iommu/amd: Don't use MSI address range for DMA addresses
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kernel / amd_iommu_init.c
blob82cbee9547c702088ef8fff80e9d33c153cadfd6
1 /*
2 * Copyright (C) 2007-2009 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/sysdev.h>
25 #include <linux/interrupt.h>
26 #include <linux/msi.h>
27 #include <asm/pci-direct.h>
28 #include <asm/amd_iommu_proto.h>
29 #include <asm/amd_iommu_types.h>
30 #include <asm/amd_iommu.h>
31 #include <asm/iommu.h>
32 #include <asm/gart.h>
33 #include <asm/x86_init.h>
36 * definitions for the ACPI scanning code
38 #define IVRS_HEADER_LENGTH 48
40 #define ACPI_IVHD_TYPE 0x10
41 #define ACPI_IVMD_TYPE_ALL 0x20
42 #define ACPI_IVMD_TYPE 0x21
43 #define ACPI_IVMD_TYPE_RANGE 0x22
45 #define IVHD_DEV_ALL 0x01
46 #define IVHD_DEV_SELECT 0x02
47 #define IVHD_DEV_SELECT_RANGE_START 0x03
48 #define IVHD_DEV_RANGE_END 0x04
49 #define IVHD_DEV_ALIAS 0x42
50 #define IVHD_DEV_ALIAS_RANGE 0x43
51 #define IVHD_DEV_EXT_SELECT 0x46
52 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
54 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
55 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
56 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
57 #define IVHD_FLAG_ISOC_EN_MASK 0x08
59 #define IVMD_FLAG_EXCL_RANGE 0x08
60 #define IVMD_FLAG_UNITY_MAP 0x01
62 #define ACPI_DEVFLAG_INITPASS 0x01
63 #define ACPI_DEVFLAG_EXTINT 0x02
64 #define ACPI_DEVFLAG_NMI 0x04
65 #define ACPI_DEVFLAG_SYSMGT1 0x10
66 #define ACPI_DEVFLAG_SYSMGT2 0x20
67 #define ACPI_DEVFLAG_LINT0 0x40
68 #define ACPI_DEVFLAG_LINT1 0x80
69 #define ACPI_DEVFLAG_ATSDIS 0x10000000
72 * ACPI table definitions
74 * These data structures are laid over the table to parse the important values
75 * out of it.
79 * structure describing one IOMMU in the ACPI table. Typically followed by one
80 * or more ivhd_entrys.
82 struct ivhd_header {
83 u8 type;
84 u8 flags;
85 u16 length;
86 u16 devid;
87 u16 cap_ptr;
88 u64 mmio_phys;
89 u16 pci_seg;
90 u16 info;
91 u32 reserved;
92 } __attribute__((packed));
95 * A device entry describing which devices a specific IOMMU translates and
96 * which requestor ids they use.
98 struct ivhd_entry {
99 u8 type;
100 u16 devid;
101 u8 flags;
102 u32 ext;
103 } __attribute__((packed));
106 * An AMD IOMMU memory definition structure. It defines things like exclusion
107 * ranges for devices and regions that should be unity mapped.
109 struct ivmd_header {
110 u8 type;
111 u8 flags;
112 u16 length;
113 u16 devid;
114 u16 aux;
115 u64 resv;
116 u64 range_start;
117 u64 range_length;
118 } __attribute__((packed));
120 bool amd_iommu_dump;
122 static int __initdata amd_iommu_detected;
123 static bool __initdata amd_iommu_disabled;
125 u16 amd_iommu_last_bdf; /* largest PCI device id we have
126 to handle */
127 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
128 we find in ACPI */
129 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
131 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
132 system */
134 /* Array to assign indices to IOMMUs*/
135 struct amd_iommu *amd_iommus[MAX_IOMMUS];
136 int amd_iommus_present;
138 /* IOMMUs have a non-present cache? */
139 bool amd_iommu_np_cache __read_mostly;
142 * The ACPI table parsing functions set this variable on an error
144 static int __initdata amd_iommu_init_err;
147 * List of protection domains - used during resume
149 LIST_HEAD(amd_iommu_pd_list);
150 spinlock_t amd_iommu_pd_lock;
153 * Pointer to the device table which is shared by all AMD IOMMUs
154 * it is indexed by the PCI device id or the HT unit id and contains
155 * information about the domain the device belongs to as well as the
156 * page table root pointer.
158 struct dev_table_entry *amd_iommu_dev_table;
161 * The alias table is a driver specific data structure which contains the
162 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
163 * More than one device can share the same requestor id.
165 u16 *amd_iommu_alias_table;
168 * The rlookup table is used to find the IOMMU which is responsible
169 * for a specific device. It is also indexed by the PCI device id.
171 struct amd_iommu **amd_iommu_rlookup_table;
174 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
175 * to know which ones are already in use.
177 unsigned long *amd_iommu_pd_alloc_bitmap;
179 static u32 dev_table_size; /* size of the device table */
180 static u32 alias_table_size; /* size of the alias table */
181 static u32 rlookup_table_size; /* size if the rlookup table */
183 static inline void update_last_devid(u16 devid)
185 if (devid > amd_iommu_last_bdf)
186 amd_iommu_last_bdf = devid;
189 static inline unsigned long tbl_size(int entry_size)
191 unsigned shift = PAGE_SHIFT +
192 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
194 return 1UL << shift;
197 /****************************************************************************
199 * AMD IOMMU MMIO register space handling functions
201 * These functions are used to program the IOMMU device registers in
202 * MMIO space required for that driver.
204 ****************************************************************************/
207 * This function set the exclusion range in the IOMMU. DMA accesses to the
208 * exclusion range are passed through untranslated
210 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
212 u64 start = iommu->exclusion_start & PAGE_MASK;
213 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
214 u64 entry;
216 if (!iommu->exclusion_start)
217 return;
219 entry = start | MMIO_EXCL_ENABLE_MASK;
220 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
221 &entry, sizeof(entry));
223 entry = limit;
224 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
225 &entry, sizeof(entry));
228 /* Programs the physical address of the device table into the IOMMU hardware */
229 static void __init iommu_set_device_table(struct amd_iommu *iommu)
231 u64 entry;
233 BUG_ON(iommu->mmio_base == NULL);
235 entry = virt_to_phys(amd_iommu_dev_table);
236 entry |= (dev_table_size >> 12) - 1;
237 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
238 &entry, sizeof(entry));
241 /* Generic functions to enable/disable certain features of the IOMMU. */
242 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
244 u32 ctrl;
246 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
247 ctrl |= (1 << bit);
248 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
251 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
253 u32 ctrl;
255 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
256 ctrl &= ~(1 << bit);
257 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
260 /* Function to enable the hardware */
261 static void iommu_enable(struct amd_iommu *iommu)
263 printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx\n",
264 dev_name(&iommu->dev->dev), iommu->cap_ptr);
266 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
269 static void iommu_disable(struct amd_iommu *iommu)
271 /* Disable command buffer */
272 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
274 /* Disable event logging and event interrupts */
275 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
276 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
278 /* Disable IOMMU hardware itself */
279 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
283 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
284 * the system has one.
286 static u8 * __init iommu_map_mmio_space(u64 address)
288 u8 *ret;
290 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
291 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
292 address);
293 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
294 return NULL;
297 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
298 if (ret != NULL)
299 return ret;
301 release_mem_region(address, MMIO_REGION_LENGTH);
303 return NULL;
306 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
308 if (iommu->mmio_base)
309 iounmap(iommu->mmio_base);
310 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
313 /****************************************************************************
315 * The functions below belong to the first pass of AMD IOMMU ACPI table
316 * parsing. In this pass we try to find out the highest device id this
317 * code has to handle. Upon this information the size of the shared data
318 * structures is determined later.
320 ****************************************************************************/
323 * This function calculates the length of a given IVHD entry
325 static inline int ivhd_entry_length(u8 *ivhd)
327 return 0x04 << (*ivhd >> 6);
331 * This function reads the last device id the IOMMU has to handle from the PCI
332 * capability header for this IOMMU
334 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
336 u32 cap;
338 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
339 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
341 return 0;
345 * After reading the highest device id from the IOMMU PCI capability header
346 * this function looks if there is a higher device id defined in the ACPI table
348 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
350 u8 *p = (void *)h, *end = (void *)h;
351 struct ivhd_entry *dev;
353 p += sizeof(*h);
354 end += h->length;
356 find_last_devid_on_pci(PCI_BUS(h->devid),
357 PCI_SLOT(h->devid),
358 PCI_FUNC(h->devid),
359 h->cap_ptr);
361 while (p < end) {
362 dev = (struct ivhd_entry *)p;
363 switch (dev->type) {
364 case IVHD_DEV_SELECT:
365 case IVHD_DEV_RANGE_END:
366 case IVHD_DEV_ALIAS:
367 case IVHD_DEV_EXT_SELECT:
368 /* all the above subfield types refer to device ids */
369 update_last_devid(dev->devid);
370 break;
371 default:
372 break;
374 p += ivhd_entry_length(p);
377 WARN_ON(p != end);
379 return 0;
383 * Iterate over all IVHD entries in the ACPI table and find the highest device
384 * id which we need to handle. This is the first of three functions which parse
385 * the ACPI table. So we check the checksum here.
387 static int __init find_last_devid_acpi(struct acpi_table_header *table)
389 int i;
390 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
391 struct ivhd_header *h;
394 * Validate checksum here so we don't need to do it when
395 * we actually parse the table
397 for (i = 0; i < table->length; ++i)
398 checksum += p[i];
399 if (checksum != 0) {
400 /* ACPI table corrupt */
401 amd_iommu_init_err = -ENODEV;
402 return 0;
405 p += IVRS_HEADER_LENGTH;
407 end += table->length;
408 while (p < end) {
409 h = (struct ivhd_header *)p;
410 switch (h->type) {
411 case ACPI_IVHD_TYPE:
412 find_last_devid_from_ivhd(h);
413 break;
414 default:
415 break;
417 p += h->length;
419 WARN_ON(p != end);
421 return 0;
424 /****************************************************************************
426 * The following functions belong the the code path which parses the ACPI table
427 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
428 * data structures, initialize the device/alias/rlookup table and also
429 * basically initialize the hardware.
431 ****************************************************************************/
434 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
435 * write commands to that buffer later and the IOMMU will execute them
436 * asynchronously
438 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
440 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
441 get_order(CMD_BUFFER_SIZE));
443 if (cmd_buf == NULL)
444 return NULL;
446 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
448 return cmd_buf;
452 * This function resets the command buffer if the IOMMU stopped fetching
453 * commands from it.
455 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
457 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
459 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
460 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
462 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
466 * This function writes the command buffer address to the hardware and
467 * enables it.
469 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
471 u64 entry;
473 BUG_ON(iommu->cmd_buf == NULL);
475 entry = (u64)virt_to_phys(iommu->cmd_buf);
476 entry |= MMIO_CMD_SIZE_512;
478 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
479 &entry, sizeof(entry));
481 amd_iommu_reset_cmd_buffer(iommu);
482 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
485 static void __init free_command_buffer(struct amd_iommu *iommu)
487 free_pages((unsigned long)iommu->cmd_buf,
488 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
491 /* allocates the memory where the IOMMU will log its events to */
492 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
494 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
495 get_order(EVT_BUFFER_SIZE));
497 if (iommu->evt_buf == NULL)
498 return NULL;
500 iommu->evt_buf_size = EVT_BUFFER_SIZE;
502 return iommu->evt_buf;
505 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
507 u64 entry;
509 BUG_ON(iommu->evt_buf == NULL);
511 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
513 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
514 &entry, sizeof(entry));
516 /* set head and tail to zero manually */
517 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
518 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
520 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
523 static void __init free_event_buffer(struct amd_iommu *iommu)
525 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
528 /* sets a specific bit in the device table entry. */
529 static void set_dev_entry_bit(u16 devid, u8 bit)
531 int i = (bit >> 5) & 0x07;
532 int _bit = bit & 0x1f;
534 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
537 static int get_dev_entry_bit(u16 devid, u8 bit)
539 int i = (bit >> 5) & 0x07;
540 int _bit = bit & 0x1f;
542 return (amd_iommu_dev_table[devid].data[i] & (1 << _bit)) >> _bit;
546 void amd_iommu_apply_erratum_63(u16 devid)
548 int sysmgt;
550 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
551 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
553 if (sysmgt == 0x01)
554 set_dev_entry_bit(devid, DEV_ENTRY_IW);
557 /* Writes the specific IOMMU for a device into the rlookup table */
558 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
560 amd_iommu_rlookup_table[devid] = iommu;
564 * This function takes the device specific flags read from the ACPI
565 * table and sets up the device table entry with that information
567 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
568 u16 devid, u32 flags, u32 ext_flags)
570 if (flags & ACPI_DEVFLAG_INITPASS)
571 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
572 if (flags & ACPI_DEVFLAG_EXTINT)
573 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
574 if (flags & ACPI_DEVFLAG_NMI)
575 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
576 if (flags & ACPI_DEVFLAG_SYSMGT1)
577 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
578 if (flags & ACPI_DEVFLAG_SYSMGT2)
579 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
580 if (flags & ACPI_DEVFLAG_LINT0)
581 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
582 if (flags & ACPI_DEVFLAG_LINT1)
583 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
585 amd_iommu_apply_erratum_63(devid);
587 set_iommu_for_device(iommu, devid);
591 * Reads the device exclusion range from ACPI and initialize IOMMU with
592 * it
594 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
596 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
598 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
599 return;
601 if (iommu) {
603 * We only can configure exclusion ranges per IOMMU, not
604 * per device. But we can enable the exclusion range per
605 * device. This is done here
607 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
608 iommu->exclusion_start = m->range_start;
609 iommu->exclusion_length = m->range_length;
614 * This function reads some important data from the IOMMU PCI space and
615 * initializes the driver data structure with it. It reads the hardware
616 * capabilities and the first/last device entries
618 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
620 int cap_ptr = iommu->cap_ptr;
621 u32 range, misc;
623 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
624 &iommu->cap);
625 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
626 &range);
627 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
628 &misc);
630 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
631 MMIO_GET_FD(range));
632 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
633 MMIO_GET_LD(range));
634 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
636 if (is_rd890_iommu(iommu->dev)) {
637 pci_read_config_dword(iommu->dev, 0xf0, &iommu->cache_cfg[0]);
638 pci_read_config_dword(iommu->dev, 0xf4, &iommu->cache_cfg[1]);
639 pci_read_config_dword(iommu->dev, 0xf8, &iommu->cache_cfg[2]);
640 pci_read_config_dword(iommu->dev, 0xfc, &iommu->cache_cfg[3]);
645 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
646 * initializes the hardware and our data structures with it.
648 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
649 struct ivhd_header *h)
651 u8 *p = (u8 *)h;
652 u8 *end = p, flags = 0;
653 u16 devid = 0, devid_start = 0, devid_to = 0;
654 u32 dev_i, ext_flags = 0;
655 bool alias = false;
656 struct ivhd_entry *e;
659 * First save the recommended feature enable bits from ACPI
661 iommu->acpi_flags = h->flags;
664 * Done. Now parse the device entries
666 p += sizeof(struct ivhd_header);
667 end += h->length;
670 while (p < end) {
671 e = (struct ivhd_entry *)p;
672 switch (e->type) {
673 case IVHD_DEV_ALL:
675 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
676 " last device %02x:%02x.%x flags: %02x\n",
677 PCI_BUS(iommu->first_device),
678 PCI_SLOT(iommu->first_device),
679 PCI_FUNC(iommu->first_device),
680 PCI_BUS(iommu->last_device),
681 PCI_SLOT(iommu->last_device),
682 PCI_FUNC(iommu->last_device),
683 e->flags);
685 for (dev_i = iommu->first_device;
686 dev_i <= iommu->last_device; ++dev_i)
687 set_dev_entry_from_acpi(iommu, dev_i,
688 e->flags, 0);
689 break;
690 case IVHD_DEV_SELECT:
692 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
693 "flags: %02x\n",
694 PCI_BUS(e->devid),
695 PCI_SLOT(e->devid),
696 PCI_FUNC(e->devid),
697 e->flags);
699 devid = e->devid;
700 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
701 break;
702 case IVHD_DEV_SELECT_RANGE_START:
704 DUMP_printk(" DEV_SELECT_RANGE_START\t "
705 "devid: %02x:%02x.%x flags: %02x\n",
706 PCI_BUS(e->devid),
707 PCI_SLOT(e->devid),
708 PCI_FUNC(e->devid),
709 e->flags);
711 devid_start = e->devid;
712 flags = e->flags;
713 ext_flags = 0;
714 alias = false;
715 break;
716 case IVHD_DEV_ALIAS:
718 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
719 "flags: %02x devid_to: %02x:%02x.%x\n",
720 PCI_BUS(e->devid),
721 PCI_SLOT(e->devid),
722 PCI_FUNC(e->devid),
723 e->flags,
724 PCI_BUS(e->ext >> 8),
725 PCI_SLOT(e->ext >> 8),
726 PCI_FUNC(e->ext >> 8));
728 devid = e->devid;
729 devid_to = e->ext >> 8;
730 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
731 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
732 amd_iommu_alias_table[devid] = devid_to;
733 break;
734 case IVHD_DEV_ALIAS_RANGE:
736 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
737 "devid: %02x:%02x.%x flags: %02x "
738 "devid_to: %02x:%02x.%x\n",
739 PCI_BUS(e->devid),
740 PCI_SLOT(e->devid),
741 PCI_FUNC(e->devid),
742 e->flags,
743 PCI_BUS(e->ext >> 8),
744 PCI_SLOT(e->ext >> 8),
745 PCI_FUNC(e->ext >> 8));
747 devid_start = e->devid;
748 flags = e->flags;
749 devid_to = e->ext >> 8;
750 ext_flags = 0;
751 alias = true;
752 break;
753 case IVHD_DEV_EXT_SELECT:
755 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
756 "flags: %02x ext: %08x\n",
757 PCI_BUS(e->devid),
758 PCI_SLOT(e->devid),
759 PCI_FUNC(e->devid),
760 e->flags, e->ext);
762 devid = e->devid;
763 set_dev_entry_from_acpi(iommu, devid, e->flags,
764 e->ext);
765 break;
766 case IVHD_DEV_EXT_SELECT_RANGE:
768 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
769 "%02x:%02x.%x flags: %02x ext: %08x\n",
770 PCI_BUS(e->devid),
771 PCI_SLOT(e->devid),
772 PCI_FUNC(e->devid),
773 e->flags, e->ext);
775 devid_start = e->devid;
776 flags = e->flags;
777 ext_flags = e->ext;
778 alias = false;
779 break;
780 case IVHD_DEV_RANGE_END:
782 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
783 PCI_BUS(e->devid),
784 PCI_SLOT(e->devid),
785 PCI_FUNC(e->devid));
787 devid = e->devid;
788 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
789 if (alias) {
790 amd_iommu_alias_table[dev_i] = devid_to;
791 set_dev_entry_from_acpi(iommu,
792 devid_to, flags, ext_flags);
794 set_dev_entry_from_acpi(iommu, dev_i,
795 flags, ext_flags);
797 break;
798 default:
799 break;
802 p += ivhd_entry_length(p);
806 /* Initializes the device->iommu mapping for the driver */
807 static int __init init_iommu_devices(struct amd_iommu *iommu)
809 u32 i;
811 for (i = iommu->first_device; i <= iommu->last_device; ++i)
812 set_iommu_for_device(iommu, i);
814 return 0;
817 static void __init free_iommu_one(struct amd_iommu *iommu)
819 free_command_buffer(iommu);
820 free_event_buffer(iommu);
821 iommu_unmap_mmio_space(iommu);
824 static void __init free_iommu_all(void)
826 struct amd_iommu *iommu, *next;
828 for_each_iommu_safe(iommu, next) {
829 list_del(&iommu->list);
830 free_iommu_one(iommu);
831 kfree(iommu);
836 * This function clues the initialization function for one IOMMU
837 * together and also allocates the command buffer and programs the
838 * hardware. It does NOT enable the IOMMU. This is done afterwards.
840 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
842 spin_lock_init(&iommu->lock);
844 /* Add IOMMU to internal data structures */
845 list_add_tail(&iommu->list, &amd_iommu_list);
846 iommu->index = amd_iommus_present++;
848 if (unlikely(iommu->index >= MAX_IOMMUS)) {
849 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
850 return -ENOSYS;
853 /* Index is fine - add IOMMU to the array */
854 amd_iommus[iommu->index] = iommu;
857 * Copy data from ACPI table entry to the iommu struct
859 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
860 if (!iommu->dev)
861 return 1;
863 iommu->cap_ptr = h->cap_ptr;
864 iommu->pci_seg = h->pci_seg;
865 iommu->mmio_phys = h->mmio_phys;
866 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
867 if (!iommu->mmio_base)
868 return -ENOMEM;
870 iommu->cmd_buf = alloc_command_buffer(iommu);
871 if (!iommu->cmd_buf)
872 return -ENOMEM;
874 iommu->evt_buf = alloc_event_buffer(iommu);
875 if (!iommu->evt_buf)
876 return -ENOMEM;
878 iommu->int_enabled = false;
880 init_iommu_from_pci(iommu);
881 init_iommu_from_acpi(iommu, h);
882 init_iommu_devices(iommu);
884 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
885 amd_iommu_np_cache = true;
887 return pci_enable_device(iommu->dev);
891 * Iterates over all IOMMU entries in the ACPI table, allocates the
892 * IOMMU structure and initializes it with init_iommu_one()
894 static int __init init_iommu_all(struct acpi_table_header *table)
896 u8 *p = (u8 *)table, *end = (u8 *)table;
897 struct ivhd_header *h;
898 struct amd_iommu *iommu;
899 int ret;
901 end += table->length;
902 p += IVRS_HEADER_LENGTH;
904 while (p < end) {
905 h = (struct ivhd_header *)p;
906 switch (*p) {
907 case ACPI_IVHD_TYPE:
909 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
910 "seg: %d flags: %01x info %04x\n",
911 PCI_BUS(h->devid), PCI_SLOT(h->devid),
912 PCI_FUNC(h->devid), h->cap_ptr,
913 h->pci_seg, h->flags, h->info);
914 DUMP_printk(" mmio-addr: %016llx\n",
915 h->mmio_phys);
917 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
918 if (iommu == NULL) {
919 amd_iommu_init_err = -ENOMEM;
920 return 0;
923 ret = init_iommu_one(iommu, h);
924 if (ret) {
925 amd_iommu_init_err = ret;
926 return 0;
928 break;
929 default:
930 break;
932 p += h->length;
935 WARN_ON(p != end);
937 return 0;
940 /****************************************************************************
942 * The following functions initialize the MSI interrupts for all IOMMUs
943 * in the system. Its a bit challenging because there could be multiple
944 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
945 * pci_dev.
947 ****************************************************************************/
949 static int iommu_setup_msi(struct amd_iommu *iommu)
951 int r;
953 if (pci_enable_msi(iommu->dev))
954 return 1;
956 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
957 IRQF_SAMPLE_RANDOM,
958 "AMD-Vi",
959 NULL);
961 if (r) {
962 pci_disable_msi(iommu->dev);
963 return 1;
966 iommu->int_enabled = true;
967 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
969 return 0;
972 static int iommu_init_msi(struct amd_iommu *iommu)
974 if (iommu->int_enabled)
975 return 0;
977 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
978 return iommu_setup_msi(iommu);
980 return 1;
983 /****************************************************************************
985 * The next functions belong to the third pass of parsing the ACPI
986 * table. In this last pass the memory mapping requirements are
987 * gathered (like exclusion and unity mapping reanges).
989 ****************************************************************************/
991 static void __init free_unity_maps(void)
993 struct unity_map_entry *entry, *next;
995 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
996 list_del(&entry->list);
997 kfree(entry);
1001 /* called when we find an exclusion range definition in ACPI */
1002 static int __init init_exclusion_range(struct ivmd_header *m)
1004 int i;
1006 switch (m->type) {
1007 case ACPI_IVMD_TYPE:
1008 set_device_exclusion_range(m->devid, m);
1009 break;
1010 case ACPI_IVMD_TYPE_ALL:
1011 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1012 set_device_exclusion_range(i, m);
1013 break;
1014 case ACPI_IVMD_TYPE_RANGE:
1015 for (i = m->devid; i <= m->aux; ++i)
1016 set_device_exclusion_range(i, m);
1017 break;
1018 default:
1019 break;
1022 return 0;
1025 /* called for unity map ACPI definition */
1026 static int __init init_unity_map_range(struct ivmd_header *m)
1028 struct unity_map_entry *e = 0;
1029 char *s;
1031 e = kzalloc(sizeof(*e), GFP_KERNEL);
1032 if (e == NULL)
1033 return -ENOMEM;
1035 switch (m->type) {
1036 default:
1037 kfree(e);
1038 return 0;
1039 case ACPI_IVMD_TYPE:
1040 s = "IVMD_TYPEi\t\t\t";
1041 e->devid_start = e->devid_end = m->devid;
1042 break;
1043 case ACPI_IVMD_TYPE_ALL:
1044 s = "IVMD_TYPE_ALL\t\t";
1045 e->devid_start = 0;
1046 e->devid_end = amd_iommu_last_bdf;
1047 break;
1048 case ACPI_IVMD_TYPE_RANGE:
1049 s = "IVMD_TYPE_RANGE\t\t";
1050 e->devid_start = m->devid;
1051 e->devid_end = m->aux;
1052 break;
1054 e->address_start = PAGE_ALIGN(m->range_start);
1055 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1056 e->prot = m->flags >> 1;
1058 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1059 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1060 PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1061 PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1062 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1063 e->address_start, e->address_end, m->flags);
1065 list_add_tail(&e->list, &amd_iommu_unity_map);
1067 return 0;
1070 /* iterates over all memory definitions we find in the ACPI table */
1071 static int __init init_memory_definitions(struct acpi_table_header *table)
1073 u8 *p = (u8 *)table, *end = (u8 *)table;
1074 struct ivmd_header *m;
1076 end += table->length;
1077 p += IVRS_HEADER_LENGTH;
1079 while (p < end) {
1080 m = (struct ivmd_header *)p;
1081 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1082 init_exclusion_range(m);
1083 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1084 init_unity_map_range(m);
1086 p += m->length;
1089 return 0;
1093 * Init the device table to not allow DMA access for devices and
1094 * suppress all page faults
1096 static void init_device_table(void)
1098 u32 devid;
1100 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1101 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1102 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1106 static void iommu_init_flags(struct amd_iommu *iommu)
1108 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1109 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1110 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1112 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1113 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1114 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1116 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1117 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1118 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1120 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1121 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1122 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1125 * make IOMMU memory accesses cache coherent
1127 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1130 static void iommu_apply_quirks(struct amd_iommu *iommu)
1132 if (is_rd890_iommu(iommu->dev)) {
1133 pci_write_config_dword(iommu->dev, 0xf0, iommu->cache_cfg[0]);
1134 pci_write_config_dword(iommu->dev, 0xf4, iommu->cache_cfg[1]);
1135 pci_write_config_dword(iommu->dev, 0xf8, iommu->cache_cfg[2]);
1136 pci_write_config_dword(iommu->dev, 0xfc, iommu->cache_cfg[3]);
1141 * This function finally enables all IOMMUs found in the system after
1142 * they have been initialized
1144 static void enable_iommus(void)
1146 struct amd_iommu *iommu;
1148 for_each_iommu(iommu) {
1149 iommu_disable(iommu);
1150 iommu_apply_quirks(iommu);
1151 iommu_init_flags(iommu);
1152 iommu_set_device_table(iommu);
1153 iommu_enable_command_buffer(iommu);
1154 iommu_enable_event_buffer(iommu);
1155 iommu_set_exclusion_range(iommu);
1156 iommu_init_msi(iommu);
1157 iommu_enable(iommu);
1161 static void disable_iommus(void)
1163 struct amd_iommu *iommu;
1165 for_each_iommu(iommu)
1166 iommu_disable(iommu);
1170 * Suspend/Resume support
1171 * disable suspend until real resume implemented
1174 static int amd_iommu_resume(struct sys_device *dev)
1176 /* re-load the hardware */
1177 enable_iommus();
1180 * we have to flush after the IOMMUs are enabled because a
1181 * disabled IOMMU will never execute the commands we send
1183 amd_iommu_flush_all_devices();
1184 amd_iommu_flush_all_domains();
1186 return 0;
1189 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
1191 /* disable IOMMUs to go out of the way for BIOS */
1192 disable_iommus();
1194 return 0;
1197 static struct sysdev_class amd_iommu_sysdev_class = {
1198 .name = "amd_iommu",
1199 .suspend = amd_iommu_suspend,
1200 .resume = amd_iommu_resume,
1203 static struct sys_device device_amd_iommu = {
1204 .id = 0,
1205 .cls = &amd_iommu_sysdev_class,
1209 * This is the core init function for AMD IOMMU hardware in the system.
1210 * This function is called from the generic x86 DMA layer initialization
1211 * code.
1213 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1214 * three times:
1216 * 1 pass) Find the highest PCI device id the driver has to handle.
1217 * Upon this information the size of the data structures is
1218 * determined that needs to be allocated.
1220 * 2 pass) Initialize the data structures just allocated with the
1221 * information in the ACPI table about available AMD IOMMUs
1222 * in the system. It also maps the PCI devices in the
1223 * system to specific IOMMUs
1225 * 3 pass) After the basic data structures are allocated and
1226 * initialized we update them with information about memory
1227 * remapping requirements parsed out of the ACPI table in
1228 * this last pass.
1230 * After that the hardware is initialized and ready to go. In the last
1231 * step we do some Linux specific things like registering the driver in
1232 * the dma_ops interface and initializing the suspend/resume support
1233 * functions. Finally it prints some information about AMD IOMMUs and
1234 * the driver state and enables the hardware.
1236 static int __init amd_iommu_init(void)
1238 int i, ret = 0;
1241 * First parse ACPI tables to find the largest Bus/Dev/Func
1242 * we need to handle. Upon this information the shared data
1243 * structures for the IOMMUs in the system will be allocated
1245 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1246 return -ENODEV;
1248 ret = amd_iommu_init_err;
1249 if (ret)
1250 goto out;
1252 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1253 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1254 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1256 ret = -ENOMEM;
1258 /* Device table - directly used by all IOMMUs */
1259 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1260 get_order(dev_table_size));
1261 if (amd_iommu_dev_table == NULL)
1262 goto out;
1265 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1266 * IOMMU see for that device
1268 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1269 get_order(alias_table_size));
1270 if (amd_iommu_alias_table == NULL)
1271 goto free;
1273 /* IOMMU rlookup table - find the IOMMU for a specific device */
1274 amd_iommu_rlookup_table = (void *)__get_free_pages(
1275 GFP_KERNEL | __GFP_ZERO,
1276 get_order(rlookup_table_size));
1277 if (amd_iommu_rlookup_table == NULL)
1278 goto free;
1280 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1281 GFP_KERNEL | __GFP_ZERO,
1282 get_order(MAX_DOMAIN_ID/8));
1283 if (amd_iommu_pd_alloc_bitmap == NULL)
1284 goto free;
1286 /* init the device table */
1287 init_device_table();
1290 * let all alias entries point to itself
1292 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1293 amd_iommu_alias_table[i] = i;
1296 * never allocate domain 0 because its used as the non-allocated and
1297 * error value placeholder
1299 amd_iommu_pd_alloc_bitmap[0] = 1;
1301 spin_lock_init(&amd_iommu_pd_lock);
1304 * now the data structures are allocated and basically initialized
1305 * start the real acpi table scan
1307 ret = -ENODEV;
1308 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1309 goto free;
1311 if (amd_iommu_init_err) {
1312 ret = amd_iommu_init_err;
1313 goto free;
1316 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1317 goto free;
1319 if (amd_iommu_init_err) {
1320 ret = amd_iommu_init_err;
1321 goto free;
1324 ret = sysdev_class_register(&amd_iommu_sysdev_class);
1325 if (ret)
1326 goto free;
1328 ret = sysdev_register(&device_amd_iommu);
1329 if (ret)
1330 goto free;
1332 ret = amd_iommu_init_devices();
1333 if (ret)
1334 goto free;
1336 enable_iommus();
1338 if (iommu_pass_through)
1339 ret = amd_iommu_init_passthrough();
1340 else
1341 ret = amd_iommu_init_dma_ops();
1343 if (ret)
1344 goto free_disable;
1346 amd_iommu_init_api();
1348 amd_iommu_init_notifier();
1350 if (iommu_pass_through)
1351 goto out;
1353 if (amd_iommu_unmap_flush)
1354 printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
1355 else
1356 printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n");
1358 x86_platform.iommu_shutdown = disable_iommus;
1359 out:
1360 return ret;
1362 free_disable:
1363 disable_iommus();
1365 free:
1366 amd_iommu_uninit_devices();
1368 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1369 get_order(MAX_DOMAIN_ID/8));
1371 free_pages((unsigned long)amd_iommu_rlookup_table,
1372 get_order(rlookup_table_size));
1374 free_pages((unsigned long)amd_iommu_alias_table,
1375 get_order(alias_table_size));
1377 free_pages((unsigned long)amd_iommu_dev_table,
1378 get_order(dev_table_size));
1380 free_iommu_all();
1382 free_unity_maps();
1384 #ifdef CONFIG_GART_IOMMU
1386 * We failed to initialize the AMD IOMMU - try fallback to GART
1387 * if possible.
1389 gart_iommu_init();
1391 #endif
1393 goto out;
1396 /****************************************************************************
1398 * Early detect code. This code runs at IOMMU detection time in the DMA
1399 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1400 * IOMMUs
1402 ****************************************************************************/
1403 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1405 return 0;
1408 void __init amd_iommu_detect(void)
1410 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
1411 return;
1413 if (amd_iommu_disabled)
1414 return;
1416 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1417 iommu_detected = 1;
1418 amd_iommu_detected = 1;
1419 x86_init.iommu.iommu_init = amd_iommu_init;
1421 /* Make sure ACS will be enabled */
1422 pci_request_acs();
1426 /****************************************************************************
1428 * Parsing functions for the AMD IOMMU specific kernel command line
1429 * options.
1431 ****************************************************************************/
1433 static int __init parse_amd_iommu_dump(char *str)
1435 amd_iommu_dump = true;
1437 return 1;
1440 static int __init parse_amd_iommu_options(char *str)
1442 for (; *str; ++str) {
1443 if (strncmp(str, "fullflush", 9) == 0)
1444 amd_iommu_unmap_flush = true;
1445 if (strncmp(str, "off", 3) == 0)
1446 amd_iommu_disabled = true;
1449 return 1;
1452 __setup("amd_iommu_dump", parse_amd_iommu_dump);
1453 __setup("amd_iommu=", parse_amd_iommu_options);