Linux 3.10.17
[linux/fpc-iii.git] / drivers / iommu / amd_iommu_init.c
blobbf51abb78deed14392a4c3e80d180c41d1e76546
1 /*
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 <acpi/acpi.h>
30 #include <asm/pci-direct.h>
31 #include <asm/iommu.h>
32 #include <asm/gart.h>
33 #include <asm/x86_init.h>
34 #include <asm/iommu_table.h>
35 #include <asm/io_apic.h>
36 #include <asm/irq_remapping.h>
38 #include "amd_iommu_proto.h"
39 #include "amd_iommu_types.h"
40 #include "irq_remapping.h"
43 * definitions for the ACPI scanning code
45 #define IVRS_HEADER_LENGTH 48
47 #define ACPI_IVHD_TYPE 0x10
48 #define ACPI_IVMD_TYPE_ALL 0x20
49 #define ACPI_IVMD_TYPE 0x21
50 #define ACPI_IVMD_TYPE_RANGE 0x22
52 #define IVHD_DEV_ALL 0x01
53 #define IVHD_DEV_SELECT 0x02
54 #define IVHD_DEV_SELECT_RANGE_START 0x03
55 #define IVHD_DEV_RANGE_END 0x04
56 #define IVHD_DEV_ALIAS 0x42
57 #define IVHD_DEV_ALIAS_RANGE 0x43
58 #define IVHD_DEV_EXT_SELECT 0x46
59 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
60 #define IVHD_DEV_SPECIAL 0x48
62 #define IVHD_SPECIAL_IOAPIC 1
63 #define IVHD_SPECIAL_HPET 2
65 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
66 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
67 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
68 #define IVHD_FLAG_ISOC_EN_MASK 0x08
70 #define IVMD_FLAG_EXCL_RANGE 0x08
71 #define IVMD_FLAG_UNITY_MAP 0x01
73 #define ACPI_DEVFLAG_INITPASS 0x01
74 #define ACPI_DEVFLAG_EXTINT 0x02
75 #define ACPI_DEVFLAG_NMI 0x04
76 #define ACPI_DEVFLAG_SYSMGT1 0x10
77 #define ACPI_DEVFLAG_SYSMGT2 0x20
78 #define ACPI_DEVFLAG_LINT0 0x40
79 #define ACPI_DEVFLAG_LINT1 0x80
80 #define ACPI_DEVFLAG_ATSDIS 0x10000000
83 * ACPI table definitions
85 * These data structures are laid over the table to parse the important values
86 * out of it.
90 * structure describing one IOMMU in the ACPI table. Typically followed by one
91 * or more ivhd_entrys.
93 struct ivhd_header {
94 u8 type;
95 u8 flags;
96 u16 length;
97 u16 devid;
98 u16 cap_ptr;
99 u64 mmio_phys;
100 u16 pci_seg;
101 u16 info;
102 u32 reserved;
103 } __attribute__((packed));
106 * A device entry describing which devices a specific IOMMU translates and
107 * which requestor ids they use.
109 struct ivhd_entry {
110 u8 type;
111 u16 devid;
112 u8 flags;
113 u32 ext;
114 } __attribute__((packed));
117 * An AMD IOMMU memory definition structure. It defines things like exclusion
118 * ranges for devices and regions that should be unity mapped.
120 struct ivmd_header {
121 u8 type;
122 u8 flags;
123 u16 length;
124 u16 devid;
125 u16 aux;
126 u64 resv;
127 u64 range_start;
128 u64 range_length;
129 } __attribute__((packed));
131 bool amd_iommu_dump;
132 bool amd_iommu_irq_remap __read_mostly;
134 static bool amd_iommu_detected;
135 static bool __initdata amd_iommu_disabled;
137 u16 amd_iommu_last_bdf; /* largest PCI device id we have
138 to handle */
139 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
140 we find in ACPI */
141 u32 amd_iommu_unmap_flush; /* if true, flush on every unmap */
143 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
144 system */
146 /* Array to assign indices to IOMMUs*/
147 struct amd_iommu *amd_iommus[MAX_IOMMUS];
148 int amd_iommus_present;
150 /* IOMMUs have a non-present cache? */
151 bool amd_iommu_np_cache __read_mostly;
152 bool amd_iommu_iotlb_sup __read_mostly = true;
154 u32 amd_iommu_max_pasids __read_mostly = ~0;
156 bool amd_iommu_v2_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
189 * quickly.
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 {
204 IOMMU_START_STATE,
205 IOMMU_IVRS_DETECTED,
206 IOMMU_ACPI_FINISHED,
207 IOMMU_ENABLED,
208 IOMMU_PCI_INIT,
209 IOMMU_INTERRUPTS_EN,
210 IOMMU_DMA_OPS,
211 IOMMU_INITIALIZED,
212 IOMMU_NOT_FOUND,
213 IOMMU_INIT_ERROR,
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);
240 return 1UL << shift;
243 /* Access to l1 and l2 indexed register spaces */
245 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
247 u32 val;
249 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
250 pci_read_config_dword(iommu->dev, 0xfc, &val);
251 return 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)
263 u32 val;
265 pci_write_config_dword(iommu->dev, 0xf0, address);
266 pci_read_config_dword(iommu->dev, 0xf4, &val);
267 return 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;
293 u64 entry;
295 if (!iommu->exclusion_start)
296 return;
298 entry = start | MMIO_EXCL_ENABLE_MASK;
299 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
300 &entry, sizeof(entry));
302 entry = limit;
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)
310 u64 entry;
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)
323 u32 ctrl;
325 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
326 ctrl |= (1 << bit);
327 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
330 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
332 u32 ctrl;
334 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
335 ctrl &= ~(1 << bit);
336 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
339 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
341 u32 ctrl;
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)
374 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
375 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
376 address);
377 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
378 return NULL;
381 return (u8 __iomem *)ioremap_nocache(address, MMIO_REGION_LENGTH);
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, MMIO_REGION_LENGTH);
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)
414 u32 cap;
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)));
419 return 0;
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;
431 p += sizeof(*h);
432 end += h->length;
434 find_last_devid_on_pci(PCI_BUS_NUM(h->devid),
435 PCI_SLOT(h->devid),
436 PCI_FUNC(h->devid),
437 h->cap_ptr);
439 while (p < end) {
440 dev = (struct ivhd_entry *)p;
441 switch (dev->type) {
442 case IVHD_DEV_SELECT:
443 case IVHD_DEV_RANGE_END:
444 case IVHD_DEV_ALIAS:
445 case IVHD_DEV_EXT_SELECT:
446 /* all the above subfield types refer to device ids */
447 update_last_devid(dev->devid);
448 break;
449 default:
450 break;
452 p += ivhd_entry_length(p);
455 WARN_ON(p != end);
457 return 0;
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)
467 int i;
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)
476 checksum += p[i];
477 if (checksum != 0)
478 /* ACPI table corrupt */
479 return -ENODEV;
481 p += IVRS_HEADER_LENGTH;
483 end += table->length;
484 while (p < end) {
485 h = (struct ivhd_header *)p;
486 switch (h->type) {
487 case ACPI_IVHD_TYPE:
488 find_last_devid_from_ivhd(h);
489 break;
490 default:
491 break;
493 p += h->length;
495 WARN_ON(p != end);
497 return 0;
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
512 * asynchronously
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));
519 if (cmd_buf == NULL)
520 return NULL;
522 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
524 return cmd_buf;
528 * This function resets the command buffer if the IOMMU stopped fetching
529 * commands from it.
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
543 * enables it.
545 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
547 u64 entry;
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)
574 return 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)
583 u64 entry;
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)
611 return NULL;
613 return iommu->ppr_log;
616 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
618 u64 entry;
620 if (iommu->ppr_log == NULL)
621 return;
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)
639 return;
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))
647 return;
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)
672 int sysmgt;
674 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
675 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
677 if (sysmgt == 0x01)
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)
720 list = &ioapic_map;
721 else if (type == IVHD_SPECIAL_HPET)
722 list = &hpet_map;
723 else
724 return -EINVAL;
726 list_for_each_entry(entry, list, list) {
727 if (!(entry->id == id && entry->cmd_line))
728 continue;
730 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
731 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
733 return 0;
736 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
737 if (!entry)
738 return -ENOMEM;
740 entry->id = id;
741 entry->devid = devid;
742 entry->cmd_line = cmd_line;
744 list_add_tail(&entry->list, list);
746 return 0;
749 static int __init add_early_maps(void)
751 int i, ret;
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);
758 if (ret)
759 return ret;
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);
767 if (ret)
768 return ret;
771 return 0;
775 * Reads the device exclusion range from ACPI and initializes the IOMMU with
776 * it
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))
783 return;
785 if (iommu) {
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(m->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)
804 u8 *p = (u8 *)h;
805 u8 *end = p, flags = 0;
806 u16 devid = 0, devid_start = 0, devid_to = 0;
807 u32 dev_i, ext_flags = 0;
808 bool alias = false;
809 struct ivhd_entry *e;
810 int ret;
813 ret = add_early_maps();
814 if (ret)
815 return ret;
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);
826 end += h->length;
829 while (p < end) {
830 e = (struct ivhd_entry *)p;
831 switch (e->type) {
832 case IVHD_DEV_ALL:
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),
842 e->flags);
844 for (dev_i = iommu->first_device;
845 dev_i <= iommu->last_device; ++dev_i)
846 set_dev_entry_from_acpi(iommu, dev_i,
847 e->flags, 0);
848 break;
849 case IVHD_DEV_SELECT:
851 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
852 "flags: %02x\n",
853 PCI_BUS_NUM(e->devid),
854 PCI_SLOT(e->devid),
855 PCI_FUNC(e->devid),
856 e->flags);
858 devid = e->devid;
859 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
860 break;
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),
866 PCI_SLOT(e->devid),
867 PCI_FUNC(e->devid),
868 e->flags);
870 devid_start = e->devid;
871 flags = e->flags;
872 ext_flags = 0;
873 alias = false;
874 break;
875 case IVHD_DEV_ALIAS:
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),
880 PCI_SLOT(e->devid),
881 PCI_FUNC(e->devid),
882 e->flags,
883 PCI_BUS_NUM(e->ext >> 8),
884 PCI_SLOT(e->ext >> 8),
885 PCI_FUNC(e->ext >> 8));
887 devid = e->devid;
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;
892 break;
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),
899 PCI_SLOT(e->devid),
900 PCI_FUNC(e->devid),
901 e->flags,
902 PCI_BUS_NUM(e->ext >> 8),
903 PCI_SLOT(e->ext >> 8),
904 PCI_FUNC(e->ext >> 8));
906 devid_start = e->devid;
907 flags = e->flags;
908 devid_to = e->ext >> 8;
909 ext_flags = 0;
910 alias = true;
911 break;
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),
917 PCI_SLOT(e->devid),
918 PCI_FUNC(e->devid),
919 e->flags, e->ext);
921 devid = e->devid;
922 set_dev_entry_from_acpi(iommu, devid, e->flags,
923 e->ext);
924 break;
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),
930 PCI_SLOT(e->devid),
931 PCI_FUNC(e->devid),
932 e->flags, e->ext);
934 devid_start = e->devid;
935 flags = e->flags;
936 ext_flags = e->ext;
937 alias = false;
938 break;
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),
943 PCI_SLOT(e->devid),
944 PCI_FUNC(e->devid));
946 devid = e->devid;
947 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
948 if (alias) {
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,
954 flags, ext_flags);
956 break;
957 case IVHD_DEV_SPECIAL: {
958 u8 handle, type;
959 const char *var;
960 u16 devid;
961 int ret;
963 handle = e->ext & 0xff;
964 devid = (e->ext >> 8) & 0xffff;
965 type = (e->ext >> 24) & 0xff;
967 if (type == IVHD_SPECIAL_IOAPIC)
968 var = "IOAPIC";
969 else if (type == IVHD_SPECIAL_HPET)
970 var = "HPET";
971 else
972 var = "UNKNOWN";
974 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
975 var, (int)handle,
976 PCI_BUS_NUM(devid),
977 PCI_SLOT(devid),
978 PCI_FUNC(devid));
980 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
981 ret = add_special_device(type, handle, devid, false);
982 if (ret)
983 return ret;
984 break;
986 default:
987 break;
990 p += ivhd_entry_length(p);
993 return 0;
996 /* Initializes the device->iommu mapping for the driver */
997 static int __init init_iommu_devices(struct amd_iommu *iommu)
999 u32 i;
1001 for (i = iommu->first_device; i <= iommu->last_device; ++i)
1002 set_iommu_for_device(iommu, i);
1004 return 0;
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);
1022 kfree(iommu);
1027 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1028 * Workaround:
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)
1034 u32 value;
1036 if ((boot_cpu_data.x86 != 0x15) ||
1037 (boot_cpu_data.x86_model < 0x10) ||
1038 (boot_cpu_data.x86_model > 0x1f))
1039 return;
1041 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1042 pci_read_config_dword(iommu->dev, 0xf4, &value);
1044 if (value & BIT(2))
1045 return;
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)
1065 int ret;
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");
1075 return -ENOSYS;
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;
1088 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
1089 if (!iommu->mmio_base)
1090 return -ENOMEM;
1092 iommu->cmd_buf = alloc_command_buffer(iommu);
1093 if (!iommu->cmd_buf)
1094 return -ENOMEM;
1096 iommu->evt_buf = alloc_event_buffer(iommu);
1097 if (!iommu->evt_buf)
1098 return -ENOMEM;
1100 iommu->int_enabled = false;
1102 ret = init_iommu_from_acpi(iommu, h);
1103 if (ret)
1104 return ret;
1107 * Make sure IOMMU is not considered to translate itself. The IVRS
1108 * table tells us so, but this is a lie!
1110 amd_iommu_rlookup_table[iommu->devid] = NULL;
1112 init_iommu_devices(iommu);
1114 return 0;
1118 * Iterates over all IOMMU entries in the ACPI table, allocates the
1119 * IOMMU structure and initializes it with init_iommu_one()
1121 static int __init init_iommu_all(struct acpi_table_header *table)
1123 u8 *p = (u8 *)table, *end = (u8 *)table;
1124 struct ivhd_header *h;
1125 struct amd_iommu *iommu;
1126 int ret;
1128 end += table->length;
1129 p += IVRS_HEADER_LENGTH;
1131 while (p < end) {
1132 h = (struct ivhd_header *)p;
1133 switch (*p) {
1134 case ACPI_IVHD_TYPE:
1136 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1137 "seg: %d flags: %01x info %04x\n",
1138 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1139 PCI_FUNC(h->devid), h->cap_ptr,
1140 h->pci_seg, h->flags, h->info);
1141 DUMP_printk(" mmio-addr: %016llx\n",
1142 h->mmio_phys);
1144 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1145 if (iommu == NULL)
1146 return -ENOMEM;
1148 ret = init_iommu_one(iommu, h);
1149 if (ret)
1150 return ret;
1151 break;
1152 default:
1153 break;
1155 p += h->length;
1158 WARN_ON(p != end);
1160 return 0;
1163 static int iommu_init_pci(struct amd_iommu *iommu)
1165 int cap_ptr = iommu->cap_ptr;
1166 u32 range, misc, low, high;
1168 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1169 iommu->devid & 0xff);
1170 if (!iommu->dev)
1171 return -ENODEV;
1173 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1174 &iommu->cap);
1175 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1176 &range);
1177 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1178 &misc);
1180 iommu->first_device = PCI_DEVID(MMIO_GET_BUS(range),
1181 MMIO_GET_FD(range));
1182 iommu->last_device = PCI_DEVID(MMIO_GET_BUS(range),
1183 MMIO_GET_LD(range));
1185 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1186 amd_iommu_iotlb_sup = false;
1188 /* read extended feature bits */
1189 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1190 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1192 iommu->features = ((u64)high << 32) | low;
1194 if (iommu_feature(iommu, FEATURE_GT)) {
1195 int glxval;
1196 u32 pasids;
1197 u64 shift;
1199 shift = iommu->features & FEATURE_PASID_MASK;
1200 shift >>= FEATURE_PASID_SHIFT;
1201 pasids = (1 << shift);
1203 amd_iommu_max_pasids = min(amd_iommu_max_pasids, pasids);
1205 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1206 glxval >>= FEATURE_GLXVAL_SHIFT;
1208 if (amd_iommu_max_glx_val == -1)
1209 amd_iommu_max_glx_val = glxval;
1210 else
1211 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1214 if (iommu_feature(iommu, FEATURE_GT) &&
1215 iommu_feature(iommu, FEATURE_PPR)) {
1216 iommu->is_iommu_v2 = true;
1217 amd_iommu_v2_present = true;
1220 if (iommu_feature(iommu, FEATURE_PPR)) {
1221 iommu->ppr_log = alloc_ppr_log(iommu);
1222 if (!iommu->ppr_log)
1223 return -ENOMEM;
1226 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1227 amd_iommu_np_cache = true;
1229 if (is_rd890_iommu(iommu->dev)) {
1230 int i, j;
1232 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1233 PCI_DEVFN(0, 0));
1236 * Some rd890 systems may not be fully reconfigured by the
1237 * BIOS, so it's necessary for us to store this information so
1238 * it can be reprogrammed on resume
1240 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1241 &iommu->stored_addr_lo);
1242 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1243 &iommu->stored_addr_hi);
1245 /* Low bit locks writes to configuration space */
1246 iommu->stored_addr_lo &= ~1;
1248 for (i = 0; i < 6; i++)
1249 for (j = 0; j < 0x12; j++)
1250 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1252 for (i = 0; i < 0x83; i++)
1253 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1256 amd_iommu_erratum_746_workaround(iommu);
1258 return pci_enable_device(iommu->dev);
1261 static void print_iommu_info(void)
1263 static const char * const feat_str[] = {
1264 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1265 "IA", "GA", "HE", "PC"
1267 struct amd_iommu *iommu;
1269 for_each_iommu(iommu) {
1270 int i;
1272 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1273 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1275 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1276 pr_info("AMD-Vi: Extended features: ");
1277 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1278 if (iommu_feature(iommu, (1ULL << i)))
1279 pr_cont(" %s", feat_str[i]);
1281 pr_cont("\n");
1284 if (irq_remapping_enabled)
1285 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1288 static int __init amd_iommu_init_pci(void)
1290 struct amd_iommu *iommu;
1291 int ret = 0;
1293 for_each_iommu(iommu) {
1294 ret = iommu_init_pci(iommu);
1295 if (ret)
1296 break;
1299 ret = amd_iommu_init_devices();
1301 print_iommu_info();
1303 return ret;
1306 /****************************************************************************
1308 * The following functions initialize the MSI interrupts for all IOMMUs
1309 * in the system. It's a bit challenging because there could be multiple
1310 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1311 * pci_dev.
1313 ****************************************************************************/
1315 static int iommu_setup_msi(struct amd_iommu *iommu)
1317 int r;
1319 r = pci_enable_msi(iommu->dev);
1320 if (r)
1321 return r;
1323 r = request_threaded_irq(iommu->dev->irq,
1324 amd_iommu_int_handler,
1325 amd_iommu_int_thread,
1326 0, "AMD-Vi",
1327 iommu);
1329 if (r) {
1330 pci_disable_msi(iommu->dev);
1331 return r;
1334 iommu->int_enabled = true;
1336 return 0;
1339 static int iommu_init_msi(struct amd_iommu *iommu)
1341 int ret;
1343 if (iommu->int_enabled)
1344 goto enable_faults;
1346 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
1347 ret = iommu_setup_msi(iommu);
1348 else
1349 ret = -ENODEV;
1351 if (ret)
1352 return ret;
1354 enable_faults:
1355 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1357 if (iommu->ppr_log != NULL)
1358 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1360 return 0;
1363 /****************************************************************************
1365 * The next functions belong to the third pass of parsing the ACPI
1366 * table. In this last pass the memory mapping requirements are
1367 * gathered (like exclusion and unity mapping ranges).
1369 ****************************************************************************/
1371 static void __init free_unity_maps(void)
1373 struct unity_map_entry *entry, *next;
1375 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1376 list_del(&entry->list);
1377 kfree(entry);
1381 /* called when we find an exclusion range definition in ACPI */
1382 static int __init init_exclusion_range(struct ivmd_header *m)
1384 int i;
1386 switch (m->type) {
1387 case ACPI_IVMD_TYPE:
1388 set_device_exclusion_range(m->devid, m);
1389 break;
1390 case ACPI_IVMD_TYPE_ALL:
1391 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1392 set_device_exclusion_range(i, m);
1393 break;
1394 case ACPI_IVMD_TYPE_RANGE:
1395 for (i = m->devid; i <= m->aux; ++i)
1396 set_device_exclusion_range(i, m);
1397 break;
1398 default:
1399 break;
1402 return 0;
1405 /* called for unity map ACPI definition */
1406 static int __init init_unity_map_range(struct ivmd_header *m)
1408 struct unity_map_entry *e = NULL;
1409 char *s;
1411 e = kzalloc(sizeof(*e), GFP_KERNEL);
1412 if (e == NULL)
1413 return -ENOMEM;
1415 switch (m->type) {
1416 default:
1417 kfree(e);
1418 return 0;
1419 case ACPI_IVMD_TYPE:
1420 s = "IVMD_TYPEi\t\t\t";
1421 e->devid_start = e->devid_end = m->devid;
1422 break;
1423 case ACPI_IVMD_TYPE_ALL:
1424 s = "IVMD_TYPE_ALL\t\t";
1425 e->devid_start = 0;
1426 e->devid_end = amd_iommu_last_bdf;
1427 break;
1428 case ACPI_IVMD_TYPE_RANGE:
1429 s = "IVMD_TYPE_RANGE\t\t";
1430 e->devid_start = m->devid;
1431 e->devid_end = m->aux;
1432 break;
1434 e->address_start = PAGE_ALIGN(m->range_start);
1435 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1436 e->prot = m->flags >> 1;
1438 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1439 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1440 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1441 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1442 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1443 e->address_start, e->address_end, m->flags);
1445 list_add_tail(&e->list, &amd_iommu_unity_map);
1447 return 0;
1450 /* iterates over all memory definitions we find in the ACPI table */
1451 static int __init init_memory_definitions(struct acpi_table_header *table)
1453 u8 *p = (u8 *)table, *end = (u8 *)table;
1454 struct ivmd_header *m;
1456 end += table->length;
1457 p += IVRS_HEADER_LENGTH;
1459 while (p < end) {
1460 m = (struct ivmd_header *)p;
1461 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1462 init_exclusion_range(m);
1463 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1464 init_unity_map_range(m);
1466 p += m->length;
1469 return 0;
1473 * Init the device table to not allow DMA access for devices and
1474 * suppress all page faults
1476 static void init_device_table_dma(void)
1478 u32 devid;
1480 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1481 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1482 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1486 static void __init uninit_device_table_dma(void)
1488 u32 devid;
1490 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1491 amd_iommu_dev_table[devid].data[0] = 0ULL;
1492 amd_iommu_dev_table[devid].data[1] = 0ULL;
1496 static void init_device_table(void)
1498 u32 devid;
1500 if (!amd_iommu_irq_remap)
1501 return;
1503 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1504 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1507 static void iommu_init_flags(struct amd_iommu *iommu)
1509 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1510 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1511 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1513 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1514 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1515 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1517 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1518 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1519 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1521 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1522 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1523 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1526 * make IOMMU memory accesses cache coherent
1528 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1530 /* Set IOTLB invalidation timeout to 1s */
1531 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1534 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1536 int i, j;
1537 u32 ioc_feature_control;
1538 struct pci_dev *pdev = iommu->root_pdev;
1540 /* RD890 BIOSes may not have completely reconfigured the iommu */
1541 if (!is_rd890_iommu(iommu->dev) || !pdev)
1542 return;
1545 * First, we need to ensure that the iommu is enabled. This is
1546 * controlled by a register in the northbridge
1549 /* Select Northbridge indirect register 0x75 and enable writing */
1550 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1551 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1553 /* Enable the iommu */
1554 if (!(ioc_feature_control & 0x1))
1555 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1557 /* Restore the iommu BAR */
1558 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1559 iommu->stored_addr_lo);
1560 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1561 iommu->stored_addr_hi);
1563 /* Restore the l1 indirect regs for each of the 6 l1s */
1564 for (i = 0; i < 6; i++)
1565 for (j = 0; j < 0x12; j++)
1566 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1568 /* Restore the l2 indirect regs */
1569 for (i = 0; i < 0x83; i++)
1570 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1572 /* Lock PCI setup registers */
1573 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1574 iommu->stored_addr_lo | 1);
1578 * This function finally enables all IOMMUs found in the system after
1579 * they have been initialized
1581 static void early_enable_iommus(void)
1583 struct amd_iommu *iommu;
1585 for_each_iommu(iommu) {
1586 iommu_disable(iommu);
1587 iommu_init_flags(iommu);
1588 iommu_set_device_table(iommu);
1589 iommu_enable_command_buffer(iommu);
1590 iommu_enable_event_buffer(iommu);
1591 iommu_set_exclusion_range(iommu);
1592 iommu_enable(iommu);
1593 iommu_flush_all_caches(iommu);
1597 static void enable_iommus_v2(void)
1599 struct amd_iommu *iommu;
1601 for_each_iommu(iommu) {
1602 iommu_enable_ppr_log(iommu);
1603 iommu_enable_gt(iommu);
1607 static void enable_iommus(void)
1609 early_enable_iommus();
1611 enable_iommus_v2();
1614 static void disable_iommus(void)
1616 struct amd_iommu *iommu;
1618 for_each_iommu(iommu)
1619 iommu_disable(iommu);
1623 * Suspend/Resume support
1624 * disable suspend until real resume implemented
1627 static void amd_iommu_resume(void)
1629 struct amd_iommu *iommu;
1631 for_each_iommu(iommu)
1632 iommu_apply_resume_quirks(iommu);
1634 /* re-load the hardware */
1635 enable_iommus();
1637 amd_iommu_enable_interrupts();
1640 static int amd_iommu_suspend(void)
1642 /* disable IOMMUs to go out of the way for BIOS */
1643 disable_iommus();
1645 return 0;
1648 static struct syscore_ops amd_iommu_syscore_ops = {
1649 .suspend = amd_iommu_suspend,
1650 .resume = amd_iommu_resume,
1653 static void __init free_on_init_error(void)
1655 free_pages((unsigned long)irq_lookup_table,
1656 get_order(rlookup_table_size));
1658 if (amd_iommu_irq_cache) {
1659 kmem_cache_destroy(amd_iommu_irq_cache);
1660 amd_iommu_irq_cache = NULL;
1664 free_pages((unsigned long)amd_iommu_rlookup_table,
1665 get_order(rlookup_table_size));
1667 free_pages((unsigned long)amd_iommu_alias_table,
1668 get_order(alias_table_size));
1670 free_pages((unsigned long)amd_iommu_dev_table,
1671 get_order(dev_table_size));
1673 free_iommu_all();
1675 #ifdef CONFIG_GART_IOMMU
1677 * We failed to initialize the AMD IOMMU - try fallback to GART
1678 * if possible.
1680 gart_iommu_init();
1682 #endif
1685 /* SB IOAPIC is always on this device in AMD systems */
1686 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1688 static bool __init check_ioapic_information(void)
1690 const char *fw_bug = FW_BUG;
1691 bool ret, has_sb_ioapic;
1692 int idx;
1694 has_sb_ioapic = false;
1695 ret = false;
1698 * If we have map overrides on the kernel command line the
1699 * messages in this function might not describe firmware bugs
1700 * anymore - so be careful
1702 if (cmdline_maps)
1703 fw_bug = "";
1705 for (idx = 0; idx < nr_ioapics; idx++) {
1706 int devid, id = mpc_ioapic_id(idx);
1708 devid = get_ioapic_devid(id);
1709 if (devid < 0) {
1710 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1711 fw_bug, id);
1712 ret = false;
1713 } else if (devid == IOAPIC_SB_DEVID) {
1714 has_sb_ioapic = true;
1715 ret = true;
1719 if (!has_sb_ioapic) {
1721 * We expect the SB IOAPIC to be listed in the IVRS
1722 * table. The system timer is connected to the SB IOAPIC
1723 * and if we don't have it in the list the system will
1724 * panic at boot time. This situation usually happens
1725 * when the BIOS is buggy and provides us the wrong
1726 * device id for the IOAPIC in the system.
1728 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
1731 if (!ret)
1732 pr_err("AMD-Vi: Disabling interrupt remapping\n");
1734 return ret;
1737 static void __init free_dma_resources(void)
1739 amd_iommu_uninit_devices();
1741 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1742 get_order(MAX_DOMAIN_ID/8));
1744 free_unity_maps();
1748 * This is the hardware init function for AMD IOMMU in the system.
1749 * This function is called either from amd_iommu_init or from the interrupt
1750 * remapping setup code.
1752 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1753 * three times:
1755 * 1 pass) Find the highest PCI device id the driver has to handle.
1756 * Upon this information the size of the data structures is
1757 * determined that needs to be allocated.
1759 * 2 pass) Initialize the data structures just allocated with the
1760 * information in the ACPI table about available AMD IOMMUs
1761 * in the system. It also maps the PCI devices in the
1762 * system to specific IOMMUs
1764 * 3 pass) After the basic data structures are allocated and
1765 * initialized we update them with information about memory
1766 * remapping requirements parsed out of the ACPI table in
1767 * this last pass.
1769 * After everything is set up the IOMMUs are enabled and the necessary
1770 * hotplug and suspend notifiers are registered.
1772 static int __init early_amd_iommu_init(void)
1774 struct acpi_table_header *ivrs_base;
1775 acpi_size ivrs_size;
1776 acpi_status status;
1777 int i, ret = 0;
1779 if (!amd_iommu_detected)
1780 return -ENODEV;
1782 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1783 if (status == AE_NOT_FOUND)
1784 return -ENODEV;
1785 else if (ACPI_FAILURE(status)) {
1786 const char *err = acpi_format_exception(status);
1787 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1788 return -EINVAL;
1792 * First parse ACPI tables to find the largest Bus/Dev/Func
1793 * we need to handle. Upon this information the shared data
1794 * structures for the IOMMUs in the system will be allocated
1796 ret = find_last_devid_acpi(ivrs_base);
1797 if (ret)
1798 goto out;
1800 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1801 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1802 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1804 /* Device table - directly used by all IOMMUs */
1805 ret = -ENOMEM;
1806 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1807 get_order(dev_table_size));
1808 if (amd_iommu_dev_table == NULL)
1809 goto out;
1812 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1813 * IOMMU see for that device
1815 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1816 get_order(alias_table_size));
1817 if (amd_iommu_alias_table == NULL)
1818 goto out;
1820 /* IOMMU rlookup table - find the IOMMU for a specific device */
1821 amd_iommu_rlookup_table = (void *)__get_free_pages(
1822 GFP_KERNEL | __GFP_ZERO,
1823 get_order(rlookup_table_size));
1824 if (amd_iommu_rlookup_table == NULL)
1825 goto out;
1827 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1828 GFP_KERNEL | __GFP_ZERO,
1829 get_order(MAX_DOMAIN_ID/8));
1830 if (amd_iommu_pd_alloc_bitmap == NULL)
1831 goto out;
1834 * let all alias entries point to itself
1836 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1837 amd_iommu_alias_table[i] = i;
1840 * never allocate domain 0 because its used as the non-allocated and
1841 * error value placeholder
1843 amd_iommu_pd_alloc_bitmap[0] = 1;
1845 spin_lock_init(&amd_iommu_pd_lock);
1848 * now the data structures are allocated and basically initialized
1849 * start the real acpi table scan
1851 ret = init_iommu_all(ivrs_base);
1852 if (ret)
1853 goto out;
1855 if (amd_iommu_irq_remap)
1856 amd_iommu_irq_remap = check_ioapic_information();
1858 if (amd_iommu_irq_remap) {
1860 * Interrupt remapping enabled, create kmem_cache for the
1861 * remapping tables.
1863 ret = -ENOMEM;
1864 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
1865 MAX_IRQS_PER_TABLE * sizeof(u32),
1866 IRQ_TABLE_ALIGNMENT,
1867 0, NULL);
1868 if (!amd_iommu_irq_cache)
1869 goto out;
1871 irq_lookup_table = (void *)__get_free_pages(
1872 GFP_KERNEL | __GFP_ZERO,
1873 get_order(rlookup_table_size));
1874 if (!irq_lookup_table)
1875 goto out;
1878 ret = init_memory_definitions(ivrs_base);
1879 if (ret)
1880 goto out;
1882 /* init the device table */
1883 init_device_table();
1885 out:
1886 /* Don't leak any ACPI memory */
1887 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1888 ivrs_base = NULL;
1890 return ret;
1893 static int amd_iommu_enable_interrupts(void)
1895 struct amd_iommu *iommu;
1896 int ret = 0;
1898 for_each_iommu(iommu) {
1899 ret = iommu_init_msi(iommu);
1900 if (ret)
1901 goto out;
1904 out:
1905 return ret;
1908 static bool detect_ivrs(void)
1910 struct acpi_table_header *ivrs_base;
1911 acpi_size ivrs_size;
1912 acpi_status status;
1914 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1915 if (status == AE_NOT_FOUND)
1916 return false;
1917 else if (ACPI_FAILURE(status)) {
1918 const char *err = acpi_format_exception(status);
1919 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1920 return false;
1923 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1925 /* Make sure ACS will be enabled during PCI probe */
1926 pci_request_acs();
1928 if (!disable_irq_remap)
1929 amd_iommu_irq_remap = true;
1931 return true;
1934 static int amd_iommu_init_dma(void)
1936 struct amd_iommu *iommu;
1937 int ret;
1939 if (iommu_pass_through)
1940 ret = amd_iommu_init_passthrough();
1941 else
1942 ret = amd_iommu_init_dma_ops();
1944 if (ret)
1945 return ret;
1947 init_device_table_dma();
1949 for_each_iommu(iommu)
1950 iommu_flush_all_caches(iommu);
1952 amd_iommu_init_api();
1954 amd_iommu_init_notifier();
1956 return 0;
1959 /****************************************************************************
1961 * AMD IOMMU Initialization State Machine
1963 ****************************************************************************/
1965 static int __init state_next(void)
1967 int ret = 0;
1969 switch (init_state) {
1970 case IOMMU_START_STATE:
1971 if (!detect_ivrs()) {
1972 init_state = IOMMU_NOT_FOUND;
1973 ret = -ENODEV;
1974 } else {
1975 init_state = IOMMU_IVRS_DETECTED;
1977 break;
1978 case IOMMU_IVRS_DETECTED:
1979 ret = early_amd_iommu_init();
1980 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
1981 break;
1982 case IOMMU_ACPI_FINISHED:
1983 early_enable_iommus();
1984 register_syscore_ops(&amd_iommu_syscore_ops);
1985 x86_platform.iommu_shutdown = disable_iommus;
1986 init_state = IOMMU_ENABLED;
1987 break;
1988 case IOMMU_ENABLED:
1989 ret = amd_iommu_init_pci();
1990 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
1991 enable_iommus_v2();
1992 break;
1993 case IOMMU_PCI_INIT:
1994 ret = amd_iommu_enable_interrupts();
1995 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
1996 break;
1997 case IOMMU_INTERRUPTS_EN:
1998 ret = amd_iommu_init_dma();
1999 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2000 break;
2001 case IOMMU_DMA_OPS:
2002 init_state = IOMMU_INITIALIZED;
2003 break;
2004 case IOMMU_INITIALIZED:
2005 /* Nothing to do */
2006 break;
2007 case IOMMU_NOT_FOUND:
2008 case IOMMU_INIT_ERROR:
2009 /* Error states => do nothing */
2010 ret = -EINVAL;
2011 break;
2012 default:
2013 /* Unknown state */
2014 BUG();
2017 return ret;
2020 static int __init iommu_go_to_state(enum iommu_init_state state)
2022 int ret = 0;
2024 while (init_state != state) {
2025 ret = state_next();
2026 if (init_state == IOMMU_NOT_FOUND ||
2027 init_state == IOMMU_INIT_ERROR)
2028 break;
2031 return ret;
2034 #ifdef CONFIG_IRQ_REMAP
2035 int __init amd_iommu_prepare(void)
2037 return iommu_go_to_state(IOMMU_ACPI_FINISHED);
2040 int __init amd_iommu_supported(void)
2042 return amd_iommu_irq_remap ? 1 : 0;
2045 int __init amd_iommu_enable(void)
2047 int ret;
2049 ret = iommu_go_to_state(IOMMU_ENABLED);
2050 if (ret)
2051 return ret;
2053 irq_remapping_enabled = 1;
2055 return 0;
2058 void amd_iommu_disable(void)
2060 amd_iommu_suspend();
2063 int amd_iommu_reenable(int mode)
2065 amd_iommu_resume();
2067 return 0;
2070 int __init amd_iommu_enable_faulting(void)
2072 /* We enable MSI later when PCI is initialized */
2073 return 0;
2075 #endif
2078 * This is the core init function for AMD IOMMU hardware in the system.
2079 * This function is called from the generic x86 DMA layer initialization
2080 * code.
2082 static int __init amd_iommu_init(void)
2084 int ret;
2086 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2087 if (ret) {
2088 free_dma_resources();
2089 if (!irq_remapping_enabled) {
2090 disable_iommus();
2091 free_on_init_error();
2092 } else {
2093 struct amd_iommu *iommu;
2095 uninit_device_table_dma();
2096 for_each_iommu(iommu)
2097 iommu_flush_all_caches(iommu);
2101 return ret;
2104 /****************************************************************************
2106 * Early detect code. This code runs at IOMMU detection time in the DMA
2107 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2108 * IOMMUs
2110 ****************************************************************************/
2111 int __init amd_iommu_detect(void)
2113 int ret;
2115 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2116 return -ENODEV;
2118 if (amd_iommu_disabled)
2119 return -ENODEV;
2121 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2122 if (ret)
2123 return ret;
2125 amd_iommu_detected = true;
2126 iommu_detected = 1;
2127 x86_init.iommu.iommu_init = amd_iommu_init;
2129 return 0;
2132 /****************************************************************************
2134 * Parsing functions for the AMD IOMMU specific kernel command line
2135 * options.
2137 ****************************************************************************/
2139 static int __init parse_amd_iommu_dump(char *str)
2141 amd_iommu_dump = true;
2143 return 1;
2146 static int __init parse_amd_iommu_options(char *str)
2148 for (; *str; ++str) {
2149 if (strncmp(str, "fullflush", 9) == 0)
2150 amd_iommu_unmap_flush = true;
2151 if (strncmp(str, "off", 3) == 0)
2152 amd_iommu_disabled = true;
2153 if (strncmp(str, "force_isolation", 15) == 0)
2154 amd_iommu_force_isolation = true;
2157 return 1;
2160 static int __init parse_ivrs_ioapic(char *str)
2162 unsigned int bus, dev, fn;
2163 int ret, id, i;
2164 u16 devid;
2166 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2168 if (ret != 4) {
2169 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2170 return 1;
2173 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2174 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2175 str);
2176 return 1;
2179 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2181 cmdline_maps = true;
2182 i = early_ioapic_map_size++;
2183 early_ioapic_map[i].id = id;
2184 early_ioapic_map[i].devid = devid;
2185 early_ioapic_map[i].cmd_line = true;
2187 return 1;
2190 static int __init parse_ivrs_hpet(char *str)
2192 unsigned int bus, dev, fn;
2193 int ret, id, i;
2194 u16 devid;
2196 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2198 if (ret != 4) {
2199 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2200 return 1;
2203 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2204 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2205 str);
2206 return 1;
2209 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2211 cmdline_maps = true;
2212 i = early_hpet_map_size++;
2213 early_hpet_map[i].id = id;
2214 early_hpet_map[i].devid = devid;
2215 early_hpet_map[i].cmd_line = true;
2217 return 1;
2220 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2221 __setup("amd_iommu=", parse_amd_iommu_options);
2222 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2223 __setup("ivrs_hpet", parse_ivrs_hpet);
2225 IOMMU_INIT_FINISH(amd_iommu_detect,
2226 gart_iommu_hole_init,
2227 NULL,
2228 NULL);
2230 bool amd_iommu_v2_supported(void)
2232 return amd_iommu_v2_present;
2234 EXPORT_SYMBOL(amd_iommu_v2_supported);