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 <linux/export.h>
29 #include <asm/pci-direct.h>
30 #include <asm/iommu.h>
32 #include <asm/x86_init.h>
33 #include <asm/iommu_table.h>
34 #include <asm/io_apic.h>
35 #include <asm/irq_remapping.h>
37 #include "amd_iommu_proto.h"
38 #include "amd_iommu_types.h"
39 #include "irq_remapping.h"
42 * definitions for the ACPI scanning code
44 #define IVRS_HEADER_LENGTH 48
46 #define ACPI_IVHD_TYPE 0x10
47 #define ACPI_IVMD_TYPE_ALL 0x20
48 #define ACPI_IVMD_TYPE 0x21
49 #define ACPI_IVMD_TYPE_RANGE 0x22
51 #define IVHD_DEV_ALL 0x01
52 #define IVHD_DEV_SELECT 0x02
53 #define IVHD_DEV_SELECT_RANGE_START 0x03
54 #define IVHD_DEV_RANGE_END 0x04
55 #define IVHD_DEV_ALIAS 0x42
56 #define IVHD_DEV_ALIAS_RANGE 0x43
57 #define IVHD_DEV_EXT_SELECT 0x46
58 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
59 #define IVHD_DEV_SPECIAL 0x48
61 #define IVHD_SPECIAL_IOAPIC 1
62 #define IVHD_SPECIAL_HPET 2
64 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
65 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
66 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
67 #define IVHD_FLAG_ISOC_EN_MASK 0x08
69 #define IVMD_FLAG_EXCL_RANGE 0x08
70 #define IVMD_FLAG_UNITY_MAP 0x01
72 #define ACPI_DEVFLAG_INITPASS 0x01
73 #define ACPI_DEVFLAG_EXTINT 0x02
74 #define ACPI_DEVFLAG_NMI 0x04
75 #define ACPI_DEVFLAG_SYSMGT1 0x10
76 #define ACPI_DEVFLAG_SYSMGT2 0x20
77 #define ACPI_DEVFLAG_LINT0 0x40
78 #define ACPI_DEVFLAG_LINT1 0x80
79 #define ACPI_DEVFLAG_ATSDIS 0x10000000
82 * ACPI table definitions
84 * These data structures are laid over the table to parse the important values
89 * structure describing one IOMMU in the ACPI table. Typically followed by one
90 * or more ivhd_entrys.
102 } __attribute__((packed
));
105 * A device entry describing which devices a specific IOMMU translates and
106 * which requestor ids they use.
113 } __attribute__((packed
));
116 * An AMD IOMMU memory definition structure. It defines things like exclusion
117 * ranges for devices and regions that should be unity mapped.
128 } __attribute__((packed
));
131 bool amd_iommu_irq_remap __read_mostly
;
133 static bool amd_iommu_detected
;
134 static bool __initdata amd_iommu_disabled
;
136 u16 amd_iommu_last_bdf
; /* largest PCI device id we have
138 LIST_HEAD(amd_iommu_unity_map
); /* a list of required unity mappings
140 u32 amd_iommu_unmap_flush
; /* if true, flush on every unmap */
142 LIST_HEAD(amd_iommu_list
); /* list of all AMD IOMMUs in the
145 /* Array to assign indices to IOMMUs*/
146 struct amd_iommu
*amd_iommus
[MAX_IOMMUS
];
147 int amd_iommus_present
;
149 /* IOMMUs have a non-present cache? */
150 bool amd_iommu_np_cache __read_mostly
;
151 bool amd_iommu_iotlb_sup __read_mostly
= true;
153 u32 amd_iommu_max_pasid __read_mostly
= ~0;
155 bool amd_iommu_v2_present __read_mostly
;
156 bool amd_iommu_pc_present __read_mostly
;
158 bool amd_iommu_force_isolation __read_mostly
;
161 * List of protection domains - used during resume
163 LIST_HEAD(amd_iommu_pd_list
);
164 spinlock_t amd_iommu_pd_lock
;
167 * Pointer to the device table which is shared by all AMD IOMMUs
168 * it is indexed by the PCI device id or the HT unit id and contains
169 * information about the domain the device belongs to as well as the
170 * page table root pointer.
172 struct dev_table_entry
*amd_iommu_dev_table
;
175 * The alias table is a driver specific data structure which contains the
176 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
177 * More than one device can share the same requestor id.
179 u16
*amd_iommu_alias_table
;
182 * The rlookup table is used to find the IOMMU which is responsible
183 * for a specific device. It is also indexed by the PCI device id.
185 struct amd_iommu
**amd_iommu_rlookup_table
;
188 * This table is used to find the irq remapping table for a given device id
191 struct irq_remap_table
**irq_lookup_table
;
194 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
195 * to know which ones are already in use.
197 unsigned long *amd_iommu_pd_alloc_bitmap
;
199 static u32 dev_table_size
; /* size of the device table */
200 static u32 alias_table_size
; /* size of the alias table */
201 static u32 rlookup_table_size
; /* size if the rlookup table */
203 enum iommu_init_state
{
216 /* Early ioapic and hpet maps from kernel command line */
217 #define EARLY_MAP_SIZE 4
218 static struct devid_map __initdata early_ioapic_map
[EARLY_MAP_SIZE
];
219 static struct devid_map __initdata early_hpet_map
[EARLY_MAP_SIZE
];
220 static int __initdata early_ioapic_map_size
;
221 static int __initdata early_hpet_map_size
;
222 static bool __initdata cmdline_maps
;
224 static enum iommu_init_state init_state
= IOMMU_START_STATE
;
226 static int amd_iommu_enable_interrupts(void);
227 static int __init
iommu_go_to_state(enum iommu_init_state state
);
229 static inline void update_last_devid(u16 devid
)
231 if (devid
> amd_iommu_last_bdf
)
232 amd_iommu_last_bdf
= devid
;
235 static inline unsigned long tbl_size(int entry_size
)
237 unsigned shift
= PAGE_SHIFT
+
238 get_order(((int)amd_iommu_last_bdf
+ 1) * entry_size
);
243 /* Access to l1 and l2 indexed register spaces */
245 static u32
iommu_read_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
)
249 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
250 pci_read_config_dword(iommu
->dev
, 0xfc, &val
);
254 static void iommu_write_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
, u32 val
)
256 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16 | 1 << 31));
257 pci_write_config_dword(iommu
->dev
, 0xfc, val
);
258 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
261 static u32
iommu_read_l2(struct amd_iommu
*iommu
, u8 address
)
265 pci_write_config_dword(iommu
->dev
, 0xf0, address
);
266 pci_read_config_dword(iommu
->dev
, 0xf4, &val
);
270 static void iommu_write_l2(struct amd_iommu
*iommu
, u8 address
, u32 val
)
272 pci_write_config_dword(iommu
->dev
, 0xf0, (address
| 1 << 8));
273 pci_write_config_dword(iommu
->dev
, 0xf4, val
);
276 /****************************************************************************
278 * AMD IOMMU MMIO register space handling functions
280 * These functions are used to program the IOMMU device registers in
281 * MMIO space required for that driver.
283 ****************************************************************************/
286 * This function set the exclusion range in the IOMMU. DMA accesses to the
287 * exclusion range are passed through untranslated
289 static void iommu_set_exclusion_range(struct amd_iommu
*iommu
)
291 u64 start
= iommu
->exclusion_start
& PAGE_MASK
;
292 u64 limit
= (start
+ iommu
->exclusion_length
) & PAGE_MASK
;
295 if (!iommu
->exclusion_start
)
298 entry
= start
| MMIO_EXCL_ENABLE_MASK
;
299 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_BASE_OFFSET
,
300 &entry
, sizeof(entry
));
303 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_LIMIT_OFFSET
,
304 &entry
, sizeof(entry
));
307 /* Programs the physical address of the device table into the IOMMU hardware */
308 static void iommu_set_device_table(struct amd_iommu
*iommu
)
312 BUG_ON(iommu
->mmio_base
== NULL
);
314 entry
= virt_to_phys(amd_iommu_dev_table
);
315 entry
|= (dev_table_size
>> 12) - 1;
316 memcpy_toio(iommu
->mmio_base
+ MMIO_DEV_TABLE_OFFSET
,
317 &entry
, sizeof(entry
));
320 /* Generic functions to enable/disable certain features of the IOMMU. */
321 static void iommu_feature_enable(struct amd_iommu
*iommu
, u8 bit
)
325 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
327 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
330 static void iommu_feature_disable(struct amd_iommu
*iommu
, u8 bit
)
334 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
336 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
339 static void iommu_set_inv_tlb_timeout(struct amd_iommu
*iommu
, int timeout
)
343 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
344 ctrl
&= ~CTRL_INV_TO_MASK
;
345 ctrl
|= (timeout
<< CONTROL_INV_TIMEOUT
) & CTRL_INV_TO_MASK
;
346 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
349 /* Function to enable the hardware */
350 static void iommu_enable(struct amd_iommu
*iommu
)
352 iommu_feature_enable(iommu
, CONTROL_IOMMU_EN
);
355 static void iommu_disable(struct amd_iommu
*iommu
)
357 /* Disable command buffer */
358 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
360 /* Disable event logging and event interrupts */
361 iommu_feature_disable(iommu
, CONTROL_EVT_INT_EN
);
362 iommu_feature_disable(iommu
, CONTROL_EVT_LOG_EN
);
364 /* Disable IOMMU hardware itself */
365 iommu_feature_disable(iommu
, CONTROL_IOMMU_EN
);
369 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
370 * the system has one.
372 static u8 __iomem
* __init
iommu_map_mmio_space(u64 address
, u64 end
)
374 if (!request_mem_region(address
, end
, "amd_iommu")) {
375 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
377 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
381 return (u8 __iomem
*)ioremap_nocache(address
, end
);
384 static void __init
iommu_unmap_mmio_space(struct amd_iommu
*iommu
)
386 if (iommu
->mmio_base
)
387 iounmap(iommu
->mmio_base
);
388 release_mem_region(iommu
->mmio_phys
, iommu
->mmio_phys_end
);
391 /****************************************************************************
393 * The functions below belong to the first pass of AMD IOMMU ACPI table
394 * parsing. In this pass we try to find out the highest device id this
395 * code has to handle. Upon this information the size of the shared data
396 * structures is determined later.
398 ****************************************************************************/
401 * This function calculates the length of a given IVHD entry
403 static inline int ivhd_entry_length(u8
*ivhd
)
405 return 0x04 << (*ivhd
>> 6);
409 * This function reads the last device id the IOMMU has to handle from the PCI
410 * capability header for this IOMMU
412 static int __init
find_last_devid_on_pci(int bus
, int dev
, int fn
, int cap_ptr
)
416 cap
= read_pci_config(bus
, dev
, fn
, cap_ptr
+MMIO_RANGE_OFFSET
);
417 update_last_devid(PCI_DEVID(MMIO_GET_BUS(cap
), MMIO_GET_LD(cap
)));
423 * After reading the highest device id from the IOMMU PCI capability header
424 * this function looks if there is a higher device id defined in the ACPI table
426 static int __init
find_last_devid_from_ivhd(struct ivhd_header
*h
)
428 u8
*p
= (void *)h
, *end
= (void *)h
;
429 struct ivhd_entry
*dev
;
434 find_last_devid_on_pci(PCI_BUS_NUM(h
->devid
),
440 dev
= (struct ivhd_entry
*)p
;
442 case IVHD_DEV_SELECT
:
443 case IVHD_DEV_RANGE_END
:
445 case IVHD_DEV_EXT_SELECT
:
446 /* all the above subfield types refer to device ids */
447 update_last_devid(dev
->devid
);
452 p
+= ivhd_entry_length(p
);
461 * Iterate over all IVHD entries in the ACPI table and find the highest device
462 * id which we need to handle. This is the first of three functions which parse
463 * the ACPI table. So we check the checksum here.
465 static int __init
find_last_devid_acpi(struct acpi_table_header
*table
)
468 u8 checksum
= 0, *p
= (u8
*)table
, *end
= (u8
*)table
;
469 struct ivhd_header
*h
;
472 * Validate checksum here so we don't need to do it when
473 * we actually parse the table
475 for (i
= 0; i
< table
->length
; ++i
)
478 /* ACPI table corrupt */
481 p
+= IVRS_HEADER_LENGTH
;
483 end
+= table
->length
;
485 h
= (struct ivhd_header
*)p
;
488 find_last_devid_from_ivhd(h
);
500 /****************************************************************************
502 * The following functions belong to the code path which parses the ACPI table
503 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
504 * data structures, initialize the device/alias/rlookup table and also
505 * basically initialize the hardware.
507 ****************************************************************************/
510 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
511 * write commands to that buffer later and the IOMMU will execute them
514 static u8
* __init
alloc_command_buffer(struct amd_iommu
*iommu
)
516 u8
*cmd_buf
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
517 get_order(CMD_BUFFER_SIZE
));
522 iommu
->cmd_buf_size
= CMD_BUFFER_SIZE
| CMD_BUFFER_UNINITIALIZED
;
528 * This function resets the command buffer if the IOMMU stopped fetching
531 void amd_iommu_reset_cmd_buffer(struct amd_iommu
*iommu
)
533 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
535 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_HEAD_OFFSET
);
536 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_TAIL_OFFSET
);
538 iommu_feature_enable(iommu
, CONTROL_CMDBUF_EN
);
542 * This function writes the command buffer address to the hardware and
545 static void iommu_enable_command_buffer(struct amd_iommu
*iommu
)
549 BUG_ON(iommu
->cmd_buf
== NULL
);
551 entry
= (u64
)virt_to_phys(iommu
->cmd_buf
);
552 entry
|= MMIO_CMD_SIZE_512
;
554 memcpy_toio(iommu
->mmio_base
+ MMIO_CMD_BUF_OFFSET
,
555 &entry
, sizeof(entry
));
557 amd_iommu_reset_cmd_buffer(iommu
);
558 iommu
->cmd_buf_size
&= ~(CMD_BUFFER_UNINITIALIZED
);
561 static void __init
free_command_buffer(struct amd_iommu
*iommu
)
563 free_pages((unsigned long)iommu
->cmd_buf
,
564 get_order(iommu
->cmd_buf_size
& ~(CMD_BUFFER_UNINITIALIZED
)));
567 /* allocates the memory where the IOMMU will log its events to */
568 static u8
* __init
alloc_event_buffer(struct amd_iommu
*iommu
)
570 iommu
->evt_buf
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
571 get_order(EVT_BUFFER_SIZE
));
573 if (iommu
->evt_buf
== NULL
)
576 iommu
->evt_buf_size
= EVT_BUFFER_SIZE
;
578 return iommu
->evt_buf
;
581 static void iommu_enable_event_buffer(struct amd_iommu
*iommu
)
585 BUG_ON(iommu
->evt_buf
== NULL
);
587 entry
= (u64
)virt_to_phys(iommu
->evt_buf
) | EVT_LEN_MASK
;
589 memcpy_toio(iommu
->mmio_base
+ MMIO_EVT_BUF_OFFSET
,
590 &entry
, sizeof(entry
));
592 /* set head and tail to zero manually */
593 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_HEAD_OFFSET
);
594 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_TAIL_OFFSET
);
596 iommu_feature_enable(iommu
, CONTROL_EVT_LOG_EN
);
599 static void __init
free_event_buffer(struct amd_iommu
*iommu
)
601 free_pages((unsigned long)iommu
->evt_buf
, get_order(EVT_BUFFER_SIZE
));
604 /* allocates the memory where the IOMMU will log its events to */
605 static u8
* __init
alloc_ppr_log(struct amd_iommu
*iommu
)
607 iommu
->ppr_log
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
608 get_order(PPR_LOG_SIZE
));
610 if (iommu
->ppr_log
== NULL
)
613 return iommu
->ppr_log
;
616 static void iommu_enable_ppr_log(struct amd_iommu
*iommu
)
620 if (iommu
->ppr_log
== NULL
)
623 entry
= (u64
)virt_to_phys(iommu
->ppr_log
) | PPR_LOG_SIZE_512
;
625 memcpy_toio(iommu
->mmio_base
+ MMIO_PPR_LOG_OFFSET
,
626 &entry
, sizeof(entry
));
628 /* set head and tail to zero manually */
629 writel(0x00, iommu
->mmio_base
+ MMIO_PPR_HEAD_OFFSET
);
630 writel(0x00, iommu
->mmio_base
+ MMIO_PPR_TAIL_OFFSET
);
632 iommu_feature_enable(iommu
, CONTROL_PPFLOG_EN
);
633 iommu_feature_enable(iommu
, CONTROL_PPR_EN
);
636 static void __init
free_ppr_log(struct amd_iommu
*iommu
)
638 if (iommu
->ppr_log
== NULL
)
641 free_pages((unsigned long)iommu
->ppr_log
, get_order(PPR_LOG_SIZE
));
644 static void iommu_enable_gt(struct amd_iommu
*iommu
)
646 if (!iommu_feature(iommu
, FEATURE_GT
))
649 iommu_feature_enable(iommu
, CONTROL_GT_EN
);
652 /* sets a specific bit in the device table entry. */
653 static void set_dev_entry_bit(u16 devid
, u8 bit
)
655 int i
= (bit
>> 6) & 0x03;
656 int _bit
= bit
& 0x3f;
658 amd_iommu_dev_table
[devid
].data
[i
] |= (1UL << _bit
);
661 static int get_dev_entry_bit(u16 devid
, u8 bit
)
663 int i
= (bit
>> 6) & 0x03;
664 int _bit
= bit
& 0x3f;
666 return (amd_iommu_dev_table
[devid
].data
[i
] & (1UL << _bit
)) >> _bit
;
670 void amd_iommu_apply_erratum_63(u16 devid
)
674 sysmgt
= get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
) |
675 (get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
) << 1);
678 set_dev_entry_bit(devid
, DEV_ENTRY_IW
);
681 /* Writes the specific IOMMU for a device into the rlookup table */
682 static void __init
set_iommu_for_device(struct amd_iommu
*iommu
, u16 devid
)
684 amd_iommu_rlookup_table
[devid
] = iommu
;
688 * This function takes the device specific flags read from the ACPI
689 * table and sets up the device table entry with that information
691 static void __init
set_dev_entry_from_acpi(struct amd_iommu
*iommu
,
692 u16 devid
, u32 flags
, u32 ext_flags
)
694 if (flags
& ACPI_DEVFLAG_INITPASS
)
695 set_dev_entry_bit(devid
, DEV_ENTRY_INIT_PASS
);
696 if (flags
& ACPI_DEVFLAG_EXTINT
)
697 set_dev_entry_bit(devid
, DEV_ENTRY_EINT_PASS
);
698 if (flags
& ACPI_DEVFLAG_NMI
)
699 set_dev_entry_bit(devid
, DEV_ENTRY_NMI_PASS
);
700 if (flags
& ACPI_DEVFLAG_SYSMGT1
)
701 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
);
702 if (flags
& ACPI_DEVFLAG_SYSMGT2
)
703 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
);
704 if (flags
& ACPI_DEVFLAG_LINT0
)
705 set_dev_entry_bit(devid
, DEV_ENTRY_LINT0_PASS
);
706 if (flags
& ACPI_DEVFLAG_LINT1
)
707 set_dev_entry_bit(devid
, DEV_ENTRY_LINT1_PASS
);
709 amd_iommu_apply_erratum_63(devid
);
711 set_iommu_for_device(iommu
, devid
);
714 static int __init
add_special_device(u8 type
, u8 id
, u16 devid
, bool cmd_line
)
716 struct devid_map
*entry
;
717 struct list_head
*list
;
719 if (type
== IVHD_SPECIAL_IOAPIC
)
721 else if (type
== IVHD_SPECIAL_HPET
)
726 list_for_each_entry(entry
, list
, list
) {
727 if (!(entry
->id
== id
&& entry
->cmd_line
))
730 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
731 type
== IVHD_SPECIAL_IOAPIC
? "IOAPIC" : "HPET", id
);
736 entry
= kzalloc(sizeof(*entry
), GFP_KERNEL
);
741 entry
->devid
= devid
;
742 entry
->cmd_line
= cmd_line
;
744 list_add_tail(&entry
->list
, list
);
749 static int __init
add_early_maps(void)
753 for (i
= 0; i
< early_ioapic_map_size
; ++i
) {
754 ret
= add_special_device(IVHD_SPECIAL_IOAPIC
,
755 early_ioapic_map
[i
].id
,
756 early_ioapic_map
[i
].devid
,
757 early_ioapic_map
[i
].cmd_line
);
762 for (i
= 0; i
< early_hpet_map_size
; ++i
) {
763 ret
= add_special_device(IVHD_SPECIAL_HPET
,
764 early_hpet_map
[i
].id
,
765 early_hpet_map
[i
].devid
,
766 early_hpet_map
[i
].cmd_line
);
775 * Reads the device exclusion range from ACPI and initializes the IOMMU with
778 static void __init
set_device_exclusion_range(u16 devid
, struct ivmd_header
*m
)
780 struct amd_iommu
*iommu
= amd_iommu_rlookup_table
[devid
];
782 if (!(m
->flags
& IVMD_FLAG_EXCL_RANGE
))
787 * We only can configure exclusion ranges per IOMMU, not
788 * per device. But we can enable the exclusion range per
789 * device. This is done here
791 set_dev_entry_bit(devid
, DEV_ENTRY_EX
);
792 iommu
->exclusion_start
= m
->range_start
;
793 iommu
->exclusion_length
= m
->range_length
;
798 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
799 * initializes the hardware and our data structures with it.
801 static int __init
init_iommu_from_acpi(struct amd_iommu
*iommu
,
802 struct ivhd_header
*h
)
805 u8
*end
= p
, flags
= 0;
806 u16 devid
= 0, devid_start
= 0, devid_to
= 0;
807 u32 dev_i
, ext_flags
= 0;
809 struct ivhd_entry
*e
;
813 ret
= add_early_maps();
818 * First save the recommended feature enable bits from ACPI
820 iommu
->acpi_flags
= h
->flags
;
823 * Done. Now parse the device entries
825 p
+= sizeof(struct ivhd_header
);
830 e
= (struct ivhd_entry
*)p
;
834 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
835 " last device %02x:%02x.%x flags: %02x\n",
836 PCI_BUS_NUM(iommu
->first_device
),
837 PCI_SLOT(iommu
->first_device
),
838 PCI_FUNC(iommu
->first_device
),
839 PCI_BUS_NUM(iommu
->last_device
),
840 PCI_SLOT(iommu
->last_device
),
841 PCI_FUNC(iommu
->last_device
),
844 for (dev_i
= iommu
->first_device
;
845 dev_i
<= iommu
->last_device
; ++dev_i
)
846 set_dev_entry_from_acpi(iommu
, dev_i
,
849 case IVHD_DEV_SELECT
:
851 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
853 PCI_BUS_NUM(e
->devid
),
859 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
861 case IVHD_DEV_SELECT_RANGE_START
:
863 DUMP_printk(" DEV_SELECT_RANGE_START\t "
864 "devid: %02x:%02x.%x flags: %02x\n",
865 PCI_BUS_NUM(e
->devid
),
870 devid_start
= e
->devid
;
877 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
878 "flags: %02x devid_to: %02x:%02x.%x\n",
879 PCI_BUS_NUM(e
->devid
),
883 PCI_BUS_NUM(e
->ext
>> 8),
884 PCI_SLOT(e
->ext
>> 8),
885 PCI_FUNC(e
->ext
>> 8));
888 devid_to
= e
->ext
>> 8;
889 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
890 set_dev_entry_from_acpi(iommu
, devid_to
, e
->flags
, 0);
891 amd_iommu_alias_table
[devid
] = devid_to
;
893 case IVHD_DEV_ALIAS_RANGE
:
895 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
896 "devid: %02x:%02x.%x flags: %02x "
897 "devid_to: %02x:%02x.%x\n",
898 PCI_BUS_NUM(e
->devid
),
902 PCI_BUS_NUM(e
->ext
>> 8),
903 PCI_SLOT(e
->ext
>> 8),
904 PCI_FUNC(e
->ext
>> 8));
906 devid_start
= e
->devid
;
908 devid_to
= e
->ext
>> 8;
912 case IVHD_DEV_EXT_SELECT
:
914 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
915 "flags: %02x ext: %08x\n",
916 PCI_BUS_NUM(e
->devid
),
922 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
,
925 case IVHD_DEV_EXT_SELECT_RANGE
:
927 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
928 "%02x:%02x.%x flags: %02x ext: %08x\n",
929 PCI_BUS_NUM(e
->devid
),
934 devid_start
= e
->devid
;
939 case IVHD_DEV_RANGE_END
:
941 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
942 PCI_BUS_NUM(e
->devid
),
947 for (dev_i
= devid_start
; dev_i
<= devid
; ++dev_i
) {
949 amd_iommu_alias_table
[dev_i
] = devid_to
;
950 set_dev_entry_from_acpi(iommu
,
951 devid_to
, flags
, ext_flags
);
953 set_dev_entry_from_acpi(iommu
, dev_i
,
957 case IVHD_DEV_SPECIAL
: {
963 handle
= e
->ext
& 0xff;
964 devid
= (e
->ext
>> 8) & 0xffff;
965 type
= (e
->ext
>> 24) & 0xff;
967 if (type
== IVHD_SPECIAL_IOAPIC
)
969 else if (type
== IVHD_SPECIAL_HPET
)
974 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
980 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
981 ret
= add_special_device(type
, handle
, devid
, false);
990 p
+= ivhd_entry_length(p
);
996 /* Initializes the device->iommu mapping for the driver */
997 static int __init
init_iommu_devices(struct amd_iommu
*iommu
)
1001 for (i
= iommu
->first_device
; i
<= iommu
->last_device
; ++i
)
1002 set_iommu_for_device(iommu
, i
);
1007 static void __init
free_iommu_one(struct amd_iommu
*iommu
)
1009 free_command_buffer(iommu
);
1010 free_event_buffer(iommu
);
1011 free_ppr_log(iommu
);
1012 iommu_unmap_mmio_space(iommu
);
1015 static void __init
free_iommu_all(void)
1017 struct amd_iommu
*iommu
, *next
;
1019 for_each_iommu_safe(iommu
, next
) {
1020 list_del(&iommu
->list
);
1021 free_iommu_one(iommu
);
1027 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1029 * BIOS should disable L2B micellaneous clock gating by setting
1030 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1032 static void amd_iommu_erratum_746_workaround(struct amd_iommu
*iommu
)
1036 if ((boot_cpu_data
.x86
!= 0x15) ||
1037 (boot_cpu_data
.x86_model
< 0x10) ||
1038 (boot_cpu_data
.x86_model
> 0x1f))
1041 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
1042 pci_read_config_dword(iommu
->dev
, 0xf4, &value
);
1047 /* Select NB indirect register 0x90 and enable writing */
1048 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90 | (1 << 8));
1050 pci_write_config_dword(iommu
->dev
, 0xf4, value
| 0x4);
1051 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1052 dev_name(&iommu
->dev
->dev
));
1054 /* Clear the enable writing bit */
1055 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
1059 * This function clues the initialization function for one IOMMU
1060 * together and also allocates the command buffer and programs the
1061 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1063 static int __init
init_iommu_one(struct amd_iommu
*iommu
, struct ivhd_header
*h
)
1067 spin_lock_init(&iommu
->lock
);
1069 /* Add IOMMU to internal data structures */
1070 list_add_tail(&iommu
->list
, &amd_iommu_list
);
1071 iommu
->index
= amd_iommus_present
++;
1073 if (unlikely(iommu
->index
>= MAX_IOMMUS
)) {
1074 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1078 /* Index is fine - add IOMMU to the array */
1079 amd_iommus
[iommu
->index
] = iommu
;
1082 * Copy data from ACPI table entry to the iommu struct
1084 iommu
->devid
= h
->devid
;
1085 iommu
->cap_ptr
= h
->cap_ptr
;
1086 iommu
->pci_seg
= h
->pci_seg
;
1087 iommu
->mmio_phys
= h
->mmio_phys
;
1089 /* Check if IVHD EFR contains proper max banks/counters */
1090 if ((h
->efr
!= 0) &&
1091 ((h
->efr
& (0xF << 13)) != 0) &&
1092 ((h
->efr
& (0x3F << 17)) != 0)) {
1093 iommu
->mmio_phys_end
= MMIO_REG_END_OFFSET
;
1095 iommu
->mmio_phys_end
= MMIO_CNTR_CONF_OFFSET
;
1098 iommu
->mmio_base
= iommu_map_mmio_space(iommu
->mmio_phys
,
1099 iommu
->mmio_phys_end
);
1100 if (!iommu
->mmio_base
)
1103 iommu
->cmd_buf
= alloc_command_buffer(iommu
);
1104 if (!iommu
->cmd_buf
)
1107 iommu
->evt_buf
= alloc_event_buffer(iommu
);
1108 if (!iommu
->evt_buf
)
1111 iommu
->int_enabled
= false;
1113 ret
= init_iommu_from_acpi(iommu
, h
);
1118 * Make sure IOMMU is not considered to translate itself. The IVRS
1119 * table tells us so, but this is a lie!
1121 amd_iommu_rlookup_table
[iommu
->devid
] = NULL
;
1123 init_iommu_devices(iommu
);
1129 * Iterates over all IOMMU entries in the ACPI table, allocates the
1130 * IOMMU structure and initializes it with init_iommu_one()
1132 static int __init
init_iommu_all(struct acpi_table_header
*table
)
1134 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1135 struct ivhd_header
*h
;
1136 struct amd_iommu
*iommu
;
1139 end
+= table
->length
;
1140 p
+= IVRS_HEADER_LENGTH
;
1143 h
= (struct ivhd_header
*)p
;
1145 case ACPI_IVHD_TYPE
:
1147 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1148 "seg: %d flags: %01x info %04x\n",
1149 PCI_BUS_NUM(h
->devid
), PCI_SLOT(h
->devid
),
1150 PCI_FUNC(h
->devid
), h
->cap_ptr
,
1151 h
->pci_seg
, h
->flags
, h
->info
);
1152 DUMP_printk(" mmio-addr: %016llx\n",
1155 iommu
= kzalloc(sizeof(struct amd_iommu
), GFP_KERNEL
);
1159 ret
= init_iommu_one(iommu
, h
);
1175 static void init_iommu_perf_ctr(struct amd_iommu
*iommu
)
1177 u64 val
= 0xabcd, val2
= 0;
1179 if (!iommu_feature(iommu
, FEATURE_PC
))
1182 amd_iommu_pc_present
= true;
1184 /* Check if the performance counters can be written to */
1185 if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val
, true)) ||
1186 (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2
, false)) ||
1188 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1189 amd_iommu_pc_present
= false;
1193 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1195 val
= readl(iommu
->mmio_base
+ MMIO_CNTR_CONF_OFFSET
);
1196 iommu
->max_banks
= (u8
) ((val
>> 12) & 0x3f);
1197 iommu
->max_counters
= (u8
) ((val
>> 7) & 0xf);
1201 static int iommu_init_pci(struct amd_iommu
*iommu
)
1203 int cap_ptr
= iommu
->cap_ptr
;
1204 u32 range
, misc
, low
, high
;
1206 iommu
->dev
= pci_get_bus_and_slot(PCI_BUS_NUM(iommu
->devid
),
1207 iommu
->devid
& 0xff);
1211 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_CAP_HDR_OFFSET
,
1213 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_RANGE_OFFSET
,
1215 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_MISC_OFFSET
,
1218 iommu
->first_device
= PCI_DEVID(MMIO_GET_BUS(range
),
1219 MMIO_GET_FD(range
));
1220 iommu
->last_device
= PCI_DEVID(MMIO_GET_BUS(range
),
1221 MMIO_GET_LD(range
));
1223 if (!(iommu
->cap
& (1 << IOMMU_CAP_IOTLB
)))
1224 amd_iommu_iotlb_sup
= false;
1226 /* read extended feature bits */
1227 low
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
);
1228 high
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
+ 4);
1230 iommu
->features
= ((u64
)high
<< 32) | low
;
1232 if (iommu_feature(iommu
, FEATURE_GT
)) {
1237 pasmax
= iommu
->features
& FEATURE_PASID_MASK
;
1238 pasmax
>>= FEATURE_PASID_SHIFT
;
1239 max_pasid
= (1 << (pasmax
+ 1)) - 1;
1241 amd_iommu_max_pasid
= min(amd_iommu_max_pasid
, max_pasid
);
1243 BUG_ON(amd_iommu_max_pasid
& ~PASID_MASK
);
1245 glxval
= iommu
->features
& FEATURE_GLXVAL_MASK
;
1246 glxval
>>= FEATURE_GLXVAL_SHIFT
;
1248 if (amd_iommu_max_glx_val
== -1)
1249 amd_iommu_max_glx_val
= glxval
;
1251 amd_iommu_max_glx_val
= min(amd_iommu_max_glx_val
, glxval
);
1254 if (iommu_feature(iommu
, FEATURE_GT
) &&
1255 iommu_feature(iommu
, FEATURE_PPR
)) {
1256 iommu
->is_iommu_v2
= true;
1257 amd_iommu_v2_present
= true;
1260 if (iommu_feature(iommu
, FEATURE_PPR
)) {
1261 iommu
->ppr_log
= alloc_ppr_log(iommu
);
1262 if (!iommu
->ppr_log
)
1266 if (iommu
->cap
& (1UL << IOMMU_CAP_NPCACHE
))
1267 amd_iommu_np_cache
= true;
1269 init_iommu_perf_ctr(iommu
);
1271 if (is_rd890_iommu(iommu
->dev
)) {
1274 iommu
->root_pdev
= pci_get_bus_and_slot(iommu
->dev
->bus
->number
,
1278 * Some rd890 systems may not be fully reconfigured by the
1279 * BIOS, so it's necessary for us to store this information so
1280 * it can be reprogrammed on resume
1282 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1283 &iommu
->stored_addr_lo
);
1284 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1285 &iommu
->stored_addr_hi
);
1287 /* Low bit locks writes to configuration space */
1288 iommu
->stored_addr_lo
&= ~1;
1290 for (i
= 0; i
< 6; i
++)
1291 for (j
= 0; j
< 0x12; j
++)
1292 iommu
->stored_l1
[i
][j
] = iommu_read_l1(iommu
, i
, j
);
1294 for (i
= 0; i
< 0x83; i
++)
1295 iommu
->stored_l2
[i
] = iommu_read_l2(iommu
, i
);
1298 amd_iommu_erratum_746_workaround(iommu
);
1300 return pci_enable_device(iommu
->dev
);
1303 static void print_iommu_info(void)
1305 static const char * const feat_str
[] = {
1306 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1307 "IA", "GA", "HE", "PC"
1309 struct amd_iommu
*iommu
;
1311 for_each_iommu(iommu
) {
1314 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1315 dev_name(&iommu
->dev
->dev
), iommu
->cap_ptr
);
1317 if (iommu
->cap
& (1 << IOMMU_CAP_EFR
)) {
1318 pr_info("AMD-Vi: Extended features: ");
1319 for (i
= 0; i
< ARRAY_SIZE(feat_str
); ++i
) {
1320 if (iommu_feature(iommu
, (1ULL << i
)))
1321 pr_cont(" %s", feat_str
[i
]);
1326 if (irq_remapping_enabled
)
1327 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1330 static int __init
amd_iommu_init_pci(void)
1332 struct amd_iommu
*iommu
;
1335 for_each_iommu(iommu
) {
1336 ret
= iommu_init_pci(iommu
);
1341 ret
= amd_iommu_init_devices();
1348 /****************************************************************************
1350 * The following functions initialize the MSI interrupts for all IOMMUs
1351 * in the system. It's a bit challenging because there could be multiple
1352 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1355 ****************************************************************************/
1357 static int iommu_setup_msi(struct amd_iommu
*iommu
)
1361 r
= pci_enable_msi(iommu
->dev
);
1365 r
= request_threaded_irq(iommu
->dev
->irq
,
1366 amd_iommu_int_handler
,
1367 amd_iommu_int_thread
,
1372 pci_disable_msi(iommu
->dev
);
1376 iommu
->int_enabled
= true;
1381 static int iommu_init_msi(struct amd_iommu
*iommu
)
1385 if (iommu
->int_enabled
)
1388 if (iommu
->dev
->msi_cap
)
1389 ret
= iommu_setup_msi(iommu
);
1397 iommu_feature_enable(iommu
, CONTROL_EVT_INT_EN
);
1399 if (iommu
->ppr_log
!= NULL
)
1400 iommu_feature_enable(iommu
, CONTROL_PPFINT_EN
);
1405 /****************************************************************************
1407 * The next functions belong to the third pass of parsing the ACPI
1408 * table. In this last pass the memory mapping requirements are
1409 * gathered (like exclusion and unity mapping ranges).
1411 ****************************************************************************/
1413 static void __init
free_unity_maps(void)
1415 struct unity_map_entry
*entry
, *next
;
1417 list_for_each_entry_safe(entry
, next
, &amd_iommu_unity_map
, list
) {
1418 list_del(&entry
->list
);
1423 /* called when we find an exclusion range definition in ACPI */
1424 static int __init
init_exclusion_range(struct ivmd_header
*m
)
1429 case ACPI_IVMD_TYPE
:
1430 set_device_exclusion_range(m
->devid
, m
);
1432 case ACPI_IVMD_TYPE_ALL
:
1433 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1434 set_device_exclusion_range(i
, m
);
1436 case ACPI_IVMD_TYPE_RANGE
:
1437 for (i
= m
->devid
; i
<= m
->aux
; ++i
)
1438 set_device_exclusion_range(i
, m
);
1447 /* called for unity map ACPI definition */
1448 static int __init
init_unity_map_range(struct ivmd_header
*m
)
1450 struct unity_map_entry
*e
= NULL
;
1453 e
= kzalloc(sizeof(*e
), GFP_KERNEL
);
1461 case ACPI_IVMD_TYPE
:
1462 s
= "IVMD_TYPEi\t\t\t";
1463 e
->devid_start
= e
->devid_end
= m
->devid
;
1465 case ACPI_IVMD_TYPE_ALL
:
1466 s
= "IVMD_TYPE_ALL\t\t";
1468 e
->devid_end
= amd_iommu_last_bdf
;
1470 case ACPI_IVMD_TYPE_RANGE
:
1471 s
= "IVMD_TYPE_RANGE\t\t";
1472 e
->devid_start
= m
->devid
;
1473 e
->devid_end
= m
->aux
;
1476 e
->address_start
= PAGE_ALIGN(m
->range_start
);
1477 e
->address_end
= e
->address_start
+ PAGE_ALIGN(m
->range_length
);
1478 e
->prot
= m
->flags
>> 1;
1480 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1481 " range_start: %016llx range_end: %016llx flags: %x\n", s
,
1482 PCI_BUS_NUM(e
->devid_start
), PCI_SLOT(e
->devid_start
),
1483 PCI_FUNC(e
->devid_start
), PCI_BUS_NUM(e
->devid_end
),
1484 PCI_SLOT(e
->devid_end
), PCI_FUNC(e
->devid_end
),
1485 e
->address_start
, e
->address_end
, m
->flags
);
1487 list_add_tail(&e
->list
, &amd_iommu_unity_map
);
1492 /* iterates over all memory definitions we find in the ACPI table */
1493 static int __init
init_memory_definitions(struct acpi_table_header
*table
)
1495 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1496 struct ivmd_header
*m
;
1498 end
+= table
->length
;
1499 p
+= IVRS_HEADER_LENGTH
;
1502 m
= (struct ivmd_header
*)p
;
1503 if (m
->flags
& IVMD_FLAG_EXCL_RANGE
)
1504 init_exclusion_range(m
);
1505 else if (m
->flags
& IVMD_FLAG_UNITY_MAP
)
1506 init_unity_map_range(m
);
1515 * Init the device table to not allow DMA access for devices and
1516 * suppress all page faults
1518 static void init_device_table_dma(void)
1522 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1523 set_dev_entry_bit(devid
, DEV_ENTRY_VALID
);
1524 set_dev_entry_bit(devid
, DEV_ENTRY_TRANSLATION
);
1528 static void __init
uninit_device_table_dma(void)
1532 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1533 amd_iommu_dev_table
[devid
].data
[0] = 0ULL;
1534 amd_iommu_dev_table
[devid
].data
[1] = 0ULL;
1538 static void init_device_table(void)
1542 if (!amd_iommu_irq_remap
)
1545 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
)
1546 set_dev_entry_bit(devid
, DEV_ENTRY_IRQ_TBL_EN
);
1549 static void iommu_init_flags(struct amd_iommu
*iommu
)
1551 iommu
->acpi_flags
& IVHD_FLAG_HT_TUN_EN_MASK
?
1552 iommu_feature_enable(iommu
, CONTROL_HT_TUN_EN
) :
1553 iommu_feature_disable(iommu
, CONTROL_HT_TUN_EN
);
1555 iommu
->acpi_flags
& IVHD_FLAG_PASSPW_EN_MASK
?
1556 iommu_feature_enable(iommu
, CONTROL_PASSPW_EN
) :
1557 iommu_feature_disable(iommu
, CONTROL_PASSPW_EN
);
1559 iommu
->acpi_flags
& IVHD_FLAG_RESPASSPW_EN_MASK
?
1560 iommu_feature_enable(iommu
, CONTROL_RESPASSPW_EN
) :
1561 iommu_feature_disable(iommu
, CONTROL_RESPASSPW_EN
);
1563 iommu
->acpi_flags
& IVHD_FLAG_ISOC_EN_MASK
?
1564 iommu_feature_enable(iommu
, CONTROL_ISOC_EN
) :
1565 iommu_feature_disable(iommu
, CONTROL_ISOC_EN
);
1568 * make IOMMU memory accesses cache coherent
1570 iommu_feature_enable(iommu
, CONTROL_COHERENT_EN
);
1572 /* Set IOTLB invalidation timeout to 1s */
1573 iommu_set_inv_tlb_timeout(iommu
, CTRL_INV_TO_1S
);
1576 static void iommu_apply_resume_quirks(struct amd_iommu
*iommu
)
1579 u32 ioc_feature_control
;
1580 struct pci_dev
*pdev
= iommu
->root_pdev
;
1582 /* RD890 BIOSes may not have completely reconfigured the iommu */
1583 if (!is_rd890_iommu(iommu
->dev
) || !pdev
)
1587 * First, we need to ensure that the iommu is enabled. This is
1588 * controlled by a register in the northbridge
1591 /* Select Northbridge indirect register 0x75 and enable writing */
1592 pci_write_config_dword(pdev
, 0x60, 0x75 | (1 << 7));
1593 pci_read_config_dword(pdev
, 0x64, &ioc_feature_control
);
1595 /* Enable the iommu */
1596 if (!(ioc_feature_control
& 0x1))
1597 pci_write_config_dword(pdev
, 0x64, ioc_feature_control
| 1);
1599 /* Restore the iommu BAR */
1600 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1601 iommu
->stored_addr_lo
);
1602 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1603 iommu
->stored_addr_hi
);
1605 /* Restore the l1 indirect regs for each of the 6 l1s */
1606 for (i
= 0; i
< 6; i
++)
1607 for (j
= 0; j
< 0x12; j
++)
1608 iommu_write_l1(iommu
, i
, j
, iommu
->stored_l1
[i
][j
]);
1610 /* Restore the l2 indirect regs */
1611 for (i
= 0; i
< 0x83; i
++)
1612 iommu_write_l2(iommu
, i
, iommu
->stored_l2
[i
]);
1614 /* Lock PCI setup registers */
1615 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1616 iommu
->stored_addr_lo
| 1);
1620 * This function finally enables all IOMMUs found in the system after
1621 * they have been initialized
1623 static void early_enable_iommus(void)
1625 struct amd_iommu
*iommu
;
1627 for_each_iommu(iommu
) {
1628 iommu_disable(iommu
);
1629 iommu_init_flags(iommu
);
1630 iommu_set_device_table(iommu
);
1631 iommu_enable_command_buffer(iommu
);
1632 iommu_enable_event_buffer(iommu
);
1633 iommu_set_exclusion_range(iommu
);
1634 iommu_enable(iommu
);
1635 iommu_flush_all_caches(iommu
);
1639 static void enable_iommus_v2(void)
1641 struct amd_iommu
*iommu
;
1643 for_each_iommu(iommu
) {
1644 iommu_enable_ppr_log(iommu
);
1645 iommu_enable_gt(iommu
);
1649 static void enable_iommus(void)
1651 early_enable_iommus();
1656 static void disable_iommus(void)
1658 struct amd_iommu
*iommu
;
1660 for_each_iommu(iommu
)
1661 iommu_disable(iommu
);
1665 * Suspend/Resume support
1666 * disable suspend until real resume implemented
1669 static void amd_iommu_resume(void)
1671 struct amd_iommu
*iommu
;
1673 for_each_iommu(iommu
)
1674 iommu_apply_resume_quirks(iommu
);
1676 /* re-load the hardware */
1679 amd_iommu_enable_interrupts();
1682 static int amd_iommu_suspend(void)
1684 /* disable IOMMUs to go out of the way for BIOS */
1690 static struct syscore_ops amd_iommu_syscore_ops
= {
1691 .suspend
= amd_iommu_suspend
,
1692 .resume
= amd_iommu_resume
,
1695 static void __init
free_on_init_error(void)
1697 free_pages((unsigned long)irq_lookup_table
,
1698 get_order(rlookup_table_size
));
1700 if (amd_iommu_irq_cache
) {
1701 kmem_cache_destroy(amd_iommu_irq_cache
);
1702 amd_iommu_irq_cache
= NULL
;
1706 free_pages((unsigned long)amd_iommu_rlookup_table
,
1707 get_order(rlookup_table_size
));
1709 free_pages((unsigned long)amd_iommu_alias_table
,
1710 get_order(alias_table_size
));
1712 free_pages((unsigned long)amd_iommu_dev_table
,
1713 get_order(dev_table_size
));
1717 #ifdef CONFIG_GART_IOMMU
1719 * We failed to initialize the AMD IOMMU - try fallback to GART
1727 /* SB IOAPIC is always on this device in AMD systems */
1728 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1730 static bool __init
check_ioapic_information(void)
1732 const char *fw_bug
= FW_BUG
;
1733 bool ret
, has_sb_ioapic
;
1736 has_sb_ioapic
= false;
1740 * If we have map overrides on the kernel command line the
1741 * messages in this function might not describe firmware bugs
1742 * anymore - so be careful
1747 for (idx
= 0; idx
< nr_ioapics
; idx
++) {
1748 int devid
, id
= mpc_ioapic_id(idx
);
1750 devid
= get_ioapic_devid(id
);
1752 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1755 } else if (devid
== IOAPIC_SB_DEVID
) {
1756 has_sb_ioapic
= true;
1761 if (!has_sb_ioapic
) {
1763 * We expect the SB IOAPIC to be listed in the IVRS
1764 * table. The system timer is connected to the SB IOAPIC
1765 * and if we don't have it in the list the system will
1766 * panic at boot time. This situation usually happens
1767 * when the BIOS is buggy and provides us the wrong
1768 * device id for the IOAPIC in the system.
1770 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug
);
1774 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1779 static void __init
free_dma_resources(void)
1781 amd_iommu_uninit_devices();
1783 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap
,
1784 get_order(MAX_DOMAIN_ID
/8));
1790 * This is the hardware init function for AMD IOMMU in the system.
1791 * This function is called either from amd_iommu_init or from the interrupt
1792 * remapping setup code.
1794 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1797 * 1 pass) Find the highest PCI device id the driver has to handle.
1798 * Upon this information the size of the data structures is
1799 * determined that needs to be allocated.
1801 * 2 pass) Initialize the data structures just allocated with the
1802 * information in the ACPI table about available AMD IOMMUs
1803 * in the system. It also maps the PCI devices in the
1804 * system to specific IOMMUs
1806 * 3 pass) After the basic data structures are allocated and
1807 * initialized we update them with information about memory
1808 * remapping requirements parsed out of the ACPI table in
1811 * After everything is set up the IOMMUs are enabled and the necessary
1812 * hotplug and suspend notifiers are registered.
1814 static int __init
early_amd_iommu_init(void)
1816 struct acpi_table_header
*ivrs_base
;
1817 acpi_size ivrs_size
;
1821 if (!amd_iommu_detected
)
1824 status
= acpi_get_table_with_size("IVRS", 0, &ivrs_base
, &ivrs_size
);
1825 if (status
== AE_NOT_FOUND
)
1827 else if (ACPI_FAILURE(status
)) {
1828 const char *err
= acpi_format_exception(status
);
1829 pr_err("AMD-Vi: IVRS table error: %s\n", err
);
1834 * First parse ACPI tables to find the largest Bus/Dev/Func
1835 * we need to handle. Upon this information the shared data
1836 * structures for the IOMMUs in the system will be allocated
1838 ret
= find_last_devid_acpi(ivrs_base
);
1842 dev_table_size
= tbl_size(DEV_TABLE_ENTRY_SIZE
);
1843 alias_table_size
= tbl_size(ALIAS_TABLE_ENTRY_SIZE
);
1844 rlookup_table_size
= tbl_size(RLOOKUP_TABLE_ENTRY_SIZE
);
1846 /* Device table - directly used by all IOMMUs */
1848 amd_iommu_dev_table
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
1849 get_order(dev_table_size
));
1850 if (amd_iommu_dev_table
== NULL
)
1854 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1855 * IOMMU see for that device
1857 amd_iommu_alias_table
= (void *)__get_free_pages(GFP_KERNEL
,
1858 get_order(alias_table_size
));
1859 if (amd_iommu_alias_table
== NULL
)
1862 /* IOMMU rlookup table - find the IOMMU for a specific device */
1863 amd_iommu_rlookup_table
= (void *)__get_free_pages(
1864 GFP_KERNEL
| __GFP_ZERO
,
1865 get_order(rlookup_table_size
));
1866 if (amd_iommu_rlookup_table
== NULL
)
1869 amd_iommu_pd_alloc_bitmap
= (void *)__get_free_pages(
1870 GFP_KERNEL
| __GFP_ZERO
,
1871 get_order(MAX_DOMAIN_ID
/8));
1872 if (amd_iommu_pd_alloc_bitmap
== NULL
)
1876 * let all alias entries point to itself
1878 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1879 amd_iommu_alias_table
[i
] = i
;
1882 * never allocate domain 0 because its used as the non-allocated and
1883 * error value placeholder
1885 amd_iommu_pd_alloc_bitmap
[0] = 1;
1887 spin_lock_init(&amd_iommu_pd_lock
);
1890 * now the data structures are allocated and basically initialized
1891 * start the real acpi table scan
1893 ret
= init_iommu_all(ivrs_base
);
1897 if (amd_iommu_irq_remap
)
1898 amd_iommu_irq_remap
= check_ioapic_information();
1900 if (amd_iommu_irq_remap
) {
1902 * Interrupt remapping enabled, create kmem_cache for the
1906 amd_iommu_irq_cache
= kmem_cache_create("irq_remap_cache",
1907 MAX_IRQS_PER_TABLE
* sizeof(u32
),
1908 IRQ_TABLE_ALIGNMENT
,
1910 if (!amd_iommu_irq_cache
)
1913 irq_lookup_table
= (void *)__get_free_pages(
1914 GFP_KERNEL
| __GFP_ZERO
,
1915 get_order(rlookup_table_size
));
1916 if (!irq_lookup_table
)
1920 ret
= init_memory_definitions(ivrs_base
);
1924 /* init the device table */
1925 init_device_table();
1928 /* Don't leak any ACPI memory */
1929 early_acpi_os_unmap_memory((char __iomem
*)ivrs_base
, ivrs_size
);
1935 static int amd_iommu_enable_interrupts(void)
1937 struct amd_iommu
*iommu
;
1940 for_each_iommu(iommu
) {
1941 ret
= iommu_init_msi(iommu
);
1950 static bool detect_ivrs(void)
1952 struct acpi_table_header
*ivrs_base
;
1953 acpi_size ivrs_size
;
1956 status
= acpi_get_table_with_size("IVRS", 0, &ivrs_base
, &ivrs_size
);
1957 if (status
== AE_NOT_FOUND
)
1959 else if (ACPI_FAILURE(status
)) {
1960 const char *err
= acpi_format_exception(status
);
1961 pr_err("AMD-Vi: IVRS table error: %s\n", err
);
1965 early_acpi_os_unmap_memory((char __iomem
*)ivrs_base
, ivrs_size
);
1967 /* Make sure ACS will be enabled during PCI probe */
1970 if (!disable_irq_remap
)
1971 amd_iommu_irq_remap
= true;
1976 static int amd_iommu_init_dma(void)
1978 struct amd_iommu
*iommu
;
1981 if (iommu_pass_through
)
1982 ret
= amd_iommu_init_passthrough();
1984 ret
= amd_iommu_init_dma_ops();
1989 init_device_table_dma();
1991 for_each_iommu(iommu
)
1992 iommu_flush_all_caches(iommu
);
1994 amd_iommu_init_api();
1996 amd_iommu_init_notifier();
2001 /****************************************************************************
2003 * AMD IOMMU Initialization State Machine
2005 ****************************************************************************/
2007 static int __init
state_next(void)
2011 switch (init_state
) {
2012 case IOMMU_START_STATE
:
2013 if (!detect_ivrs()) {
2014 init_state
= IOMMU_NOT_FOUND
;
2017 init_state
= IOMMU_IVRS_DETECTED
;
2020 case IOMMU_IVRS_DETECTED
:
2021 ret
= early_amd_iommu_init();
2022 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_ACPI_FINISHED
;
2024 case IOMMU_ACPI_FINISHED
:
2025 early_enable_iommus();
2026 register_syscore_ops(&amd_iommu_syscore_ops
);
2027 x86_platform
.iommu_shutdown
= disable_iommus
;
2028 init_state
= IOMMU_ENABLED
;
2031 ret
= amd_iommu_init_pci();
2032 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_PCI_INIT
;
2035 case IOMMU_PCI_INIT
:
2036 ret
= amd_iommu_enable_interrupts();
2037 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_INTERRUPTS_EN
;
2039 case IOMMU_INTERRUPTS_EN
:
2040 ret
= amd_iommu_init_dma();
2041 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_DMA_OPS
;
2044 init_state
= IOMMU_INITIALIZED
;
2046 case IOMMU_INITIALIZED
:
2049 case IOMMU_NOT_FOUND
:
2050 case IOMMU_INIT_ERROR
:
2051 /* Error states => do nothing */
2062 static int __init
iommu_go_to_state(enum iommu_init_state state
)
2066 while (init_state
!= state
) {
2068 if (init_state
== IOMMU_NOT_FOUND
||
2069 init_state
== IOMMU_INIT_ERROR
)
2076 #ifdef CONFIG_IRQ_REMAP
2077 int __init
amd_iommu_prepare(void)
2079 return iommu_go_to_state(IOMMU_ACPI_FINISHED
);
2082 int __init
amd_iommu_supported(void)
2084 return amd_iommu_irq_remap
? 1 : 0;
2087 int __init
amd_iommu_enable(void)
2091 ret
= iommu_go_to_state(IOMMU_ENABLED
);
2095 irq_remapping_enabled
= 1;
2100 void amd_iommu_disable(void)
2102 amd_iommu_suspend();
2105 int amd_iommu_reenable(int mode
)
2112 int __init
amd_iommu_enable_faulting(void)
2114 /* We enable MSI later when PCI is initialized */
2120 * This is the core init function for AMD IOMMU hardware in the system.
2121 * This function is called from the generic x86 DMA layer initialization
2124 static int __init
amd_iommu_init(void)
2128 ret
= iommu_go_to_state(IOMMU_INITIALIZED
);
2130 free_dma_resources();
2131 if (!irq_remapping_enabled
) {
2133 free_on_init_error();
2135 struct amd_iommu
*iommu
;
2137 uninit_device_table_dma();
2138 for_each_iommu(iommu
)
2139 iommu_flush_all_caches(iommu
);
2146 /****************************************************************************
2148 * Early detect code. This code runs at IOMMU detection time in the DMA
2149 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2152 ****************************************************************************/
2153 int __init
amd_iommu_detect(void)
2157 if (no_iommu
|| (iommu_detected
&& !gart_iommu_aperture
))
2160 if (amd_iommu_disabled
)
2163 ret
= iommu_go_to_state(IOMMU_IVRS_DETECTED
);
2167 amd_iommu_detected
= true;
2169 x86_init
.iommu
.iommu_init
= amd_iommu_init
;
2174 /****************************************************************************
2176 * Parsing functions for the AMD IOMMU specific kernel command line
2179 ****************************************************************************/
2181 static int __init
parse_amd_iommu_dump(char *str
)
2183 amd_iommu_dump
= true;
2188 static int __init
parse_amd_iommu_options(char *str
)
2190 for (; *str
; ++str
) {
2191 if (strncmp(str
, "fullflush", 9) == 0)
2192 amd_iommu_unmap_flush
= true;
2193 if (strncmp(str
, "off", 3) == 0)
2194 amd_iommu_disabled
= true;
2195 if (strncmp(str
, "force_isolation", 15) == 0)
2196 amd_iommu_force_isolation
= true;
2202 static int __init
parse_ivrs_ioapic(char *str
)
2204 unsigned int bus
, dev
, fn
;
2208 ret
= sscanf(str
, "[%d]=%x:%x.%x", &id
, &bus
, &dev
, &fn
);
2211 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str
);
2215 if (early_ioapic_map_size
== EARLY_MAP_SIZE
) {
2216 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2221 devid
= ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2223 cmdline_maps
= true;
2224 i
= early_ioapic_map_size
++;
2225 early_ioapic_map
[i
].id
= id
;
2226 early_ioapic_map
[i
].devid
= devid
;
2227 early_ioapic_map
[i
].cmd_line
= true;
2232 static int __init
parse_ivrs_hpet(char *str
)
2234 unsigned int bus
, dev
, fn
;
2238 ret
= sscanf(str
, "[%d]=%x:%x.%x", &id
, &bus
, &dev
, &fn
);
2241 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str
);
2245 if (early_hpet_map_size
== EARLY_MAP_SIZE
) {
2246 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2251 devid
= ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2253 cmdline_maps
= true;
2254 i
= early_hpet_map_size
++;
2255 early_hpet_map
[i
].id
= id
;
2256 early_hpet_map
[i
].devid
= devid
;
2257 early_hpet_map
[i
].cmd_line
= true;
2262 __setup("amd_iommu_dump", parse_amd_iommu_dump
);
2263 __setup("amd_iommu=", parse_amd_iommu_options
);
2264 __setup("ivrs_ioapic", parse_ivrs_ioapic
);
2265 __setup("ivrs_hpet", parse_ivrs_hpet
);
2267 IOMMU_INIT_FINISH(amd_iommu_detect
,
2268 gart_iommu_hole_init
,
2272 bool amd_iommu_v2_supported(void)
2274 return amd_iommu_v2_present
;
2276 EXPORT_SYMBOL(amd_iommu_v2_supported
);
2278 /****************************************************************************
2280 * IOMMU EFR Performance Counter support functionality. This code allows
2281 * access to the IOMMU PC functionality.
2283 ****************************************************************************/
2285 u8
amd_iommu_pc_get_max_banks(u16 devid
)
2287 struct amd_iommu
*iommu
;
2290 /* locate the iommu governing the devid */
2291 iommu
= amd_iommu_rlookup_table
[devid
];
2293 ret
= iommu
->max_banks
;
2297 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks
);
2299 bool amd_iommu_pc_supported(void)
2301 return amd_iommu_pc_present
;
2303 EXPORT_SYMBOL(amd_iommu_pc_supported
);
2305 u8
amd_iommu_pc_get_max_counters(u16 devid
)
2307 struct amd_iommu
*iommu
;
2310 /* locate the iommu governing the devid */
2311 iommu
= amd_iommu_rlookup_table
[devid
];
2313 ret
= iommu
->max_counters
;
2317 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters
);
2319 int amd_iommu_pc_get_set_reg_val(u16 devid
, u8 bank
, u8 cntr
, u8 fxn
,
2320 u64
*value
, bool is_write
)
2322 struct amd_iommu
*iommu
;
2326 /* Make sure the IOMMU PC resource is available */
2327 if (!amd_iommu_pc_present
)
2330 /* Locate the iommu associated with the device ID */
2331 iommu
= amd_iommu_rlookup_table
[devid
];
2333 /* Check for valid iommu and pc register indexing */
2334 if (WARN_ON((iommu
== NULL
) || (fxn
> 0x28) || (fxn
& 7)))
2337 offset
= (u32
)(((0x40|bank
) << 12) | (cntr
<< 8) | fxn
);
2339 /* Limit the offset to the hw defined mmio region aperture */
2340 max_offset_lim
= (u32
)(((0x40|iommu
->max_banks
) << 12) |
2341 (iommu
->max_counters
<< 8) | 0x28);
2342 if ((offset
< MMIO_CNTR_REG_OFFSET
) ||
2343 (offset
> max_offset_lim
))
2347 writel((u32
)*value
, iommu
->mmio_base
+ offset
);
2348 writel((*value
>> 32), iommu
->mmio_base
+ offset
+ 4);
2350 *value
= readl(iommu
->mmio_base
+ offset
+ 4);
2352 *value
= readl(iommu
->mmio_base
+ offset
);
2357 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val
);