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 <asm/pci-direct.h>
28 #include <asm/amd_iommu_proto.h>
29 #include <asm/amd_iommu_types.h>
30 #include <asm/amd_iommu.h>
31 #include <asm/iommu.h>
33 #include <asm/x86_init.h>
34 #include <asm/iommu_table.h>
36 * definitions for the ACPI scanning code
38 #define IVRS_HEADER_LENGTH 48
40 #define ACPI_IVHD_TYPE 0x10
41 #define ACPI_IVMD_TYPE_ALL 0x20
42 #define ACPI_IVMD_TYPE 0x21
43 #define ACPI_IVMD_TYPE_RANGE 0x22
45 #define IVHD_DEV_ALL 0x01
46 #define IVHD_DEV_SELECT 0x02
47 #define IVHD_DEV_SELECT_RANGE_START 0x03
48 #define IVHD_DEV_RANGE_END 0x04
49 #define IVHD_DEV_ALIAS 0x42
50 #define IVHD_DEV_ALIAS_RANGE 0x43
51 #define IVHD_DEV_EXT_SELECT 0x46
52 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
54 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
55 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
56 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
57 #define IVHD_FLAG_ISOC_EN_MASK 0x08
59 #define IVMD_FLAG_EXCL_RANGE 0x08
60 #define IVMD_FLAG_UNITY_MAP 0x01
62 #define ACPI_DEVFLAG_INITPASS 0x01
63 #define ACPI_DEVFLAG_EXTINT 0x02
64 #define ACPI_DEVFLAG_NMI 0x04
65 #define ACPI_DEVFLAG_SYSMGT1 0x10
66 #define ACPI_DEVFLAG_SYSMGT2 0x20
67 #define ACPI_DEVFLAG_LINT0 0x40
68 #define ACPI_DEVFLAG_LINT1 0x80
69 #define ACPI_DEVFLAG_ATSDIS 0x10000000
72 * ACPI table definitions
74 * These data structures are laid over the table to parse the important values
79 * structure describing one IOMMU in the ACPI table. Typically followed by one
80 * or more ivhd_entrys.
92 } __attribute__((packed
));
95 * A device entry describing which devices a specific IOMMU translates and
96 * which requestor ids they use.
103 } __attribute__((packed
));
106 * An AMD IOMMU memory definition structure. It defines things like exclusion
107 * ranges for devices and regions that should be unity mapped.
118 } __attribute__((packed
));
122 static int __initdata amd_iommu_detected
;
123 static bool __initdata amd_iommu_disabled
;
125 u16 amd_iommu_last_bdf
; /* largest PCI device id we have
127 LIST_HEAD(amd_iommu_unity_map
); /* a list of required unity mappings
129 bool amd_iommu_unmap_flush
; /* if true, flush on every unmap */
131 LIST_HEAD(amd_iommu_list
); /* list of all AMD IOMMUs in the
134 /* Array to assign indices to IOMMUs*/
135 struct amd_iommu
*amd_iommus
[MAX_IOMMUS
];
136 int amd_iommus_present
;
138 /* IOMMUs have a non-present cache? */
139 bool amd_iommu_np_cache __read_mostly
;
142 * The ACPI table parsing functions set this variable on an error
144 static int __initdata amd_iommu_init_err
;
147 * List of protection domains - used during resume
149 LIST_HEAD(amd_iommu_pd_list
);
150 spinlock_t amd_iommu_pd_lock
;
153 * Pointer to the device table which is shared by all AMD IOMMUs
154 * it is indexed by the PCI device id or the HT unit id and contains
155 * information about the domain the device belongs to as well as the
156 * page table root pointer.
158 struct dev_table_entry
*amd_iommu_dev_table
;
161 * The alias table is a driver specific data structure which contains the
162 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
163 * More than one device can share the same requestor id.
165 u16
*amd_iommu_alias_table
;
168 * The rlookup table is used to find the IOMMU which is responsible
169 * for a specific device. It is also indexed by the PCI device id.
171 struct amd_iommu
**amd_iommu_rlookup_table
;
174 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
175 * to know which ones are already in use.
177 unsigned long *amd_iommu_pd_alloc_bitmap
;
179 static u32 dev_table_size
; /* size of the device table */
180 static u32 alias_table_size
; /* size of the alias table */
181 static u32 rlookup_table_size
; /* size if the rlookup table */
184 * This function flushes all internal caches of
185 * the IOMMU used by this driver.
187 extern void iommu_flush_all_caches(struct amd_iommu
*iommu
);
189 static inline void update_last_devid(u16 devid
)
191 if (devid
> amd_iommu_last_bdf
)
192 amd_iommu_last_bdf
= devid
;
195 static inline unsigned long tbl_size(int entry_size
)
197 unsigned shift
= PAGE_SHIFT
+
198 get_order(((int)amd_iommu_last_bdf
+ 1) * entry_size
);
203 /* Access to l1 and l2 indexed register spaces */
205 static u32
iommu_read_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
)
209 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
210 pci_read_config_dword(iommu
->dev
, 0xfc, &val
);
214 static void iommu_write_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
, u32 val
)
216 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16 | 1 << 31));
217 pci_write_config_dword(iommu
->dev
, 0xfc, val
);
218 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
221 static u32
iommu_read_l2(struct amd_iommu
*iommu
, u8 address
)
225 pci_write_config_dword(iommu
->dev
, 0xf0, address
);
226 pci_read_config_dword(iommu
->dev
, 0xf4, &val
);
230 static void iommu_write_l2(struct amd_iommu
*iommu
, u8 address
, u32 val
)
232 pci_write_config_dword(iommu
->dev
, 0xf0, (address
| 1 << 8));
233 pci_write_config_dword(iommu
->dev
, 0xf4, val
);
236 /****************************************************************************
238 * AMD IOMMU MMIO register space handling functions
240 * These functions are used to program the IOMMU device registers in
241 * MMIO space required for that driver.
243 ****************************************************************************/
246 * This function set the exclusion range in the IOMMU. DMA accesses to the
247 * exclusion range are passed through untranslated
249 static void iommu_set_exclusion_range(struct amd_iommu
*iommu
)
251 u64 start
= iommu
->exclusion_start
& PAGE_MASK
;
252 u64 limit
= (start
+ iommu
->exclusion_length
) & PAGE_MASK
;
255 if (!iommu
->exclusion_start
)
258 entry
= start
| MMIO_EXCL_ENABLE_MASK
;
259 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_BASE_OFFSET
,
260 &entry
, sizeof(entry
));
263 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_LIMIT_OFFSET
,
264 &entry
, sizeof(entry
));
267 /* Programs the physical address of the device table into the IOMMU hardware */
268 static void __init
iommu_set_device_table(struct amd_iommu
*iommu
)
272 BUG_ON(iommu
->mmio_base
== NULL
);
274 entry
= virt_to_phys(amd_iommu_dev_table
);
275 entry
|= (dev_table_size
>> 12) - 1;
276 memcpy_toio(iommu
->mmio_base
+ MMIO_DEV_TABLE_OFFSET
,
277 &entry
, sizeof(entry
));
280 /* Generic functions to enable/disable certain features of the IOMMU. */
281 static void iommu_feature_enable(struct amd_iommu
*iommu
, u8 bit
)
285 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
287 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
290 static void iommu_feature_disable(struct amd_iommu
*iommu
, u8 bit
)
294 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
296 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
299 /* Function to enable the hardware */
300 static void iommu_enable(struct amd_iommu
*iommu
)
302 static const char * const feat_str
[] = {
303 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
304 "IA", "GA", "HE", "PC", NULL
308 printk(KERN_INFO
"AMD-Vi: Enabling IOMMU at %s cap 0x%hx",
309 dev_name(&iommu
->dev
->dev
), iommu
->cap_ptr
);
311 if (iommu
->cap
& (1 << IOMMU_CAP_EFR
)) {
312 printk(KERN_CONT
" extended features: ");
313 for (i
= 0; feat_str
[i
]; ++i
)
314 if (iommu_feature(iommu
, (1ULL << i
)))
315 printk(KERN_CONT
" %s", feat_str
[i
]);
317 printk(KERN_CONT
"\n");
319 iommu_feature_enable(iommu
, CONTROL_IOMMU_EN
);
322 static void iommu_disable(struct amd_iommu
*iommu
)
324 /* Disable command buffer */
325 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
327 /* Disable event logging and event interrupts */
328 iommu_feature_disable(iommu
, CONTROL_EVT_INT_EN
);
329 iommu_feature_disable(iommu
, CONTROL_EVT_LOG_EN
);
331 /* Disable IOMMU hardware itself */
332 iommu_feature_disable(iommu
, CONTROL_IOMMU_EN
);
336 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
337 * the system has one.
339 static u8
* __init
iommu_map_mmio_space(u64 address
)
343 if (!request_mem_region(address
, MMIO_REGION_LENGTH
, "amd_iommu")) {
344 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
346 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
350 ret
= ioremap_nocache(address
, MMIO_REGION_LENGTH
);
354 release_mem_region(address
, MMIO_REGION_LENGTH
);
359 static void __init
iommu_unmap_mmio_space(struct amd_iommu
*iommu
)
361 if (iommu
->mmio_base
)
362 iounmap(iommu
->mmio_base
);
363 release_mem_region(iommu
->mmio_phys
, MMIO_REGION_LENGTH
);
366 /****************************************************************************
368 * The functions below belong to the first pass of AMD IOMMU ACPI table
369 * parsing. In this pass we try to find out the highest device id this
370 * code has to handle. Upon this information the size of the shared data
371 * structures is determined later.
373 ****************************************************************************/
376 * This function calculates the length of a given IVHD entry
378 static inline int ivhd_entry_length(u8
*ivhd
)
380 return 0x04 << (*ivhd
>> 6);
384 * This function reads the last device id the IOMMU has to handle from the PCI
385 * capability header for this IOMMU
387 static int __init
find_last_devid_on_pci(int bus
, int dev
, int fn
, int cap_ptr
)
391 cap
= read_pci_config(bus
, dev
, fn
, cap_ptr
+MMIO_RANGE_OFFSET
);
392 update_last_devid(calc_devid(MMIO_GET_BUS(cap
), MMIO_GET_LD(cap
)));
398 * After reading the highest device id from the IOMMU PCI capability header
399 * this function looks if there is a higher device id defined in the ACPI table
401 static int __init
find_last_devid_from_ivhd(struct ivhd_header
*h
)
403 u8
*p
= (void *)h
, *end
= (void *)h
;
404 struct ivhd_entry
*dev
;
409 find_last_devid_on_pci(PCI_BUS(h
->devid
),
415 dev
= (struct ivhd_entry
*)p
;
417 case IVHD_DEV_SELECT
:
418 case IVHD_DEV_RANGE_END
:
420 case IVHD_DEV_EXT_SELECT
:
421 /* all the above subfield types refer to device ids */
422 update_last_devid(dev
->devid
);
427 p
+= ivhd_entry_length(p
);
436 * Iterate over all IVHD entries in the ACPI table and find the highest device
437 * id which we need to handle. This is the first of three functions which parse
438 * the ACPI table. So we check the checksum here.
440 static int __init
find_last_devid_acpi(struct acpi_table_header
*table
)
443 u8 checksum
= 0, *p
= (u8
*)table
, *end
= (u8
*)table
;
444 struct ivhd_header
*h
;
447 * Validate checksum here so we don't need to do it when
448 * we actually parse the table
450 for (i
= 0; i
< table
->length
; ++i
)
453 /* ACPI table corrupt */
454 amd_iommu_init_err
= -ENODEV
;
458 p
+= IVRS_HEADER_LENGTH
;
460 end
+= table
->length
;
462 h
= (struct ivhd_header
*)p
;
465 find_last_devid_from_ivhd(h
);
477 /****************************************************************************
479 * The following functions belong the the code path which parses the ACPI table
480 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
481 * data structures, initialize the device/alias/rlookup table and also
482 * basically initialize the hardware.
484 ****************************************************************************/
487 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
488 * write commands to that buffer later and the IOMMU will execute them
491 static u8
* __init
alloc_command_buffer(struct amd_iommu
*iommu
)
493 u8
*cmd_buf
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
494 get_order(CMD_BUFFER_SIZE
));
499 iommu
->cmd_buf_size
= CMD_BUFFER_SIZE
| CMD_BUFFER_UNINITIALIZED
;
505 * This function resets the command buffer if the IOMMU stopped fetching
508 void amd_iommu_reset_cmd_buffer(struct amd_iommu
*iommu
)
510 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
512 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_HEAD_OFFSET
);
513 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_TAIL_OFFSET
);
515 iommu_feature_enable(iommu
, CONTROL_CMDBUF_EN
);
519 * This function writes the command buffer address to the hardware and
522 static void iommu_enable_command_buffer(struct amd_iommu
*iommu
)
526 BUG_ON(iommu
->cmd_buf
== NULL
);
528 entry
= (u64
)virt_to_phys(iommu
->cmd_buf
);
529 entry
|= MMIO_CMD_SIZE_512
;
531 memcpy_toio(iommu
->mmio_base
+ MMIO_CMD_BUF_OFFSET
,
532 &entry
, sizeof(entry
));
534 amd_iommu_reset_cmd_buffer(iommu
);
535 iommu
->cmd_buf_size
&= ~(CMD_BUFFER_UNINITIALIZED
);
538 static void __init
free_command_buffer(struct amd_iommu
*iommu
)
540 free_pages((unsigned long)iommu
->cmd_buf
,
541 get_order(iommu
->cmd_buf_size
& ~(CMD_BUFFER_UNINITIALIZED
)));
544 /* allocates the memory where the IOMMU will log its events to */
545 static u8
* __init
alloc_event_buffer(struct amd_iommu
*iommu
)
547 iommu
->evt_buf
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
548 get_order(EVT_BUFFER_SIZE
));
550 if (iommu
->evt_buf
== NULL
)
553 iommu
->evt_buf_size
= EVT_BUFFER_SIZE
;
555 return iommu
->evt_buf
;
558 static void iommu_enable_event_buffer(struct amd_iommu
*iommu
)
562 BUG_ON(iommu
->evt_buf
== NULL
);
564 entry
= (u64
)virt_to_phys(iommu
->evt_buf
) | EVT_LEN_MASK
;
566 memcpy_toio(iommu
->mmio_base
+ MMIO_EVT_BUF_OFFSET
,
567 &entry
, sizeof(entry
));
569 /* set head and tail to zero manually */
570 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_HEAD_OFFSET
);
571 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_TAIL_OFFSET
);
573 iommu_feature_enable(iommu
, CONTROL_EVT_LOG_EN
);
576 static void __init
free_event_buffer(struct amd_iommu
*iommu
)
578 free_pages((unsigned long)iommu
->evt_buf
, get_order(EVT_BUFFER_SIZE
));
581 /* sets a specific bit in the device table entry. */
582 static void set_dev_entry_bit(u16 devid
, u8 bit
)
584 int i
= (bit
>> 5) & 0x07;
585 int _bit
= bit
& 0x1f;
587 amd_iommu_dev_table
[devid
].data
[i
] |= (1 << _bit
);
590 static int get_dev_entry_bit(u16 devid
, u8 bit
)
592 int i
= (bit
>> 5) & 0x07;
593 int _bit
= bit
& 0x1f;
595 return (amd_iommu_dev_table
[devid
].data
[i
] & (1 << _bit
)) >> _bit
;
599 void amd_iommu_apply_erratum_63(u16 devid
)
603 sysmgt
= get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
) |
604 (get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
) << 1);
607 set_dev_entry_bit(devid
, DEV_ENTRY_IW
);
610 /* Writes the specific IOMMU for a device into the rlookup table */
611 static void __init
set_iommu_for_device(struct amd_iommu
*iommu
, u16 devid
)
613 amd_iommu_rlookup_table
[devid
] = iommu
;
617 * This function takes the device specific flags read from the ACPI
618 * table and sets up the device table entry with that information
620 static void __init
set_dev_entry_from_acpi(struct amd_iommu
*iommu
,
621 u16 devid
, u32 flags
, u32 ext_flags
)
623 if (flags
& ACPI_DEVFLAG_INITPASS
)
624 set_dev_entry_bit(devid
, DEV_ENTRY_INIT_PASS
);
625 if (flags
& ACPI_DEVFLAG_EXTINT
)
626 set_dev_entry_bit(devid
, DEV_ENTRY_EINT_PASS
);
627 if (flags
& ACPI_DEVFLAG_NMI
)
628 set_dev_entry_bit(devid
, DEV_ENTRY_NMI_PASS
);
629 if (flags
& ACPI_DEVFLAG_SYSMGT1
)
630 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
);
631 if (flags
& ACPI_DEVFLAG_SYSMGT2
)
632 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
);
633 if (flags
& ACPI_DEVFLAG_LINT0
)
634 set_dev_entry_bit(devid
, DEV_ENTRY_LINT0_PASS
);
635 if (flags
& ACPI_DEVFLAG_LINT1
)
636 set_dev_entry_bit(devid
, DEV_ENTRY_LINT1_PASS
);
638 amd_iommu_apply_erratum_63(devid
);
640 set_iommu_for_device(iommu
, devid
);
644 * Reads the device exclusion range from ACPI and initialize IOMMU with
647 static void __init
set_device_exclusion_range(u16 devid
, struct ivmd_header
*m
)
649 struct amd_iommu
*iommu
= amd_iommu_rlookup_table
[devid
];
651 if (!(m
->flags
& IVMD_FLAG_EXCL_RANGE
))
656 * We only can configure exclusion ranges per IOMMU, not
657 * per device. But we can enable the exclusion range per
658 * device. This is done here
660 set_dev_entry_bit(m
->devid
, DEV_ENTRY_EX
);
661 iommu
->exclusion_start
= m
->range_start
;
662 iommu
->exclusion_length
= m
->range_length
;
667 * This function reads some important data from the IOMMU PCI space and
668 * initializes the driver data structure with it. It reads the hardware
669 * capabilities and the first/last device entries
671 static void __init
init_iommu_from_pci(struct amd_iommu
*iommu
)
673 int cap_ptr
= iommu
->cap_ptr
;
674 u32 range
, misc
, low
, high
;
677 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_CAP_HDR_OFFSET
,
679 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_RANGE_OFFSET
,
681 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_MISC_OFFSET
,
684 iommu
->first_device
= calc_devid(MMIO_GET_BUS(range
),
686 iommu
->last_device
= calc_devid(MMIO_GET_BUS(range
),
688 iommu
->evt_msi_num
= MMIO_MSI_NUM(misc
);
690 /* read extended feature bits */
691 low
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
);
692 high
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
+ 4);
694 iommu
->features
= ((u64
)high
<< 32) | low
;
696 if (!is_rd890_iommu(iommu
->dev
))
700 * Some rd890 systems may not be fully reconfigured by the BIOS, so
701 * it's necessary for us to store this information so it can be
702 * reprogrammed on resume
705 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
706 &iommu
->stored_addr_lo
);
707 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
708 &iommu
->stored_addr_hi
);
710 /* Low bit locks writes to configuration space */
711 iommu
->stored_addr_lo
&= ~1;
713 for (i
= 0; i
< 6; i
++)
714 for (j
= 0; j
< 0x12; j
++)
715 iommu
->stored_l1
[i
][j
] = iommu_read_l1(iommu
, i
, j
);
717 for (i
= 0; i
< 0x83; i
++)
718 iommu
->stored_l2
[i
] = iommu_read_l2(iommu
, i
);
722 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
723 * initializes the hardware and our data structures with it.
725 static void __init
init_iommu_from_acpi(struct amd_iommu
*iommu
,
726 struct ivhd_header
*h
)
729 u8
*end
= p
, flags
= 0;
730 u16 dev_i
, devid
= 0, devid_start
= 0, devid_to
= 0;
733 struct ivhd_entry
*e
;
736 * First save the recommended feature enable bits from ACPI
738 iommu
->acpi_flags
= h
->flags
;
741 * Done. Now parse the device entries
743 p
+= sizeof(struct ivhd_header
);
748 e
= (struct ivhd_entry
*)p
;
752 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
753 " last device %02x:%02x.%x flags: %02x\n",
754 PCI_BUS(iommu
->first_device
),
755 PCI_SLOT(iommu
->first_device
),
756 PCI_FUNC(iommu
->first_device
),
757 PCI_BUS(iommu
->last_device
),
758 PCI_SLOT(iommu
->last_device
),
759 PCI_FUNC(iommu
->last_device
),
762 for (dev_i
= iommu
->first_device
;
763 dev_i
<= iommu
->last_device
; ++dev_i
)
764 set_dev_entry_from_acpi(iommu
, dev_i
,
767 case IVHD_DEV_SELECT
:
769 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
777 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
779 case IVHD_DEV_SELECT_RANGE_START
:
781 DUMP_printk(" DEV_SELECT_RANGE_START\t "
782 "devid: %02x:%02x.%x flags: %02x\n",
788 devid_start
= e
->devid
;
795 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
796 "flags: %02x devid_to: %02x:%02x.%x\n",
801 PCI_BUS(e
->ext
>> 8),
802 PCI_SLOT(e
->ext
>> 8),
803 PCI_FUNC(e
->ext
>> 8));
806 devid_to
= e
->ext
>> 8;
807 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
808 set_dev_entry_from_acpi(iommu
, devid_to
, e
->flags
, 0);
809 amd_iommu_alias_table
[devid
] = devid_to
;
811 case IVHD_DEV_ALIAS_RANGE
:
813 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
814 "devid: %02x:%02x.%x flags: %02x "
815 "devid_to: %02x:%02x.%x\n",
820 PCI_BUS(e
->ext
>> 8),
821 PCI_SLOT(e
->ext
>> 8),
822 PCI_FUNC(e
->ext
>> 8));
824 devid_start
= e
->devid
;
826 devid_to
= e
->ext
>> 8;
830 case IVHD_DEV_EXT_SELECT
:
832 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
833 "flags: %02x ext: %08x\n",
840 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
,
843 case IVHD_DEV_EXT_SELECT_RANGE
:
845 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
846 "%02x:%02x.%x flags: %02x ext: %08x\n",
852 devid_start
= e
->devid
;
857 case IVHD_DEV_RANGE_END
:
859 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
865 for (dev_i
= devid_start
; dev_i
<= devid
; ++dev_i
) {
867 amd_iommu_alias_table
[dev_i
] = devid_to
;
868 set_dev_entry_from_acpi(iommu
,
869 devid_to
, flags
, ext_flags
);
871 set_dev_entry_from_acpi(iommu
, dev_i
,
879 p
+= ivhd_entry_length(p
);
883 /* Initializes the device->iommu mapping for the driver */
884 static int __init
init_iommu_devices(struct amd_iommu
*iommu
)
888 for (i
= iommu
->first_device
; i
<= iommu
->last_device
; ++i
)
889 set_iommu_for_device(iommu
, i
);
894 static void __init
free_iommu_one(struct amd_iommu
*iommu
)
896 free_command_buffer(iommu
);
897 free_event_buffer(iommu
);
898 iommu_unmap_mmio_space(iommu
);
901 static void __init
free_iommu_all(void)
903 struct amd_iommu
*iommu
, *next
;
905 for_each_iommu_safe(iommu
, next
) {
906 list_del(&iommu
->list
);
907 free_iommu_one(iommu
);
913 * This function clues the initialization function for one IOMMU
914 * together and also allocates the command buffer and programs the
915 * hardware. It does NOT enable the IOMMU. This is done afterwards.
917 static int __init
init_iommu_one(struct amd_iommu
*iommu
, struct ivhd_header
*h
)
919 spin_lock_init(&iommu
->lock
);
921 /* Add IOMMU to internal data structures */
922 list_add_tail(&iommu
->list
, &amd_iommu_list
);
923 iommu
->index
= amd_iommus_present
++;
925 if (unlikely(iommu
->index
>= MAX_IOMMUS
)) {
926 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
930 /* Index is fine - add IOMMU to the array */
931 amd_iommus
[iommu
->index
] = iommu
;
934 * Copy data from ACPI table entry to the iommu struct
936 iommu
->dev
= pci_get_bus_and_slot(PCI_BUS(h
->devid
), h
->devid
& 0xff);
940 iommu
->cap_ptr
= h
->cap_ptr
;
941 iommu
->pci_seg
= h
->pci_seg
;
942 iommu
->mmio_phys
= h
->mmio_phys
;
943 iommu
->mmio_base
= iommu_map_mmio_space(h
->mmio_phys
);
944 if (!iommu
->mmio_base
)
947 iommu
->cmd_buf
= alloc_command_buffer(iommu
);
951 iommu
->evt_buf
= alloc_event_buffer(iommu
);
955 iommu
->int_enabled
= false;
957 init_iommu_from_pci(iommu
);
958 init_iommu_from_acpi(iommu
, h
);
959 init_iommu_devices(iommu
);
961 if (iommu
->cap
& (1UL << IOMMU_CAP_NPCACHE
))
962 amd_iommu_np_cache
= true;
964 return pci_enable_device(iommu
->dev
);
968 * Iterates over all IOMMU entries in the ACPI table, allocates the
969 * IOMMU structure and initializes it with init_iommu_one()
971 static int __init
init_iommu_all(struct acpi_table_header
*table
)
973 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
974 struct ivhd_header
*h
;
975 struct amd_iommu
*iommu
;
978 end
+= table
->length
;
979 p
+= IVRS_HEADER_LENGTH
;
982 h
= (struct ivhd_header
*)p
;
986 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
987 "seg: %d flags: %01x info %04x\n",
988 PCI_BUS(h
->devid
), PCI_SLOT(h
->devid
),
989 PCI_FUNC(h
->devid
), h
->cap_ptr
,
990 h
->pci_seg
, h
->flags
, h
->info
);
991 DUMP_printk(" mmio-addr: %016llx\n",
994 iommu
= kzalloc(sizeof(struct amd_iommu
), GFP_KERNEL
);
996 amd_iommu_init_err
= -ENOMEM
;
1000 ret
= init_iommu_one(iommu
, h
);
1002 amd_iommu_init_err
= ret
;
1017 /****************************************************************************
1019 * The following functions initialize the MSI interrupts for all IOMMUs
1020 * in the system. Its a bit challenging because there could be multiple
1021 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1024 ****************************************************************************/
1026 static int iommu_setup_msi(struct amd_iommu
*iommu
)
1030 if (pci_enable_msi(iommu
->dev
))
1033 r
= request_irq(iommu
->dev
->irq
, amd_iommu_int_handler
,
1039 pci_disable_msi(iommu
->dev
);
1043 iommu
->int_enabled
= true;
1044 iommu_feature_enable(iommu
, CONTROL_EVT_INT_EN
);
1049 static int iommu_init_msi(struct amd_iommu
*iommu
)
1051 if (iommu
->int_enabled
)
1054 if (pci_find_capability(iommu
->dev
, PCI_CAP_ID_MSI
))
1055 return iommu_setup_msi(iommu
);
1060 /****************************************************************************
1062 * The next functions belong to the third pass of parsing the ACPI
1063 * table. In this last pass the memory mapping requirements are
1064 * gathered (like exclusion and unity mapping reanges).
1066 ****************************************************************************/
1068 static void __init
free_unity_maps(void)
1070 struct unity_map_entry
*entry
, *next
;
1072 list_for_each_entry_safe(entry
, next
, &amd_iommu_unity_map
, list
) {
1073 list_del(&entry
->list
);
1078 /* called when we find an exclusion range definition in ACPI */
1079 static int __init
init_exclusion_range(struct ivmd_header
*m
)
1084 case ACPI_IVMD_TYPE
:
1085 set_device_exclusion_range(m
->devid
, m
);
1087 case ACPI_IVMD_TYPE_ALL
:
1088 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1089 set_device_exclusion_range(i
, m
);
1091 case ACPI_IVMD_TYPE_RANGE
:
1092 for (i
= m
->devid
; i
<= m
->aux
; ++i
)
1093 set_device_exclusion_range(i
, m
);
1102 /* called for unity map ACPI definition */
1103 static int __init
init_unity_map_range(struct ivmd_header
*m
)
1105 struct unity_map_entry
*e
= 0;
1108 e
= kzalloc(sizeof(*e
), GFP_KERNEL
);
1116 case ACPI_IVMD_TYPE
:
1117 s
= "IVMD_TYPEi\t\t\t";
1118 e
->devid_start
= e
->devid_end
= m
->devid
;
1120 case ACPI_IVMD_TYPE_ALL
:
1121 s
= "IVMD_TYPE_ALL\t\t";
1123 e
->devid_end
= amd_iommu_last_bdf
;
1125 case ACPI_IVMD_TYPE_RANGE
:
1126 s
= "IVMD_TYPE_RANGE\t\t";
1127 e
->devid_start
= m
->devid
;
1128 e
->devid_end
= m
->aux
;
1131 e
->address_start
= PAGE_ALIGN(m
->range_start
);
1132 e
->address_end
= e
->address_start
+ PAGE_ALIGN(m
->range_length
);
1133 e
->prot
= m
->flags
>> 1;
1135 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1136 " range_start: %016llx range_end: %016llx flags: %x\n", s
,
1137 PCI_BUS(e
->devid_start
), PCI_SLOT(e
->devid_start
),
1138 PCI_FUNC(e
->devid_start
), PCI_BUS(e
->devid_end
),
1139 PCI_SLOT(e
->devid_end
), PCI_FUNC(e
->devid_end
),
1140 e
->address_start
, e
->address_end
, m
->flags
);
1142 list_add_tail(&e
->list
, &amd_iommu_unity_map
);
1147 /* iterates over all memory definitions we find in the ACPI table */
1148 static int __init
init_memory_definitions(struct acpi_table_header
*table
)
1150 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1151 struct ivmd_header
*m
;
1153 end
+= table
->length
;
1154 p
+= IVRS_HEADER_LENGTH
;
1157 m
= (struct ivmd_header
*)p
;
1158 if (m
->flags
& IVMD_FLAG_EXCL_RANGE
)
1159 init_exclusion_range(m
);
1160 else if (m
->flags
& IVMD_FLAG_UNITY_MAP
)
1161 init_unity_map_range(m
);
1170 * Init the device table to not allow DMA access for devices and
1171 * suppress all page faults
1173 static void init_device_table(void)
1177 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1178 set_dev_entry_bit(devid
, DEV_ENTRY_VALID
);
1179 set_dev_entry_bit(devid
, DEV_ENTRY_TRANSLATION
);
1183 static void iommu_init_flags(struct amd_iommu
*iommu
)
1185 iommu
->acpi_flags
& IVHD_FLAG_HT_TUN_EN_MASK
?
1186 iommu_feature_enable(iommu
, CONTROL_HT_TUN_EN
) :
1187 iommu_feature_disable(iommu
, CONTROL_HT_TUN_EN
);
1189 iommu
->acpi_flags
& IVHD_FLAG_PASSPW_EN_MASK
?
1190 iommu_feature_enable(iommu
, CONTROL_PASSPW_EN
) :
1191 iommu_feature_disable(iommu
, CONTROL_PASSPW_EN
);
1193 iommu
->acpi_flags
& IVHD_FLAG_RESPASSPW_EN_MASK
?
1194 iommu_feature_enable(iommu
, CONTROL_RESPASSPW_EN
) :
1195 iommu_feature_disable(iommu
, CONTROL_RESPASSPW_EN
);
1197 iommu
->acpi_flags
& IVHD_FLAG_ISOC_EN_MASK
?
1198 iommu_feature_enable(iommu
, CONTROL_ISOC_EN
) :
1199 iommu_feature_disable(iommu
, CONTROL_ISOC_EN
);
1202 * make IOMMU memory accesses cache coherent
1204 iommu_feature_enable(iommu
, CONTROL_COHERENT_EN
);
1207 static void iommu_apply_resume_quirks(struct amd_iommu
*iommu
)
1210 u32 ioc_feature_control
;
1211 struct pci_dev
*pdev
= NULL
;
1213 /* RD890 BIOSes may not have completely reconfigured the iommu */
1214 if (!is_rd890_iommu(iommu
->dev
))
1218 * First, we need to ensure that the iommu is enabled. This is
1219 * controlled by a register in the northbridge
1221 pdev
= pci_get_bus_and_slot(iommu
->dev
->bus
->number
, PCI_DEVFN(0, 0));
1226 /* Select Northbridge indirect register 0x75 and enable writing */
1227 pci_write_config_dword(pdev
, 0x60, 0x75 | (1 << 7));
1228 pci_read_config_dword(pdev
, 0x64, &ioc_feature_control
);
1230 /* Enable the iommu */
1231 if (!(ioc_feature_control
& 0x1))
1232 pci_write_config_dword(pdev
, 0x64, ioc_feature_control
| 1);
1236 /* Restore the iommu BAR */
1237 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1238 iommu
->stored_addr_lo
);
1239 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1240 iommu
->stored_addr_hi
);
1242 /* Restore the l1 indirect regs for each of the 6 l1s */
1243 for (i
= 0; i
< 6; i
++)
1244 for (j
= 0; j
< 0x12; j
++)
1245 iommu_write_l1(iommu
, i
, j
, iommu
->stored_l1
[i
][j
]);
1247 /* Restore the l2 indirect regs */
1248 for (i
= 0; i
< 0x83; i
++)
1249 iommu_write_l2(iommu
, i
, iommu
->stored_l2
[i
]);
1251 /* Lock PCI setup registers */
1252 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1253 iommu
->stored_addr_lo
| 1);
1257 * This function finally enables all IOMMUs found in the system after
1258 * they have been initialized
1260 static void enable_iommus(void)
1262 struct amd_iommu
*iommu
;
1264 for_each_iommu(iommu
) {
1265 iommu_disable(iommu
);
1266 iommu_init_flags(iommu
);
1267 iommu_set_device_table(iommu
);
1268 iommu_enable_command_buffer(iommu
);
1269 iommu_enable_event_buffer(iommu
);
1270 iommu_set_exclusion_range(iommu
);
1271 iommu_init_msi(iommu
);
1272 iommu_enable(iommu
);
1273 iommu_flush_all_caches(iommu
);
1277 static void disable_iommus(void)
1279 struct amd_iommu
*iommu
;
1281 for_each_iommu(iommu
)
1282 iommu_disable(iommu
);
1286 * Suspend/Resume support
1287 * disable suspend until real resume implemented
1290 static void amd_iommu_resume(void)
1292 struct amd_iommu
*iommu
;
1294 for_each_iommu(iommu
)
1295 iommu_apply_resume_quirks(iommu
);
1297 /* re-load the hardware */
1301 * we have to flush after the IOMMUs are enabled because a
1302 * disabled IOMMU will never execute the commands we send
1304 for_each_iommu(iommu
)
1305 iommu_flush_all_caches(iommu
);
1308 static int amd_iommu_suspend(void)
1310 /* disable IOMMUs to go out of the way for BIOS */
1316 static struct syscore_ops amd_iommu_syscore_ops
= {
1317 .suspend
= amd_iommu_suspend
,
1318 .resume
= amd_iommu_resume
,
1322 * This is the core init function for AMD IOMMU hardware in the system.
1323 * This function is called from the generic x86 DMA layer initialization
1326 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1329 * 1 pass) Find the highest PCI device id the driver has to handle.
1330 * Upon this information the size of the data structures is
1331 * determined that needs to be allocated.
1333 * 2 pass) Initialize the data structures just allocated with the
1334 * information in the ACPI table about available AMD IOMMUs
1335 * in the system. It also maps the PCI devices in the
1336 * system to specific IOMMUs
1338 * 3 pass) After the basic data structures are allocated and
1339 * initialized we update them with information about memory
1340 * remapping requirements parsed out of the ACPI table in
1343 * After that the hardware is initialized and ready to go. In the last
1344 * step we do some Linux specific things like registering the driver in
1345 * the dma_ops interface and initializing the suspend/resume support
1346 * functions. Finally it prints some information about AMD IOMMUs and
1347 * the driver state and enables the hardware.
1349 static int __init
amd_iommu_init(void)
1354 * First parse ACPI tables to find the largest Bus/Dev/Func
1355 * we need to handle. Upon this information the shared data
1356 * structures for the IOMMUs in the system will be allocated
1358 if (acpi_table_parse("IVRS", find_last_devid_acpi
) != 0)
1361 ret
= amd_iommu_init_err
;
1365 dev_table_size
= tbl_size(DEV_TABLE_ENTRY_SIZE
);
1366 alias_table_size
= tbl_size(ALIAS_TABLE_ENTRY_SIZE
);
1367 rlookup_table_size
= tbl_size(RLOOKUP_TABLE_ENTRY_SIZE
);
1371 /* Device table - directly used by all IOMMUs */
1372 amd_iommu_dev_table
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
1373 get_order(dev_table_size
));
1374 if (amd_iommu_dev_table
== NULL
)
1378 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1379 * IOMMU see for that device
1381 amd_iommu_alias_table
= (void *)__get_free_pages(GFP_KERNEL
,
1382 get_order(alias_table_size
));
1383 if (amd_iommu_alias_table
== NULL
)
1386 /* IOMMU rlookup table - find the IOMMU for a specific device */
1387 amd_iommu_rlookup_table
= (void *)__get_free_pages(
1388 GFP_KERNEL
| __GFP_ZERO
,
1389 get_order(rlookup_table_size
));
1390 if (amd_iommu_rlookup_table
== NULL
)
1393 amd_iommu_pd_alloc_bitmap
= (void *)__get_free_pages(
1394 GFP_KERNEL
| __GFP_ZERO
,
1395 get_order(MAX_DOMAIN_ID
/8));
1396 if (amd_iommu_pd_alloc_bitmap
== NULL
)
1399 /* init the device table */
1400 init_device_table();
1403 * let all alias entries point to itself
1405 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1406 amd_iommu_alias_table
[i
] = i
;
1409 * never allocate domain 0 because its used as the non-allocated and
1410 * error value placeholder
1412 amd_iommu_pd_alloc_bitmap
[0] = 1;
1414 spin_lock_init(&amd_iommu_pd_lock
);
1417 * now the data structures are allocated and basically initialized
1418 * start the real acpi table scan
1421 if (acpi_table_parse("IVRS", init_iommu_all
) != 0)
1424 if (amd_iommu_init_err
) {
1425 ret
= amd_iommu_init_err
;
1429 if (acpi_table_parse("IVRS", init_memory_definitions
) != 0)
1432 if (amd_iommu_init_err
) {
1433 ret
= amd_iommu_init_err
;
1437 ret
= amd_iommu_init_devices();
1443 if (iommu_pass_through
)
1444 ret
= amd_iommu_init_passthrough();
1446 ret
= amd_iommu_init_dma_ops();
1451 amd_iommu_init_api();
1453 amd_iommu_init_notifier();
1455 register_syscore_ops(&amd_iommu_syscore_ops
);
1457 if (iommu_pass_through
)
1460 if (amd_iommu_unmap_flush
)
1461 printk(KERN_INFO
"AMD-Vi: IO/TLB flush on unmap enabled\n");
1463 printk(KERN_INFO
"AMD-Vi: Lazy IO/TLB flushing enabled\n");
1465 x86_platform
.iommu_shutdown
= disable_iommus
;
1473 amd_iommu_uninit_devices();
1475 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap
,
1476 get_order(MAX_DOMAIN_ID
/8));
1478 free_pages((unsigned long)amd_iommu_rlookup_table
,
1479 get_order(rlookup_table_size
));
1481 free_pages((unsigned long)amd_iommu_alias_table
,
1482 get_order(alias_table_size
));
1484 free_pages((unsigned long)amd_iommu_dev_table
,
1485 get_order(dev_table_size
));
1491 #ifdef CONFIG_GART_IOMMU
1493 * We failed to initialize the AMD IOMMU - try fallback to GART
1503 /****************************************************************************
1505 * Early detect code. This code runs at IOMMU detection time in the DMA
1506 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1509 ****************************************************************************/
1510 static int __init
early_amd_iommu_detect(struct acpi_table_header
*table
)
1515 int __init
amd_iommu_detect(void)
1517 if (no_iommu
|| (iommu_detected
&& !gart_iommu_aperture
))
1520 if (amd_iommu_disabled
)
1523 if (acpi_table_parse("IVRS", early_amd_iommu_detect
) == 0) {
1525 amd_iommu_detected
= 1;
1526 x86_init
.iommu
.iommu_init
= amd_iommu_init
;
1528 /* Make sure ACS will be enabled */
1535 /****************************************************************************
1537 * Parsing functions for the AMD IOMMU specific kernel command line
1540 ****************************************************************************/
1542 static int __init
parse_amd_iommu_dump(char *str
)
1544 amd_iommu_dump
= true;
1549 static int __init
parse_amd_iommu_options(char *str
)
1551 for (; *str
; ++str
) {
1552 if (strncmp(str
, "fullflush", 9) == 0)
1553 amd_iommu_unmap_flush
= true;
1554 if (strncmp(str
, "off", 3) == 0)
1555 amd_iommu_disabled
= true;
1561 __setup("amd_iommu_dump", parse_amd_iommu_dump
);
1562 __setup("amd_iommu=", parse_amd_iommu_options
);
1564 IOMMU_INIT_FINISH(amd_iommu_detect
,
1565 gart_iommu_hole_init
,