Merge branch 'mini2440-dev-unlikely' into mini2440-dev
[linux-2.6/mini2440.git] / arch / x86 / kernel / amd_iommu_init.c
blobc20001e4f5562cf6199c546cbd9fa4f307dab122
1 /*
2 * Copyright (C) 2007-2008 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/gfp.h>
23 #include <linux/list.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_types.h>
29 #include <asm/amd_iommu.h>
30 #include <asm/iommu.h>
31 #include <asm/gart.h>
34 * definitions for the ACPI scanning code
36 #define IVRS_HEADER_LENGTH 48
38 #define ACPI_IVHD_TYPE 0x10
39 #define ACPI_IVMD_TYPE_ALL 0x20
40 #define ACPI_IVMD_TYPE 0x21
41 #define ACPI_IVMD_TYPE_RANGE 0x22
43 #define IVHD_DEV_ALL 0x01
44 #define IVHD_DEV_SELECT 0x02
45 #define IVHD_DEV_SELECT_RANGE_START 0x03
46 #define IVHD_DEV_RANGE_END 0x04
47 #define IVHD_DEV_ALIAS 0x42
48 #define IVHD_DEV_ALIAS_RANGE 0x43
49 #define IVHD_DEV_EXT_SELECT 0x46
50 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
52 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
53 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
54 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
55 #define IVHD_FLAG_ISOC_EN_MASK 0x08
57 #define IVMD_FLAG_EXCL_RANGE 0x08
58 #define IVMD_FLAG_UNITY_MAP 0x01
60 #define ACPI_DEVFLAG_INITPASS 0x01
61 #define ACPI_DEVFLAG_EXTINT 0x02
62 #define ACPI_DEVFLAG_NMI 0x04
63 #define ACPI_DEVFLAG_SYSMGT1 0x10
64 #define ACPI_DEVFLAG_SYSMGT2 0x20
65 #define ACPI_DEVFLAG_LINT0 0x40
66 #define ACPI_DEVFLAG_LINT1 0x80
67 #define ACPI_DEVFLAG_ATSDIS 0x10000000
70 * ACPI table definitions
72 * These data structures are laid over the table to parse the important values
73 * out of it.
77 * structure describing one IOMMU in the ACPI table. Typically followed by one
78 * or more ivhd_entrys.
80 struct ivhd_header {
81 u8 type;
82 u8 flags;
83 u16 length;
84 u16 devid;
85 u16 cap_ptr;
86 u64 mmio_phys;
87 u16 pci_seg;
88 u16 info;
89 u32 reserved;
90 } __attribute__((packed));
93 * A device entry describing which devices a specific IOMMU translates and
94 * which requestor ids they use.
96 struct ivhd_entry {
97 u8 type;
98 u16 devid;
99 u8 flags;
100 u32 ext;
101 } __attribute__((packed));
104 * An AMD IOMMU memory definition structure. It defines things like exclusion
105 * ranges for devices and regions that should be unity mapped.
107 struct ivmd_header {
108 u8 type;
109 u8 flags;
110 u16 length;
111 u16 devid;
112 u16 aux;
113 u64 resv;
114 u64 range_start;
115 u64 range_length;
116 } __attribute__((packed));
118 bool amd_iommu_dump;
120 static int __initdata amd_iommu_detected;
122 u16 amd_iommu_last_bdf; /* largest PCI device id we have
123 to handle */
124 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
125 we find in ACPI */
126 #ifdef CONFIG_IOMMU_STRESS
127 bool amd_iommu_isolate = false;
128 #else
129 bool amd_iommu_isolate = true; /* if true, device isolation is
130 enabled */
131 #endif
133 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
135 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
136 system */
139 * Pointer to the device table which is shared by all AMD IOMMUs
140 * it is indexed by the PCI device id or the HT unit id and contains
141 * information about the domain the device belongs to as well as the
142 * page table root pointer.
144 struct dev_table_entry *amd_iommu_dev_table;
147 * The alias table is a driver specific data structure which contains the
148 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
149 * More than one device can share the same requestor id.
151 u16 *amd_iommu_alias_table;
154 * The rlookup table is used to find the IOMMU which is responsible
155 * for a specific device. It is also indexed by the PCI device id.
157 struct amd_iommu **amd_iommu_rlookup_table;
160 * The pd table (protection domain table) is used to find the protection domain
161 * data structure a device belongs to. Indexed with the PCI device id too.
163 struct protection_domain **amd_iommu_pd_table;
166 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
167 * to know which ones are already in use.
169 unsigned long *amd_iommu_pd_alloc_bitmap;
171 static u32 dev_table_size; /* size of the device table */
172 static u32 alias_table_size; /* size of the alias table */
173 static u32 rlookup_table_size; /* size if the rlookup table */
175 static inline void update_last_devid(u16 devid)
177 if (devid > amd_iommu_last_bdf)
178 amd_iommu_last_bdf = devid;
181 static inline unsigned long tbl_size(int entry_size)
183 unsigned shift = PAGE_SHIFT +
184 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
186 return 1UL << shift;
189 /****************************************************************************
191 * AMD IOMMU MMIO register space handling functions
193 * These functions are used to program the IOMMU device registers in
194 * MMIO space required for that driver.
196 ****************************************************************************/
199 * This function set the exclusion range in the IOMMU. DMA accesses to the
200 * exclusion range are passed through untranslated
202 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
204 u64 start = iommu->exclusion_start & PAGE_MASK;
205 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
206 u64 entry;
208 if (!iommu->exclusion_start)
209 return;
211 entry = start | MMIO_EXCL_ENABLE_MASK;
212 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
213 &entry, sizeof(entry));
215 entry = limit;
216 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
217 &entry, sizeof(entry));
220 /* Programs the physical address of the device table into the IOMMU hardware */
221 static void __init iommu_set_device_table(struct amd_iommu *iommu)
223 u64 entry;
225 BUG_ON(iommu->mmio_base == NULL);
227 entry = virt_to_phys(amd_iommu_dev_table);
228 entry |= (dev_table_size >> 12) - 1;
229 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
230 &entry, sizeof(entry));
233 /* Generic functions to enable/disable certain features of the IOMMU. */
234 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
236 u32 ctrl;
238 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
239 ctrl |= (1 << bit);
240 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
243 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
245 u32 ctrl;
247 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
248 ctrl &= ~(1 << bit);
249 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
252 /* Function to enable the hardware */
253 static void iommu_enable(struct amd_iommu *iommu)
255 printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx\n",
256 dev_name(&iommu->dev->dev), iommu->cap_ptr);
258 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
261 static void iommu_disable(struct amd_iommu *iommu)
263 /* Disable command buffer */
264 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
266 /* Disable event logging and event interrupts */
267 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
268 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
270 /* Disable IOMMU hardware itself */
271 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
275 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
276 * the system has one.
278 static u8 * __init iommu_map_mmio_space(u64 address)
280 u8 *ret;
282 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu"))
283 return NULL;
285 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
286 if (ret != NULL)
287 return ret;
289 release_mem_region(address, MMIO_REGION_LENGTH);
291 return NULL;
294 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
296 if (iommu->mmio_base)
297 iounmap(iommu->mmio_base);
298 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
301 /****************************************************************************
303 * The functions below belong to the first pass of AMD IOMMU ACPI table
304 * parsing. In this pass we try to find out the highest device id this
305 * code has to handle. Upon this information the size of the shared data
306 * structures is determined later.
308 ****************************************************************************/
311 * This function calculates the length of a given IVHD entry
313 static inline int ivhd_entry_length(u8 *ivhd)
315 return 0x04 << (*ivhd >> 6);
319 * This function reads the last device id the IOMMU has to handle from the PCI
320 * capability header for this IOMMU
322 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
324 u32 cap;
326 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
327 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
329 return 0;
333 * After reading the highest device id from the IOMMU PCI capability header
334 * this function looks if there is a higher device id defined in the ACPI table
336 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
338 u8 *p = (void *)h, *end = (void *)h;
339 struct ivhd_entry *dev;
341 p += sizeof(*h);
342 end += h->length;
344 find_last_devid_on_pci(PCI_BUS(h->devid),
345 PCI_SLOT(h->devid),
346 PCI_FUNC(h->devid),
347 h->cap_ptr);
349 while (p < end) {
350 dev = (struct ivhd_entry *)p;
351 switch (dev->type) {
352 case IVHD_DEV_SELECT:
353 case IVHD_DEV_RANGE_END:
354 case IVHD_DEV_ALIAS:
355 case IVHD_DEV_EXT_SELECT:
356 /* all the above subfield types refer to device ids */
357 update_last_devid(dev->devid);
358 break;
359 default:
360 break;
362 p += ivhd_entry_length(p);
365 WARN_ON(p != end);
367 return 0;
371 * Iterate over all IVHD entries in the ACPI table and find the highest device
372 * id which we need to handle. This is the first of three functions which parse
373 * the ACPI table. So we check the checksum here.
375 static int __init find_last_devid_acpi(struct acpi_table_header *table)
377 int i;
378 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
379 struct ivhd_header *h;
382 * Validate checksum here so we don't need to do it when
383 * we actually parse the table
385 for (i = 0; i < table->length; ++i)
386 checksum += p[i];
387 if (checksum != 0)
388 /* ACPI table corrupt */
389 return -ENODEV;
391 p += IVRS_HEADER_LENGTH;
393 end += table->length;
394 while (p < end) {
395 h = (struct ivhd_header *)p;
396 switch (h->type) {
397 case ACPI_IVHD_TYPE:
398 find_last_devid_from_ivhd(h);
399 break;
400 default:
401 break;
403 p += h->length;
405 WARN_ON(p != end);
407 return 0;
410 /****************************************************************************
412 * The following functions belong the the code path which parses the ACPI table
413 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
414 * data structures, initialize the device/alias/rlookup table and also
415 * basically initialize the hardware.
417 ****************************************************************************/
420 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
421 * write commands to that buffer later and the IOMMU will execute them
422 * asynchronously
424 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
426 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
427 get_order(CMD_BUFFER_SIZE));
429 if (cmd_buf == NULL)
430 return NULL;
432 iommu->cmd_buf_size = CMD_BUFFER_SIZE;
434 return cmd_buf;
438 * This function resets the command buffer if the IOMMU stopped fetching
439 * commands from it.
441 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
443 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
445 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
446 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
448 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
452 * This function writes the command buffer address to the hardware and
453 * enables it.
455 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
457 u64 entry;
459 BUG_ON(iommu->cmd_buf == NULL);
461 entry = (u64)virt_to_phys(iommu->cmd_buf);
462 entry |= MMIO_CMD_SIZE_512;
464 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
465 &entry, sizeof(entry));
467 amd_iommu_reset_cmd_buffer(iommu);
470 static void __init free_command_buffer(struct amd_iommu *iommu)
472 free_pages((unsigned long)iommu->cmd_buf,
473 get_order(iommu->cmd_buf_size));
476 /* allocates the memory where the IOMMU will log its events to */
477 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
479 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
480 get_order(EVT_BUFFER_SIZE));
482 if (iommu->evt_buf == NULL)
483 return NULL;
485 iommu->evt_buf_size = EVT_BUFFER_SIZE;
487 return iommu->evt_buf;
490 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
492 u64 entry;
494 BUG_ON(iommu->evt_buf == NULL);
496 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
498 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
499 &entry, sizeof(entry));
501 /* set head and tail to zero manually */
502 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
503 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
505 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
508 static void __init free_event_buffer(struct amd_iommu *iommu)
510 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
513 /* sets a specific bit in the device table entry. */
514 static void set_dev_entry_bit(u16 devid, u8 bit)
516 int i = (bit >> 5) & 0x07;
517 int _bit = bit & 0x1f;
519 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
522 static int get_dev_entry_bit(u16 devid, u8 bit)
524 int i = (bit >> 5) & 0x07;
525 int _bit = bit & 0x1f;
527 return (amd_iommu_dev_table[devid].data[i] & (1 << _bit)) >> _bit;
531 void amd_iommu_apply_erratum_63(u16 devid)
533 int sysmgt;
535 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
536 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
538 if (sysmgt == 0x01)
539 set_dev_entry_bit(devid, DEV_ENTRY_IW);
542 /* Writes the specific IOMMU for a device into the rlookup table */
543 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
545 amd_iommu_rlookup_table[devid] = iommu;
549 * This function takes the device specific flags read from the ACPI
550 * table and sets up the device table entry with that information
552 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
553 u16 devid, u32 flags, u32 ext_flags)
555 if (flags & ACPI_DEVFLAG_INITPASS)
556 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
557 if (flags & ACPI_DEVFLAG_EXTINT)
558 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
559 if (flags & ACPI_DEVFLAG_NMI)
560 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
561 if (flags & ACPI_DEVFLAG_SYSMGT1)
562 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
563 if (flags & ACPI_DEVFLAG_SYSMGT2)
564 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
565 if (flags & ACPI_DEVFLAG_LINT0)
566 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
567 if (flags & ACPI_DEVFLAG_LINT1)
568 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
570 amd_iommu_apply_erratum_63(devid);
572 set_iommu_for_device(iommu, devid);
576 * Reads the device exclusion range from ACPI and initialize IOMMU with
577 * it
579 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
581 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
583 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
584 return;
586 if (iommu) {
588 * We only can configure exclusion ranges per IOMMU, not
589 * per device. But we can enable the exclusion range per
590 * device. This is done here
592 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
593 iommu->exclusion_start = m->range_start;
594 iommu->exclusion_length = m->range_length;
599 * This function reads some important data from the IOMMU PCI space and
600 * initializes the driver data structure with it. It reads the hardware
601 * capabilities and the first/last device entries
603 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
605 int cap_ptr = iommu->cap_ptr;
606 u32 range, misc;
608 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
609 &iommu->cap);
610 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
611 &range);
612 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
613 &misc);
615 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
616 MMIO_GET_FD(range));
617 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
618 MMIO_GET_LD(range));
619 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
623 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
624 * initializes the hardware and our data structures with it.
626 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
627 struct ivhd_header *h)
629 u8 *p = (u8 *)h;
630 u8 *end = p, flags = 0;
631 u16 dev_i, devid = 0, devid_start = 0, devid_to = 0;
632 u32 ext_flags = 0;
633 bool alias = false;
634 struct ivhd_entry *e;
637 * First set the recommended feature enable bits from ACPI
638 * into the IOMMU control registers
640 h->flags & IVHD_FLAG_HT_TUN_EN_MASK ?
641 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
642 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
644 h->flags & IVHD_FLAG_PASSPW_EN_MASK ?
645 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
646 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
648 h->flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
649 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
650 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
652 h->flags & IVHD_FLAG_ISOC_EN_MASK ?
653 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
654 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
657 * make IOMMU memory accesses cache coherent
659 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
662 * Done. Now parse the device entries
664 p += sizeof(struct ivhd_header);
665 end += h->length;
668 while (p < end) {
669 e = (struct ivhd_entry *)p;
670 switch (e->type) {
671 case IVHD_DEV_ALL:
673 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
674 " last device %02x:%02x.%x flags: %02x\n",
675 PCI_BUS(iommu->first_device),
676 PCI_SLOT(iommu->first_device),
677 PCI_FUNC(iommu->first_device),
678 PCI_BUS(iommu->last_device),
679 PCI_SLOT(iommu->last_device),
680 PCI_FUNC(iommu->last_device),
681 e->flags);
683 for (dev_i = iommu->first_device;
684 dev_i <= iommu->last_device; ++dev_i)
685 set_dev_entry_from_acpi(iommu, dev_i,
686 e->flags, 0);
687 break;
688 case IVHD_DEV_SELECT:
690 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
691 "flags: %02x\n",
692 PCI_BUS(e->devid),
693 PCI_SLOT(e->devid),
694 PCI_FUNC(e->devid),
695 e->flags);
697 devid = e->devid;
698 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
699 break;
700 case IVHD_DEV_SELECT_RANGE_START:
702 DUMP_printk(" DEV_SELECT_RANGE_START\t "
703 "devid: %02x:%02x.%x flags: %02x\n",
704 PCI_BUS(e->devid),
705 PCI_SLOT(e->devid),
706 PCI_FUNC(e->devid),
707 e->flags);
709 devid_start = e->devid;
710 flags = e->flags;
711 ext_flags = 0;
712 alias = false;
713 break;
714 case IVHD_DEV_ALIAS:
716 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
717 "flags: %02x devid_to: %02x:%02x.%x\n",
718 PCI_BUS(e->devid),
719 PCI_SLOT(e->devid),
720 PCI_FUNC(e->devid),
721 e->flags,
722 PCI_BUS(e->ext >> 8),
723 PCI_SLOT(e->ext >> 8),
724 PCI_FUNC(e->ext >> 8));
726 devid = e->devid;
727 devid_to = e->ext >> 8;
728 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
729 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
730 amd_iommu_alias_table[devid] = devid_to;
731 break;
732 case IVHD_DEV_ALIAS_RANGE:
734 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
735 "devid: %02x:%02x.%x flags: %02x "
736 "devid_to: %02x:%02x.%x\n",
737 PCI_BUS(e->devid),
738 PCI_SLOT(e->devid),
739 PCI_FUNC(e->devid),
740 e->flags,
741 PCI_BUS(e->ext >> 8),
742 PCI_SLOT(e->ext >> 8),
743 PCI_FUNC(e->ext >> 8));
745 devid_start = e->devid;
746 flags = e->flags;
747 devid_to = e->ext >> 8;
748 ext_flags = 0;
749 alias = true;
750 break;
751 case IVHD_DEV_EXT_SELECT:
753 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
754 "flags: %02x ext: %08x\n",
755 PCI_BUS(e->devid),
756 PCI_SLOT(e->devid),
757 PCI_FUNC(e->devid),
758 e->flags, e->ext);
760 devid = e->devid;
761 set_dev_entry_from_acpi(iommu, devid, e->flags,
762 e->ext);
763 break;
764 case IVHD_DEV_EXT_SELECT_RANGE:
766 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
767 "%02x:%02x.%x flags: %02x ext: %08x\n",
768 PCI_BUS(e->devid),
769 PCI_SLOT(e->devid),
770 PCI_FUNC(e->devid),
771 e->flags, e->ext);
773 devid_start = e->devid;
774 flags = e->flags;
775 ext_flags = e->ext;
776 alias = false;
777 break;
778 case IVHD_DEV_RANGE_END:
780 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
781 PCI_BUS(e->devid),
782 PCI_SLOT(e->devid),
783 PCI_FUNC(e->devid));
785 devid = e->devid;
786 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
787 if (alias) {
788 amd_iommu_alias_table[dev_i] = devid_to;
789 set_dev_entry_from_acpi(iommu,
790 devid_to, flags, ext_flags);
792 set_dev_entry_from_acpi(iommu, dev_i,
793 flags, ext_flags);
795 break;
796 default:
797 break;
800 p += ivhd_entry_length(p);
804 /* Initializes the device->iommu mapping for the driver */
805 static int __init init_iommu_devices(struct amd_iommu *iommu)
807 u16 i;
809 for (i = iommu->first_device; i <= iommu->last_device; ++i)
810 set_iommu_for_device(iommu, i);
812 return 0;
815 static void __init free_iommu_one(struct amd_iommu *iommu)
817 free_command_buffer(iommu);
818 free_event_buffer(iommu);
819 iommu_unmap_mmio_space(iommu);
822 static void __init free_iommu_all(void)
824 struct amd_iommu *iommu, *next;
826 for_each_iommu_safe(iommu, next) {
827 list_del(&iommu->list);
828 free_iommu_one(iommu);
829 kfree(iommu);
834 * This function clues the initialization function for one IOMMU
835 * together and also allocates the command buffer and programs the
836 * hardware. It does NOT enable the IOMMU. This is done afterwards.
838 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
840 spin_lock_init(&iommu->lock);
841 list_add_tail(&iommu->list, &amd_iommu_list);
844 * Copy data from ACPI table entry to the iommu struct
846 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
847 if (!iommu->dev)
848 return 1;
850 iommu->cap_ptr = h->cap_ptr;
851 iommu->pci_seg = h->pci_seg;
852 iommu->mmio_phys = h->mmio_phys;
853 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
854 if (!iommu->mmio_base)
855 return -ENOMEM;
857 iommu->cmd_buf = alloc_command_buffer(iommu);
858 if (!iommu->cmd_buf)
859 return -ENOMEM;
861 iommu->evt_buf = alloc_event_buffer(iommu);
862 if (!iommu->evt_buf)
863 return -ENOMEM;
865 iommu->int_enabled = false;
867 init_iommu_from_pci(iommu);
868 init_iommu_from_acpi(iommu, h);
869 init_iommu_devices(iommu);
871 return pci_enable_device(iommu->dev);
875 * Iterates over all IOMMU entries in the ACPI table, allocates the
876 * IOMMU structure and initializes it with init_iommu_one()
878 static int __init init_iommu_all(struct acpi_table_header *table)
880 u8 *p = (u8 *)table, *end = (u8 *)table;
881 struct ivhd_header *h;
882 struct amd_iommu *iommu;
883 int ret;
885 end += table->length;
886 p += IVRS_HEADER_LENGTH;
888 while (p < end) {
889 h = (struct ivhd_header *)p;
890 switch (*p) {
891 case ACPI_IVHD_TYPE:
893 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
894 "seg: %d flags: %01x info %04x\n",
895 PCI_BUS(h->devid), PCI_SLOT(h->devid),
896 PCI_FUNC(h->devid), h->cap_ptr,
897 h->pci_seg, h->flags, h->info);
898 DUMP_printk(" mmio-addr: %016llx\n",
899 h->mmio_phys);
901 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
902 if (iommu == NULL)
903 return -ENOMEM;
904 ret = init_iommu_one(iommu, h);
905 if (ret)
906 return ret;
907 break;
908 default:
909 break;
911 p += h->length;
914 WARN_ON(p != end);
916 return 0;
919 /****************************************************************************
921 * The following functions initialize the MSI interrupts for all IOMMUs
922 * in the system. Its a bit challenging because there could be multiple
923 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
924 * pci_dev.
926 ****************************************************************************/
928 static int __init iommu_setup_msi(struct amd_iommu *iommu)
930 int r;
932 if (pci_enable_msi(iommu->dev))
933 return 1;
935 r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
936 IRQF_SAMPLE_RANDOM,
937 "AMD-Vi",
938 NULL);
940 if (r) {
941 pci_disable_msi(iommu->dev);
942 return 1;
945 iommu->int_enabled = true;
946 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
948 return 0;
951 static int iommu_init_msi(struct amd_iommu *iommu)
953 if (iommu->int_enabled)
954 return 0;
956 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
957 return iommu_setup_msi(iommu);
959 return 1;
962 /****************************************************************************
964 * The next functions belong to the third pass of parsing the ACPI
965 * table. In this last pass the memory mapping requirements are
966 * gathered (like exclusion and unity mapping reanges).
968 ****************************************************************************/
970 static void __init free_unity_maps(void)
972 struct unity_map_entry *entry, *next;
974 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
975 list_del(&entry->list);
976 kfree(entry);
980 /* called when we find an exclusion range definition in ACPI */
981 static int __init init_exclusion_range(struct ivmd_header *m)
983 int i;
985 switch (m->type) {
986 case ACPI_IVMD_TYPE:
987 set_device_exclusion_range(m->devid, m);
988 break;
989 case ACPI_IVMD_TYPE_ALL:
990 for (i = 0; i <= amd_iommu_last_bdf; ++i)
991 set_device_exclusion_range(i, m);
992 break;
993 case ACPI_IVMD_TYPE_RANGE:
994 for (i = m->devid; i <= m->aux; ++i)
995 set_device_exclusion_range(i, m);
996 break;
997 default:
998 break;
1001 return 0;
1004 /* called for unity map ACPI definition */
1005 static int __init init_unity_map_range(struct ivmd_header *m)
1007 struct unity_map_entry *e = 0;
1008 char *s;
1010 e = kzalloc(sizeof(*e), GFP_KERNEL);
1011 if (e == NULL)
1012 return -ENOMEM;
1014 switch (m->type) {
1015 default:
1016 kfree(e);
1017 return 0;
1018 case ACPI_IVMD_TYPE:
1019 s = "IVMD_TYPEi\t\t\t";
1020 e->devid_start = e->devid_end = m->devid;
1021 break;
1022 case ACPI_IVMD_TYPE_ALL:
1023 s = "IVMD_TYPE_ALL\t\t";
1024 e->devid_start = 0;
1025 e->devid_end = amd_iommu_last_bdf;
1026 break;
1027 case ACPI_IVMD_TYPE_RANGE:
1028 s = "IVMD_TYPE_RANGE\t\t";
1029 e->devid_start = m->devid;
1030 e->devid_end = m->aux;
1031 break;
1033 e->address_start = PAGE_ALIGN(m->range_start);
1034 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1035 e->prot = m->flags >> 1;
1037 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1038 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1039 PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1040 PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1041 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1042 e->address_start, e->address_end, m->flags);
1044 list_add_tail(&e->list, &amd_iommu_unity_map);
1046 return 0;
1049 /* iterates over all memory definitions we find in the ACPI table */
1050 static int __init init_memory_definitions(struct acpi_table_header *table)
1052 u8 *p = (u8 *)table, *end = (u8 *)table;
1053 struct ivmd_header *m;
1055 end += table->length;
1056 p += IVRS_HEADER_LENGTH;
1058 while (p < end) {
1059 m = (struct ivmd_header *)p;
1060 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1061 init_exclusion_range(m);
1062 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1063 init_unity_map_range(m);
1065 p += m->length;
1068 return 0;
1072 * Init the device table to not allow DMA access for devices and
1073 * suppress all page faults
1075 static void init_device_table(void)
1077 u16 devid;
1079 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1080 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1081 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1086 * This function finally enables all IOMMUs found in the system after
1087 * they have been initialized
1089 static void enable_iommus(void)
1091 struct amd_iommu *iommu;
1093 for_each_iommu(iommu) {
1094 iommu_disable(iommu);
1095 iommu_set_device_table(iommu);
1096 iommu_enable_command_buffer(iommu);
1097 iommu_enable_event_buffer(iommu);
1098 iommu_set_exclusion_range(iommu);
1099 iommu_init_msi(iommu);
1100 iommu_enable(iommu);
1104 static void disable_iommus(void)
1106 struct amd_iommu *iommu;
1108 for_each_iommu(iommu)
1109 iommu_disable(iommu);
1113 * Suspend/Resume support
1114 * disable suspend until real resume implemented
1117 static int amd_iommu_resume(struct sys_device *dev)
1119 /* re-load the hardware */
1120 enable_iommus();
1123 * we have to flush after the IOMMUs are enabled because a
1124 * disabled IOMMU will never execute the commands we send
1126 amd_iommu_flush_all_devices();
1127 amd_iommu_flush_all_domains();
1129 return 0;
1132 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
1134 /* disable IOMMUs to go out of the way for BIOS */
1135 disable_iommus();
1137 return 0;
1140 static struct sysdev_class amd_iommu_sysdev_class = {
1141 .name = "amd_iommu",
1142 .suspend = amd_iommu_suspend,
1143 .resume = amd_iommu_resume,
1146 static struct sys_device device_amd_iommu = {
1147 .id = 0,
1148 .cls = &amd_iommu_sysdev_class,
1152 * This is the core init function for AMD IOMMU hardware in the system.
1153 * This function is called from the generic x86 DMA layer initialization
1154 * code.
1156 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1157 * three times:
1159 * 1 pass) Find the highest PCI device id the driver has to handle.
1160 * Upon this information the size of the data structures is
1161 * determined that needs to be allocated.
1163 * 2 pass) Initialize the data structures just allocated with the
1164 * information in the ACPI table about available AMD IOMMUs
1165 * in the system. It also maps the PCI devices in the
1166 * system to specific IOMMUs
1168 * 3 pass) After the basic data structures are allocated and
1169 * initialized we update them with information about memory
1170 * remapping requirements parsed out of the ACPI table in
1171 * this last pass.
1173 * After that the hardware is initialized and ready to go. In the last
1174 * step we do some Linux specific things like registering the driver in
1175 * the dma_ops interface and initializing the suspend/resume support
1176 * functions. Finally it prints some information about AMD IOMMUs and
1177 * the driver state and enables the hardware.
1179 int __init amd_iommu_init(void)
1181 int i, ret = 0;
1184 if (no_iommu) {
1185 printk(KERN_INFO "AMD-Vi disabled by kernel command line\n");
1186 return 0;
1189 if (!amd_iommu_detected)
1190 return -ENODEV;
1193 * First parse ACPI tables to find the largest Bus/Dev/Func
1194 * we need to handle. Upon this information the shared data
1195 * structures for the IOMMUs in the system will be allocated
1197 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1198 return -ENODEV;
1200 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1201 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1202 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1204 ret = -ENOMEM;
1206 /* Device table - directly used by all IOMMUs */
1207 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1208 get_order(dev_table_size));
1209 if (amd_iommu_dev_table == NULL)
1210 goto out;
1213 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1214 * IOMMU see for that device
1216 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1217 get_order(alias_table_size));
1218 if (amd_iommu_alias_table == NULL)
1219 goto free;
1221 /* IOMMU rlookup table - find the IOMMU for a specific device */
1222 amd_iommu_rlookup_table = (void *)__get_free_pages(
1223 GFP_KERNEL | __GFP_ZERO,
1224 get_order(rlookup_table_size));
1225 if (amd_iommu_rlookup_table == NULL)
1226 goto free;
1229 * Protection Domain table - maps devices to protection domains
1230 * This table has the same size as the rlookup_table
1232 amd_iommu_pd_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1233 get_order(rlookup_table_size));
1234 if (amd_iommu_pd_table == NULL)
1235 goto free;
1237 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1238 GFP_KERNEL | __GFP_ZERO,
1239 get_order(MAX_DOMAIN_ID/8));
1240 if (amd_iommu_pd_alloc_bitmap == NULL)
1241 goto free;
1243 /* init the device table */
1244 init_device_table();
1247 * let all alias entries point to itself
1249 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1250 amd_iommu_alias_table[i] = i;
1253 * never allocate domain 0 because its used as the non-allocated and
1254 * error value placeholder
1256 amd_iommu_pd_alloc_bitmap[0] = 1;
1259 * now the data structures are allocated and basically initialized
1260 * start the real acpi table scan
1262 ret = -ENODEV;
1263 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1264 goto free;
1266 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1267 goto free;
1269 ret = sysdev_class_register(&amd_iommu_sysdev_class);
1270 if (ret)
1271 goto free;
1273 ret = sysdev_register(&device_amd_iommu);
1274 if (ret)
1275 goto free;
1277 if (iommu_pass_through)
1278 ret = amd_iommu_init_passthrough();
1279 else
1280 ret = amd_iommu_init_dma_ops();
1281 if (ret)
1282 goto free;
1284 enable_iommus();
1286 if (iommu_pass_through)
1287 goto out;
1289 printk(KERN_INFO "AMD-Vi: device isolation ");
1290 if (amd_iommu_isolate)
1291 printk("enabled\n");
1292 else
1293 printk("disabled\n");
1295 if (amd_iommu_unmap_flush)
1296 printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
1297 else
1298 printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n");
1300 out:
1301 return ret;
1303 free:
1304 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1305 get_order(MAX_DOMAIN_ID/8));
1307 free_pages((unsigned long)amd_iommu_pd_table,
1308 get_order(rlookup_table_size));
1310 free_pages((unsigned long)amd_iommu_rlookup_table,
1311 get_order(rlookup_table_size));
1313 free_pages((unsigned long)amd_iommu_alias_table,
1314 get_order(alias_table_size));
1316 free_pages((unsigned long)amd_iommu_dev_table,
1317 get_order(dev_table_size));
1319 free_iommu_all();
1321 free_unity_maps();
1323 goto out;
1326 void amd_iommu_shutdown(void)
1328 disable_iommus();
1331 /****************************************************************************
1333 * Early detect code. This code runs at IOMMU detection time in the DMA
1334 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1335 * IOMMUs
1337 ****************************************************************************/
1338 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1340 return 0;
1343 void __init amd_iommu_detect(void)
1345 if (swiotlb || no_iommu || (iommu_detected && !gart_iommu_aperture))
1346 return;
1348 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1349 iommu_detected = 1;
1350 amd_iommu_detected = 1;
1351 #ifdef CONFIG_GART_IOMMU
1352 gart_iommu_aperture_disabled = 1;
1353 gart_iommu_aperture = 0;
1354 #endif
1358 /****************************************************************************
1360 * Parsing functions for the AMD IOMMU specific kernel command line
1361 * options.
1363 ****************************************************************************/
1365 static int __init parse_amd_iommu_dump(char *str)
1367 amd_iommu_dump = true;
1369 return 1;
1372 static int __init parse_amd_iommu_options(char *str)
1374 for (; *str; ++str) {
1375 if (strncmp(str, "isolate", 7) == 0)
1376 amd_iommu_isolate = true;
1377 if (strncmp(str, "share", 5) == 0)
1378 amd_iommu_isolate = false;
1379 if (strncmp(str, "fullflush", 9) == 0)
1380 amd_iommu_unmap_flush = true;
1383 return 1;
1386 __setup("amd_iommu_dump", parse_amd_iommu_dump);
1387 __setup("amd_iommu=", parse_amd_iommu_options);