
834 u16 devid
, u32 flags
, u32 ext_flags
)
836 if (flags
& ACPI_DEVFLAG_INITPASS
)
837 set_dev_entry_bit(devid
, DEV_ENTRY_INIT_PASS
);
838 if (flags
& ACPI_DEVFLAG_EXTINT
)
839 set_dev_entry_bit(devid
, DEV_ENTRY_EINT_PASS
);
840 if (flags
& ACPI_DEVFLAG_NMI
)
841 set_dev_entry_bit(devid
, DEV_ENTRY_NMI_PASS
);
842 if (flags
& ACPI_DEVFLAG_SYSMGT1
)
843 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
);
844 if (flags
& ACPI_DEVFLAG_SYSMGT2
)
845 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
);
846 if (flags
& ACPI_DEVFLAG_LINT0
)
847 set_dev_entry_bit(devid
, DEV_ENTRY_LINT0_PASS
);
848 if (flags
& ACPI_DEVFLAG_LINT1
)
849 set_dev_entry_bit(devid
, DEV_ENTRY_LINT1_PASS
);
851 amd_iommu_apply_erratum_63(devid
);
853 set_iommu_for_device(iommu
, devid
);
856 static int __init
add_special_device(u8 type
, u8 id
, u16
*devid
, bool cmd_line
)
858 struct devid_map
*entry
;
859 struct list_head
*list
;
861 if (type
== IVHD_SPECIAL_IOAPIC
)
863 else if (type
== IVHD_SPECIAL_HPET
)
868 list_for_each_entry(entry
, list
, list
) {
869 if (!(entry
->id
== id
&& entry
->cmd_line
))
872 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
873 type
== IVHD_SPECIAL_IOAPIC
? "IOAPIC" : "HPET", id
);
875 *devid
= entry
->devid
;
880 entry
= kzalloc(sizeof(*entry
), GFP_KERNEL
);
885 entry
->devid
= *devid
;
886 entry
->cmd_line
= cmd_line
;
888 list_add_tail(&entry
->list
, list
);
893 static int __init
add_acpi_hid_device(u8
*hid
, u8
*uid
, u16
*devid
,
896 struct acpihid_map_entry
*entry
;
897 struct list_head
*list
= &acpihid_map
;
899 list_for_each_entry(entry
, list
, list
) {
900 if (strcmp(entry
->hid
, hid
) ||
901 (*uid
&& *entry
->uid
&& strcmp(entry
->uid
, uid
)) ||
905 pr_info("AMD-Vi: Command-line override for hid:%s uid:%s\n",
907 *devid
= entry
->devid
;
911 entry
= kzalloc(sizeof(*entry
), GFP_KERNEL
);
915 memcpy(entry
->uid
, uid
, strlen(uid
));
916 memcpy(entry
->hid
, hid
, strlen(hid
));
917 entry
->devid
= *devid
;
918 entry
->cmd_line
= cmd_line
;
919 entry
->root_devid
= (entry
->devid
& (~0x7));
921 pr_info("AMD-Vi:%s, add hid:%s, uid:%s, rdevid:%d\n",
922 entry
->cmd_line
? "cmd" : "ivrs",
923 entry
->hid
, entry
->uid
, entry
->root_devid
);
925 list_add_tail(&entry
->list
, list
);
929 static int __init
add_early_maps(void)
933 for (i
= 0; i
< early_ioapic_map_size
; ++i
) {
934 ret
= add_special_device(IVHD_SPECIAL_IOAPIC
,
935 early_ioapic_map
[i
].id
,
936 &early_ioapic_map
[i
].devid
,
937 early_ioapic_map
[i
].cmd_line
);
942 for (i
= 0; i
< early_hpet_map_size
; ++i
) {
943 ret
= add_special_device(IVHD_SPECIAL_HPET
,
944 early_hpet_map
[i
].id
,
945 &early_hpet_map
[i
].devid
,
946 early_hpet_map
[i
].cmd_line
);
951 for (i
= 0; i
< early_acpihid_map_size
; ++i
) {
952 ret
= add_acpi_hid_device(early_acpihid_map
[i
].hid
,
953 early_acpihid_map
[i
].uid
,
954 &early_acpihid_map
[i
].devid
,
955 early_acpihid_map
[i
].cmd_line
);
964 * Reads the device exclusion range from ACPI and initializes the IOMMU with
967 static void __init
set_device_exclusion_range(u16 devid
, struct ivmd_header
*m
)
969 struct amd_iommu
*iommu
= amd_iommu_rlookup_table
[devid
];
971 if (!(m
->flags
& IVMD_FLAG_EXCL_RANGE
))
976 * We only can configure exclusion ranges per IOMMU, not
977 * per device. But we can enable the exclusion range per
978 * device. This is done here
980 set_dev_entry_bit(devid
, DEV_ENTRY_EX
);
981 iommu
->exclusion_start
= m
->range_start
;
982 iommu
->exclusion_length
= m
->range_length
;
987 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
988 * initializes the hardware and our data structures with it.
990 static int __init
init_iommu_from_acpi(struct amd_iommu
*iommu
,
991 struct ivhd_header
*h
)
994 u8
*end
= p
, flags
= 0;
995 u16 devid
= 0, devid_start
= 0, devid_to
= 0;
996 u32 dev_i
, ext_flags
= 0;
998 struct ivhd_entry
*e
;
1003 ret
= add_early_maps();
1008 * First save the recommended feature enable bits from ACPI
1010 iommu
->acpi_flags
= h
->flags
;
1013 * Done. Now parse the device entries
1015 ivhd_size
= get_ivhd_header_size(h
);
1017 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h
->type
);
1027 e
= (struct ivhd_entry
*)p
;
1031 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e
->flags
);
1033 for (dev_i
= 0; dev_i
<= amd_iommu_last_bdf
; ++dev_i
)
1034 set_dev_entry_from_acpi(iommu
, dev_i
, e
->flags
, 0);
1036 case IVHD_DEV_SELECT
:
1038 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1040 PCI_BUS_NUM(e
->devid
),
1046 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1048 case IVHD_DEV_SELECT_RANGE_START
:
1050 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1051 "devid: %02x:%02x.%x flags: %02x\n",
1052 PCI_BUS_NUM(e
->devid
),
1057 devid_start
= e
->devid
;
1062 case IVHD_DEV_ALIAS
:
1064 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1065 "flags: %02x devid_to: %02x:%02x.%x\n",
1066 PCI_BUS_NUM(e
->devid
),
1070 PCI_BUS_NUM(e
->ext
>> 8),
1071 PCI_SLOT(e
->ext
>> 8),
1072 PCI_FUNC(e
->ext
>> 8));
1075 devid_to
= e
->ext
>> 8;
1076 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1077 set_dev_entry_from_acpi(iommu
, devid_to
, e
->flags
, 0);
1078 amd_iommu_alias_table
[devid
] = devid_to
;
1080 case IVHD_DEV_ALIAS_RANGE
:
1082 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1083 "devid: %02x:%02x.%x flags: %02x "
1084 "devid_to: %02x:%02x.%x\n",
1085 PCI_BUS_NUM(e
->devid
),
1089 PCI_BUS_NUM(e
->ext
>> 8),
1090 PCI_SLOT(e
->ext
>> 8),
1091 PCI_FUNC(e
->ext
>> 8));
1093 devid_start
= e
->devid
;
1095 devid_to
= e
->ext
>> 8;
1099 case IVHD_DEV_EXT_SELECT
:
1101 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1102 "flags: %02x ext: %08x\n",
1103 PCI_BUS_NUM(e
->devid
),
1109 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
,
1112 case IVHD_DEV_EXT_SELECT_RANGE
:
1114 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1115 "%02x:%02x.%x flags: %02x ext: %08x\n",
1116 PCI_BUS_NUM(e
->devid
),
1121 devid_start
= e
->devid
;
1126 case IVHD_DEV_RANGE_END
:
1128 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1129 PCI_BUS_NUM(e
->devid
),
1131 PCI_FUNC(e
->devid
));
1134 for (dev_i
= devid_start
; dev_i
<= devid
; ++dev_i
) {
1136 amd_iommu_alias_table
[dev_i
] = devid_to
;
1137 set_dev_entry_from_acpi(iommu
,
1138 devid_to
, flags
, ext_flags
);
1140 set_dev_entry_from_acpi(iommu
, dev_i
,
1144 case IVHD_DEV_SPECIAL
: {
1150 handle
= e
->ext
& 0xff;
1151 devid
= (e
->ext
>> 8) & 0xffff;
1152 type
= (e
->ext
>> 24) & 0xff;
1154 if (type
== IVHD_SPECIAL_IOAPIC
)
1156 else if (type
== IVHD_SPECIAL_HPET
)
1161 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1167 ret
= add_special_device(type
, handle
, &devid
, false);
1172 * add_special_device might update the devid in case a
1173 * command-line override is present. So call
1174 * set_dev_entry_from_acpi after add_special_device.
1176 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1180 case IVHD_DEV_ACPI_HID
: {
1182 u8 hid
[ACPIHID_HID_LEN
] = {0};
1183 u8 uid
[ACPIHID_UID_LEN
] = {0};
1186 if (h
->type
!= 0x40) {
1187 pr_err(FW_BUG
"Invalid IVHD device type %#x\n",
1192 memcpy(hid
, (u8
*)(&e
->ext
), ACPIHID_HID_LEN
- 1);
1193 hid
[ACPIHID_HID_LEN
- 1] = '\0';
1196 pr_err(FW_BUG
"Invalid HID.\n");
1201 case UID_NOT_PRESENT
:
1204 pr_warn(FW_BUG
"Invalid UID length.\n");
1207 case UID_IS_INTEGER
:
1209 sprintf(uid
, "%d", e
->uid
);
1212 case UID_IS_CHARACTER
:
1214 memcpy(uid
, (u8
*)(&e
->uid
), ACPIHID_UID_LEN
- 1);
1215 uid
[ACPIHID_UID_LEN
- 1] = '\0';
1223 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1231 ret
= add_acpi_hid_device(hid
, uid
, &devid
, false);
1236 * add_special_device might update the devid in case a
1237 * command-line override is present. So call
1238 * set_dev_entry_from_acpi after add_special_device.
1240 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1248 p
+= ivhd_entry_length(p
);
1254 static void __init
free_iommu_one(struct amd_iommu
*iommu
)
1256 free_command_buffer(iommu
);
1257 free_event_buffer(iommu
);
1258 free_ppr_log(iommu
);
1260 iommu_unmap_mmio_space(iommu
);
1263 static void __init
free_iommu_all(void)
1265 struct amd_iommu
*iommu
, *next
;
1267 for_each_iommu_safe(iommu
, next
) {
1268 list_del(&iommu
->list
);
1269 free_iommu_one(iommu
);
1275 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1277 * BIOS should disable L2B micellaneous clock gating by setting
1278 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1280 static void amd_iommu_erratum_746_workaround(struct amd_iommu
*iommu
)
1284 if ((boot_cpu_data
.x86
!= 0x15) ||
1285 (boot_cpu_data
.x86_model
< 0x10) ||
1286 (boot_cpu_data
.x86_model
> 0x1f))
1289 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
1290 pci_read_config_dword(iommu
->dev
, 0xf4, &value
);
1295 /* Select NB indirect register 0x90 and enable writing */
1296 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90 | (1 << 8));
1298 pci_write_config_dword(iommu
->dev
, 0xf4, value
| 0x4);
1299 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1300 dev_name(&iommu
->dev
->dev
));
1302 /* Clear the enable writing bit */
1303 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
1307 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1309 * BIOS should enable ATS write permission check by setting
1310 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1312 static void amd_iommu_ats_write_check_workaround(struct amd_iommu
*iommu
)
1316 if ((boot_cpu_data
.x86
!= 0x15) ||
1317 (boot_cpu_data
.x86_model
< 0x30) ||
1318 (boot_cpu_data
.x86_model
> 0x3f))
1321 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1322 value
= iommu_read_l2(iommu
, 0x47);
1327 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1328 iommu_write_l2(iommu
, 0x47, value
| BIT(0));
1330 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1331 dev_name(&iommu
->dev
->dev
));
1335 * This function clues the initialization function for one IOMMU
1336 * together and also allocates the command buffer and programs the
1337 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1339 static int __init
init_iommu_one(struct amd_iommu
*iommu
, struct ivhd_header
*h
)
1343 spin_lock_init(&iommu
->lock
);
1345 /* Add IOMMU to internal data structures */
1346 list_add_tail(&iommu
->list
, &amd_iommu_list
);
1347 iommu
->index
= amd_iommus_present
++;
1349 if (unlikely(iommu
->index
>= MAX_IOMMUS
)) {
1350 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1354 /* Index is fine - add IOMMU to the array */
1355 amd_iommus
[iommu
->index
] = iommu
;
1358 * Copy data from ACPI table entry to the iommu struct
1360 iommu
->devid
= h
->devid
;
1361 iommu
->cap_ptr
= h
->cap_ptr
;
1362 iommu
->pci_seg
= h
->pci_seg
;
1363 iommu
->mmio_phys
= h
->mmio_phys
;
1367 /* Check if IVHD EFR contains proper max banks/counters */
1368 if ((h
->efr_attr
!= 0) &&
1369 ((h
->efr_attr
& (0xF << 13)) != 0) &&
1370 ((h
->efr_attr
& (0x3F << 17)) != 0))
1371 iommu
->mmio_phys_end
= MMIO_REG_END_OFFSET
;
1373 iommu
->mmio_phys_end
= MMIO_CNTR_CONF_OFFSET
;
1374 if (((h
->efr_attr
& (0x1 << IOMMU_FEAT_GASUP_SHIFT
)) == 0))
1375 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY
;
1379 if (h
->efr_reg
& (1 << 9))
1380 iommu
->mmio_phys_end
= MMIO_REG_END_OFFSET
;
1382 iommu
->mmio_phys_end
= MMIO_CNTR_CONF_OFFSET
;
1383 if (((h
->efr_reg
& (0x1 << IOMMU_EFR_GASUP_SHIFT
)) == 0))
1384 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY
;
1390 iommu
->mmio_base
= iommu_map_mmio_space(iommu
->mmio_phys
,
1391 iommu
->mmio_phys_end
);
1392 if (!iommu
->mmio_base
)
1395 if (alloc_command_buffer(iommu
))
1398 if (alloc_event_buffer(iommu
))
1401 iommu
->int_enabled
= false;
1403 ret
= init_iommu_from_acpi(iommu
, h
);
1407 ret
= amd_iommu_create_irq_domain(iommu
);
1412 * Make sure IOMMU is not considered to translate itself. The IVRS
1413 * table tells us so, but this is a lie!
1415 amd_iommu_rlookup_table
[iommu
->devid
] = NULL
;
1421 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1422 * @ivrs Pointer to the IVRS header
1424 * This function search through all IVDB of the maximum supported IVHD
1426 static u8
get_highest_supported_ivhd_type(struct acpi_table_header
*ivrs
)
1428 u8
*base
= (u8
*)ivrs
;
1429 struct ivhd_header
*ivhd
= (struct ivhd_header
*)
1430 (base
+ IVRS_HEADER_LENGTH
);
1431 u8 last_type
= ivhd
->type
;
1432 u16 devid
= ivhd
->devid
;
1434 while (((u8
*)ivhd
- base
< ivrs
->length
) &&
1435 (ivhd
->type
<= ACPI_IVHD_TYPE_MAX_SUPPORTED
)) {
1436 u8
*p
= (u8
*) ivhd
;
1438 if (ivhd
->devid
== devid
)
1439 last_type
= ivhd
->type
;
1440 ivhd
= (struct ivhd_header
*)(p
+ ivhd
->length
);
1447 * Iterates over all IOMMU entries in the ACPI table, allocates the
1448 * IOMMU structure and initializes it with init_iommu_one()
1450 static int __init
init_iommu_all(struct acpi_table_header
*table
)
1452 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1453 struct ivhd_header
*h
;
1454 struct amd_iommu
*iommu
;
1457 end
+= table
->length
;
1458 p
+= IVRS_HEADER_LENGTH
;
1461 h
= (struct ivhd_header
*)p
;
1462 if (*p
== amd_iommu_target_ivhd_type
) {
1464 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1465 "seg: %d flags: %01x info %04x\n",
1466 PCI_BUS_NUM(h
->devid
), PCI_SLOT(h
->devid
),
1467 PCI_FUNC(h
->devid
), h
->cap_ptr
,
1468 h
->pci_seg
, h
->flags
, h
->info
);
1469 DUMP_printk(" mmio-addr: %016llx\n",
1472 iommu
= kzalloc(sizeof(struct amd_iommu
), GFP_KERNEL
);
1476 ret
= init_iommu_one(iommu
, h
);
1488 static int iommu_pc_get_set_reg(struct amd_iommu
*iommu
, u8 bank
, u8 cntr
,
1489 u8 fxn
, u64
*value
, bool is_write
);
1491 static void init_iommu_perf_ctr(struct amd_iommu
*iommu
)
1493 u64 val
= 0xabcd, val2
= 0;
1495 if (!iommu_feature(iommu
, FEATURE_PC
))
1498 amd_iommu_pc_present
= true;
1500 /* Check if the performance counters can be written to */
1501 if ((iommu_pc_get_set_reg(iommu
, 0, 0, 0, &val
, true)) ||
1502 (iommu_pc_get_set_reg(iommu
, 0, 0, 0, &val2
, false)) ||
1504 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1505 amd_iommu_pc_present
= false;
1509 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1511 val
= readl(iommu
->mmio_base
+ MMIO_CNTR_CONF_OFFSET
);
1512 iommu
->max_banks
= (u8
) ((val
>> 12) & 0x3f);
1513 iommu
->max_counters
= (u8
) ((val
>> 7) & 0xf);
1516 static ssize_t
amd_iommu_show_cap(struct device
*dev
,
1517 struct device_attribute
*attr
,
1520 struct amd_iommu
*iommu
= dev_to_amd_iommu(dev
);
1521 return sprintf(buf
, "%x\n", iommu
->cap
);
1523 static DEVICE_ATTR(cap
, S_IRUGO
, amd_iommu_show_cap
, NULL
);
1525 static ssize_t
amd_iommu_show_features(struct device
*dev
,
1526 struct device_attribute
*attr
,
1529 struct amd_iommu
*iommu
= dev_to_amd_iommu(dev
);
1530 return sprintf(buf
, "%llx\n", iommu
->features
);
1532 static DEVICE_ATTR(features
, S_IRUGO
, amd_iommu_show_features
, NULL
);
1534 static struct attribute
*amd_iommu_attrs
[] = {
1536 &dev_attr_features
.attr
,
1540 static struct attribute_group amd_iommu_group
= {
1541 .name
= "amd-iommu",
1542 .attrs
= amd_iommu_attrs
,
1545 static const struct attribute_group
*amd_iommu_groups
[] = {
1550 static int iommu_init_pci(struct amd_iommu
*iommu
)
1552 int cap_ptr
= iommu
->cap_ptr
;
1553 u32 range
, misc
, low
, high
;
1556 iommu
->dev
= pci_get_bus_and_slot(PCI_BUS_NUM(iommu
->devid
),
1557 iommu
->devid
& 0xff);
1561 /* Prevent binding other PCI device drivers to IOMMU devices */
1562 iommu
->dev
->match_driver
= false;
1564 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_CAP_HDR_OFFSET
,
1566 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_RANGE_OFFSET
,
1568 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_MISC_OFFSET
,
1571 if (!(iommu
->cap
& (1 << IOMMU_CAP_IOTLB
)))
1572 amd_iommu_iotlb_sup
= false;
1574 /* read extended feature bits */
1575 low
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
);
1576 high
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
+ 4);
1578 iommu
->features
= ((u64
)high
<< 32) | low
;
1580 if (iommu_feature(iommu
, FEATURE_GT
)) {
1585 pasmax
= iommu
->features
& FEATURE_PASID_MASK
;
1586 pasmax
>>= FEATURE_PASID_SHIFT
;
1587 max_pasid
= (1 << (pasmax
+ 1)) - 1;
1589 amd_iommu_max_pasid
= min(amd_iommu_max_pasid
, max_pasid
);
1591 BUG_ON(amd_iommu_max_pasid
& ~PASID_MASK
);
1593 glxval
= iommu
->features
& FEATURE_GLXVAL_MASK
;
1594 glxval
>>= FEATURE_GLXVAL_SHIFT
;
1596 if (amd_iommu_max_glx_val
== -1)
1597 amd_iommu_max_glx_val
= glxval
;
1599 amd_iommu_max_glx_val
= min(amd_iommu_max_glx_val
, glxval
);
1602 if (iommu_feature(iommu
, FEATURE_GT
) &&
1603 iommu_feature(iommu
, FEATURE_PPR
)) {
1604 iommu
->is_iommu_v2
= true;
1605 amd_iommu_v2_present
= true;
1608 if (iommu_feature(iommu
, FEATURE_PPR
) && alloc_ppr_log(iommu
))
1611 ret
= iommu_init_ga(iommu
);
1615 if (iommu
->cap
& (1UL << IOMMU_CAP_NPCACHE
))
1616 amd_iommu_np_cache
= true;
1618 init_iommu_perf_ctr(iommu
);
1620 if (is_rd890_iommu(iommu
->dev
)) {
1623 iommu
->root_pdev
= pci_get_bus_and_slot(iommu
->dev
->bus
->number
,
1627 * Some rd890 systems may not be fully reconfigured by the
1628 * BIOS, so it's necessary for us to store this information so
1629 * it can be reprogrammed on resume
1631 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1632 &iommu
->stored_addr_lo
);
1633 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1634 &iommu
->stored_addr_hi
);
1636 /* Low bit locks writes to configuration space */
1637 iommu
->stored_addr_lo
&= ~1;
1639 for (i
= 0; i
< 6; i
++)
1640 for (j
= 0; j
< 0x12; j
++)
1641 iommu
->stored_l1
[i
][j
] = iommu_read_l1(iommu
, i
, j
);
1643 for (i
= 0; i
< 0x83; i
++)
1644 iommu
->stored_l2
[i
] = iommu_read_l2(iommu
, i
);
1647 amd_iommu_erratum_746_workaround(iommu
);
1648 amd_iommu_ats_write_check_workaround(iommu
);
1650 iommu_device_sysfs_add(&iommu
->iommu
, &iommu
->dev
->dev
,
1651 amd_iommu_groups
, "ivhd%d", iommu
->index
);
1652 iommu_device_set_ops(&iommu
->iommu
, &amd_iommu_ops
);
1653 iommu_device_register(&iommu
->iommu
);
1655 return pci_enable_device(iommu
->dev
);
1658 static void print_iommu_info(void)
1660 static const char * const feat_str
[] = {
1661 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1662 "IA", "GA", "HE", "PC"
1664 struct amd_iommu
*iommu
;
1666 for_each_iommu(iommu
) {
1669 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1670 dev_name(&iommu
->dev
->dev
), iommu
->cap_ptr
);
1672 if (iommu
->cap
& (1 << IOMMU_CAP_EFR
)) {
1673 pr_info("AMD-Vi: Extended features (%#llx):\n",
1675 for (i
= 0; i
< ARRAY_SIZE(feat_str
); ++i
) {
1676 if (iommu_feature(iommu
, (1ULL << i
)))
1677 pr_cont(" %s", feat_str
[i
]);
1680 if (iommu
->features
& FEATURE_GAM_VAPIC
)
1681 pr_cont(" GA_vAPIC");
1686 if (irq_remapping_enabled
) {
1687 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1688 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
1689 pr_info("AMD-Vi: virtual APIC enabled\n");
1693 static int __init
amd_iommu_init_pci(void)
1695 struct amd_iommu
*iommu
;
1698 for_each_iommu(iommu
) {
1699 ret
= iommu_init_pci(iommu
);
1705 * Order is important here to make sure any unity map requirements are
1706 * fulfilled. The unity mappings are created and written to the device
1707 * table during the amd_iommu_init_api() call.
1709 * After that we call init_device_table_dma() to make sure any
1710 * uninitialized DTE will block DMA, and in the end we flush the caches
1711 * of all IOMMUs to make sure the changes to the device table are
1714 ret
= amd_iommu_init_api();
1716 init_device_table_dma();
1718 for_each_iommu(iommu
)
1719 iommu_flush_all_caches(iommu
);
1727 /****************************************************************************
1729 * The following functions initialize the MSI interrupts for all IOMMUs
1730 * in the system. It's a bit challenging because there could be multiple
1731 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1734 ****************************************************************************/
1736 static int iommu_setup_msi(struct amd_iommu
*iommu
)
1740 r
= pci_enable_msi(iommu
->dev
);
1744 r
= request_threaded_irq(iommu
->dev
->irq
,
1745 amd_iommu_int_handler
,
1746 amd_iommu_int_thread
,
1751 pci_disable_msi(iommu
->dev
);
1755 iommu
->int_enabled
= true;
1760 static int iommu_init_msi(struct amd_iommu
*iommu
)
1764 if (iommu
->int_enabled
)
1767 if (iommu
->dev
->msi_cap
)
1768 ret
= iommu_setup_msi(iommu
);
1776 iommu_feature_enable(iommu
, CONTROL_EVT_INT_EN
);
1778 if (iommu
->ppr_log
!= NULL
)
1779 iommu_feature_enable(iommu
, CONTROL_PPFINT_EN
);
1781 iommu_ga_log_enable(iommu
);
1786 /****************************************************************************
1788 * The next functions belong to the third pass of parsing the ACPI
1789 * table. In this last pass the memory mapping requirements are
1790 * gathered (like exclusion and unity mapping ranges).
1792 ****************************************************************************/
1794 static void __init
free_unity_maps(void)
1796 struct unity_map_entry
*entry
, *next
;
1798 list_for_each_entry_safe(entry
, next
, &amd_iommu_unity_map
, list
) {
1799 list_del(&entry
->list
);
1804 /* called when we find an exclusion range definition in ACPI */
1805 static int __init
init_exclusion_range(struct ivmd_header
*m
)
1810 case ACPI_IVMD_TYPE
:
1811 set_device_exclusion_range(m
->devid
, m
);
1813 case ACPI_IVMD_TYPE_ALL
:
1814 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1815 set_device_exclusion_range(i
, m
);
1817 case ACPI_IVMD_TYPE_RANGE
:
1818 for (i
= m
->devid
; i
<= m
->aux
; ++i
)
1819 set_device_exclusion_range(i
, m
);
1828 /* called for unity map ACPI definition */
1829 static int __init
init_unity_map_range(struct ivmd_header
*m
)
1831 struct unity_map_entry
*e
= NULL
;
1834 e
= kzalloc(sizeof(*e
), GFP_KERNEL
);
1842 case ACPI_IVMD_TYPE
:
1843 s
= "IVMD_TYPEi\t\t\t";
1844 e
->devid_start
= e
->devid_end
= m
->devid
;
1846 case ACPI_IVMD_TYPE_ALL
:
1847 s
= "IVMD_TYPE_ALL\t\t";
1849 e
->devid_end
= amd_iommu_last_bdf
;
1851 case ACPI_IVMD_TYPE_RANGE
:
1852 s
= "IVMD_TYPE_RANGE\t\t";
1853 e
->devid_start
= m
->devid
;
1854 e
->devid_end
= m
->aux
;
1857 e
->address_start
= PAGE_ALIGN(m
->range_start
);
1858 e
->address_end
= e
->address_start
+ PAGE_ALIGN(m
->range_length
);
1859 e
->prot
= m
->flags
>> 1;
1861 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1862 " range_start: %016llx range_end: %016llx flags: %x\n", s
,
1863 PCI_BUS_NUM(e
->devid_start
), PCI_SLOT(e
->devid_start
),
1864 PCI_FUNC(e
->devid_start
), PCI_BUS_NUM(e
->devid_end
),
1865 PCI_SLOT(e
->devid_end
), PCI_FUNC(e
->devid_end
),
1866 e
->address_start
, e
->address_end
, m
->flags
);
1868 list_add_tail(&e
->list
, &amd_iommu_unity_map
);
1873 /* iterates over all memory definitions we find in the ACPI table */
1874 static int __init
init_memory_definitions(struct acpi_table_header
*table
)
1876 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1877 struct ivmd_header
*m
;
1879 end
+= table
->length
;
1880 p
+= IVRS_HEADER_LENGTH
;
1883 m
= (struct ivmd_header
*)p
;
1884 if (m
->flags
& IVMD_FLAG_EXCL_RANGE
)
1885 init_exclusion_range(m
);
1886 else if (m
->flags
& IVMD_FLAG_UNITY_MAP
)
1887 init_unity_map_range(m
);
1896 * Init the device table to not allow DMA access for devices and
1897 * suppress all page faults
1899 static void init_device_table_dma(void)
1903 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1904 set_dev_entry_bit(devid
, DEV_ENTRY_VALID
);
1905 set_dev_entry_bit(devid
, DEV_ENTRY_TRANSLATION
);
1907 * In kdump kernels in-flight DMA from the old kernel might
1908 * cause IO_PAGE_FAULTs. There are no reports that a kdump
1909 * actually failed because of that, so just disable fault
1910 * reporting in the hardware to get rid of the messages
1912 if (is_kdump_kernel())
1913 set_dev_entry_bit(devid
, DEV_ENTRY_NO_PAGE_FAULT
);
1917 static void __init
uninit_device_table_dma(void)
1921 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1922 amd_iommu_dev_table
[devid
].data
[0] = 0ULL;
1923 amd_iommu_dev_table
[devid
].data
[1] = 0ULL;
1927 static void init_device_table(void)
1931 if (!amd_iommu_irq_remap
)
1934 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
)
1935 set_dev_entry_bit(devid
, DEV_ENTRY_IRQ_TBL_EN
);
1938 static void iommu_init_flags(struct amd_iommu
*iommu
)
1940 iommu
->acpi_flags
& IVHD_FLAG_HT_TUN_EN_MASK
?
1941 iommu_feature_enable(iommu
, CONTROL_HT_TUN_EN
) :
1942 iommu_feature_disable(iommu
, CONTROL_HT_TUN_EN
);
1944 iommu
->acpi_flags
& IVHD_FLAG_PASSPW_EN_MASK
?
1945 iommu_feature_enable(iommu
, CONTROL_PASSPW_EN
) :
1946 iommu_feature_disable(iommu
, CONTROL_PASSPW_EN
);
1948 iommu
->acpi_flags
& IVHD_FLAG_RESPASSPW_EN_MASK
?
1949 iommu_feature_enable(iommu
, CONTROL_RESPASSPW_EN
) :
1950 iommu_feature_disable(iommu
, CONTROL_RESPASSPW_EN
);
1952 iommu
->acpi_flags
& IVHD_FLAG_ISOC_EN_MASK
?
1953 iommu_feature_enable(iommu
, CONTROL_ISOC_EN
) :
1954 iommu_feature_disable(iommu
, CONTROL_ISOC_EN
);
1957 * make IOMMU memory accesses cache coherent
1959 iommu_feature_enable(iommu
, CONTROL_COHERENT_EN
);
1961 /* Set IOTLB invalidation timeout to 1s */
1962 iommu_set_inv_tlb_timeout(iommu
, CTRL_INV_TO_1S
);
1965 static void iommu_apply_resume_quirks(struct amd_iommu
*iommu
)
1968 u32 ioc_feature_control
;
1969 struct pci_dev
*pdev
= iommu
->root_pdev
;
1971 /* RD890 BIOSes may not have completely reconfigured the iommu */
1972 if (!is_rd890_iommu(iommu
->dev
) || !pdev
)
1976 * First, we need to ensure that the iommu is enabled. This is
1977 * controlled by a register in the northbridge
1980 /* Select Northbridge indirect register 0x75 and enable writing */
1981 pci_write_config_dword(pdev
, 0x60, 0x75 | (1 << 7));
1982 pci_read_config_dword(pdev
, 0x64, &ioc_feature_control
);
1984 /* Enable the iommu */
1985 if (!(ioc_feature_control
& 0x1))
1986 pci_write_config_dword(pdev
, 0x64, ioc_feature_control
| 1);
1988 /* Restore the iommu BAR */
1989 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1990 iommu
->stored_addr_lo
);
1991 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1992 iommu
->stored_addr_hi
);
1994 /* Restore the l1 indirect regs for each of the 6 l1s */
1995 for (i
= 0; i
< 6; i
++)
1996 for (j
= 0; j
< 0x12; j
++)
1997 iommu_write_l1(iommu
, i
, j
, iommu
->stored_l1
[i
][j
]);
1999 /* Restore the l2 indirect regs */
2000 for (i
= 0; i
< 0x83; i
++)
2001 iommu_write_l2(iommu
, i
, iommu
->stored_l2
[i
]);
2003 /* Lock PCI setup registers */
2004 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
2005 iommu
->stored_addr_lo
| 1);
2008 static void iommu_enable_ga(struct amd_iommu
*iommu
)
2010 #ifdef CONFIG_IRQ_REMAP
2011 switch (amd_iommu_guest_ir
) {
2012 case AMD_IOMMU_GUEST_IR_VAPIC
:
2013 iommu_feature_enable(iommu
, CONTROL_GAM_EN
);
2015 case AMD_IOMMU_GUEST_IR_LEGACY_GA
:
2016 iommu_feature_enable(iommu
, CONTROL_GA_EN
);
2017 iommu
->irte_ops
= &irte_128_ops
;
2020 iommu
->irte_ops
= &irte_32_ops
;
2027 * This function finally enables all IOMMUs found in the system after
2028 * they have been initialized
2030 static void early_enable_iommus(void)
2032 struct amd_iommu
*iommu
;
2034 for_each_iommu(iommu
) {
2035 iommu_disable(iommu
);
2036 iommu_init_flags(iommu
);
2037 iommu_set_device_table(iommu
);
2038 iommu_enable_command_buffer(iommu
);
2039 iommu_enable_event_buffer(iommu
);
2040 iommu_set_exclusion_range(iommu
);
2041 iommu_enable_ga(iommu
);
2042 iommu_enable(iommu
);
2043 iommu_flush_all_caches(iommu
);
2046 #ifdef CONFIG_IRQ_REMAP
2047 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
2048 amd_iommu_irq_ops
.capability
|= (1 << IRQ_POSTING_CAP
);
2052 static void enable_iommus_v2(void)
2054 struct amd_iommu
*iommu
;
2056 for_each_iommu(iommu
) {
2057 iommu_enable_ppr_log(iommu
);
2058 iommu_enable_gt(iommu
);
2062 static void enable_iommus(void)
2064 early_enable_iommus();
2069 static void disable_iommus(void)
2071 struct amd_iommu
*iommu
;
2073 for_each_iommu(iommu
)
2074 iommu_disable(iommu
);
2076 #ifdef CONFIG_IRQ_REMAP
2077 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
2078 amd_iommu_irq_ops
.capability
&= ~(1 << IRQ_POSTING_CAP
);
2083 * Suspend/Resume support
2084 * disable suspend until real resume implemented
2087 static void amd_iommu_resume(void)
2089 struct amd_iommu
*iommu
;
2091 for_each_iommu(iommu
)
2092 iommu_apply_resume_quirks(iommu
);
2094 /* re-load the hardware */
2097 amd_iommu_enable_interrupts();
2100 static int amd_iommu_suspend(void)
2102 /* disable IOMMUs to go out of the way for BIOS */
2108 static struct syscore_ops amd_iommu_syscore_ops
= {
2109 .suspend
= amd_iommu_suspend
,
2110 .resume
= amd_iommu_resume
,
2113 static void __init
free_iommu_resources(void)
2115 kmemleak_free(irq_lookup_table
);
2116 free_pages((unsigned long)irq_lookup_table
,
2117 get_order(rlookup_table_size
));
2118 irq_lookup_table
= NULL
;
2120 kmem_cache_destroy(amd_iommu_irq_cache
);
2121 amd_iommu_irq_cache
= NULL
;
2123 free_pages((unsigned long)amd_iommu_rlookup_table
,
2124 get_order(rlookup_table_size
));
2125 amd_iommu_rlookup_table
= NULL
;
2127 free_pages((unsigned long)amd_iommu_alias_table
,
2128 get_order(alias_table_size
));
2129 amd_iommu_alias_table
= NULL
;
2131 free_pages((unsigned long)amd_iommu_dev_table
,
2132 get_order(dev_table_size
));
2133 amd_iommu_dev_table
= NULL
;
2137 #ifdef CONFIG_GART_IOMMU
2139 * We failed to initialize the AMD IOMMU - try fallback to GART
2147 /* SB IOAPIC is always on this device in AMD systems */
2148 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2150 static bool __init
check_ioapic_information(void)
2152 const char *fw_bug
= FW_BUG
;
2153 bool ret
, has_sb_ioapic
;
2156 has_sb_ioapic
= false;
2160 * If we have map overrides on the kernel command line the
2161 * messages in this function might not describe firmware bugs
2162 * anymore - so be careful
2167 for (idx
= 0; idx
< nr_ioapics
; idx
++) {
2168 int devid
, id
= mpc_ioapic_id(idx
);
2170 devid
= get_ioapic_devid(id
);
2172 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
2175 } else if (devid
== IOAPIC_SB_DEVID
) {
2176 has_sb_ioapic
= true;
2181 if (!has_sb_ioapic
) {
2183 * We expect the SB IOAPIC to be listed in the IVRS
2184 * table. The system timer is connected to the SB IOAPIC
2185 * and if we don't have it in the list the system will
2186 * panic at boot time. This situation usually happens
2187 * when the BIOS is buggy and provides us the wrong
2188 * device id for the IOAPIC in the system.
2190 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug
);
2194 pr_err("AMD-Vi: Disabling interrupt remapping\n");
2199 static void __init
free_dma_resources(void)
2201 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap
,
2202 get_order(MAX_DOMAIN_ID
/8));
2203 amd_iommu_pd_alloc_bitmap
= NULL
;
2209 * This is the hardware init function for AMD IOMMU in the system.
2210 * This function is called either from amd_iommu_init or from the interrupt
2211 * remapping setup code.
2213 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2216 * 1 pass) Discover the most comprehensive IVHD type to use.
2218 * 2 pass) Find the highest PCI device id the driver has to handle.
2219 * Upon this information the size of the data structures is
2220 * determined that needs to be allocated.
2222 * 3 pass) Initialize the data structures just allocated with the
2223 * information in the ACPI table about available AMD IOMMUs
2224 * in the system. It also maps the PCI devices in the
2225 * system to specific IOMMUs
2227 * 4 pass) After the basic data structures are allocated and
2228 * initialized we update them with information about memory
2229 * remapping requirements parsed out of the ACPI table in
2232 * After everything is set up the IOMMUs are enabled and the necessary
2233 * hotplug and suspend notifiers are registered.
2235 static int __init
early_amd_iommu_init(void)
2237 struct acpi_table_header
*ivrs_base
;
2239 int i
, remap_cache_sz
, ret
= 0;
2241 if (!amd_iommu_detected
)
2244 status
= acpi_get_table("IVRS", 0, &ivrs_base
);
2245 if (status
== AE_NOT_FOUND
)
2247 else if (ACPI_FAILURE(status
)) {
2248 const char *err
= acpi_format_exception(status
);
2249 pr_err("AMD-Vi: IVRS table error: %s\n", err
);
2254 * Validate checksum here so we don't need to do it when
2255 * we actually parse the table
2257 ret
= check_ivrs_checksum(ivrs_base
);
2261 amd_iommu_target_ivhd_type
= get_highest_supported_ivhd_type(ivrs_base
);
2262 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type
);
2265 * First parse ACPI tables to find the largest Bus/Dev/Func
2266 * we need to handle. Upon this information the shared data
2267 * structures for the IOMMUs in the system will be allocated
2269 ret
= find_last_devid_acpi(ivrs_base
);
2273 dev_table_size
= tbl_size(DEV_TABLE_ENTRY_SIZE
);
2274 alias_table_size
= tbl_size(ALIAS_TABLE_ENTRY_SIZE
);
2275 rlookup_table_size
= tbl_size(RLOOKUP_TABLE_ENTRY_SIZE
);
2277 /* Device table - directly used by all IOMMUs */
2279 amd_iommu_dev_table
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
2280 get_order(dev_table_size
));
2281 if (amd_iommu_dev_table
== NULL
)
2285 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2286 * IOMMU see for that device
2288 amd_iommu_alias_table
= (void *)__get_free_pages(GFP_KERNEL
,
2289 get_order(alias_table_size
));
2290 if (amd_iommu_alias_table
== NULL
)
2293 /* IOMMU rlookup table - find the IOMMU for a specific device */
2294 amd_iommu_rlookup_table
= (void *)__get_free_pages(
2295 GFP_KERNEL
| __GFP_ZERO
,
2296 get_order(rlookup_table_size
));
2297 if (amd_iommu_rlookup_table
== NULL
)
2300 amd_iommu_pd_alloc_bitmap
= (void *)__get_free_pages(
2301 GFP_KERNEL
| __GFP_ZERO
,
2302 get_order(MAX_DOMAIN_ID
/8));
2303 if (amd_iommu_pd_alloc_bitmap
== NULL
)
2307 * let all alias entries point to itself
2309 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
2310 amd_iommu_alias_table
[i
] = i
;
2313 * never allocate domain 0 because its used as the non-allocated and
2314 * error value placeholder
2316 __set_bit(0, amd_iommu_pd_alloc_bitmap
);
2318 spin_lock_init(&amd_iommu_pd_lock
);
2321 * now the data structures are allocated and basically initialized
2322 * start the real acpi table scan
2324 ret
= init_iommu_all(ivrs_base
);
2328 /* Disable any previously enabled IOMMUs */
2331 if (amd_iommu_irq_remap
)
2332 amd_iommu_irq_remap
= check_ioapic_information();
2334 if (amd_iommu_irq_remap
) {
2336 * Interrupt remapping enabled, create kmem_cache for the
2340 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir
))
2341 remap_cache_sz
= MAX_IRQS_PER_TABLE
* sizeof(u32
);
2343 remap_cache_sz
= MAX_IRQS_PER_TABLE
* (sizeof(u64
) * 2);
2344 amd_iommu_irq_cache
= kmem_cache_create("irq_remap_cache",
2346 IRQ_TABLE_ALIGNMENT
,
2348 if (!amd_iommu_irq_cache
)
2351 irq_lookup_table
= (void *)__get_free_pages(
2352 GFP_KERNEL
| __GFP_ZERO
,
2353 get_order(rlookup_table_size
));
2354 kmemleak_alloc(irq_lookup_table
, rlookup_table_size
,
2356 if (!irq_lookup_table
)
2360 ret
= init_memory_definitions(ivrs_base
);
2364 /* init the device table */
2365 init_device_table();
2368 /* Don't leak any ACPI memory */
2369 acpi_put_table(ivrs_base
);
2375 static int amd_iommu_enable_interrupts(void)
2377 struct amd_iommu
*iommu
;
2380 for_each_iommu(iommu
) {
2381 ret
= iommu_init_msi(iommu
);
2390 static bool detect_ivrs(void)
2392 struct acpi_table_header
*ivrs_base
;
2395 status
= acpi_get_table("IVRS", 0, &ivrs_base
);
2396 if (status
== AE_NOT_FOUND
)
2398 else if (ACPI_FAILURE(status
)) {
2399 const char *err
= acpi_format_exception(status
);
2400 pr_err("AMD-Vi: IVRS table error: %s\n", err
);
2404 acpi_put_table(ivrs_base
);
2406 /* Make sure ACS will be enabled during PCI probe */
2412 /****************************************************************************
2414 * AMD IOMMU Initialization State Machine
2416 ****************************************************************************/
2418 static int __init
state_next(void)
2422 switch (init_state
) {
2423 case IOMMU_START_STATE
:
2424 if (!detect_ivrs()) {
2425 init_state
= IOMMU_NOT_FOUND
;
2428 init_state
= IOMMU_IVRS_DETECTED
;
2431 case IOMMU_IVRS_DETECTED
:
2432 ret
= early_amd_iommu_init();
2433 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_ACPI_FINISHED
;
2434 if (init_state
== IOMMU_ACPI_FINISHED
&& amd_iommu_disabled
) {
2435 pr_info("AMD-Vi: AMD IOMMU disabled on kernel command-line\n");
2436 free_dma_resources();
2437 free_iommu_resources();
2438 init_state
= IOMMU_CMDLINE_DISABLED
;
2442 case IOMMU_ACPI_FINISHED
:
2443 early_enable_iommus();
2444 register_syscore_ops(&amd_iommu_syscore_ops
);
2445 x86_platform
.iommu_shutdown
= disable_iommus
;
2446 init_state
= IOMMU_ENABLED
;
2449 ret
= amd_iommu_init_pci();
2450 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_PCI_INIT
;
2453 case IOMMU_PCI_INIT
:
2454 ret
= amd_iommu_enable_interrupts();
2455 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_INTERRUPTS_EN
;
2457 case IOMMU_INTERRUPTS_EN
:
2458 ret
= amd_iommu_init_dma_ops();
2459 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_DMA_OPS
;
2462 init_state
= IOMMU_INITIALIZED
;
2464 case IOMMU_INITIALIZED
:
2467 case IOMMU_NOT_FOUND
:
2468 case IOMMU_INIT_ERROR
:
2469 case IOMMU_CMDLINE_DISABLED
:
2470 /* Error states => do nothing */
2481 static int __init
iommu_go_to_state(enum iommu_init_state state
)
2485 while (init_state
!= state
) {
2486 if (init_state
== IOMMU_NOT_FOUND
||
2487 init_state
== IOMMU_INIT_ERROR
||
2488 init_state
== IOMMU_CMDLINE_DISABLED
)
2496 #ifdef CONFIG_IRQ_REMAP
2497 int __init
amd_iommu_prepare(void)
2501 amd_iommu_irq_remap
= true;
2503 ret
= iommu_go_to_state(IOMMU_ACPI_FINISHED
);
2506 return amd_iommu_irq_remap
? 0 : -ENODEV
;
2509 int __init
amd_iommu_enable(void)
2513 ret
= iommu_go_to_state(IOMMU_ENABLED
);
2517 irq_remapping_enabled
= 1;
2522 void amd_iommu_disable(void)
2524 amd_iommu_suspend();
2527 int amd_iommu_reenable(int mode
)
2534 int __init
amd_iommu_enable_faulting(void)
2536 /* We enable MSI later when PCI is initialized */
2542 * This is the core init function for AMD IOMMU hardware in the system.
2543 * This function is called from the generic x86 DMA layer initialization
2546 static int __init
amd_iommu_init(void)
2550 ret
= iommu_go_to_state(IOMMU_INITIALIZED
);
2552 free_dma_resources();
2553 if (!irq_remapping_enabled
) {
2555 free_iommu_resources();
2557 struct amd_iommu
*iommu
;
2559 uninit_device_table_dma();
2560 for_each_iommu(iommu
)
2561 iommu_flush_all_caches(iommu
);
2568 static bool amd_iommu_sme_check(void)
2570 if (!sme_active() || (boot_cpu_data
.x86
!= 0x17))
2573 /* For Fam17h, a specific level of support is required */
2574 if (boot_cpu_data
.microcode
>= 0x08001205)
2577 if ((boot_cpu_data
.microcode
>= 0x08001126) &&
2578 (boot_cpu_data
.microcode
<= 0x080011ff))
2581 pr_notice("AMD-Vi: IOMMU not currently supported when SME is active\n");
2586 /****************************************************************************
2588 * Early detect code. This code runs at IOMMU detection time in the DMA
2589 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2592 ****************************************************************************/
2593 int __init
amd_iommu_detect(void)
2597 if (no_iommu
|| (iommu_detected
&& !gart_iommu_aperture
))
2600 if (!amd_iommu_sme_check())
2603 ret
= iommu_go_to_state(IOMMU_IVRS_DETECTED
);
2607 amd_iommu_detected
= true;
2609 x86_init
.iommu
.iommu_init
= amd_iommu_init
;
2614 /****************************************************************************
2616 * Parsing functions for the AMD IOMMU specific kernel command line
2619 ****************************************************************************/
2621 static int __init
parse_amd_iommu_dump(char *str
)
2623 amd_iommu_dump
= true;
2628 static int __init
parse_amd_iommu_intr(char *str
)
2630 for (; *str
; ++str
) {
2631 if (strncmp(str
, "legacy", 6) == 0) {
2632 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY
;
2635 if (strncmp(str
, "vapic", 5) == 0) {
2636 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_VAPIC
;
2643 static int __init
parse_amd_iommu_options(char *str
)
2645 for (; *str
; ++str
) {
2646 if (strncmp(str
, "fullflush", 9) == 0)
2647 amd_iommu_unmap_flush
= true;
2648 if (strncmp(str
, "off", 3) == 0)
2649 amd_iommu_disabled
= true;
2650 if (strncmp(str
, "force_isolation", 15) == 0)
2651 amd_iommu_force_isolation
= true;
2657 static int __init
parse_ivrs_ioapic(char *str
)
2659 unsigned int bus
, dev
, fn
;
2663 ret
= sscanf(str
, "[%d]=%x:%x.%x", &id
, &bus
, &dev
, &fn
);
2666 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str
);
2670 if (early_ioapic_map_size
== EARLY_MAP_SIZE
) {
2671 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2676 devid
= ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2678 cmdline_maps
= true;
2679 i
= early_ioapic_map_size
++;
2680 early_ioapic_map
[i
].id
= id
;
2681 early_ioapic_map
[i
].devid
= devid
;
2682 early_ioapic_map
[i
].cmd_line
= true;
2687 static int __init
parse_ivrs_hpet(char *str
)
2689 unsigned int bus
, dev
, fn
;
2693 ret
= sscanf(str
, "[%d]=%x:%x.%x", &id
, &bus
, &dev
, &fn
);
2696 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str
);
2700 if (early_hpet_map_size
== EARLY_MAP_SIZE
) {
2701 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2706 devid
= ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2708 cmdline_maps
= true;
2709 i
= early_hpet_map_size
++;
2710 early_hpet_map
[i
].id
= id
;
2711 early_hpet_map
[i
].devid
= devid
;
2712 early_hpet_map
[i
].cmd_line
= true;
2717 static int __init
parse_ivrs_acpihid(char *str
)
2720 char *hid
, *uid
, *p
;
2721 char acpiid
[ACPIHID_UID_LEN
+ ACPIHID_HID_LEN
] = {0};
2724 ret
= sscanf(str
, "[%x:%x.%x]=%s", &bus
, &dev
, &fn
, acpiid
);
2726 pr_err("AMD-Vi: Invalid command line: ivrs_acpihid(%s)\n", str
);
2731 hid
= strsep(&p
, ":");
2734 if (!hid
|| !(*hid
) || !uid
) {
2735 pr_err("AMD-Vi: Invalid command line: hid or uid\n");
2739 i
= early_acpihid_map_size
++;
2740 memcpy(early_acpihid_map
[i
].hid
, hid
, strlen(hid
));
2741 memcpy(early_acpihid_map
[i
].uid
, uid
, strlen(uid
));
2742 early_acpihid_map
[i
].devid
=
2743 ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2744 early_acpihid_map
[i
].cmd_line
= true;
2749 __setup("amd_iommu_dump", parse_amd_iommu_dump
);
2750 __setup("amd_iommu=", parse_amd_iommu_options
);
2751 __setup("amd_iommu_intr=", parse_amd_iommu_intr
);
2752 __setup("ivrs_ioapic", parse_ivrs_ioapic
);
2753 __setup("ivrs_hpet", parse_ivrs_hpet
);
2754 __setup("ivrs_acpihid", parse_ivrs_acpihid
);
2756 IOMMU_INIT_FINISH(amd_iommu_detect
,
2757 gart_iommu_hole_init
,
2761 bool amd_iommu_v2_supported(void)
2763 return amd_iommu_v2_present
;
2765 EXPORT_SYMBOL(amd_iommu_v2_supported
);
2767 struct amd_iommu
*get_amd_iommu(unsigned int idx
)
2770 struct amd_iommu
*iommu
;
2772 for_each_iommu(iommu
)
2777 EXPORT_SYMBOL(get_amd_iommu
);
2779 /****************************************************************************
2781 * IOMMU EFR Performance Counter support functionality. This code allows
2782 * access to the IOMMU PC functionality.
2784 ****************************************************************************/
2786 u8
amd_iommu_pc_get_max_banks(unsigned int idx
)
2788 struct amd_iommu
*iommu
= get_amd_iommu(idx
);
2791 return iommu
->max_banks
;
2795 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks
);
2797 bool amd_iommu_pc_supported(void)
2799 return amd_iommu_pc_present
;
2801 EXPORT_SYMBOL(amd_iommu_pc_supported
);
2803 u8
amd_iommu_pc_get_max_counters(unsigned int idx
)
2805 struct amd_iommu
*iommu
= get_amd_iommu(idx
);
2808 return iommu
->max_counters
;
2812 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters
);
2814 static int iommu_pc_get_set_reg(struct amd_iommu
*iommu
, u8 bank
, u8 cntr
,
2815 u8 fxn
, u64
*value
, bool is_write
)
2820 /* Make sure the IOMMU PC resource is available */
2821 if (!amd_iommu_pc_present
)
2824 /* Check for valid iommu and pc register indexing */
2825 if (WARN_ON(!iommu
|| (fxn
> 0x28) || (fxn
& 7)))
2828 offset
= (u32
)(((0x40 | bank
) << 12) | (cntr
<< 8) | fxn
);
2830 /* Limit the offset to the hw defined mmio region aperture */
2831 max_offset_lim
= (u32
)(((0x40 | iommu
->max_banks
) << 12) |
2832 (iommu
->max_counters
<< 8) | 0x28);
2833 if ((offset
< MMIO_CNTR_REG_OFFSET
) ||
2834 (offset
> max_offset_lim
))
2838 u64 val
= *value
& GENMASK_ULL(47, 0);
2840 writel((u32
)val
, iommu
->mmio_base
+ offset
);
2841 writel((val
>> 32), iommu
->mmio_base
+ offset
+ 4);
2843 *value
= readl(iommu
->mmio_base
+ offset
+ 4);
2845 *value
|= readl(iommu
->mmio_base
+ offset
);
2846 *value
&= GENMASK_ULL(47, 0);
2852 int amd_iommu_pc_get_reg(struct amd_iommu
*iommu
, u8 bank
, u8 cntr
, u8 fxn
, u64
*value
)
2857 return iommu_pc_get_set_reg(iommu
, bank
, cntr
, fxn
, value
, false);
2859 EXPORT_SYMBOL(amd_iommu_pc_get_reg
);
2861 int amd_iommu_pc_set_reg(struct amd_iommu
*iommu
, u8 bank
, u8 cntr
, u8 fxn
, u64
*value
)
2866 return iommu_pc_get_set_reg(iommu
, bank
, cntr
, fxn
, value
, true);
2868 EXPORT_SYMBOL(amd_iommu_pc_set_reg
);