2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
4 * Leo Duran <leo.duran@amd.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/pci.h>
21 #include <linux/acpi.h>
22 #include <linux/list.h>
23 #include <linux/bitmap.h>
24 #include <linux/slab.h>
25 #include <linux/syscore_ops.h>
26 #include <linux/interrupt.h>
27 #include <linux/msi.h>
28 #include <linux/amd-iommu.h>
29 #include <linux/export.h>
30 #include <linux/iommu.h>
31 #include <asm/pci-direct.h>
32 #include <asm/iommu.h>
34 #include <asm/x86_init.h>
35 #include <asm/iommu_table.h>
36 #include <asm/io_apic.h>
37 #include <asm/irq_remapping.h>
39 #include "amd_iommu_proto.h"
40 #include "amd_iommu_types.h"
41 #include "irq_remapping.h"
44 * definitions for the ACPI scanning code
46 #define IVRS_HEADER_LENGTH 48
48 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
49 #define ACPI_IVMD_TYPE_ALL 0x20
50 #define ACPI_IVMD_TYPE 0x21
51 #define ACPI_IVMD_TYPE_RANGE 0x22
53 #define IVHD_DEV_ALL 0x01
54 #define IVHD_DEV_SELECT 0x02
55 #define IVHD_DEV_SELECT_RANGE_START 0x03
56 #define IVHD_DEV_RANGE_END 0x04
57 #define IVHD_DEV_ALIAS 0x42
58 #define IVHD_DEV_ALIAS_RANGE 0x43
59 #define IVHD_DEV_EXT_SELECT 0x46
60 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
61 #define IVHD_DEV_SPECIAL 0x48
62 #define IVHD_DEV_ACPI_HID 0xf0
64 #define UID_NOT_PRESENT 0
65 #define UID_IS_INTEGER 1
66 #define UID_IS_CHARACTER 2
68 #define IVHD_SPECIAL_IOAPIC 1
69 #define IVHD_SPECIAL_HPET 2
71 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
72 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
73 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
74 #define IVHD_FLAG_ISOC_EN_MASK 0x08
76 #define IVMD_FLAG_EXCL_RANGE 0x08
77 #define IVMD_FLAG_UNITY_MAP 0x01
79 #define ACPI_DEVFLAG_INITPASS 0x01
80 #define ACPI_DEVFLAG_EXTINT 0x02
81 #define ACPI_DEVFLAG_NMI 0x04
82 #define ACPI_DEVFLAG_SYSMGT1 0x10
83 #define ACPI_DEVFLAG_SYSMGT2 0x20
84 #define ACPI_DEVFLAG_LINT0 0x40
85 #define ACPI_DEVFLAG_LINT1 0x80
86 #define ACPI_DEVFLAG_ATSDIS 0x10000000
88 #define LOOP_TIMEOUT 100000
90 * ACPI table definitions
92 * These data structures are laid over the table to parse the important values
97 * structure describing one IOMMU in the ACPI table. Typically followed by one
98 * or more ivhd_entrys.
111 /* Following only valid on IVHD type 11h and 40h */
112 u64 efr_reg
; /* Exact copy of MMIO_EXT_FEATURES */
114 } __attribute__((packed
));
117 * A device entry describing which devices a specific IOMMU translates and
118 * which requestor ids they use.
130 } __attribute__((packed
));
133 * An AMD IOMMU memory definition structure. It defines things like exclusion
134 * ranges for devices and regions that should be unity mapped.
145 } __attribute__((packed
));
148 bool amd_iommu_irq_remap __read_mostly
;
150 int amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_VAPIC
;
152 static bool amd_iommu_detected
;
153 static bool __initdata amd_iommu_disabled
;
154 static int amd_iommu_target_ivhd_type
;
156 u16 amd_iommu_last_bdf
; /* largest PCI device id we have
158 LIST_HEAD(amd_iommu_unity_map
); /* a list of required unity mappings
160 bool amd_iommu_unmap_flush
; /* if true, flush on every unmap */
162 LIST_HEAD(amd_iommu_list
); /* list of all AMD IOMMUs in the
165 /* Array to assign indices to IOMMUs*/
166 struct amd_iommu
*amd_iommus
[MAX_IOMMUS
];
167 int amd_iommus_present
;
169 /* IOMMUs have a non-present cache? */
170 bool amd_iommu_np_cache __read_mostly
;
171 bool amd_iommu_iotlb_sup __read_mostly
= true;
173 u32 amd_iommu_max_pasid __read_mostly
= ~0;
175 bool amd_iommu_v2_present __read_mostly
;
176 static bool amd_iommu_pc_present __read_mostly
;
178 bool amd_iommu_force_isolation __read_mostly
;
181 * List of protection domains - used during resume
183 LIST_HEAD(amd_iommu_pd_list
);
184 spinlock_t amd_iommu_pd_lock
;
187 * Pointer to the device table which is shared by all AMD IOMMUs
188 * it is indexed by the PCI device id or the HT unit id and contains
189 * information about the domain the device belongs to as well as the
190 * page table root pointer.
192 struct dev_table_entry
*amd_iommu_dev_table
;
195 * The alias table is a driver specific data structure which contains the
196 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
197 * More than one device can share the same requestor id.
199 u16
*amd_iommu_alias_table
;
202 * The rlookup table is used to find the IOMMU which is responsible
203 * for a specific device. It is also indexed by the PCI device id.
205 struct amd_iommu
**amd_iommu_rlookup_table
;
208 * This table is used to find the irq remapping table for a given device id
211 struct irq_remap_table
**irq_lookup_table
;
214 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
215 * to know which ones are already in use.
217 unsigned long *amd_iommu_pd_alloc_bitmap
;
219 static u32 dev_table_size
; /* size of the device table */
220 static u32 alias_table_size
; /* size of the alias table */
221 static u32 rlookup_table_size
; /* size if the rlookup table */
223 enum iommu_init_state
{
236 /* Early ioapic and hpet maps from kernel command line */
237 #define EARLY_MAP_SIZE 4
238 static struct devid_map __initdata early_ioapic_map
[EARLY_MAP_SIZE
];
239 static struct devid_map __initdata early_hpet_map
[EARLY_MAP_SIZE
];
240 static struct acpihid_map_entry __initdata early_acpihid_map
[EARLY_MAP_SIZE
];
242 static int __initdata early_ioapic_map_size
;
243 static int __initdata early_hpet_map_size
;
244 static int __initdata early_acpihid_map_size
;
246 static bool __initdata cmdline_maps
;
248 static enum iommu_init_state init_state
= IOMMU_START_STATE
;
250 static int amd_iommu_enable_interrupts(void);
251 static int __init
iommu_go_to_state(enum iommu_init_state state
);
252 static void init_device_table_dma(void);
254 static int iommu_pc_get_set_reg_val(struct amd_iommu
*iommu
,
255 u8 bank
, u8 cntr
, u8 fxn
,
256 u64
*value
, bool is_write
);
258 static inline void update_last_devid(u16 devid
)
260 if (devid
> amd_iommu_last_bdf
)
261 amd_iommu_last_bdf
= devid
;
264 static inline unsigned long tbl_size(int entry_size
)
266 unsigned shift
= PAGE_SHIFT
+
267 get_order(((int)amd_iommu_last_bdf
+ 1) * entry_size
);
272 /* Access to l1 and l2 indexed register spaces */
274 static u32
iommu_read_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
)
278 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
279 pci_read_config_dword(iommu
->dev
, 0xfc, &val
);
283 static void iommu_write_l1(struct amd_iommu
*iommu
, u16 l1
, u8 address
, u32 val
)
285 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16 | 1 << 31));
286 pci_write_config_dword(iommu
->dev
, 0xfc, val
);
287 pci_write_config_dword(iommu
->dev
, 0xf8, (address
| l1
<< 16));
290 static u32
iommu_read_l2(struct amd_iommu
*iommu
, u8 address
)
294 pci_write_config_dword(iommu
->dev
, 0xf0, address
);
295 pci_read_config_dword(iommu
->dev
, 0xf4, &val
);
299 static void iommu_write_l2(struct amd_iommu
*iommu
, u8 address
, u32 val
)
301 pci_write_config_dword(iommu
->dev
, 0xf0, (address
| 1 << 8));
302 pci_write_config_dword(iommu
->dev
, 0xf4, val
);
305 /****************************************************************************
307 * AMD IOMMU MMIO register space handling functions
309 * These functions are used to program the IOMMU device registers in
310 * MMIO space required for that driver.
312 ****************************************************************************/
315 * This function set the exclusion range in the IOMMU. DMA accesses to the
316 * exclusion range are passed through untranslated
318 static void iommu_set_exclusion_range(struct amd_iommu
*iommu
)
320 u64 start
= iommu
->exclusion_start
& PAGE_MASK
;
321 u64 limit
= (start
+ iommu
->exclusion_length
) & PAGE_MASK
;
324 if (!iommu
->exclusion_start
)
327 entry
= start
| MMIO_EXCL_ENABLE_MASK
;
328 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_BASE_OFFSET
,
329 &entry
, sizeof(entry
));
332 memcpy_toio(iommu
->mmio_base
+ MMIO_EXCL_LIMIT_OFFSET
,
333 &entry
, sizeof(entry
));
336 /* Programs the physical address of the device table into the IOMMU hardware */
337 static void iommu_set_device_table(struct amd_iommu
*iommu
)
341 BUG_ON(iommu
->mmio_base
== NULL
);
343 entry
= virt_to_phys(amd_iommu_dev_table
);
344 entry
|= (dev_table_size
>> 12) - 1;
345 memcpy_toio(iommu
->mmio_base
+ MMIO_DEV_TABLE_OFFSET
,
346 &entry
, sizeof(entry
));
349 /* Generic functions to enable/disable certain features of the IOMMU. */
350 static void iommu_feature_enable(struct amd_iommu
*iommu
, u8 bit
)
354 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
356 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
359 static void iommu_feature_disable(struct amd_iommu
*iommu
, u8 bit
)
363 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
365 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
368 static void iommu_set_inv_tlb_timeout(struct amd_iommu
*iommu
, int timeout
)
372 ctrl
= readl(iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
373 ctrl
&= ~CTRL_INV_TO_MASK
;
374 ctrl
|= (timeout
<< CONTROL_INV_TIMEOUT
) & CTRL_INV_TO_MASK
;
375 writel(ctrl
, iommu
->mmio_base
+ MMIO_CONTROL_OFFSET
);
378 /* Function to enable the hardware */
379 static void iommu_enable(struct amd_iommu
*iommu
)
381 iommu_feature_enable(iommu
, CONTROL_IOMMU_EN
);
384 static void iommu_disable(struct amd_iommu
*iommu
)
386 /* Disable command buffer */
387 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
389 /* Disable event logging and event interrupts */
390 iommu_feature_disable(iommu
, CONTROL_EVT_INT_EN
);
391 iommu_feature_disable(iommu
, CONTROL_EVT_LOG_EN
);
393 /* Disable IOMMU GA_LOG */
394 iommu_feature_disable(iommu
, CONTROL_GALOG_EN
);
395 iommu_feature_disable(iommu
, CONTROL_GAINT_EN
);
397 /* Disable IOMMU hardware itself */
398 iommu_feature_disable(iommu
, CONTROL_IOMMU_EN
);
402 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
403 * the system has one.
405 static u8 __iomem
* __init
iommu_map_mmio_space(u64 address
, u64 end
)
407 if (!request_mem_region(address
, end
, "amd_iommu")) {
408 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
410 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
414 return (u8 __iomem
*)ioremap_nocache(address
, end
);
417 static void __init
iommu_unmap_mmio_space(struct amd_iommu
*iommu
)
419 if (iommu
->mmio_base
)
420 iounmap(iommu
->mmio_base
);
421 release_mem_region(iommu
->mmio_phys
, iommu
->mmio_phys_end
);
424 static inline u32
get_ivhd_header_size(struct ivhd_header
*h
)
440 /****************************************************************************
442 * The functions below belong to the first pass of AMD IOMMU ACPI table
443 * parsing. In this pass we try to find out the highest device id this
444 * code has to handle. Upon this information the size of the shared data
445 * structures is determined later.
447 ****************************************************************************/
450 * This function calculates the length of a given IVHD entry
452 static inline int ivhd_entry_length(u8
*ivhd
)
454 u32 type
= ((struct ivhd_entry
*)ivhd
)->type
;
457 return 0x04 << (*ivhd
>> 6);
458 } else if (type
== IVHD_DEV_ACPI_HID
) {
459 /* For ACPI_HID, offset 21 is uid len */
460 return *((u8
*)ivhd
+ 21) + 22;
466 * After reading the highest device id from the IOMMU PCI capability header
467 * this function looks if there is a higher device id defined in the ACPI table
469 static int __init
find_last_devid_from_ivhd(struct ivhd_header
*h
)
471 u8
*p
= (void *)h
, *end
= (void *)h
;
472 struct ivhd_entry
*dev
;
474 u32 ivhd_size
= get_ivhd_header_size(h
);
477 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h
->type
);
485 dev
= (struct ivhd_entry
*)p
;
488 /* Use maximum BDF value for DEV_ALL */
489 update_last_devid(0xffff);
491 case IVHD_DEV_SELECT
:
492 case IVHD_DEV_RANGE_END
:
494 case IVHD_DEV_EXT_SELECT
:
495 /* all the above subfield types refer to device ids */
496 update_last_devid(dev
->devid
);
501 p
+= ivhd_entry_length(p
);
509 static int __init
check_ivrs_checksum(struct acpi_table_header
*table
)
512 u8 checksum
= 0, *p
= (u8
*)table
;
514 for (i
= 0; i
< table
->length
; ++i
)
517 /* ACPI table corrupt */
518 pr_err(FW_BUG
"AMD-Vi: IVRS invalid checksum\n");
526 * Iterate over all IVHD entries in the ACPI table and find the highest device
527 * id which we need to handle. This is the first of three functions which parse
528 * the ACPI table. So we check the checksum here.
530 static int __init
find_last_devid_acpi(struct acpi_table_header
*table
)
532 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
533 struct ivhd_header
*h
;
535 p
+= IVRS_HEADER_LENGTH
;
537 end
+= table
->length
;
539 h
= (struct ivhd_header
*)p
;
540 if (h
->type
== amd_iommu_target_ivhd_type
) {
541 int ret
= find_last_devid_from_ivhd(h
);
553 /****************************************************************************
555 * The following functions belong to the code path which parses the ACPI table
556 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
557 * data structures, initialize the device/alias/rlookup table and also
558 * basically initialize the hardware.
560 ****************************************************************************/
563 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
564 * write commands to that buffer later and the IOMMU will execute them
567 static int __init
alloc_command_buffer(struct amd_iommu
*iommu
)
569 iommu
->cmd_buf
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
570 get_order(CMD_BUFFER_SIZE
));
572 return iommu
->cmd_buf
? 0 : -ENOMEM
;
576 * This function resets the command buffer if the IOMMU stopped fetching
579 void amd_iommu_reset_cmd_buffer(struct amd_iommu
*iommu
)
581 iommu_feature_disable(iommu
, CONTROL_CMDBUF_EN
);
583 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_HEAD_OFFSET
);
584 writel(0x00, iommu
->mmio_base
+ MMIO_CMD_TAIL_OFFSET
);
586 iommu_feature_enable(iommu
, CONTROL_CMDBUF_EN
);
590 * This function writes the command buffer address to the hardware and
593 static void iommu_enable_command_buffer(struct amd_iommu
*iommu
)
597 BUG_ON(iommu
->cmd_buf
== NULL
);
599 entry
= (u64
)virt_to_phys(iommu
->cmd_buf
);
600 entry
|= MMIO_CMD_SIZE_512
;
602 memcpy_toio(iommu
->mmio_base
+ MMIO_CMD_BUF_OFFSET
,
603 &entry
, sizeof(entry
));
605 amd_iommu_reset_cmd_buffer(iommu
);
608 static void __init
free_command_buffer(struct amd_iommu
*iommu
)
610 free_pages((unsigned long)iommu
->cmd_buf
, get_order(CMD_BUFFER_SIZE
));
613 /* allocates the memory where the IOMMU will log its events to */
614 static int __init
alloc_event_buffer(struct amd_iommu
*iommu
)
616 iommu
->evt_buf
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
617 get_order(EVT_BUFFER_SIZE
));
619 return iommu
->evt_buf
? 0 : -ENOMEM
;
622 static void iommu_enable_event_buffer(struct amd_iommu
*iommu
)
626 BUG_ON(iommu
->evt_buf
== NULL
);
628 entry
= (u64
)virt_to_phys(iommu
->evt_buf
) | EVT_LEN_MASK
;
630 memcpy_toio(iommu
->mmio_base
+ MMIO_EVT_BUF_OFFSET
,
631 &entry
, sizeof(entry
));
633 /* set head and tail to zero manually */
634 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_HEAD_OFFSET
);
635 writel(0x00, iommu
->mmio_base
+ MMIO_EVT_TAIL_OFFSET
);
637 iommu_feature_enable(iommu
, CONTROL_EVT_LOG_EN
);
640 static void __init
free_event_buffer(struct amd_iommu
*iommu
)
642 free_pages((unsigned long)iommu
->evt_buf
, get_order(EVT_BUFFER_SIZE
));
645 /* allocates the memory where the IOMMU will log its events to */
646 static int __init
alloc_ppr_log(struct amd_iommu
*iommu
)
648 iommu
->ppr_log
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
649 get_order(PPR_LOG_SIZE
));
651 return iommu
->ppr_log
? 0 : -ENOMEM
;
654 static void iommu_enable_ppr_log(struct amd_iommu
*iommu
)
658 if (iommu
->ppr_log
== NULL
)
661 entry
= (u64
)virt_to_phys(iommu
->ppr_log
) | PPR_LOG_SIZE_512
;
663 memcpy_toio(iommu
->mmio_base
+ MMIO_PPR_LOG_OFFSET
,
664 &entry
, sizeof(entry
));
666 /* set head and tail to zero manually */
667 writel(0x00, iommu
->mmio_base
+ MMIO_PPR_HEAD_OFFSET
);
668 writel(0x00, iommu
->mmio_base
+ MMIO_PPR_TAIL_OFFSET
);
670 iommu_feature_enable(iommu
, CONTROL_PPFLOG_EN
);
671 iommu_feature_enable(iommu
, CONTROL_PPR_EN
);
674 static void __init
free_ppr_log(struct amd_iommu
*iommu
)
676 if (iommu
->ppr_log
== NULL
)
679 free_pages((unsigned long)iommu
->ppr_log
, get_order(PPR_LOG_SIZE
));
682 static void free_ga_log(struct amd_iommu
*iommu
)
684 #ifdef CONFIG_IRQ_REMAP
686 free_pages((unsigned long)iommu
->ga_log
,
687 get_order(GA_LOG_SIZE
));
688 if (iommu
->ga_log_tail
)
689 free_pages((unsigned long)iommu
->ga_log_tail
,
694 static int iommu_ga_log_enable(struct amd_iommu
*iommu
)
696 #ifdef CONFIG_IRQ_REMAP
702 status
= readl(iommu
->mmio_base
+ MMIO_STATUS_OFFSET
);
704 /* Check if already running */
705 if (status
& (MMIO_STATUS_GALOG_RUN_MASK
))
708 iommu_feature_enable(iommu
, CONTROL_GAINT_EN
);
709 iommu_feature_enable(iommu
, CONTROL_GALOG_EN
);
711 for (i
= 0; i
< LOOP_TIMEOUT
; ++i
) {
712 status
= readl(iommu
->mmio_base
+ MMIO_STATUS_OFFSET
);
713 if (status
& (MMIO_STATUS_GALOG_RUN_MASK
))
717 if (i
>= LOOP_TIMEOUT
)
719 #endif /* CONFIG_IRQ_REMAP */
723 #ifdef CONFIG_IRQ_REMAP
724 static int iommu_init_ga_log(struct amd_iommu
*iommu
)
728 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
731 iommu
->ga_log
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
732 get_order(GA_LOG_SIZE
));
736 iommu
->ga_log_tail
= (u8
*)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
738 if (!iommu
->ga_log_tail
)
741 entry
= (u64
)virt_to_phys(iommu
->ga_log
) | GA_LOG_SIZE_512
;
742 memcpy_toio(iommu
->mmio_base
+ MMIO_GA_LOG_BASE_OFFSET
,
743 &entry
, sizeof(entry
));
744 entry
= ((u64
)virt_to_phys(iommu
->ga_log
) & 0xFFFFFFFFFFFFFULL
) & ~7ULL;
745 memcpy_toio(iommu
->mmio_base
+ MMIO_GA_LOG_TAIL_OFFSET
,
746 &entry
, sizeof(entry
));
747 writel(0x00, iommu
->mmio_base
+ MMIO_GA_HEAD_OFFSET
);
748 writel(0x00, iommu
->mmio_base
+ MMIO_GA_TAIL_OFFSET
);
755 #endif /* CONFIG_IRQ_REMAP */
757 static int iommu_init_ga(struct amd_iommu
*iommu
)
761 #ifdef CONFIG_IRQ_REMAP
762 /* Note: We have already checked GASup from IVRS table.
763 * Now, we need to make sure that GAMSup is set.
765 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
) &&
766 !iommu_feature(iommu
, FEATURE_GAM_VAPIC
))
767 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY_GA
;
769 ret
= iommu_init_ga_log(iommu
);
770 #endif /* CONFIG_IRQ_REMAP */
775 static void iommu_enable_gt(struct amd_iommu
*iommu
)
777 if (!iommu_feature(iommu
, FEATURE_GT
))
780 iommu_feature_enable(iommu
, CONTROL_GT_EN
);
783 /* sets a specific bit in the device table entry. */
784 static void set_dev_entry_bit(u16 devid
, u8 bit
)
786 int i
= (bit
>> 6) & 0x03;
787 int _bit
= bit
& 0x3f;
789 amd_iommu_dev_table
[devid
].data
[i
] |= (1UL << _bit
);
792 static int get_dev_entry_bit(u16 devid
, u8 bit
)
794 int i
= (bit
>> 6) & 0x03;
795 int _bit
= bit
& 0x3f;
797 return (amd_iommu_dev_table
[devid
].data
[i
] & (1UL << _bit
)) >> _bit
;
801 void amd_iommu_apply_erratum_63(u16 devid
)
805 sysmgt
= get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
) |
806 (get_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
) << 1);
809 set_dev_entry_bit(devid
, DEV_ENTRY_IW
);
812 /* Writes the specific IOMMU for a device into the rlookup table */
813 static void __init
set_iommu_for_device(struct amd_iommu
*iommu
, u16 devid
)
815 amd_iommu_rlookup_table
[devid
] = iommu
;
819 * This function takes the device specific flags read from the ACPI
820 * table and sets up the device table entry with that information
822 static void __init
set_dev_entry_from_acpi(struct amd_iommu
*iommu
,
823 u16 devid
, u32 flags
, u32 ext_flags
)
825 if (flags
& ACPI_DEVFLAG_INITPASS
)
826 set_dev_entry_bit(devid
, DEV_ENTRY_INIT_PASS
);
827 if (flags
& ACPI_DEVFLAG_EXTINT
)
828 set_dev_entry_bit(devid
, DEV_ENTRY_EINT_PASS
);
829 if (flags
& ACPI_DEVFLAG_NMI
)
830 set_dev_entry_bit(devid
, DEV_ENTRY_NMI_PASS
);
831 if (flags
& ACPI_DEVFLAG_SYSMGT1
)
832 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT1
);
833 if (flags
& ACPI_DEVFLAG_SYSMGT2
)
834 set_dev_entry_bit(devid
, DEV_ENTRY_SYSMGT2
);
835 if (flags
& ACPI_DEVFLAG_LINT0
)
836 set_dev_entry_bit(devid
, DEV_ENTRY_LINT0_PASS
);
837 if (flags
& ACPI_DEVFLAG_LINT1
)
838 set_dev_entry_bit(devid
, DEV_ENTRY_LINT1_PASS
);
840 amd_iommu_apply_erratum_63(devid
);
842 set_iommu_for_device(iommu
, devid
);
845 static int __init
add_special_device(u8 type
, u8 id
, u16
*devid
, bool cmd_line
)
847 struct devid_map
*entry
;
848 struct list_head
*list
;
850 if (type
== IVHD_SPECIAL_IOAPIC
)
852 else if (type
== IVHD_SPECIAL_HPET
)
857 list_for_each_entry(entry
, list
, list
) {
858 if (!(entry
->id
== id
&& entry
->cmd_line
))
861 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
862 type
== IVHD_SPECIAL_IOAPIC
? "IOAPIC" : "HPET", id
);
864 *devid
= entry
->devid
;
869 entry
= kzalloc(sizeof(*entry
), GFP_KERNEL
);
874 entry
->devid
= *devid
;
875 entry
->cmd_line
= cmd_line
;
877 list_add_tail(&entry
->list
, list
);
882 static int __init
add_acpi_hid_device(u8
*hid
, u8
*uid
, u16
*devid
,
885 struct acpihid_map_entry
*entry
;
886 struct list_head
*list
= &acpihid_map
;
888 list_for_each_entry(entry
, list
, list
) {
889 if (strcmp(entry
->hid
, hid
) ||
890 (*uid
&& *entry
->uid
&& strcmp(entry
->uid
, uid
)) ||
894 pr_info("AMD-Vi: Command-line override for hid:%s uid:%s\n",
896 *devid
= entry
->devid
;
900 entry
= kzalloc(sizeof(*entry
), GFP_KERNEL
);
904 memcpy(entry
->uid
, uid
, strlen(uid
));
905 memcpy(entry
->hid
, hid
, strlen(hid
));
906 entry
->devid
= *devid
;
907 entry
->cmd_line
= cmd_line
;
908 entry
->root_devid
= (entry
->devid
& (~0x7));
910 pr_info("AMD-Vi:%s, add hid:%s, uid:%s, rdevid:%d\n",
911 entry
->cmd_line
? "cmd" : "ivrs",
912 entry
->hid
, entry
->uid
, entry
->root_devid
);
914 list_add_tail(&entry
->list
, list
);
918 static int __init
add_early_maps(void)
922 for (i
= 0; i
< early_ioapic_map_size
; ++i
) {
923 ret
= add_special_device(IVHD_SPECIAL_IOAPIC
,
924 early_ioapic_map
[i
].id
,
925 &early_ioapic_map
[i
].devid
,
926 early_ioapic_map
[i
].cmd_line
);
931 for (i
= 0; i
< early_hpet_map_size
; ++i
) {
932 ret
= add_special_device(IVHD_SPECIAL_HPET
,
933 early_hpet_map
[i
].id
,
934 &early_hpet_map
[i
].devid
,
935 early_hpet_map
[i
].cmd_line
);
940 for (i
= 0; i
< early_acpihid_map_size
; ++i
) {
941 ret
= add_acpi_hid_device(early_acpihid_map
[i
].hid
,
942 early_acpihid_map
[i
].uid
,
943 &early_acpihid_map
[i
].devid
,
944 early_acpihid_map
[i
].cmd_line
);
953 * Reads the device exclusion range from ACPI and initializes the IOMMU with
956 static void __init
set_device_exclusion_range(u16 devid
, struct ivmd_header
*m
)
958 struct amd_iommu
*iommu
= amd_iommu_rlookup_table
[devid
];
960 if (!(m
->flags
& IVMD_FLAG_EXCL_RANGE
))
965 * We only can configure exclusion ranges per IOMMU, not
966 * per device. But we can enable the exclusion range per
967 * device. This is done here
969 set_dev_entry_bit(devid
, DEV_ENTRY_EX
);
970 iommu
->exclusion_start
= m
->range_start
;
971 iommu
->exclusion_length
= m
->range_length
;
976 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
977 * initializes the hardware and our data structures with it.
979 static int __init
init_iommu_from_acpi(struct amd_iommu
*iommu
,
980 struct ivhd_header
*h
)
983 u8
*end
= p
, flags
= 0;
984 u16 devid
= 0, devid_start
= 0, devid_to
= 0;
985 u32 dev_i
, ext_flags
= 0;
987 struct ivhd_entry
*e
;
992 ret
= add_early_maps();
997 * First save the recommended feature enable bits from ACPI
999 iommu
->acpi_flags
= h
->flags
;
1002 * Done. Now parse the device entries
1004 ivhd_size
= get_ivhd_header_size(h
);
1006 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h
->type
);
1016 e
= (struct ivhd_entry
*)p
;
1020 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e
->flags
);
1022 for (dev_i
= 0; dev_i
<= amd_iommu_last_bdf
; ++dev_i
)
1023 set_dev_entry_from_acpi(iommu
, dev_i
, e
->flags
, 0);
1025 case IVHD_DEV_SELECT
:
1027 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1029 PCI_BUS_NUM(e
->devid
),
1035 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1037 case IVHD_DEV_SELECT_RANGE_START
:
1039 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1040 "devid: %02x:%02x.%x flags: %02x\n",
1041 PCI_BUS_NUM(e
->devid
),
1046 devid_start
= e
->devid
;
1051 case IVHD_DEV_ALIAS
:
1053 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1054 "flags: %02x devid_to: %02x:%02x.%x\n",
1055 PCI_BUS_NUM(e
->devid
),
1059 PCI_BUS_NUM(e
->ext
>> 8),
1060 PCI_SLOT(e
->ext
>> 8),
1061 PCI_FUNC(e
->ext
>> 8));
1064 devid_to
= e
->ext
>> 8;
1065 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1066 set_dev_entry_from_acpi(iommu
, devid_to
, e
->flags
, 0);
1067 amd_iommu_alias_table
[devid
] = devid_to
;
1069 case IVHD_DEV_ALIAS_RANGE
:
1071 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1072 "devid: %02x:%02x.%x flags: %02x "
1073 "devid_to: %02x:%02x.%x\n",
1074 PCI_BUS_NUM(e
->devid
),
1078 PCI_BUS_NUM(e
->ext
>> 8),
1079 PCI_SLOT(e
->ext
>> 8),
1080 PCI_FUNC(e
->ext
>> 8));
1082 devid_start
= e
->devid
;
1084 devid_to
= e
->ext
>> 8;
1088 case IVHD_DEV_EXT_SELECT
:
1090 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1091 "flags: %02x ext: %08x\n",
1092 PCI_BUS_NUM(e
->devid
),
1098 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
,
1101 case IVHD_DEV_EXT_SELECT_RANGE
:
1103 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1104 "%02x:%02x.%x flags: %02x ext: %08x\n",
1105 PCI_BUS_NUM(e
->devid
),
1110 devid_start
= e
->devid
;
1115 case IVHD_DEV_RANGE_END
:
1117 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1118 PCI_BUS_NUM(e
->devid
),
1120 PCI_FUNC(e
->devid
));
1123 for (dev_i
= devid_start
; dev_i
<= devid
; ++dev_i
) {
1125 amd_iommu_alias_table
[dev_i
] = devid_to
;
1126 set_dev_entry_from_acpi(iommu
,
1127 devid_to
, flags
, ext_flags
);
1129 set_dev_entry_from_acpi(iommu
, dev_i
,
1133 case IVHD_DEV_SPECIAL
: {
1139 handle
= e
->ext
& 0xff;
1140 devid
= (e
->ext
>> 8) & 0xffff;
1141 type
= (e
->ext
>> 24) & 0xff;
1143 if (type
== IVHD_SPECIAL_IOAPIC
)
1145 else if (type
== IVHD_SPECIAL_HPET
)
1150 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1156 ret
= add_special_device(type
, handle
, &devid
, false);
1161 * add_special_device might update the devid in case a
1162 * command-line override is present. So call
1163 * set_dev_entry_from_acpi after add_special_device.
1165 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1169 case IVHD_DEV_ACPI_HID
: {
1171 u8 hid
[ACPIHID_HID_LEN
] = {0};
1172 u8 uid
[ACPIHID_UID_LEN
] = {0};
1175 if (h
->type
!= 0x40) {
1176 pr_err(FW_BUG
"Invalid IVHD device type %#x\n",
1181 memcpy(hid
, (u8
*)(&e
->ext
), ACPIHID_HID_LEN
- 1);
1182 hid
[ACPIHID_HID_LEN
- 1] = '\0';
1185 pr_err(FW_BUG
"Invalid HID.\n");
1190 case UID_NOT_PRESENT
:
1193 pr_warn(FW_BUG
"Invalid UID length.\n");
1196 case UID_IS_INTEGER
:
1198 sprintf(uid
, "%d", e
->uid
);
1201 case UID_IS_CHARACTER
:
1203 memcpy(uid
, (u8
*)(&e
->uid
), ACPIHID_UID_LEN
- 1);
1204 uid
[ACPIHID_UID_LEN
- 1] = '\0';
1212 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1220 ret
= add_acpi_hid_device(hid
, uid
, &devid
, false);
1225 * add_special_device might update the devid in case a
1226 * command-line override is present. So call
1227 * set_dev_entry_from_acpi after add_special_device.
1229 set_dev_entry_from_acpi(iommu
, devid
, e
->flags
, 0);
1237 p
+= ivhd_entry_length(p
);
1243 static void __init
free_iommu_one(struct amd_iommu
*iommu
)
1245 free_command_buffer(iommu
);
1246 free_event_buffer(iommu
);
1247 free_ppr_log(iommu
);
1249 iommu_unmap_mmio_space(iommu
);
1252 static void __init
free_iommu_all(void)
1254 struct amd_iommu
*iommu
, *next
;
1256 for_each_iommu_safe(iommu
, next
) {
1257 list_del(&iommu
->list
);
1258 free_iommu_one(iommu
);
1264 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1266 * BIOS should disable L2B micellaneous clock gating by setting
1267 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1269 static void amd_iommu_erratum_746_workaround(struct amd_iommu
*iommu
)
1273 if ((boot_cpu_data
.x86
!= 0x15) ||
1274 (boot_cpu_data
.x86_model
< 0x10) ||
1275 (boot_cpu_data
.x86_model
> 0x1f))
1278 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
1279 pci_read_config_dword(iommu
->dev
, 0xf4, &value
);
1284 /* Select NB indirect register 0x90 and enable writing */
1285 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90 | (1 << 8));
1287 pci_write_config_dword(iommu
->dev
, 0xf4, value
| 0x4);
1288 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1289 dev_name(&iommu
->dev
->dev
));
1291 /* Clear the enable writing bit */
1292 pci_write_config_dword(iommu
->dev
, 0xf0, 0x90);
1296 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1298 * BIOS should enable ATS write permission check by setting
1299 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1301 static void amd_iommu_ats_write_check_workaround(struct amd_iommu
*iommu
)
1305 if ((boot_cpu_data
.x86
!= 0x15) ||
1306 (boot_cpu_data
.x86_model
< 0x30) ||
1307 (boot_cpu_data
.x86_model
> 0x3f))
1310 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1311 value
= iommu_read_l2(iommu
, 0x47);
1316 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1317 iommu_write_l2(iommu
, 0x47, value
| BIT(0));
1319 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1320 dev_name(&iommu
->dev
->dev
));
1324 * This function clues the initialization function for one IOMMU
1325 * together and also allocates the command buffer and programs the
1326 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1328 static int __init
init_iommu_one(struct amd_iommu
*iommu
, struct ivhd_header
*h
)
1332 spin_lock_init(&iommu
->lock
);
1334 /* Add IOMMU to internal data structures */
1335 list_add_tail(&iommu
->list
, &amd_iommu_list
);
1336 iommu
->index
= amd_iommus_present
++;
1338 if (unlikely(iommu
->index
>= MAX_IOMMUS
)) {
1339 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1343 /* Index is fine - add IOMMU to the array */
1344 amd_iommus
[iommu
->index
] = iommu
;
1347 * Copy data from ACPI table entry to the iommu struct
1349 iommu
->devid
= h
->devid
;
1350 iommu
->cap_ptr
= h
->cap_ptr
;
1351 iommu
->pci_seg
= h
->pci_seg
;
1352 iommu
->mmio_phys
= h
->mmio_phys
;
1356 /* Check if IVHD EFR contains proper max banks/counters */
1357 if ((h
->efr_attr
!= 0) &&
1358 ((h
->efr_attr
& (0xF << 13)) != 0) &&
1359 ((h
->efr_attr
& (0x3F << 17)) != 0))
1360 iommu
->mmio_phys_end
= MMIO_REG_END_OFFSET
;
1362 iommu
->mmio_phys_end
= MMIO_CNTR_CONF_OFFSET
;
1363 if (((h
->efr_attr
& (0x1 << IOMMU_FEAT_GASUP_SHIFT
)) == 0))
1364 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY
;
1368 if (h
->efr_reg
& (1 << 9))
1369 iommu
->mmio_phys_end
= MMIO_REG_END_OFFSET
;
1371 iommu
->mmio_phys_end
= MMIO_CNTR_CONF_OFFSET
;
1372 if (((h
->efr_reg
& (0x1 << IOMMU_EFR_GASUP_SHIFT
)) == 0))
1373 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY
;
1379 iommu
->mmio_base
= iommu_map_mmio_space(iommu
->mmio_phys
,
1380 iommu
->mmio_phys_end
);
1381 if (!iommu
->mmio_base
)
1384 if (alloc_command_buffer(iommu
))
1387 if (alloc_event_buffer(iommu
))
1390 iommu
->int_enabled
= false;
1392 ret
= init_iommu_from_acpi(iommu
, h
);
1396 ret
= amd_iommu_create_irq_domain(iommu
);
1401 * Make sure IOMMU is not considered to translate itself. The IVRS
1402 * table tells us so, but this is a lie!
1404 amd_iommu_rlookup_table
[iommu
->devid
] = NULL
;
1410 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1411 * @ivrs Pointer to the IVRS header
1413 * This function search through all IVDB of the maximum supported IVHD
1415 static u8
get_highest_supported_ivhd_type(struct acpi_table_header
*ivrs
)
1417 u8
*base
= (u8
*)ivrs
;
1418 struct ivhd_header
*ivhd
= (struct ivhd_header
*)
1419 (base
+ IVRS_HEADER_LENGTH
);
1420 u8 last_type
= ivhd
->type
;
1421 u16 devid
= ivhd
->devid
;
1423 while (((u8
*)ivhd
- base
< ivrs
->length
) &&
1424 (ivhd
->type
<= ACPI_IVHD_TYPE_MAX_SUPPORTED
)) {
1425 u8
*p
= (u8
*) ivhd
;
1427 if (ivhd
->devid
== devid
)
1428 last_type
= ivhd
->type
;
1429 ivhd
= (struct ivhd_header
*)(p
+ ivhd
->length
);
1436 * Iterates over all IOMMU entries in the ACPI table, allocates the
1437 * IOMMU structure and initializes it with init_iommu_one()
1439 static int __init
init_iommu_all(struct acpi_table_header
*table
)
1441 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1442 struct ivhd_header
*h
;
1443 struct amd_iommu
*iommu
;
1446 end
+= table
->length
;
1447 p
+= IVRS_HEADER_LENGTH
;
1450 h
= (struct ivhd_header
*)p
;
1451 if (*p
== amd_iommu_target_ivhd_type
) {
1453 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1454 "seg: %d flags: %01x info %04x\n",
1455 PCI_BUS_NUM(h
->devid
), PCI_SLOT(h
->devid
),
1456 PCI_FUNC(h
->devid
), h
->cap_ptr
,
1457 h
->pci_seg
, h
->flags
, h
->info
);
1458 DUMP_printk(" mmio-addr: %016llx\n",
1461 iommu
= kzalloc(sizeof(struct amd_iommu
), GFP_KERNEL
);
1465 ret
= init_iommu_one(iommu
, h
);
1478 static void init_iommu_perf_ctr(struct amd_iommu
*iommu
)
1480 u64 val
= 0xabcd, val2
= 0;
1482 if (!iommu_feature(iommu
, FEATURE_PC
))
1485 amd_iommu_pc_present
= true;
1487 /* Check if the performance counters can be written to */
1488 if ((0 != iommu_pc_get_set_reg_val(iommu
, 0, 0, 0, &val
, true)) ||
1489 (0 != iommu_pc_get_set_reg_val(iommu
, 0, 0, 0, &val2
, false)) ||
1491 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1492 amd_iommu_pc_present
= false;
1496 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1498 val
= readl(iommu
->mmio_base
+ MMIO_CNTR_CONF_OFFSET
);
1499 iommu
->max_banks
= (u8
) ((val
>> 12) & 0x3f);
1500 iommu
->max_counters
= (u8
) ((val
>> 7) & 0xf);
1503 static ssize_t
amd_iommu_show_cap(struct device
*dev
,
1504 struct device_attribute
*attr
,
1507 struct amd_iommu
*iommu
= dev_get_drvdata(dev
);
1508 return sprintf(buf
, "%x\n", iommu
->cap
);
1510 static DEVICE_ATTR(cap
, S_IRUGO
, amd_iommu_show_cap
, NULL
);
1512 static ssize_t
amd_iommu_show_features(struct device
*dev
,
1513 struct device_attribute
*attr
,
1516 struct amd_iommu
*iommu
= dev_get_drvdata(dev
);
1517 return sprintf(buf
, "%llx\n", iommu
->features
);
1519 static DEVICE_ATTR(features
, S_IRUGO
, amd_iommu_show_features
, NULL
);
1521 static struct attribute
*amd_iommu_attrs
[] = {
1523 &dev_attr_features
.attr
,
1527 static struct attribute_group amd_iommu_group
= {
1528 .name
= "amd-iommu",
1529 .attrs
= amd_iommu_attrs
,
1532 static const struct attribute_group
*amd_iommu_groups
[] = {
1537 static int iommu_init_pci(struct amd_iommu
*iommu
)
1539 int cap_ptr
= iommu
->cap_ptr
;
1540 u32 range
, misc
, low
, high
;
1543 iommu
->dev
= pci_get_bus_and_slot(PCI_BUS_NUM(iommu
->devid
),
1544 iommu
->devid
& 0xff);
1548 /* Prevent binding other PCI device drivers to IOMMU devices */
1549 iommu
->dev
->match_driver
= false;
1551 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_CAP_HDR_OFFSET
,
1553 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_RANGE_OFFSET
,
1555 pci_read_config_dword(iommu
->dev
, cap_ptr
+ MMIO_MISC_OFFSET
,
1558 if (!(iommu
->cap
& (1 << IOMMU_CAP_IOTLB
)))
1559 amd_iommu_iotlb_sup
= false;
1561 /* read extended feature bits */
1562 low
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
);
1563 high
= readl(iommu
->mmio_base
+ MMIO_EXT_FEATURES
+ 4);
1565 iommu
->features
= ((u64
)high
<< 32) | low
;
1567 if (iommu_feature(iommu
, FEATURE_GT
)) {
1572 pasmax
= iommu
->features
& FEATURE_PASID_MASK
;
1573 pasmax
>>= FEATURE_PASID_SHIFT
;
1574 max_pasid
= (1 << (pasmax
+ 1)) - 1;
1576 amd_iommu_max_pasid
= min(amd_iommu_max_pasid
, max_pasid
);
1578 BUG_ON(amd_iommu_max_pasid
& ~PASID_MASK
);
1580 glxval
= iommu
->features
& FEATURE_GLXVAL_MASK
;
1581 glxval
>>= FEATURE_GLXVAL_SHIFT
;
1583 if (amd_iommu_max_glx_val
== -1)
1584 amd_iommu_max_glx_val
= glxval
;
1586 amd_iommu_max_glx_val
= min(amd_iommu_max_glx_val
, glxval
);
1589 if (iommu_feature(iommu
, FEATURE_GT
) &&
1590 iommu_feature(iommu
, FEATURE_PPR
)) {
1591 iommu
->is_iommu_v2
= true;
1592 amd_iommu_v2_present
= true;
1595 if (iommu_feature(iommu
, FEATURE_PPR
) && alloc_ppr_log(iommu
))
1598 ret
= iommu_init_ga(iommu
);
1602 if (iommu
->cap
& (1UL << IOMMU_CAP_NPCACHE
))
1603 amd_iommu_np_cache
= true;
1605 init_iommu_perf_ctr(iommu
);
1607 if (is_rd890_iommu(iommu
->dev
)) {
1610 iommu
->root_pdev
= pci_get_bus_and_slot(iommu
->dev
->bus
->number
,
1614 * Some rd890 systems may not be fully reconfigured by the
1615 * BIOS, so it's necessary for us to store this information so
1616 * it can be reprogrammed on resume
1618 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1619 &iommu
->stored_addr_lo
);
1620 pci_read_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1621 &iommu
->stored_addr_hi
);
1623 /* Low bit locks writes to configuration space */
1624 iommu
->stored_addr_lo
&= ~1;
1626 for (i
= 0; i
< 6; i
++)
1627 for (j
= 0; j
< 0x12; j
++)
1628 iommu
->stored_l1
[i
][j
] = iommu_read_l1(iommu
, i
, j
);
1630 for (i
= 0; i
< 0x83; i
++)
1631 iommu
->stored_l2
[i
] = iommu_read_l2(iommu
, i
);
1634 amd_iommu_erratum_746_workaround(iommu
);
1635 amd_iommu_ats_write_check_workaround(iommu
);
1637 iommu
->iommu_dev
= iommu_device_create(&iommu
->dev
->dev
, iommu
,
1638 amd_iommu_groups
, "ivhd%d",
1641 return pci_enable_device(iommu
->dev
);
1644 static void print_iommu_info(void)
1646 static const char * const feat_str
[] = {
1647 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1648 "IA", "GA", "HE", "PC"
1650 struct amd_iommu
*iommu
;
1652 for_each_iommu(iommu
) {
1655 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1656 dev_name(&iommu
->dev
->dev
), iommu
->cap_ptr
);
1658 if (iommu
->cap
& (1 << IOMMU_CAP_EFR
)) {
1659 pr_info("AMD-Vi: Extended features (%#llx):\n",
1661 for (i
= 0; i
< ARRAY_SIZE(feat_str
); ++i
) {
1662 if (iommu_feature(iommu
, (1ULL << i
)))
1663 pr_cont(" %s", feat_str
[i
]);
1666 if (iommu
->features
& FEATURE_GAM_VAPIC
)
1667 pr_cont(" GA_vAPIC");
1672 if (irq_remapping_enabled
) {
1673 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1674 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
1675 pr_info("AMD-Vi: virtual APIC enabled\n");
1679 static int __init
amd_iommu_init_pci(void)
1681 struct amd_iommu
*iommu
;
1684 for_each_iommu(iommu
) {
1685 ret
= iommu_init_pci(iommu
);
1691 * Order is important here to make sure any unity map requirements are
1692 * fulfilled. The unity mappings are created and written to the device
1693 * table during the amd_iommu_init_api() call.
1695 * After that we call init_device_table_dma() to make sure any
1696 * uninitialized DTE will block DMA, and in the end we flush the caches
1697 * of all IOMMUs to make sure the changes to the device table are
1700 ret
= amd_iommu_init_api();
1702 init_device_table_dma();
1704 for_each_iommu(iommu
)
1705 iommu_flush_all_caches(iommu
);
1713 /****************************************************************************
1715 * The following functions initialize the MSI interrupts for all IOMMUs
1716 * in the system. It's a bit challenging because there could be multiple
1717 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1720 ****************************************************************************/
1722 static int iommu_setup_msi(struct amd_iommu
*iommu
)
1726 r
= pci_enable_msi(iommu
->dev
);
1730 r
= request_threaded_irq(iommu
->dev
->irq
,
1731 amd_iommu_int_handler
,
1732 amd_iommu_int_thread
,
1737 pci_disable_msi(iommu
->dev
);
1741 iommu
->int_enabled
= true;
1746 static int iommu_init_msi(struct amd_iommu
*iommu
)
1750 if (iommu
->int_enabled
)
1753 if (iommu
->dev
->msi_cap
)
1754 ret
= iommu_setup_msi(iommu
);
1762 iommu_feature_enable(iommu
, CONTROL_EVT_INT_EN
);
1764 if (iommu
->ppr_log
!= NULL
)
1765 iommu_feature_enable(iommu
, CONTROL_PPFINT_EN
);
1767 iommu_ga_log_enable(iommu
);
1772 /****************************************************************************
1774 * The next functions belong to the third pass of parsing the ACPI
1775 * table. In this last pass the memory mapping requirements are
1776 * gathered (like exclusion and unity mapping ranges).
1778 ****************************************************************************/
1780 static void __init
free_unity_maps(void)
1782 struct unity_map_entry
*entry
, *next
;
1784 list_for_each_entry_safe(entry
, next
, &amd_iommu_unity_map
, list
) {
1785 list_del(&entry
->list
);
1790 /* called when we find an exclusion range definition in ACPI */
1791 static int __init
init_exclusion_range(struct ivmd_header
*m
)
1796 case ACPI_IVMD_TYPE
:
1797 set_device_exclusion_range(m
->devid
, m
);
1799 case ACPI_IVMD_TYPE_ALL
:
1800 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
1801 set_device_exclusion_range(i
, m
);
1803 case ACPI_IVMD_TYPE_RANGE
:
1804 for (i
= m
->devid
; i
<= m
->aux
; ++i
)
1805 set_device_exclusion_range(i
, m
);
1814 /* called for unity map ACPI definition */
1815 static int __init
init_unity_map_range(struct ivmd_header
*m
)
1817 struct unity_map_entry
*e
= NULL
;
1820 e
= kzalloc(sizeof(*e
), GFP_KERNEL
);
1828 case ACPI_IVMD_TYPE
:
1829 s
= "IVMD_TYPEi\t\t\t";
1830 e
->devid_start
= e
->devid_end
= m
->devid
;
1832 case ACPI_IVMD_TYPE_ALL
:
1833 s
= "IVMD_TYPE_ALL\t\t";
1835 e
->devid_end
= amd_iommu_last_bdf
;
1837 case ACPI_IVMD_TYPE_RANGE
:
1838 s
= "IVMD_TYPE_RANGE\t\t";
1839 e
->devid_start
= m
->devid
;
1840 e
->devid_end
= m
->aux
;
1843 e
->address_start
= PAGE_ALIGN(m
->range_start
);
1844 e
->address_end
= e
->address_start
+ PAGE_ALIGN(m
->range_length
);
1845 e
->prot
= m
->flags
>> 1;
1847 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1848 " range_start: %016llx range_end: %016llx flags: %x\n", s
,
1849 PCI_BUS_NUM(e
->devid_start
), PCI_SLOT(e
->devid_start
),
1850 PCI_FUNC(e
->devid_start
), PCI_BUS_NUM(e
->devid_end
),
1851 PCI_SLOT(e
->devid_end
), PCI_FUNC(e
->devid_end
),
1852 e
->address_start
, e
->address_end
, m
->flags
);
1854 list_add_tail(&e
->list
, &amd_iommu_unity_map
);
1859 /* iterates over all memory definitions we find in the ACPI table */
1860 static int __init
init_memory_definitions(struct acpi_table_header
*table
)
1862 u8
*p
= (u8
*)table
, *end
= (u8
*)table
;
1863 struct ivmd_header
*m
;
1865 end
+= table
->length
;
1866 p
+= IVRS_HEADER_LENGTH
;
1869 m
= (struct ivmd_header
*)p
;
1870 if (m
->flags
& IVMD_FLAG_EXCL_RANGE
)
1871 init_exclusion_range(m
);
1872 else if (m
->flags
& IVMD_FLAG_UNITY_MAP
)
1873 init_unity_map_range(m
);
1882 * Init the device table to not allow DMA access for devices and
1883 * suppress all page faults
1885 static void init_device_table_dma(void)
1889 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1890 set_dev_entry_bit(devid
, DEV_ENTRY_VALID
);
1891 set_dev_entry_bit(devid
, DEV_ENTRY_TRANSLATION
);
1895 static void __init
uninit_device_table_dma(void)
1899 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
) {
1900 amd_iommu_dev_table
[devid
].data
[0] = 0ULL;
1901 amd_iommu_dev_table
[devid
].data
[1] = 0ULL;
1905 static void init_device_table(void)
1909 if (!amd_iommu_irq_remap
)
1912 for (devid
= 0; devid
<= amd_iommu_last_bdf
; ++devid
)
1913 set_dev_entry_bit(devid
, DEV_ENTRY_IRQ_TBL_EN
);
1916 static void iommu_init_flags(struct amd_iommu
*iommu
)
1918 iommu
->acpi_flags
& IVHD_FLAG_HT_TUN_EN_MASK
?
1919 iommu_feature_enable(iommu
, CONTROL_HT_TUN_EN
) :
1920 iommu_feature_disable(iommu
, CONTROL_HT_TUN_EN
);
1922 iommu
->acpi_flags
& IVHD_FLAG_PASSPW_EN_MASK
?
1923 iommu_feature_enable(iommu
, CONTROL_PASSPW_EN
) :
1924 iommu_feature_disable(iommu
, CONTROL_PASSPW_EN
);
1926 iommu
->acpi_flags
& IVHD_FLAG_RESPASSPW_EN_MASK
?
1927 iommu_feature_enable(iommu
, CONTROL_RESPASSPW_EN
) :
1928 iommu_feature_disable(iommu
, CONTROL_RESPASSPW_EN
);
1930 iommu
->acpi_flags
& IVHD_FLAG_ISOC_EN_MASK
?
1931 iommu_feature_enable(iommu
, CONTROL_ISOC_EN
) :
1932 iommu_feature_disable(iommu
, CONTROL_ISOC_EN
);
1935 * make IOMMU memory accesses cache coherent
1937 iommu_feature_enable(iommu
, CONTROL_COHERENT_EN
);
1939 /* Set IOTLB invalidation timeout to 1s */
1940 iommu_set_inv_tlb_timeout(iommu
, CTRL_INV_TO_1S
);
1943 static void iommu_apply_resume_quirks(struct amd_iommu
*iommu
)
1946 u32 ioc_feature_control
;
1947 struct pci_dev
*pdev
= iommu
->root_pdev
;
1949 /* RD890 BIOSes may not have completely reconfigured the iommu */
1950 if (!is_rd890_iommu(iommu
->dev
) || !pdev
)
1954 * First, we need to ensure that the iommu is enabled. This is
1955 * controlled by a register in the northbridge
1958 /* Select Northbridge indirect register 0x75 and enable writing */
1959 pci_write_config_dword(pdev
, 0x60, 0x75 | (1 << 7));
1960 pci_read_config_dword(pdev
, 0x64, &ioc_feature_control
);
1962 /* Enable the iommu */
1963 if (!(ioc_feature_control
& 0x1))
1964 pci_write_config_dword(pdev
, 0x64, ioc_feature_control
| 1);
1966 /* Restore the iommu BAR */
1967 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1968 iommu
->stored_addr_lo
);
1969 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 8,
1970 iommu
->stored_addr_hi
);
1972 /* Restore the l1 indirect regs for each of the 6 l1s */
1973 for (i
= 0; i
< 6; i
++)
1974 for (j
= 0; j
< 0x12; j
++)
1975 iommu_write_l1(iommu
, i
, j
, iommu
->stored_l1
[i
][j
]);
1977 /* Restore the l2 indirect regs */
1978 for (i
= 0; i
< 0x83; i
++)
1979 iommu_write_l2(iommu
, i
, iommu
->stored_l2
[i
]);
1981 /* Lock PCI setup registers */
1982 pci_write_config_dword(iommu
->dev
, iommu
->cap_ptr
+ 4,
1983 iommu
->stored_addr_lo
| 1);
1986 static void iommu_enable_ga(struct amd_iommu
*iommu
)
1988 #ifdef CONFIG_IRQ_REMAP
1989 switch (amd_iommu_guest_ir
) {
1990 case AMD_IOMMU_GUEST_IR_VAPIC
:
1991 iommu_feature_enable(iommu
, CONTROL_GAM_EN
);
1993 case AMD_IOMMU_GUEST_IR_LEGACY_GA
:
1994 iommu_feature_enable(iommu
, CONTROL_GA_EN
);
1995 iommu
->irte_ops
= &irte_128_ops
;
1998 iommu
->irte_ops
= &irte_32_ops
;
2005 * This function finally enables all IOMMUs found in the system after
2006 * they have been initialized
2008 static void early_enable_iommus(void)
2010 struct amd_iommu
*iommu
;
2012 for_each_iommu(iommu
) {
2013 iommu_disable(iommu
);
2014 iommu_init_flags(iommu
);
2015 iommu_set_device_table(iommu
);
2016 iommu_enable_command_buffer(iommu
);
2017 iommu_enable_event_buffer(iommu
);
2018 iommu_set_exclusion_range(iommu
);
2019 iommu_enable_ga(iommu
);
2020 iommu_enable(iommu
);
2021 iommu_flush_all_caches(iommu
);
2024 #ifdef CONFIG_IRQ_REMAP
2025 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
2026 amd_iommu_irq_ops
.capability
|= (1 << IRQ_POSTING_CAP
);
2030 static void enable_iommus_v2(void)
2032 struct amd_iommu
*iommu
;
2034 for_each_iommu(iommu
) {
2035 iommu_enable_ppr_log(iommu
);
2036 iommu_enable_gt(iommu
);
2040 static void enable_iommus(void)
2042 early_enable_iommus();
2047 static void disable_iommus(void)
2049 struct amd_iommu
*iommu
;
2051 for_each_iommu(iommu
)
2052 iommu_disable(iommu
);
2054 #ifdef CONFIG_IRQ_REMAP
2055 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir
))
2056 amd_iommu_irq_ops
.capability
&= ~(1 << IRQ_POSTING_CAP
);
2061 * Suspend/Resume support
2062 * disable suspend until real resume implemented
2065 static void amd_iommu_resume(void)
2067 struct amd_iommu
*iommu
;
2069 for_each_iommu(iommu
)
2070 iommu_apply_resume_quirks(iommu
);
2072 /* re-load the hardware */
2075 amd_iommu_enable_interrupts();
2078 static int amd_iommu_suspend(void)
2080 /* disable IOMMUs to go out of the way for BIOS */
2086 static struct syscore_ops amd_iommu_syscore_ops
= {
2087 .suspend
= amd_iommu_suspend
,
2088 .resume
= amd_iommu_resume
,
2091 static void __init
free_on_init_error(void)
2093 free_pages((unsigned long)irq_lookup_table
,
2094 get_order(rlookup_table_size
));
2096 kmem_cache_destroy(amd_iommu_irq_cache
);
2097 amd_iommu_irq_cache
= NULL
;
2099 free_pages((unsigned long)amd_iommu_rlookup_table
,
2100 get_order(rlookup_table_size
));
2102 free_pages((unsigned long)amd_iommu_alias_table
,
2103 get_order(alias_table_size
));
2105 free_pages((unsigned long)amd_iommu_dev_table
,
2106 get_order(dev_table_size
));
2110 #ifdef CONFIG_GART_IOMMU
2112 * We failed to initialize the AMD IOMMU - try fallback to GART
2120 /* SB IOAPIC is always on this device in AMD systems */
2121 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2123 static bool __init
check_ioapic_information(void)
2125 const char *fw_bug
= FW_BUG
;
2126 bool ret
, has_sb_ioapic
;
2129 has_sb_ioapic
= false;
2133 * If we have map overrides on the kernel command line the
2134 * messages in this function might not describe firmware bugs
2135 * anymore - so be careful
2140 for (idx
= 0; idx
< nr_ioapics
; idx
++) {
2141 int devid
, id
= mpc_ioapic_id(idx
);
2143 devid
= get_ioapic_devid(id
);
2145 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
2148 } else if (devid
== IOAPIC_SB_DEVID
) {
2149 has_sb_ioapic
= true;
2154 if (!has_sb_ioapic
) {
2156 * We expect the SB IOAPIC to be listed in the IVRS
2157 * table. The system timer is connected to the SB IOAPIC
2158 * and if we don't have it in the list the system will
2159 * panic at boot time. This situation usually happens
2160 * when the BIOS is buggy and provides us the wrong
2161 * device id for the IOAPIC in the system.
2163 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug
);
2167 pr_err("AMD-Vi: Disabling interrupt remapping\n");
2172 static void __init
free_dma_resources(void)
2174 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap
,
2175 get_order(MAX_DOMAIN_ID
/8));
2181 * This is the hardware init function for AMD IOMMU in the system.
2182 * This function is called either from amd_iommu_init or from the interrupt
2183 * remapping setup code.
2185 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2188 * 1 pass) Discover the most comprehensive IVHD type to use.
2190 * 2 pass) Find the highest PCI device id the driver has to handle.
2191 * Upon this information the size of the data structures is
2192 * determined that needs to be allocated.
2194 * 3 pass) Initialize the data structures just allocated with the
2195 * information in the ACPI table about available AMD IOMMUs
2196 * in the system. It also maps the PCI devices in the
2197 * system to specific IOMMUs
2199 * 4 pass) After the basic data structures are allocated and
2200 * initialized we update them with information about memory
2201 * remapping requirements parsed out of the ACPI table in
2204 * After everything is set up the IOMMUs are enabled and the necessary
2205 * hotplug and suspend notifiers are registered.
2207 static int __init
early_amd_iommu_init(void)
2209 struct acpi_table_header
*ivrs_base
;
2210 acpi_size ivrs_size
;
2212 int i
, remap_cache_sz
, ret
= 0;
2214 if (!amd_iommu_detected
)
2217 status
= acpi_get_table_with_size("IVRS", 0, &ivrs_base
, &ivrs_size
);
2218 if (status
== AE_NOT_FOUND
)
2220 else if (ACPI_FAILURE(status
)) {
2221 const char *err
= acpi_format_exception(status
);
2222 pr_err("AMD-Vi: IVRS table error: %s\n", err
);
2227 * Validate checksum here so we don't need to do it when
2228 * we actually parse the table
2230 ret
= check_ivrs_checksum(ivrs_base
);
2234 amd_iommu_target_ivhd_type
= get_highest_supported_ivhd_type(ivrs_base
);
2235 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type
);
2238 * First parse ACPI tables to find the largest Bus/Dev/Func
2239 * we need to handle. Upon this information the shared data
2240 * structures for the IOMMUs in the system will be allocated
2242 ret
= find_last_devid_acpi(ivrs_base
);
2246 dev_table_size
= tbl_size(DEV_TABLE_ENTRY_SIZE
);
2247 alias_table_size
= tbl_size(ALIAS_TABLE_ENTRY_SIZE
);
2248 rlookup_table_size
= tbl_size(RLOOKUP_TABLE_ENTRY_SIZE
);
2250 /* Device table - directly used by all IOMMUs */
2252 amd_iommu_dev_table
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
2253 get_order(dev_table_size
));
2254 if (amd_iommu_dev_table
== NULL
)
2258 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2259 * IOMMU see for that device
2261 amd_iommu_alias_table
= (void *)__get_free_pages(GFP_KERNEL
,
2262 get_order(alias_table_size
));
2263 if (amd_iommu_alias_table
== NULL
)
2266 /* IOMMU rlookup table - find the IOMMU for a specific device */
2267 amd_iommu_rlookup_table
= (void *)__get_free_pages(
2268 GFP_KERNEL
| __GFP_ZERO
,
2269 get_order(rlookup_table_size
));
2270 if (amd_iommu_rlookup_table
== NULL
)
2273 amd_iommu_pd_alloc_bitmap
= (void *)__get_free_pages(
2274 GFP_KERNEL
| __GFP_ZERO
,
2275 get_order(MAX_DOMAIN_ID
/8));
2276 if (amd_iommu_pd_alloc_bitmap
== NULL
)
2280 * let all alias entries point to itself
2282 for (i
= 0; i
<= amd_iommu_last_bdf
; ++i
)
2283 amd_iommu_alias_table
[i
] = i
;
2286 * never allocate domain 0 because its used as the non-allocated and
2287 * error value placeholder
2289 __set_bit(0, amd_iommu_pd_alloc_bitmap
);
2291 spin_lock_init(&amd_iommu_pd_lock
);
2294 * now the data structures are allocated and basically initialized
2295 * start the real acpi table scan
2297 ret
= init_iommu_all(ivrs_base
);
2301 if (amd_iommu_irq_remap
)
2302 amd_iommu_irq_remap
= check_ioapic_information();
2304 if (amd_iommu_irq_remap
) {
2306 * Interrupt remapping enabled, create kmem_cache for the
2310 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir
))
2311 remap_cache_sz
= MAX_IRQS_PER_TABLE
* sizeof(u32
);
2313 remap_cache_sz
= MAX_IRQS_PER_TABLE
* (sizeof(u64
) * 2);
2314 amd_iommu_irq_cache
= kmem_cache_create("irq_remap_cache",
2316 IRQ_TABLE_ALIGNMENT
,
2318 if (!amd_iommu_irq_cache
)
2321 irq_lookup_table
= (void *)__get_free_pages(
2322 GFP_KERNEL
| __GFP_ZERO
,
2323 get_order(rlookup_table_size
));
2324 if (!irq_lookup_table
)
2328 ret
= init_memory_definitions(ivrs_base
);
2332 /* init the device table */
2333 init_device_table();
2336 /* Don't leak any ACPI memory */
2337 early_acpi_os_unmap_memory((char __iomem
*)ivrs_base
, ivrs_size
);
2343 static int amd_iommu_enable_interrupts(void)
2345 struct amd_iommu
*iommu
;
2348 for_each_iommu(iommu
) {
2349 ret
= iommu_init_msi(iommu
);
2358 static bool detect_ivrs(void)
2360 struct acpi_table_header
*ivrs_base
;
2361 acpi_size ivrs_size
;
2364 status
= acpi_get_table_with_size("IVRS", 0, &ivrs_base
, &ivrs_size
);
2365 if (status
== AE_NOT_FOUND
)
2367 else if (ACPI_FAILURE(status
)) {
2368 const char *err
= acpi_format_exception(status
);
2369 pr_err("AMD-Vi: IVRS table error: %s\n", err
);
2373 early_acpi_os_unmap_memory((char __iomem
*)ivrs_base
, ivrs_size
);
2375 /* Make sure ACS will be enabled during PCI probe */
2381 /****************************************************************************
2383 * AMD IOMMU Initialization State Machine
2385 ****************************************************************************/
2387 static int __init
state_next(void)
2391 switch (init_state
) {
2392 case IOMMU_START_STATE
:
2393 if (!detect_ivrs()) {
2394 init_state
= IOMMU_NOT_FOUND
;
2397 init_state
= IOMMU_IVRS_DETECTED
;
2400 case IOMMU_IVRS_DETECTED
:
2401 ret
= early_amd_iommu_init();
2402 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_ACPI_FINISHED
;
2404 case IOMMU_ACPI_FINISHED
:
2405 early_enable_iommus();
2406 register_syscore_ops(&amd_iommu_syscore_ops
);
2407 x86_platform
.iommu_shutdown
= disable_iommus
;
2408 init_state
= IOMMU_ENABLED
;
2411 ret
= amd_iommu_init_pci();
2412 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_PCI_INIT
;
2415 case IOMMU_PCI_INIT
:
2416 ret
= amd_iommu_enable_interrupts();
2417 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_INTERRUPTS_EN
;
2419 case IOMMU_INTERRUPTS_EN
:
2420 ret
= amd_iommu_init_dma_ops();
2421 init_state
= ret
? IOMMU_INIT_ERROR
: IOMMU_DMA_OPS
;
2424 init_state
= IOMMU_INITIALIZED
;
2426 case IOMMU_INITIALIZED
:
2429 case IOMMU_NOT_FOUND
:
2430 case IOMMU_INIT_ERROR
:
2431 /* Error states => do nothing */
2442 static int __init
iommu_go_to_state(enum iommu_init_state state
)
2446 while (init_state
!= state
) {
2448 if (init_state
== IOMMU_NOT_FOUND
||
2449 init_state
== IOMMU_INIT_ERROR
)
2456 #ifdef CONFIG_IRQ_REMAP
2457 int __init
amd_iommu_prepare(void)
2461 amd_iommu_irq_remap
= true;
2463 ret
= iommu_go_to_state(IOMMU_ACPI_FINISHED
);
2466 return amd_iommu_irq_remap
? 0 : -ENODEV
;
2469 int __init
amd_iommu_enable(void)
2473 ret
= iommu_go_to_state(IOMMU_ENABLED
);
2477 irq_remapping_enabled
= 1;
2482 void amd_iommu_disable(void)
2484 amd_iommu_suspend();
2487 int amd_iommu_reenable(int mode
)
2494 int __init
amd_iommu_enable_faulting(void)
2496 /* We enable MSI later when PCI is initialized */
2502 * This is the core init function for AMD IOMMU hardware in the system.
2503 * This function is called from the generic x86 DMA layer initialization
2506 static int __init
amd_iommu_init(void)
2510 ret
= iommu_go_to_state(IOMMU_INITIALIZED
);
2512 free_dma_resources();
2513 if (!irq_remapping_enabled
) {
2515 free_on_init_error();
2517 struct amd_iommu
*iommu
;
2519 uninit_device_table_dma();
2520 for_each_iommu(iommu
)
2521 iommu_flush_all_caches(iommu
);
2528 /****************************************************************************
2530 * Early detect code. This code runs at IOMMU detection time in the DMA
2531 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2534 ****************************************************************************/
2535 int __init
amd_iommu_detect(void)
2539 if (no_iommu
|| (iommu_detected
&& !gart_iommu_aperture
))
2542 if (amd_iommu_disabled
)
2545 ret
= iommu_go_to_state(IOMMU_IVRS_DETECTED
);
2549 amd_iommu_detected
= true;
2551 x86_init
.iommu
.iommu_init
= amd_iommu_init
;
2556 /****************************************************************************
2558 * Parsing functions for the AMD IOMMU specific kernel command line
2561 ****************************************************************************/
2563 static int __init
parse_amd_iommu_dump(char *str
)
2565 amd_iommu_dump
= true;
2570 static int __init
parse_amd_iommu_intr(char *str
)
2572 for (; *str
; ++str
) {
2573 if (strncmp(str
, "legacy", 6) == 0) {
2574 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_LEGACY
;
2577 if (strncmp(str
, "vapic", 5) == 0) {
2578 amd_iommu_guest_ir
= AMD_IOMMU_GUEST_IR_VAPIC
;
2585 static int __init
parse_amd_iommu_options(char *str
)
2587 for (; *str
; ++str
) {
2588 if (strncmp(str
, "fullflush", 9) == 0)
2589 amd_iommu_unmap_flush
= true;
2590 if (strncmp(str
, "off", 3) == 0)
2591 amd_iommu_disabled
= true;
2592 if (strncmp(str
, "force_isolation", 15) == 0)
2593 amd_iommu_force_isolation
= true;
2599 static int __init
parse_ivrs_ioapic(char *str
)
2601 unsigned int bus
, dev
, fn
;
2605 ret
= sscanf(str
, "[%d]=%x:%x.%x", &id
, &bus
, &dev
, &fn
);
2608 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str
);
2612 if (early_ioapic_map_size
== EARLY_MAP_SIZE
) {
2613 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2618 devid
= ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2620 cmdline_maps
= true;
2621 i
= early_ioapic_map_size
++;
2622 early_ioapic_map
[i
].id
= id
;
2623 early_ioapic_map
[i
].devid
= devid
;
2624 early_ioapic_map
[i
].cmd_line
= true;
2629 static int __init
parse_ivrs_hpet(char *str
)
2631 unsigned int bus
, dev
, fn
;
2635 ret
= sscanf(str
, "[%d]=%x:%x.%x", &id
, &bus
, &dev
, &fn
);
2638 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str
);
2642 if (early_hpet_map_size
== EARLY_MAP_SIZE
) {
2643 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2648 devid
= ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2650 cmdline_maps
= true;
2651 i
= early_hpet_map_size
++;
2652 early_hpet_map
[i
].id
= id
;
2653 early_hpet_map
[i
].devid
= devid
;
2654 early_hpet_map
[i
].cmd_line
= true;
2659 static int __init
parse_ivrs_acpihid(char *str
)
2662 char *hid
, *uid
, *p
;
2663 char acpiid
[ACPIHID_UID_LEN
+ ACPIHID_HID_LEN
] = {0};
2666 ret
= sscanf(str
, "[%x:%x.%x]=%s", &bus
, &dev
, &fn
, acpiid
);
2668 pr_err("AMD-Vi: Invalid command line: ivrs_acpihid(%s)\n", str
);
2673 hid
= strsep(&p
, ":");
2676 if (!hid
|| !(*hid
) || !uid
) {
2677 pr_err("AMD-Vi: Invalid command line: hid or uid\n");
2681 i
= early_acpihid_map_size
++;
2682 memcpy(early_acpihid_map
[i
].hid
, hid
, strlen(hid
));
2683 memcpy(early_acpihid_map
[i
].uid
, uid
, strlen(uid
));
2684 early_acpihid_map
[i
].devid
=
2685 ((bus
& 0xff) << 8) | ((dev
& 0x1f) << 3) | (fn
& 0x7);
2686 early_acpihid_map
[i
].cmd_line
= true;
2691 __setup("amd_iommu_dump", parse_amd_iommu_dump
);
2692 __setup("amd_iommu=", parse_amd_iommu_options
);
2693 __setup("amd_iommu_intr=", parse_amd_iommu_intr
);
2694 __setup("ivrs_ioapic", parse_ivrs_ioapic
);
2695 __setup("ivrs_hpet", parse_ivrs_hpet
);
2696 __setup("ivrs_acpihid", parse_ivrs_acpihid
);
2698 IOMMU_INIT_FINISH(amd_iommu_detect
,
2699 gart_iommu_hole_init
,
2703 bool amd_iommu_v2_supported(void)
2705 return amd_iommu_v2_present
;
2707 EXPORT_SYMBOL(amd_iommu_v2_supported
);
2709 /****************************************************************************
2711 * IOMMU EFR Performance Counter support functionality. This code allows
2712 * access to the IOMMU PC functionality.
2714 ****************************************************************************/
2716 u8
amd_iommu_pc_get_max_banks(u16 devid
)
2718 struct amd_iommu
*iommu
;
2721 /* locate the iommu governing the devid */
2722 iommu
= amd_iommu_rlookup_table
[devid
];
2724 ret
= iommu
->max_banks
;
2728 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks
);
2730 bool amd_iommu_pc_supported(void)
2732 return amd_iommu_pc_present
;
2734 EXPORT_SYMBOL(amd_iommu_pc_supported
);
2736 u8
amd_iommu_pc_get_max_counters(u16 devid
)
2738 struct amd_iommu
*iommu
;
2741 /* locate the iommu governing the devid */
2742 iommu
= amd_iommu_rlookup_table
[devid
];
2744 ret
= iommu
->max_counters
;
2748 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters
);
2750 static int iommu_pc_get_set_reg_val(struct amd_iommu
*iommu
,
2751 u8 bank
, u8 cntr
, u8 fxn
,
2752 u64
*value
, bool is_write
)
2757 /* Check for valid iommu and pc register indexing */
2758 if (WARN_ON((fxn
> 0x28) || (fxn
& 7)))
2761 offset
= (u32
)(((0x40|bank
) << 12) | (cntr
<< 8) | fxn
);
2763 /* Limit the offset to the hw defined mmio region aperture */
2764 max_offset_lim
= (u32
)(((0x40|iommu
->max_banks
) << 12) |
2765 (iommu
->max_counters
<< 8) | 0x28);
2766 if ((offset
< MMIO_CNTR_REG_OFFSET
) ||
2767 (offset
> max_offset_lim
))
2771 writel((u32
)*value
, iommu
->mmio_base
+ offset
);
2772 writel((*value
>> 32), iommu
->mmio_base
+ offset
+ 4);
2774 *value
= readl(iommu
->mmio_base
+ offset
+ 4);
2776 *value
= readl(iommu
->mmio_base
+ offset
);
2781 EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val
);
2783 int amd_iommu_pc_get_set_reg_val(u16 devid
, u8 bank
, u8 cntr
, u8 fxn
,
2784 u64
*value
, bool is_write
)
2786 struct amd_iommu
*iommu
= amd_iommu_rlookup_table
[devid
];
2788 /* Make sure the IOMMU PC resource is available */
2789 if (!amd_iommu_pc_present
|| iommu
== NULL
)
2792 return iommu_pc_get_set_reg_val(iommu
, bank
, cntr
, fxn
,