2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
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/slab.h>
24 #include <linux/syscore_ops.h>
25 #include <linux/interrupt.h>
26 #include <linux/msi.h>
27 #include <linux/amd-iommu.h>
28 #include <asm/pci-direct.h>
29 #include <asm/iommu.h>
31 #include <asm/x86_init.h>
32 #include <asm/iommu_table.h>
34 #include "amd_iommu_proto.h"
35 #include "amd_iommu_types.h"
38 * definitions for the ACPI scanning code
40 #define IVRS_HEADER_LENGTH 48
42 #define ACPI_IVHD_TYPE 0x10
43 #define ACPI_IVMD_TYPE_ALL 0x20
44 #define ACPI_IVMD_TYPE 0x21
45 #define ACPI_IVMD_TYPE_RANGE 0x22
47 #define IVHD_DEV_ALL 0x01
48 #define IVHD_DEV_SELECT 0x02
49 #define IVHD_DEV_SELECT_RANGE_START 0x03
50 #define IVHD_DEV_RANGE_END 0x04
51 #define IVHD_DEV_ALIAS 0x42
52 #define IVHD_DEV_ALIAS_RANGE 0x43
53 #define IVHD_DEV_EXT_SELECT 0x46
54 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
56 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
57 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
58 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
59 #define IVHD_FLAG_ISOC_EN_MASK 0x08
61 #define IVMD_FLAG_EXCL_RANGE 0x08
62 #define IVMD_FLAG_UNITY_MAP 0x01
64 #define ACPI_DEVFLAG_INITPASS 0x01
65 #define ACPI_DEVFLAG_EXTINT 0x02
66 #define ACPI_DEVFLAG_NMI 0x04
67 #define ACPI_DEVFLAG_SYSMGT1 0x10
68 #define ACPI_DEVFLAG_SYSMGT2 0x20
69 #define ACPI_DEVFLAG_LINT0 0x40
70 #define ACPI_DEVFLAG_LINT1 0x80
71 #define ACPI_DEVFLAG_ATSDIS 0x10000000
74 * ACPI table definitions
76 * These data structures are laid over the table to parse the important values
81 * structure describing one IOMMU in the ACPI table. Typically followed by one
82 * or more ivhd_entrys.
94 } __attribute__((packed
));
97 * A device entry describing which devices a specific IOMMU translates and
98 * which requestor ids they use.
105 } __attribute__((packed
));
108 * An AMD IOMMU memory definition structure. It defines things like exclusion
109 * ranges for devices and regions that should be unity mapped.
120 } __attribute__((packed
));
124 static int __initdata amd_iommu_detected
;
125 static bool __initdata amd_iommu_disabled
;
127 u16 amd_iommu_last_bdf
; /* largest PCI device id we have
129 LIST_HEAD(amd_iommu_unity_map
); /* a list of required unity mappings
131 bool amd_iommu_unmap_flush
; /* if true, flush on every unmap */
133 LIST_HEAD(amd_iommu_list
); /* list of all AMD IOMMUs in the
136 /* Array to assign indices to IOMMUs*/
137 struct amd_iommu
*amd_iommus
[MAX_IOMMUS
];
138 int amd_iommus_present
;
140 /* IOMMUs have a non-present cache? */
141 bool amd_iommu_np_cache __read_mostly
;
142 bool amd_iommu_iotlb_sup __read_mostly
= true;
145 * The ACPI table parsing functions set this variable on an error
147 static int __initdata amd_iommu_init_err
;
150 * List of protection domains - used during resume
152 LIST_HEAD(amd_iommu_pd_list
);
153 spinlock_t amd_iommu_pd_lock
;
156 * Pointer to the device table which is shared by all AMD IOMMUs
157 * it is indexed by the PCI device id or the HT unit id and contains
158 * information about the domain the device belongs to as well as the
159 * page table root pointer.
161 struct dev_table_entry
*amd_iommu_dev_table
;
164 * The alias table is a driver specific data structure which contains the
165 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
166 * More than one device can share the same requestor id.
168 u16
*amd_iommu_alias_table
;
171 * The rlookup table is used to find the IOMMU which is responsible
172 * for a specific device. It is also indexed by the PCI device id.
174 struct amd_iommu
**amd_iommu_rlookup_table
;
177 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
178 * to know which ones are already in use.
180 unsigned long *amd_iommu_pd_alloc_bitmap
;
182 static u32 dev_table_size
; /* size of the device table */
183 static u32 alias_table_size
; /* size of the alias table */
184 static u32 rlookup_table_size
; /* size if the rlookup table */
187 * This function flushes all internal caches of
188 * the IOMMU used by this driver.
190 extern void iommu_flush_all_caches(struct amd_iommu
*iommu
);
192 static inline void update_last_devid(u16 devid
)
194 if (devid
> amd_iommu_last_bdf
)
195 amd_iommu_last_bdf
= devid
;
198 static inline unsigned long tbl_size(int entry_size
)
200 unsigned shift
= PAGE_SHIFT
+
201 get_order(((int)amd_iommu_last_bdf
+ 1) * entry_size
);
206 /* Access to l1 and l2 indexed register spaces */
208 static u32
iommu_read_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
)
212 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
213 pci_read_config_dword(iommu
->dev
, 0xfc, &val
);
217 static void iommu_write_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
, u32 val
)
219 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16 | 1 << 31));
220 pci_write_config_dword(iommu
->dev
, 0xfc, val
);
221 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
224 static u32
iommu_read_l2(struct amd_iommu
*iommu
, u8 address
)
228 pci_write_config_dword(iommu
->dev
, 0xf0, address
);
229 pci_read_config_dword(iommu
->dev
, 0xf4, &val
);
233 static void iommu_write_l2(struct amd_iommu
*iommu
, u8 address
, u32 val
)
235 pci_write_config_dword(iommu
->dev
, 0xf0, (address
| 1 << 8));
236 pci_write_config_dword(iommu
->dev
, 0xf4, val
);
239 /****************************************************************************
241 * AMD IOMMU MMIO register space handling functions
243 * These functions are used to program the IOMMU device registers in
244 * MMIO space required for that driver.
246 ****************************************************************************/
249 * This function set the exclusion range in the IOMMU. DMA accesses to the
250 * exclusion range are passed through untranslated
252 static void iommu_set_exclusion_range(struct amd_iommu
*iommu
)
254 u64 start
= iommu
->exclusion_start
& PAGE_MASK
;
255 u64 limit
= (start
+ iommu
->exclusion_length
) & PAGE_MASK
;
258 if (!iommu
->exclusion_start
)
261 entry
= start
| MMIO_EXCL_ENABLE_MASK
;
262 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_BASE_OFFSET
,
263 &entry
, sizeof(entry
));
266 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_LIMIT_OFFSET
,
267 &entry
, sizeof(entry
));
270 /* Programs the physical address of the device table into the IOMMU hardware */
271 static void iommu_set_device_table(struct amd_iommu
*iommu
)
275 BUG_ON(iommu
->mmio_base
== NULL
);
277 entry
= virt_to_phys(amd_iommu_dev_table
);
278 entry
|= (dev_table_size
>> 12) - 1;
279 memcpy_toio(iommu
->mmio_base
+ MMIO_DEV_TABLE_OFFSET
,
280 &entry
, sizeof(entry
));
283 /* Generic functions to enable/disable certain features of the IOMMU. */
284 static void iommu_feature_enable(struct amd_iommu
*iommu
, u8 bit
)
288 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
290 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
293 static void iommu_feature_disable(struct amd_iommu
*iommu
, u8 bit
)
297 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
299 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
302 /* Function to enable the hardware */
303 static void iommu_enable(struct amd_iommu
*iommu
)
305 static const char * const feat_str
[] = {
306 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
307 "IA", "GA", "HE", "PC", NULL
311 printk(KERN_INFO
"AMD-Vi: Enabling IOMMU at %s cap 0x%hx",
312 dev_name(&iommu
->dev
->dev
), iommu
->cap_ptr
);
314 if (iommu
->cap
& (1 << IOMMU_CAP_EFR
)) {
315 printk(KERN_CONT
" extended features: ");
316 for (i
= 0; feat_str
[i
]; ++i
)
317 if (iommu_feature(iommu
, (1ULL << i
)))
318 printk(KERN_CONT
" %s", feat_str
[i
]);
320 printk(KERN_CONT
"\n");
322 iommu_feature_enable(iommu
, CONTROL_IOMMU_EN
);
325 static void iommu_disable(struct amd_iommu
*iommu
)
327 /* Disable command buffer */
328 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
330 /* Disable event logging and event interrupts */
331 iommu_feature_disable(iommu
, CONTROL_EVT_INT_EN
);
332 iommu_feature_disable(iommu
, CONTROL_EVT_LOG_EN
);
334 /* Disable IOMMU hardware itself */
335 iommu_feature_disable(iommu
, CONTROL_IOMMU_EN
);
339 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
340 * the system has one.
342 static u8
* __init
iommu_map_mmio_space(u64 address
)
346 if (!request_mem_region(address
, MMIO_REGION_LENGTH
, "amd_iommu")) {
347 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
349 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
353 ret
= ioremap_nocache(address
, MMIO_REGION_LENGTH
);
357 release_mem_region(address
, MMIO_REGION_LENGTH
);
362 static void __init
iommu_unmap_mmio_space(struct amd_iommu
*iommu
)
364 if (iommu
->mmio_base
)
365 iounmap(iommu
->mmio_base
);
366 release_mem_region(iommu
->mmio_phys
, MMIO_REGION_LENGTH
);
369 /****************************************************************************
371 * The functions below belong to the first pass of AMD IOMMU ACPI table
372 * parsing. In this pass we try to find out the highest device id this
373 * code has to handle. Upon this information the size of the shared data
374 * structures is determined later.
376 ****************************************************************************/
379 * This function calculates the length of a given IVHD entry
381 static inline int ivhd_entry_length(u8
*ivhd
)
383 return 0x04 << (*ivhd
>> 6);
387 * This function reads the last device id the IOMMU has to handle from the PCI
388 * capability header for this IOMMU
390 static int __init
find_last_devid_on_pci(int bus
, int dev
, int fn
, int cap_ptr
)
394 cap
= read_pci_config(bus
, dev
, fn
, cap_ptr
+MMIO_RANGE_OFFSET
);
395 update_last_devid(calc_devid(MMIO_GET_BUS(cap
), MMIO_GET_LD(cap
)));
401 * After reading the highest device id from the IOMMU PCI capability header
402 * this function looks if there is a higher device id defined in the ACPI table
404 static int __init
find_last_devid_from_ivhd(struct ivhd_header
*h
)
406 u8
*p
= (void *)h
, *end
= (void *)h
;
407 struct ivhd_entry
*dev
;
412 find_last_devid_on_pci(PCI_BUS(h
->devid
),
418 dev
= (struct ivhd_entry
*)p
;
420 case IVHD_DEV_SELECT
:
421 case IVHD_DEV_RANGE_END
:
423 case IVHD_DEV_EXT_SELECT
:
424 /* all the above subfield types refer to device ids */
425 update_last_devid(dev
->devid
);
430 p
+= ivhd_entry_length(p
);
439 * Iterate over all IVHD entries in the ACPI table and find the highest device
440 * id which we need to handle. This is the first of three functions which parse
441 * the ACPI table. So we check the checksum here.
443 static int __init
find_last_devid_acpi(struct acpi_table_header
*table
)
446 u8 checksum
= 0, *p
= (u8
*)table
, *end
= (u8
*)table
;
447 struct ivhd_header
*h
;
450 * Validate checksum here so we don't need to do it when
451 * we actually parse the table
453 for (i
= 0; i
< table
->length
; ++i
)
456 /* ACPI table corrupt */
457 amd_iommu_init_err
= -ENODEV
;
461 p
+= IVRS_HEADER_LENGTH
;
463 end
+= table
->length
;
465 h
= (struct ivhd_header
*)p
;
468 find_last_devid_from_ivhd(h
);
480 /****************************************************************************
482 * The following functions belong the the code path which parses the ACPI table
483 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
484 * data structures, initialize the device/alias/rlookup table and also
485 * basically initialize the hardware.
487 ****************************************************************************/
490 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
491 * write commands to that buffer later and the IOMMU will execute them
494 static u8
* __init
alloc_command_buffer(struct amd_iommu
*iommu
)
496 u8
*cmd_buf
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
497 get_order(CMD_BUFFER_SIZE
));
502 iommu
->cmd_buf_size
= CMD_BUFFER_SIZE
| CMD_BUFFER_UNINITIALIZED
;
508 * This function resets the command buffer if the IOMMU stopped fetching
511 void amd_iommu_reset_cmd_buffer(struct amd_iommu
*iommu
)
513 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
515 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_HEAD_OFFSET
);
516 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_TAIL_OFFSET
);
518 iommu_feature_enable(iommu
, CONTROL_CMDBUF_EN
);
522 * This function writes the command buffer address to the hardware and
525 static void iommu_enable_command_buffer(struct amd_iommu
*iommu
)
529 BUG_ON(iommu
->cmd_buf
== NULL
);
531 entry
= (u64
)virt_to_phys(iommu
->cmd_buf
);
532 entry
|= MMIO_CMD_SIZE_512
;
534 memcpy_toio(iommu
->mmio_base
+ MMIO_CMD_BUF_OFFSET
,
535 &entry
, sizeof(entry
));
537 amd_iommu_reset_cmd_buffer(iommu
);
538 iommu
->cmd_buf_size
&= ~(CMD_BUFFER_UNINITIALIZED
);
541 static void __init
free_command_buffer(struct amd_iommu
*iommu
)
543 free_pages((unsigned long)iommu
->cmd_buf
,
544 get_order(iommu
->cmd_buf_size
& ~(CMD_BUFFER_UNINITIALIZED
)));
547 /* allocates the memory where the IOMMU will log its events to */
548 static u8
* __init
alloc_event_buffer(struct amd_iommu
*iommu
)
550 iommu
->evt_buf
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
551 get_order(EVT_BUFFER_SIZE
));
553 if (iommu
->evt_buf
== NULL
)
556 iommu
->evt_buf_size
= EVT_BUFFER_SIZE
;
558 return iommu
->evt_buf
;
561 static void iommu_enable_event_buffer(struct amd_iommu
*iommu
)
565 BUG_ON(iommu
->evt_buf
== NULL
);
567 entry
= (u64
)virt_to_phys(iommu
->evt_buf
) | EVT_LEN_MASK
;
569 memcpy_toio(iommu
->mmio_base
+ MMIO_EVT_BUF_OFFSET
,
570 &entry
, sizeof(entry
));
572 /* set head and tail to zero manually */
573 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_HEAD_OFFSET
);
574 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_TAIL_OFFSET
);
576 iommu_feature_enable(iommu
, CONTROL_EVT_LOG_EN
);
579 static void __init
free_event_buffer(struct amd_iommu
*iommu
)
581 free_pages((unsigned long)iommu
->evt_buf
, get_order(EVT_BUFFER_SIZE
));
584 /* sets a specific bit in the device table entry. */
585 static void set_dev_entry_bit(u16 devid
, u8 bit
)
587 int i
= (bit
>> 5) & 0x07;
588 int _bit
= bit
& 0x1f;
590 amd_iommu_dev_table
[devid
].data
[i
] |= (1 << _bit
);
593 static int get_dev_entry_bit(u16 devid
, u8 bit
)
595 int i
= (bit
>> 5) & 0x07;
596 int _bit
= bit
& 0x1f;
598 return (amd_iommu_dev_table
[devid
].data
[i
] & (1 << _bit
)) >> _bit
;
602 void amd_iommu_apply_erratum_63(u16 devid
)
606 sysmgt
= get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
) |
607 (get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
) << 1);
610 set_dev_entry_bit(devid
, DEV_ENTRY_IW
);
613 /* Writes the specific IOMMU for a device into the rlookup table */
614 static void __init
set_iommu_for_device(struct amd_iommu
*iommu
, u16 devid
)
616 amd_iommu_rlookup_table
[devid
] = iommu
;
620 * This function takes the device specific flags read from the ACPI
621 * table and sets up the device table entry with that information
623 static void __init
set_dev_entry_from_acpi(struct amd_iommu
*iommu
,
624 u16 devid
, u32 flags
, u32 ext_flags
)
626 if (flags
& ACPI_DEVFLAG_INITPASS
)
627 set_dev_entry_bit(devid
, DEV_ENTRY_INIT_PASS
);
628 if (flags
& ACPI_DEVFLAG_EXTINT
)
629 set_dev_entry_bit(devid
, DEV_ENTRY_EINT_PASS
);
630 if (flags
& ACPI_DEVFLAG_NMI
)
631 set_dev_entry_bit(devid
, DEV_ENTRY_NMI_PASS
);
632 if (flags
& ACPI_DEVFLAG_SYSMGT1
)
633 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
);
634 if (flags
& ACPI_DEVFLAG_SYSMGT2
)
635 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
);
636 if (flags
& ACPI_DEVFLAG_LINT0
)
637 set_dev_entry_bit(devid
, DEV_ENTRY_LINT0_PASS
);
638 if (flags
& ACPI_DEVFLAG_LINT1
)
639 set_dev_entry_bit(devid
, DEV_ENTRY_LINT1_PASS
);
641 amd_iommu_apply_erratum_63(devid
);
643 set_iommu_for_device(iommu
, devid
);
647 * Reads the device exclusion range from ACPI and initialize IOMMU with
650 static void __init
set_device_exclusion_range(u16 devid
, struct ivmd_header
*m
)
652 struct amd_iommu
*iommu
= amd_iommu_rlookup_table
[devid
];
654 if (!(m
->flags
& IVMD_FLAG_EXCL_RANGE
))
659 * We only can configure exclusion ranges per IOMMU, not
660 * per device. But we can enable the exclusion range per
661 * device. This is done here
663 set_dev_entry_bit(m
->devid
, DEV_ENTRY_EX
);
664 iommu
->exclusion_start
= m
->range_start
;
665 iommu
->exclusion_length
= m
->range_length
;
670 * This function reads some important data from the IOMMU PCI space and
671 * initializes the driver data structure with it. It reads the hardware
672 * capabilities and the first/last device entries
674 static void __init
init_iommu_from_pci(struct amd_iommu
*iommu
)
676 int cap_ptr
= iommu
->cap_ptr
;
677 u32 range
, misc
, low
, high
;
680 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_CAP_HDR_OFFSET
,
682 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_RANGE_OFFSET
,
684 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_MISC_OFFSET
,
687 iommu
->first_device
= calc_devid(MMIO_GET_BUS(range
),
689 iommu
->last_device
= calc_devid(MMIO_GET_BUS(range
),
691 iommu
->evt_msi_num
= MMIO_MSI_NUM(misc
);
693 if (!(iommu
->cap
& (1 << IOMMU_CAP_IOTLB
)))
694 amd_iommu_iotlb_sup
= false;
696 /* read extended feature bits */
697 low
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
);
698 high
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
+ 4);
700 iommu
->features
= ((u64
)high
<< 32) | low
;
702 if (!is_rd890_iommu(iommu
->dev
))
706 * Some rd890 systems may not be fully reconfigured by the BIOS, so
707 * it's necessary for us to store this information so it can be
708 * reprogrammed on resume
711 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
712 &iommu
->stored_addr_lo
);
713 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
714 &iommu
->stored_addr_hi
);
716 /* Low bit locks writes to configuration space */
717 iommu
->stored_addr_lo
&= ~1;
719 for (i
= 0; i
< 6; i
++)
720 for (j
= 0; j
< 0x12; j
++)
721 iommu
->stored_l1
[i
][j
] = iommu_read_l1(iommu
, i
, j
);
723 for (i
= 0; i
< 0x83; i
++)
724 iommu
->stored_l2
[i
] = iommu_read_l2(iommu
, i
);
728 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
729 * initializes the hardware and our data structures with it.
731 static void __init
init_iommu_from_acpi(struct amd_iommu
*iommu
,
732 struct ivhd_header
*h
)
735 u8
*end
= p
, flags
= 0;
736 u16 devid
= 0, devid_start
= 0, devid_to
= 0;
737 u32 dev_i
, ext_flags
= 0;
739 struct ivhd_entry
*e
;
742 * First save the recommended feature enable bits from ACPI
744 iommu
->acpi_flags
= h
->flags
;
747 * Done. Now parse the device entries
749 p
+= sizeof(struct ivhd_header
);
754 e
= (struct ivhd_entry
*)p
;
758 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
759 " last device %02x:%02x.%x flags: %02x\n",
760 PCI_BUS(iommu
->first_device
),
761 PCI_SLOT(iommu
->first_device
),
762 PCI_FUNC(iommu
->first_device
),
763 PCI_BUS(iommu
->last_device
),
764 PCI_SLOT(iommu
->last_device
),
765 PCI_FUNC(iommu
->last_device
),
768 for (dev_i
= iommu
->first_device
;
769 dev_i
<= iommu
->last_device
; ++dev_i
)
770 set_dev_entry_from_acpi(iommu
, dev_i
,
773 case IVHD_DEV_SELECT
:
775 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
783 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
785 case IVHD_DEV_SELECT_RANGE_START
:
787 DUMP_printk(" DEV_SELECT_RANGE_START\t "
788 "devid: %02x:%02x.%x flags: %02x\n",
794 devid_start
= e
->devid
;
801 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
802 "flags: %02x devid_to: %02x:%02x.%x\n",
807 PCI_BUS(e
->ext
>> 8),
808 PCI_SLOT(e
->ext
>> 8),
809 PCI_FUNC(e
->ext
>> 8));
812 devid_to
= e
->ext
>> 8;
813 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
814 set_dev_entry_from_acpi(iommu
, devid_to
, e
->flags
, 0);
815 amd_iommu_alias_table
[devid
] = devid_to
;
817 case IVHD_DEV_ALIAS_RANGE
:
819 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
820 "devid: %02x:%02x.%x flags: %02x "
821 "devid_to: %02x:%02x.%x\n",
826 PCI_BUS(e
->ext
>> 8),
827 PCI_SLOT(e
->ext
>> 8),
828 PCI_FUNC(e
->ext
>> 8));
830 devid_start
= e
->devid
;
832 devid_to
= e
->ext
>> 8;
836 case IVHD_DEV_EXT_SELECT
:
838 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
839 "flags: %02x ext: %08x\n",
846 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
,
849 case IVHD_DEV_EXT_SELECT_RANGE
:
851 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
852 "%02x:%02x.%x flags: %02x ext: %08x\n",
858 devid_start
= e
->devid
;
863 case IVHD_DEV_RANGE_END
:
865 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
871 for (dev_i
= devid_start
; dev_i
<= devid
; ++dev_i
) {
873 amd_iommu_alias_table
[dev_i
] = devid_to
;
874 set_dev_entry_from_acpi(iommu
,
875 devid_to
, flags
, ext_flags
);
877 set_dev_entry_from_acpi(iommu
, dev_i
,
885 p
+= ivhd_entry_length(p
);
889 /* Initializes the device->iommu mapping for the driver */
890 static int __init
init_iommu_devices(struct amd_iommu
*iommu
)
894 for (i
= iommu
->first_device
; i
<= iommu
->last_device
; ++i
)
895 set_iommu_for_device(iommu
, i
);
900 static void __init
free_iommu_one(struct amd_iommu
*iommu
)
902 free_command_buffer(iommu
);
903 free_event_buffer(iommu
);
904 iommu_unmap_mmio_space(iommu
);
907 static void __init
free_iommu_all(void)
909 struct amd_iommu
*iommu
, *next
;
911 for_each_iommu_safe(iommu
, next
) {
912 list_del(&iommu
->list
);
913 free_iommu_one(iommu
);
919 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
921 * BIOS should disable L2B micellaneous clock gating by setting
922 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
924 static void __init
amd_iommu_erratum_746_workaround(struct amd_iommu
*iommu
)
928 if ((boot_cpu_data
.x86
!= 0x15) ||
929 (boot_cpu_data
.x86_model
< 0x10) ||
930 (boot_cpu_data
.x86_model
> 0x1f))
933 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
934 pci_read_config_dword(iommu
->dev
, 0xf4, &value
);
939 /* Select NB indirect register 0x90 and enable writing */
940 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90 | (1 << 8));
942 pci_write_config_dword(iommu
->dev
, 0xf4, value
| 0x4);
943 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
944 dev_name(&iommu
->dev
->dev
));
946 /* Clear the enable writing bit */
947 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
951 * This function clues the initialization function for one IOMMU
952 * together and also allocates the command buffer and programs the
953 * hardware. It does NOT enable the IOMMU. This is done afterwards.
955 static int __init
init_iommu_one(struct amd_iommu
*iommu
, struct ivhd_header
*h
)
957 spin_lock_init(&iommu
->lock
);
959 /* Add IOMMU to internal data structures */
960 list_add_tail(&iommu
->list
, &amd_iommu_list
);
961 iommu
->index
= amd_iommus_present
++;
963 if (unlikely(iommu
->index
>= MAX_IOMMUS
)) {
964 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
968 /* Index is fine - add IOMMU to the array */
969 amd_iommus
[iommu
->index
] = iommu
;
972 * Copy data from ACPI table entry to the iommu struct
974 iommu
->dev
= pci_get_bus_and_slot(PCI_BUS(h
->devid
), h
->devid
& 0xff);
978 iommu
->root_pdev
= pci_get_bus_and_slot(iommu
->dev
->bus
->number
,
981 iommu
->cap_ptr
= h
->cap_ptr
;
982 iommu
->pci_seg
= h
->pci_seg
;
983 iommu
->mmio_phys
= h
->mmio_phys
;
984 iommu
->mmio_base
= iommu_map_mmio_space(h
->mmio_phys
);
985 if (!iommu
->mmio_base
)
988 iommu
->cmd_buf
= alloc_command_buffer(iommu
);
992 iommu
->evt_buf
= alloc_event_buffer(iommu
);
996 iommu
->int_enabled
= false;
998 init_iommu_from_pci(iommu
);
999 init_iommu_from_acpi(iommu
, h
);
1000 init_iommu_devices(iommu
);
1002 if (iommu
->cap
& (1UL << IOMMU_CAP_NPCACHE
))
1003 amd_iommu_np_cache
= true;
1005 amd_iommu_erratum_746_workaround(iommu
);
1007 return pci_enable_device(iommu
->dev
);
1011 * Iterates over all IOMMU entries in the ACPI table, allocates the
1012 * IOMMU structure and initializes it with init_iommu_one()
1014 static int __init
init_iommu_all(struct acpi_table_header
*table
)
1016 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1017 struct ivhd_header
*h
;
1018 struct amd_iommu
*iommu
;
1021 end
+= table
->length
;
1022 p
+= IVRS_HEADER_LENGTH
;
1025 h
= (struct ivhd_header
*)p
;
1027 case ACPI_IVHD_TYPE
:
1029 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1030 "seg: %d flags: %01x info %04x\n",
1031 PCI_BUS(h
->devid
), PCI_SLOT(h
->devid
),
1032 PCI_FUNC(h
->devid
), h
->cap_ptr
,
1033 h
->pci_seg
, h
->flags
, h
->info
);
1034 DUMP_printk(" mmio-addr: %016llx\n",
1037 iommu
= kzalloc(sizeof(struct amd_iommu
), GFP_KERNEL
);
1038 if (iommu
== NULL
) {
1039 amd_iommu_init_err
= -ENOMEM
;
1043 ret
= init_iommu_one(iommu
, h
);
1045 amd_iommu_init_err
= ret
;
1060 /****************************************************************************
1062 * The following functions initialize the MSI interrupts for all IOMMUs
1063 * in the system. Its a bit challenging because there could be multiple
1064 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1067 ****************************************************************************/
1069 static int iommu_setup_msi(struct amd_iommu
*iommu
)
1073 r
= pci_enable_msi(iommu
->dev
);
1077 r
= request_threaded_irq(iommu
->dev
->irq
,
1078 amd_iommu_int_handler
,
1079 amd_iommu_int_thread
,
1084 pci_disable_msi(iommu
->dev
);
1088 iommu
->int_enabled
= true;
1093 static int iommu_init_msi(struct amd_iommu
*iommu
)
1097 if (iommu
->int_enabled
)
1100 if (pci_find_capability(iommu
->dev
, PCI_CAP_ID_MSI
))
1101 ret
= iommu_setup_msi(iommu
);
1109 iommu_feature_enable(iommu
, CONTROL_EVT_INT_EN
);
1114 /****************************************************************************
1116 * The next functions belong to the third pass of parsing the ACPI
1117 * table. In this last pass the memory mapping requirements are
1118 * gathered (like exclusion and unity mapping reanges).
1120 ****************************************************************************/
1122 static void __init
free_unity_maps(void)
1124 struct unity_map_entry
*entry
, *next
;
1126 list_for_each_entry_safe(entry
, next
, &amd_iommu_unity_map
, list
) {
1127 list_del(&entry
->list
);
1132 /* called when we find an exclusion range definition in ACPI */
1133 static int __init
init_exclusion_range(struct ivmd_header
*m
)
1138 case ACPI_IVMD_TYPE
:
1139 set_device_exclusion_range(m
->devid
, m
);
1141 case ACPI_IVMD_TYPE_ALL
:
1142 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1143 set_device_exclusion_range(i
, m
);
1145 case ACPI_IVMD_TYPE_RANGE
:
1146 for (i
= m
->devid
; i
<= m
->aux
; ++i
)
1147 set_device_exclusion_range(i
, m
);
1156 /* called for unity map ACPI definition */
1157 static int __init
init_unity_map_range(struct ivmd_header
*m
)
1159 struct unity_map_entry
*e
= 0;
1162 e
= kzalloc(sizeof(*e
), GFP_KERNEL
);
1170 case ACPI_IVMD_TYPE
:
1171 s
= "IVMD_TYPEi\t\t\t";
1172 e
->devid_start
= e
->devid_end
= m
->devid
;
1174 case ACPI_IVMD_TYPE_ALL
:
1175 s
= "IVMD_TYPE_ALL\t\t";
1177 e
->devid_end
= amd_iommu_last_bdf
;
1179 case ACPI_IVMD_TYPE_RANGE
:
1180 s
= "IVMD_TYPE_RANGE\t\t";
1181 e
->devid_start
= m
->devid
;
1182 e
->devid_end
= m
->aux
;
1185 e
->address_start
= PAGE_ALIGN(m
->range_start
);
1186 e
->address_end
= e
->address_start
+ PAGE_ALIGN(m
->range_length
);
1187 e
->prot
= m
->flags
>> 1;
1189 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1190 " range_start: %016llx range_end: %016llx flags: %x\n", s
,
1191 PCI_BUS(e
->devid_start
), PCI_SLOT(e
->devid_start
),
1192 PCI_FUNC(e
->devid_start
), PCI_BUS(e
->devid_end
),
1193 PCI_SLOT(e
->devid_end
), PCI_FUNC(e
->devid_end
),
1194 e
->address_start
, e
->address_end
, m
->flags
);
1196 list_add_tail(&e
->list
, &amd_iommu_unity_map
);
1201 /* iterates over all memory definitions we find in the ACPI table */
1202 static int __init
init_memory_definitions(struct acpi_table_header
*table
)
1204 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1205 struct ivmd_header
*m
;
1207 end
+= table
->length
;
1208 p
+= IVRS_HEADER_LENGTH
;
1211 m
= (struct ivmd_header
*)p
;
1212 if (m
->flags
& IVMD_FLAG_EXCL_RANGE
)
1213 init_exclusion_range(m
);
1214 else if (m
->flags
& IVMD_FLAG_UNITY_MAP
)
1215 init_unity_map_range(m
);
1224 * Init the device table to not allow DMA access for devices and
1225 * suppress all page faults
1227 static void init_device_table(void)
1231 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1232 set_dev_entry_bit(devid
, DEV_ENTRY_VALID
);
1233 set_dev_entry_bit(devid
, DEV_ENTRY_TRANSLATION
);
1237 static void iommu_init_flags(struct amd_iommu
*iommu
)
1239 iommu
->acpi_flags
& IVHD_FLAG_HT_TUN_EN_MASK
?
1240 iommu_feature_enable(iommu
, CONTROL_HT_TUN_EN
) :
1241 iommu_feature_disable(iommu
, CONTROL_HT_TUN_EN
);
1243 iommu
->acpi_flags
& IVHD_FLAG_PASSPW_EN_MASK
?
1244 iommu_feature_enable(iommu
, CONTROL_PASSPW_EN
) :
1245 iommu_feature_disable(iommu
, CONTROL_PASSPW_EN
);
1247 iommu
->acpi_flags
& IVHD_FLAG_RESPASSPW_EN_MASK
?
1248 iommu_feature_enable(iommu
, CONTROL_RESPASSPW_EN
) :
1249 iommu_feature_disable(iommu
, CONTROL_RESPASSPW_EN
);
1251 iommu
->acpi_flags
& IVHD_FLAG_ISOC_EN_MASK
?
1252 iommu_feature_enable(iommu
, CONTROL_ISOC_EN
) :
1253 iommu_feature_disable(iommu
, CONTROL_ISOC_EN
);
1256 * make IOMMU memory accesses cache coherent
1258 iommu_feature_enable(iommu
, CONTROL_COHERENT_EN
);
1261 static void iommu_apply_resume_quirks(struct amd_iommu
*iommu
)
1264 u32 ioc_feature_control
;
1265 struct pci_dev
*pdev
= iommu
->root_pdev
;
1267 /* RD890 BIOSes may not have completely reconfigured the iommu */
1268 if (!is_rd890_iommu(iommu
->dev
) || !pdev
)
1272 * First, we need to ensure that the iommu is enabled. This is
1273 * controlled by a register in the northbridge
1276 /* Select Northbridge indirect register 0x75 and enable writing */
1277 pci_write_config_dword(pdev
, 0x60, 0x75 | (1 << 7));
1278 pci_read_config_dword(pdev
, 0x64, &ioc_feature_control
);
1280 /* Enable the iommu */
1281 if (!(ioc_feature_control
& 0x1))
1282 pci_write_config_dword(pdev
, 0x64, ioc_feature_control
| 1);
1284 /* Restore the iommu BAR */
1285 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1286 iommu
->stored_addr_lo
);
1287 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1288 iommu
->stored_addr_hi
);
1290 /* Restore the l1 indirect regs for each of the 6 l1s */
1291 for (i
= 0; i
< 6; i
++)
1292 for (j
= 0; j
< 0x12; j
++)
1293 iommu_write_l1(iommu
, i
, j
, iommu
->stored_l1
[i
][j
]);
1295 /* Restore the l2 indirect regs */
1296 for (i
= 0; i
< 0x83; i
++)
1297 iommu_write_l2(iommu
, i
, iommu
->stored_l2
[i
]);
1299 /* Lock PCI setup registers */
1300 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1301 iommu
->stored_addr_lo
| 1);
1305 * This function finally enables all IOMMUs found in the system after
1306 * they have been initialized
1308 static void enable_iommus(void)
1310 struct amd_iommu
*iommu
;
1312 for_each_iommu(iommu
) {
1313 iommu_disable(iommu
);
1314 iommu_init_flags(iommu
);
1315 iommu_set_device_table(iommu
);
1316 iommu_enable_command_buffer(iommu
);
1317 iommu_enable_event_buffer(iommu
);
1318 iommu_set_exclusion_range(iommu
);
1319 iommu_init_msi(iommu
);
1320 iommu_enable(iommu
);
1321 iommu_flush_all_caches(iommu
);
1325 static void disable_iommus(void)
1327 struct amd_iommu
*iommu
;
1329 for_each_iommu(iommu
)
1330 iommu_disable(iommu
);
1334 * Suspend/Resume support
1335 * disable suspend until real resume implemented
1338 static void amd_iommu_resume(void)
1340 struct amd_iommu
*iommu
;
1342 for_each_iommu(iommu
)
1343 iommu_apply_resume_quirks(iommu
);
1345 /* re-load the hardware */
1349 * we have to flush after the IOMMUs are enabled because a
1350 * disabled IOMMU will never execute the commands we send
1352 for_each_iommu(iommu
)
1353 iommu_flush_all_caches(iommu
);
1356 static int amd_iommu_suspend(void)
1358 /* disable IOMMUs to go out of the way for BIOS */
1364 static struct syscore_ops amd_iommu_syscore_ops
= {
1365 .suspend
= amd_iommu_suspend
,
1366 .resume
= amd_iommu_resume
,
1370 * This is the core init function for AMD IOMMU hardware in the system.
1371 * This function is called from the generic x86 DMA layer initialization
1374 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1377 * 1 pass) Find the highest PCI device id the driver has to handle.
1378 * Upon this information the size of the data structures is
1379 * determined that needs to be allocated.
1381 * 2 pass) Initialize the data structures just allocated with the
1382 * information in the ACPI table about available AMD IOMMUs
1383 * in the system. It also maps the PCI devices in the
1384 * system to specific IOMMUs
1386 * 3 pass) After the basic data structures are allocated and
1387 * initialized we update them with information about memory
1388 * remapping requirements parsed out of the ACPI table in
1391 * After that the hardware is initialized and ready to go. In the last
1392 * step we do some Linux specific things like registering the driver in
1393 * the dma_ops interface and initializing the suspend/resume support
1394 * functions. Finally it prints some information about AMD IOMMUs and
1395 * the driver state and enables the hardware.
1397 static int __init
amd_iommu_init(void)
1399 struct amd_iommu
*iommu
;
1403 * First parse ACPI tables to find the largest Bus/Dev/Func
1404 * we need to handle. Upon this information the shared data
1405 * structures for the IOMMUs in the system will be allocated
1407 if (acpi_table_parse("IVRS", find_last_devid_acpi
) != 0)
1410 ret
= amd_iommu_init_err
;
1414 dev_table_size
= tbl_size(DEV_TABLE_ENTRY_SIZE
);
1415 alias_table_size
= tbl_size(ALIAS_TABLE_ENTRY_SIZE
);
1416 rlookup_table_size
= tbl_size(RLOOKUP_TABLE_ENTRY_SIZE
);
1420 /* Device table - directly used by all IOMMUs */
1421 amd_iommu_dev_table
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
1422 get_order(dev_table_size
));
1423 if (amd_iommu_dev_table
== NULL
)
1427 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1428 * IOMMU see for that device
1430 amd_iommu_alias_table
= (void *)__get_free_pages(GFP_KERNEL
,
1431 get_order(alias_table_size
));
1432 if (amd_iommu_alias_table
== NULL
)
1435 /* IOMMU rlookup table - find the IOMMU for a specific device */
1436 amd_iommu_rlookup_table
= (void *)__get_free_pages(
1437 GFP_KERNEL
| __GFP_ZERO
,
1438 get_order(rlookup_table_size
));
1439 if (amd_iommu_rlookup_table
== NULL
)
1442 amd_iommu_pd_alloc_bitmap
= (void *)__get_free_pages(
1443 GFP_KERNEL
| __GFP_ZERO
,
1444 get_order(MAX_DOMAIN_ID
/8));
1445 if (amd_iommu_pd_alloc_bitmap
== NULL
)
1449 * let all alias entries point to itself
1451 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1452 amd_iommu_alias_table
[i
] = i
;
1455 * never allocate domain 0 because its used as the non-allocated and
1456 * error value placeholder
1458 amd_iommu_pd_alloc_bitmap
[0] = 1;
1460 spin_lock_init(&amd_iommu_pd_lock
);
1463 * now the data structures are allocated and basically initialized
1464 * start the real acpi table scan
1467 if (acpi_table_parse("IVRS", init_iommu_all
) != 0)
1470 if (amd_iommu_init_err
) {
1471 ret
= amd_iommu_init_err
;
1475 if (acpi_table_parse("IVRS", init_memory_definitions
) != 0)
1478 if (amd_iommu_init_err
) {
1479 ret
= amd_iommu_init_err
;
1483 ret
= amd_iommu_init_devices();
1489 if (iommu_pass_through
)
1490 ret
= amd_iommu_init_passthrough();
1492 ret
= amd_iommu_init_dma_ops();
1497 /* init the device table */
1498 init_device_table();
1500 for_each_iommu(iommu
)
1501 iommu_flush_all_caches(iommu
);
1503 amd_iommu_init_api();
1505 amd_iommu_init_notifier();
1507 register_syscore_ops(&amd_iommu_syscore_ops
);
1509 x86_platform
.iommu_shutdown
= disable_iommus
;
1511 if (iommu_pass_through
)
1514 if (amd_iommu_unmap_flush
)
1515 printk(KERN_INFO
"AMD-Vi: IO/TLB flush on unmap enabled\n");
1517 printk(KERN_INFO
"AMD-Vi: Lazy IO/TLB flushing enabled\n");
1526 amd_iommu_uninit_devices();
1528 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap
,
1529 get_order(MAX_DOMAIN_ID
/8));
1531 free_pages((unsigned long)amd_iommu_rlookup_table
,
1532 get_order(rlookup_table_size
));
1534 free_pages((unsigned long)amd_iommu_alias_table
,
1535 get_order(alias_table_size
));
1537 free_pages((unsigned long)amd_iommu_dev_table
,
1538 get_order(dev_table_size
));
1544 #ifdef CONFIG_GART_IOMMU
1546 * We failed to initialize the AMD IOMMU - try fallback to GART
1556 /****************************************************************************
1558 * Early detect code. This code runs at IOMMU detection time in the DMA
1559 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1562 ****************************************************************************/
1563 static int __init
early_amd_iommu_detect(struct acpi_table_header
*table
)
1568 int __init
amd_iommu_detect(void)
1570 if (no_iommu
|| (iommu_detected
&& !gart_iommu_aperture
))
1573 if (amd_iommu_disabled
)
1576 if (acpi_table_parse("IVRS", early_amd_iommu_detect
) == 0) {
1578 amd_iommu_detected
= 1;
1579 x86_init
.iommu
.iommu_init
= amd_iommu_init
;
1581 /* Make sure ACS will be enabled */
1588 /****************************************************************************
1590 * Parsing functions for the AMD IOMMU specific kernel command line
1593 ****************************************************************************/
1595 static int __init
parse_amd_iommu_dump(char *str
)
1597 amd_iommu_dump
= true;
1602 static int __init
parse_amd_iommu_options(char *str
)
1604 for (; *str
; ++str
) {
1605 if (strncmp(str
, "fullflush", 9) == 0)
1606 amd_iommu_unmap_flush
= true;
1607 if (strncmp(str
, "off", 3) == 0)
1608 amd_iommu_disabled
= true;
1614 __setup("amd_iommu_dump", parse_amd_iommu_dump
);
1615 __setup("amd_iommu=", parse_amd_iommu_options
);
1617 IOMMU_INIT_FINISH(amd_iommu_detect
,
1618 gart_iommu_hole_init
,