2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
4 * Leo Duran <leo.duran@amd.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/pci.h>
21 #include <linux/acpi.h>
22 #include <linux/list.h>
23 #include <linux/slab.h>
24 #include <linux/syscore_ops.h>
25 #include <linux/interrupt.h>
26 #include <linux/msi.h>
27 #include <linux/amd-iommu.h>
28 #include <linux/export.h>
29 #include <asm/pci-direct.h>
30 #include <asm/iommu.h>
32 #include <asm/x86_init.h>
33 #include <asm/iommu_table.h>
35 #include "amd_iommu_proto.h"
36 #include "amd_iommu_types.h"
39 * definitions for the ACPI scanning code
41 #define IVRS_HEADER_LENGTH 48
43 #define ACPI_IVHD_TYPE 0x10
44 #define ACPI_IVMD_TYPE_ALL 0x20
45 #define ACPI_IVMD_TYPE 0x21
46 #define ACPI_IVMD_TYPE_RANGE 0x22
48 #define IVHD_DEV_ALL 0x01
49 #define IVHD_DEV_SELECT 0x02
50 #define IVHD_DEV_SELECT_RANGE_START 0x03
51 #define IVHD_DEV_RANGE_END 0x04
52 #define IVHD_DEV_ALIAS 0x42
53 #define IVHD_DEV_ALIAS_RANGE 0x43
54 #define IVHD_DEV_EXT_SELECT 0x46
55 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
57 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
58 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
59 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
60 #define IVHD_FLAG_ISOC_EN_MASK 0x08
62 #define IVMD_FLAG_EXCL_RANGE 0x08
63 #define IVMD_FLAG_UNITY_MAP 0x01
65 #define ACPI_DEVFLAG_INITPASS 0x01
66 #define ACPI_DEVFLAG_EXTINT 0x02
67 #define ACPI_DEVFLAG_NMI 0x04
68 #define ACPI_DEVFLAG_SYSMGT1 0x10
69 #define ACPI_DEVFLAG_SYSMGT2 0x20
70 #define ACPI_DEVFLAG_LINT0 0x40
71 #define ACPI_DEVFLAG_LINT1 0x80
72 #define ACPI_DEVFLAG_ATSDIS 0x10000000
75 * ACPI table definitions
77 * These data structures are laid over the table to parse the important values
82 * structure describing one IOMMU in the ACPI table. Typically followed by one
83 * or more ivhd_entrys.
95 } __attribute__((packed));
98 * A device entry describing which devices a specific IOMMU translates and
99 * which requestor ids they use.
106 } __attribute__((packed));
109 * An AMD IOMMU memory definition structure. It defines things like exclusion
110 * ranges for devices and regions that should be unity mapped.
121 } __attribute__((packed));
125 static int __initdata amd_iommu_detected;
126 static bool __initdata amd_iommu_disabled;
128 u16 amd_iommu_last_bdf; /* largest PCI device id we have
130 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
132 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
134 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
137 /* Array to assign indices to IOMMUs*/
138 struct amd_iommu *amd_iommus[MAX_IOMMUS];
139 int amd_iommus_present;
141 /* IOMMUs have a non-present cache? */
142 bool amd_iommu_np_cache __read_mostly;
143 bool amd_iommu_iotlb_sup __read_mostly = true;
145 u32 amd_iommu_max_pasids __read_mostly = ~0;
147 bool amd_iommu_v2_present __read_mostly;
149 bool amd_iommu_force_isolation __read_mostly;
152 * The ACPI table parsing functions set this variable on an error
154 static int __initdata amd_iommu_init_err;
157 * List of protection domains - used during resume
159 LIST_HEAD(amd_iommu_pd_list);
160 spinlock_t amd_iommu_pd_lock;
163 * Pointer to the device table which is shared by all AMD IOMMUs
164 * it is indexed by the PCI device id or the HT unit id and contains
165 * information about the domain the device belongs to as well as the
166 * page table root pointer.
168 struct dev_table_entry *amd_iommu_dev_table;
171 * The alias table is a driver specific data structure which contains the
172 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
173 * More than one device can share the same requestor id.
175 u16 *amd_iommu_alias_table;
178 * The rlookup table is used to find the IOMMU which is responsible
179 * for a specific device. It is also indexed by the PCI device id.
181 struct amd_iommu **amd_iommu_rlookup_table;
184 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
185 * to know which ones are already in use.
187 unsigned long *amd_iommu_pd_alloc_bitmap;
189 static u32 dev_table_size; /* size of the device table */
190 static u32 alias_table_size; /* size of the alias table */
191 static u32 rlookup_table_size; /* size if the rlookup table */
194 * This function flushes all internal caches of
195 * the IOMMU used by this driver.
197 extern void iommu_flush_all_caches(struct amd_iommu *iommu);
199 static inline void update_last_devid(u16 devid)
201 if (devid > amd_iommu_last_bdf)
202 amd_iommu_last_bdf = devid;
205 static inline unsigned long tbl_size(int entry_size)
207 unsigned shift = PAGE_SHIFT +
208 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
213 /* Access to l1 and l2 indexed register spaces */
215 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
219 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
220 pci_read_config_dword(iommu->dev, 0xfc, &val);
224 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
226 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
227 pci_write_config_dword(iommu->dev, 0xfc, val);
228 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
231 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
235 pci_write_config_dword(iommu->dev, 0xf0, address);
236 pci_read_config_dword(iommu->dev, 0xf4, &val);
240 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
242 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
243 pci_write_config_dword(iommu->dev, 0xf4, val);
246 /****************************************************************************
248 * AMD IOMMU MMIO register space handling functions
250 * These functions are used to program the IOMMU device registers in
251 * MMIO space required for that driver.
253 ****************************************************************************/
256 * This function set the exclusion range in the IOMMU. DMA accesses to the
257 * exclusion range are passed through untranslated
259 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
261 u64 start = iommu->exclusion_start & PAGE_MASK;
262 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
265 if (!iommu->exclusion_start)
268 entry = start | MMIO_EXCL_ENABLE_MASK;
269 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
270 &entry, sizeof(entry));
273 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
274 &entry, sizeof(entry));
277 /* Programs the physical address of the device table into the IOMMU hardware */
278 static void __init iommu_set_device_table(struct amd_iommu *iommu)
282 BUG_ON(iommu->mmio_base == NULL);
284 entry = virt_to_phys(amd_iommu_dev_table);
285 entry |= (dev_table_size >> 12) - 1;
286 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
287 &entry, sizeof(entry));
290 /* Generic functions to enable/disable certain features of the IOMMU. */
291 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
295 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
297 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
300 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
304 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
306 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
309 /* Function to enable the hardware */
310 static void iommu_enable(struct amd_iommu *iommu)
312 static const char * const feat_str[] = {
313 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
314 "IA", "GA", "HE", "PC", NULL
318 printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx",
319 dev_name(&iommu->dev->dev), iommu->cap_ptr);
321 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
322 printk(KERN_CONT " extended features: ");
323 for (i = 0; feat_str[i]; ++i)
324 if (iommu_feature(iommu, (1ULL << i)))
325 printk(KERN_CONT " %s", feat_str[i]);
327 printk(KERN_CONT "\n");
329 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
332 static void iommu_disable(struct amd_iommu *iommu)
334 /* Disable command buffer */
335 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
337 /* Disable event logging and event interrupts */
338 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
339 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
341 /* Disable IOMMU hardware itself */
342 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
346 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
347 * the system has one.
349 static u8 * __init iommu_map_mmio_space(u64 address)
353 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
354 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
356 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
360 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
364 release_mem_region(address, MMIO_REGION_LENGTH);
369 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
371 if (iommu->mmio_base)
372 iounmap(iommu->mmio_base);
373 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
376 /****************************************************************************
378 * The functions below belong to the first pass of AMD IOMMU ACPI table
379 * parsing. In this pass we try to find out the highest device id this
380 * code has to handle. Upon this information the size of the shared data
381 * structures is determined later.
383 ****************************************************************************/
386 * This function calculates the length of a given IVHD entry
388 static inline int ivhd_entry_length(u8 *ivhd)
390 return 0x04 << (*ivhd >> 6);
394 * This function reads the last device id the IOMMU has to handle from the PCI
395 * capability header for this IOMMU
397 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
401 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
402 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
408 * After reading the highest device id from the IOMMU PCI capability header
409 * this function looks if there is a higher device id defined in the ACPI table
411 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
413 u8 *p = (void *)h, *end = (void *)h;
414 struct ivhd_entry *dev;
419 find_last_devid_on_pci(PCI_BUS(h->devid),
425 dev = (struct ivhd_entry *)p;
427 case IVHD_DEV_SELECT:
428 case IVHD_DEV_RANGE_END:
430 case IVHD_DEV_EXT_SELECT:
431 /* all the above subfield types refer to device ids */
432 update_last_devid(dev->devid);
437 p += ivhd_entry_length(p);
446 * Iterate over all IVHD entries in the ACPI table and find the highest device
447 * id which we need to handle. This is the first of three functions which parse
448 * the ACPI table. So we check the checksum here.
450 static int __init find_last_devid_acpi(struct acpi_table_header *table)
453 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
454 struct ivhd_header *h;
457 * Validate checksum here so we don't need to do it when
458 * we actually parse the table
460 for (i = 0; i < table->length; ++i)
463 /* ACPI table corrupt */
464 amd_iommu_init_err = -ENODEV;
468 p += IVRS_HEADER_LENGTH;
470 end += table->length;
472 h = (struct ivhd_header *)p;
475 find_last_devid_from_ivhd(h);
487 /****************************************************************************
489 * The following functions belong the the code path which parses the ACPI table
490 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
491 * data structures, initialize the device/alias/rlookup table and also
492 * basically initialize the hardware.
494 ****************************************************************************/
497 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
498 * write commands to that buffer later and the IOMMU will execute them
501 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
503 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
504 get_order(CMD_BUFFER_SIZE));
509 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
515 * This function resets the command buffer if the IOMMU stopped fetching
518 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
520 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
522 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
523 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
525 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
529 * This function writes the command buffer address to the hardware and
532 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
536 BUG_ON(iommu->cmd_buf == NULL);
538 entry = (u64)virt_to_phys(iommu->cmd_buf);
539 entry |= MMIO_CMD_SIZE_512;
541 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
542 &entry, sizeof(entry));
544 amd_iommu_reset_cmd_buffer(iommu);
545 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
548 static void __init free_command_buffer(struct amd_iommu *iommu)
550 free_pages((unsigned long)iommu->cmd_buf,
551 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
554 /* allocates the memory where the IOMMU will log its events to */
555 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
557 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
558 get_order(EVT_BUFFER_SIZE));
560 if (iommu->evt_buf == NULL)
563 iommu->evt_buf_size = EVT_BUFFER_SIZE;
565 return iommu->evt_buf;
568 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
572 BUG_ON(iommu->evt_buf == NULL);
574 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
576 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
577 &entry, sizeof(entry));
579 /* set head and tail to zero manually */
580 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
581 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
583 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
586 static void __init free_event_buffer(struct amd_iommu *iommu)
588 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
591 /* allocates the memory where the IOMMU will log its events to */
592 static u8 * __init alloc_ppr_log(struct amd_iommu *iommu)
594 iommu->ppr_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
595 get_order(PPR_LOG_SIZE));
597 if (iommu->ppr_log == NULL)
600 return iommu->ppr_log;
603 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
607 if (iommu->ppr_log == NULL)
610 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
612 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
613 &entry, sizeof(entry));
615 /* set head and tail to zero manually */
616 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
617 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
619 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
620 iommu_feature_enable(iommu, CONTROL_PPR_EN);
623 static void __init free_ppr_log(struct amd_iommu *iommu)
625 if (iommu->ppr_log == NULL)
628 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
631 static void iommu_enable_gt(struct amd_iommu *iommu)
633 if (!iommu_feature(iommu, FEATURE_GT))
636 iommu_feature_enable(iommu, CONTROL_GT_EN);
639 /* sets a specific bit in the device table entry. */
640 static void set_dev_entry_bit(u16 devid, u8 bit)
642 int i = (bit >> 6) & 0x03;
643 int _bit = bit & 0x3f;
645 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
648 static int get_dev_entry_bit(u16 devid, u8 bit)
650 int i = (bit >> 6) & 0x03;
651 int _bit = bit & 0x3f;
653 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
657 void amd_iommu_apply_erratum_63(u16 devid)
661 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
662 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
665 set_dev_entry_bit(devid, DEV_ENTRY_IW);
668 /* Writes the specific IOMMU for a device into the rlookup table */
669 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
671 amd_iommu_rlookup_table[devid] = iommu;
675 * This function takes the device specific flags read from the ACPI
676 * table and sets up the device table entry with that information
678 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
679 u16 devid, u32 flags, u32 ext_flags)
681 if (flags & ACPI_DEVFLAG_INITPASS)
682 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
683 if (flags & ACPI_DEVFLAG_EXTINT)
684 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
685 if (flags & ACPI_DEVFLAG_NMI)
686 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
687 if (flags & ACPI_DEVFLAG_SYSMGT1)
688 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
689 if (flags & ACPI_DEVFLAG_SYSMGT2)
690 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
691 if (flags & ACPI_DEVFLAG_LINT0)
692 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
693 if (flags & ACPI_DEVFLAG_LINT1)
694 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
696 amd_iommu_apply_erratum_63(devid);
698 set_iommu_for_device(iommu, devid);
702 * Reads the device exclusion range from ACPI and initialize IOMMU with
705 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
707 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
709 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
714 * We only can configure exclusion ranges per IOMMU, not
715 * per device. But we can enable the exclusion range per
716 * device. This is done here
718 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
719 iommu->exclusion_start = m->range_start;
720 iommu->exclusion_length = m->range_length;
725 * This function reads some important data from the IOMMU PCI space and
726 * initializes the driver data structure with it. It reads the hardware
727 * capabilities and the first/last device entries
729 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
731 int cap_ptr = iommu->cap_ptr;
732 u32 range, misc, low, high;
735 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
737 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
739 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
742 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
744 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
746 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
748 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
749 amd_iommu_iotlb_sup = false;
751 /* read extended feature bits */
752 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
753 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
755 iommu->features = ((u64)high << 32) | low;
757 if (iommu_feature(iommu, FEATURE_GT)) {
761 shift = iommu->features & FEATURE_PASID_MASK;
762 shift >>= FEATURE_PASID_SHIFT;
763 pasids = (1 << shift);
765 amd_iommu_max_pasids = min(amd_iommu_max_pasids, pasids);
768 if (iommu_feature(iommu, FEATURE_GT) &&
769 iommu_feature(iommu, FEATURE_PPR)) {
770 iommu->is_iommu_v2 = true;
771 amd_iommu_v2_present = true;
774 if (!is_rd890_iommu(iommu->dev))
778 * Some rd890 systems may not be fully reconfigured by the BIOS, so
779 * it's necessary for us to store this information so it can be
780 * reprogrammed on resume
783 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
784 &iommu->stored_addr_lo);
785 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
786 &iommu->stored_addr_hi);
788 /* Low bit locks writes to configuration space */
789 iommu->stored_addr_lo &= ~1;
791 for (i = 0; i < 6; i++)
792 for (j = 0; j < 0x12; j++)
793 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
795 for (i = 0; i < 0x83; i++)
796 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
800 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
801 * initializes the hardware and our data structures with it.
803 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
804 struct ivhd_header *h)
807 u8 *end = p, flags = 0;
808 u16 devid = 0, devid_start = 0, devid_to = 0;
809 u32 dev_i, ext_flags = 0;
811 struct ivhd_entry *e;
814 * First save the recommended feature enable bits from ACPI
816 iommu->acpi_flags = h->flags;
819 * Done. Now parse the device entries
821 p += sizeof(struct ivhd_header);
826 e = (struct ivhd_entry *)p;
830 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
831 " last device %02x:%02x.%x flags: %02x\n",
832 PCI_BUS(iommu->first_device),
833 PCI_SLOT(iommu->first_device),
834 PCI_FUNC(iommu->first_device),
835 PCI_BUS(iommu->last_device),
836 PCI_SLOT(iommu->last_device),
837 PCI_FUNC(iommu->last_device),
840 for (dev_i = iommu->first_device;
841 dev_i <= iommu->last_device; ++dev_i)
842 set_dev_entry_from_acpi(iommu, dev_i,
845 case IVHD_DEV_SELECT:
847 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
855 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
857 case IVHD_DEV_SELECT_RANGE_START:
859 DUMP_printk(" DEV_SELECT_RANGE_START\t "
860 "devid: %02x:%02x.%x flags: %02x\n",
866 devid_start = e->devid;
873 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
874 "flags: %02x devid_to: %02x:%02x.%x\n",
879 PCI_BUS(e->ext >> 8),
880 PCI_SLOT(e->ext >> 8),
881 PCI_FUNC(e->ext >> 8));
884 devid_to = e->ext >> 8;
885 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
886 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
887 amd_iommu_alias_table[devid] = devid_to;
889 case IVHD_DEV_ALIAS_RANGE:
891 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
892 "devid: %02x:%02x.%x flags: %02x "
893 "devid_to: %02x:%02x.%x\n",
898 PCI_BUS(e->ext >> 8),
899 PCI_SLOT(e->ext >> 8),
900 PCI_FUNC(e->ext >> 8));
902 devid_start = e->devid;
904 devid_to = e->ext >> 8;
908 case IVHD_DEV_EXT_SELECT:
910 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
911 "flags: %02x ext: %08x\n",
918 set_dev_entry_from_acpi(iommu, devid, e->flags,
921 case IVHD_DEV_EXT_SELECT_RANGE:
923 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
924 "%02x:%02x.%x flags: %02x ext: %08x\n",
930 devid_start = e->devid;
935 case IVHD_DEV_RANGE_END:
937 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
943 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
945 amd_iommu_alias_table[dev_i] = devid_to;
946 set_dev_entry_from_acpi(iommu,
947 devid_to, flags, ext_flags);
949 set_dev_entry_from_acpi(iommu, dev_i,
957 p += ivhd_entry_length(p);
961 /* Initializes the device->iommu mapping for the driver */
962 static int __init init_iommu_devices(struct amd_iommu *iommu)
966 for (i = iommu->first_device; i <= iommu->last_device; ++i)
967 set_iommu_for_device(iommu, i);
972 static void __init free_iommu_one(struct amd_iommu *iommu)
974 free_command_buffer(iommu);
975 free_event_buffer(iommu);
977 iommu_unmap_mmio_space(iommu);
980 static void __init free_iommu_all(void)
982 struct amd_iommu *iommu, *next;
984 for_each_iommu_safe(iommu, next) {
985 list_del(&iommu->list);
986 free_iommu_one(iommu);
992 * This function clues the initialization function for one IOMMU
993 * together and also allocates the command buffer and programs the
994 * hardware. It does NOT enable the IOMMU. This is done afterwards.
996 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
998 spin_lock_init(&iommu->lock);
1000 /* Add IOMMU to internal data structures */
1001 list_add_tail(&iommu->list, &amd_iommu_list);
1002 iommu->index = amd_iommus_present++;
1004 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1005 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1009 /* Index is fine - add IOMMU to the array */
1010 amd_iommus[iommu->index] = iommu;
1013 * Copy data from ACPI table entry to the iommu struct
1015 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
1019 iommu->cap_ptr = h->cap_ptr;
1020 iommu->pci_seg = h->pci_seg;
1021 iommu->mmio_phys = h->mmio_phys;
1022 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
1023 if (!iommu->mmio_base)
1026 iommu->cmd_buf = alloc_command_buffer(iommu);
1027 if (!iommu->cmd_buf)
1030 iommu->evt_buf = alloc_event_buffer(iommu);
1031 if (!iommu->evt_buf)
1034 iommu->int_enabled = false;
1036 init_iommu_from_pci(iommu);
1037 init_iommu_from_acpi(iommu, h);
1038 init_iommu_devices(iommu);
1040 if (iommu_feature(iommu, FEATURE_PPR)) {
1041 iommu->ppr_log = alloc_ppr_log(iommu);
1042 if (!iommu->ppr_log)
1046 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1047 amd_iommu_np_cache = true;
1049 return pci_enable_device(iommu->dev);
1053 * Iterates over all IOMMU entries in the ACPI table, allocates the
1054 * IOMMU structure and initializes it with init_iommu_one()
1056 static int __init init_iommu_all(struct acpi_table_header *table)
1058 u8 *p = (u8 *)table, *end = (u8 *)table;
1059 struct ivhd_header *h;
1060 struct amd_iommu *iommu;
1063 end += table->length;
1064 p += IVRS_HEADER_LENGTH;
1067 h = (struct ivhd_header *)p;
1069 case ACPI_IVHD_TYPE:
1071 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1072 "seg: %d flags: %01x info %04x\n",
1073 PCI_BUS(h->devid), PCI_SLOT(h->devid),
1074 PCI_FUNC(h->devid), h->cap_ptr,
1075 h->pci_seg, h->flags, h->info);
1076 DUMP_printk(" mmio-addr: %016llx\n",
1079 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1080 if (iommu == NULL) {
1081 amd_iommu_init_err = -ENOMEM;
1085 ret = init_iommu_one(iommu, h);
1087 amd_iommu_init_err = ret;
1102 /****************************************************************************
1104 * The following functions initialize the MSI interrupts for all IOMMUs
1105 * in the system. Its a bit challenging because there could be multiple
1106 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1109 ****************************************************************************/
1111 static int iommu_setup_msi(struct amd_iommu *iommu)
1115 if (pci_enable_msi(iommu->dev))
1118 r = request_threaded_irq(iommu->dev->irq,
1119 amd_iommu_int_handler,
1120 amd_iommu_int_thread,
1125 pci_disable_msi(iommu->dev);
1129 iommu->int_enabled = true;
1130 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1132 if (iommu->ppr_log != NULL)
1133 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1138 static int iommu_init_msi(struct amd_iommu *iommu)
1140 if (iommu->int_enabled)
1143 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
1144 return iommu_setup_msi(iommu);
1149 /****************************************************************************
1151 * The next functions belong to the third pass of parsing the ACPI
1152 * table. In this last pass the memory mapping requirements are
1153 * gathered (like exclusion and unity mapping reanges).
1155 ****************************************************************************/
1157 static void __init free_unity_maps(void)
1159 struct unity_map_entry *entry, *next;
1161 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1162 list_del(&entry->list);
1167 /* called when we find an exclusion range definition in ACPI */
1168 static int __init init_exclusion_range(struct ivmd_header *m)
1173 case ACPI_IVMD_TYPE:
1174 set_device_exclusion_range(m->devid, m);
1176 case ACPI_IVMD_TYPE_ALL:
1177 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1178 set_device_exclusion_range(i, m);
1180 case ACPI_IVMD_TYPE_RANGE:
1181 for (i = m->devid; i <= m->aux; ++i)
1182 set_device_exclusion_range(i, m);
1191 /* called for unity map ACPI definition */
1192 static int __init init_unity_map_range(struct ivmd_header *m)
1194 struct unity_map_entry *e = 0;
1197 e = kzalloc(sizeof(*e), GFP_KERNEL);
1205 case ACPI_IVMD_TYPE:
1206 s = "IVMD_TYPEi\t\t\t";
1207 e->devid_start = e->devid_end = m->devid;
1209 case ACPI_IVMD_TYPE_ALL:
1210 s = "IVMD_TYPE_ALL\t\t";
1212 e->devid_end = amd_iommu_last_bdf;
1214 case ACPI_IVMD_TYPE_RANGE:
1215 s = "IVMD_TYPE_RANGE\t\t";
1216 e->devid_start = m->devid;
1217 e->devid_end = m->aux;
1220 e->address_start = PAGE_ALIGN(m->range_start);
1221 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1222 e->prot = m->flags >> 1;
1224 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1225 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1226 PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1227 PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1228 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1229 e->address_start, e->address_end, m->flags);
1231 list_add_tail(&e->list, &amd_iommu_unity_map);
1236 /* iterates over all memory definitions we find in the ACPI table */
1237 static int __init init_memory_definitions(struct acpi_table_header *table)
1239 u8 *p = (u8 *)table, *end = (u8 *)table;
1240 struct ivmd_header *m;
1242 end += table->length;
1243 p += IVRS_HEADER_LENGTH;
1246 m = (struct ivmd_header *)p;
1247 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1248 init_exclusion_range(m);
1249 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1250 init_unity_map_range(m);
1259 * Init the device table to not allow DMA access for devices and
1260 * suppress all page faults
1262 static void init_device_table(void)
1266 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1267 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1268 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1272 static void iommu_init_flags(struct amd_iommu *iommu)
1274 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1275 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1276 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1278 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1279 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1280 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1282 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1283 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1284 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1286 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1287 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1288 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1291 * make IOMMU memory accesses cache coherent
1293 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1296 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1299 u32 ioc_feature_control;
1300 struct pci_dev *pdev = NULL;
1302 /* RD890 BIOSes may not have completely reconfigured the iommu */
1303 if (!is_rd890_iommu(iommu->dev))
1307 * First, we need to ensure that the iommu is enabled. This is
1308 * controlled by a register in the northbridge
1310 pdev = pci_get_bus_and_slot(iommu->dev->bus->number, PCI_DEVFN(0, 0));
1315 /* Select Northbridge indirect register 0x75 and enable writing */
1316 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1317 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1319 /* Enable the iommu */
1320 if (!(ioc_feature_control & 0x1))
1321 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1325 /* Restore the iommu BAR */
1326 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1327 iommu->stored_addr_lo);
1328 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1329 iommu->stored_addr_hi);
1331 /* Restore the l1 indirect regs for each of the 6 l1s */
1332 for (i = 0; i < 6; i++)
1333 for (j = 0; j < 0x12; j++)
1334 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1336 /* Restore the l2 indirect regs */
1337 for (i = 0; i < 0x83; i++)
1338 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1340 /* Lock PCI setup registers */
1341 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1342 iommu->stored_addr_lo | 1);
1346 * This function finally enables all IOMMUs found in the system after
1347 * they have been initialized
1349 static void enable_iommus(void)
1351 struct amd_iommu *iommu;
1353 for_each_iommu(iommu) {
1354 iommu_disable(iommu);
1355 iommu_init_flags(iommu);
1356 iommu_set_device_table(iommu);
1357 iommu_enable_command_buffer(iommu);
1358 iommu_enable_event_buffer(iommu);
1359 iommu_enable_ppr_log(iommu);
1360 iommu_enable_gt(iommu);
1361 iommu_set_exclusion_range(iommu);
1362 iommu_init_msi(iommu);
1363 iommu_enable(iommu);
1364 iommu_flush_all_caches(iommu);
1368 static void disable_iommus(void)
1370 struct amd_iommu *iommu;
1372 for_each_iommu(iommu)
1373 iommu_disable(iommu);
1377 * Suspend/Resume support
1378 * disable suspend until real resume implemented
1381 static void amd_iommu_resume(void)
1383 struct amd_iommu *iommu;
1385 for_each_iommu(iommu)
1386 iommu_apply_resume_quirks(iommu);
1388 /* re-load the hardware */
1392 * we have to flush after the IOMMUs are enabled because a
1393 * disabled IOMMU will never execute the commands we send
1395 for_each_iommu(iommu)
1396 iommu_flush_all_caches(iommu);
1399 static int amd_iommu_suspend(void)
1401 /* disable IOMMUs to go out of the way for BIOS */
1407 static struct syscore_ops amd_iommu_syscore_ops = {
1408 .suspend = amd_iommu_suspend,
1409 .resume = amd_iommu_resume,
1413 * This is the core init function for AMD IOMMU hardware in the system.
1414 * This function is called from the generic x86 DMA layer initialization
1417 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1420 * 1 pass) Find the highest PCI device id the driver has to handle.
1421 * Upon this information the size of the data structures is
1422 * determined that needs to be allocated.
1424 * 2 pass) Initialize the data structures just allocated with the
1425 * information in the ACPI table about available AMD IOMMUs
1426 * in the system. It also maps the PCI devices in the
1427 * system to specific IOMMUs
1429 * 3 pass) After the basic data structures are allocated and
1430 * initialized we update them with information about memory
1431 * remapping requirements parsed out of the ACPI table in
1434 * After that the hardware is initialized and ready to go. In the last
1435 * step we do some Linux specific things like registering the driver in
1436 * the dma_ops interface and initializing the suspend/resume support
1437 * functions. Finally it prints some information about AMD IOMMUs and
1438 * the driver state and enables the hardware.
1440 static int __init amd_iommu_init(void)
1445 * First parse ACPI tables to find the largest Bus/Dev/Func
1446 * we need to handle. Upon this information the shared data
1447 * structures for the IOMMUs in the system will be allocated
1449 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1452 ret = amd_iommu_init_err;
1456 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1457 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1458 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1462 /* Device table - directly used by all IOMMUs */
1463 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1464 get_order(dev_table_size));
1465 if (amd_iommu_dev_table == NULL)
1469 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1470 * IOMMU see for that device
1472 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1473 get_order(alias_table_size));
1474 if (amd_iommu_alias_table == NULL)
1477 /* IOMMU rlookup table - find the IOMMU for a specific device */
1478 amd_iommu_rlookup_table = (void *)__get_free_pages(
1479 GFP_KERNEL | __GFP_ZERO,
1480 get_order(rlookup_table_size));
1481 if (amd_iommu_rlookup_table == NULL)
1484 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1485 GFP_KERNEL | __GFP_ZERO,
1486 get_order(MAX_DOMAIN_ID/8));
1487 if (amd_iommu_pd_alloc_bitmap == NULL)
1490 /* init the device table */
1491 init_device_table();
1494 * let all alias entries point to itself
1496 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1497 amd_iommu_alias_table[i] = i;
1500 * never allocate domain 0 because its used as the non-allocated and
1501 * error value placeholder
1503 amd_iommu_pd_alloc_bitmap[0] = 1;
1505 spin_lock_init(&amd_iommu_pd_lock);
1508 * now the data structures are allocated and basically initialized
1509 * start the real acpi table scan
1512 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1515 if (amd_iommu_init_err) {
1516 ret = amd_iommu_init_err;
1520 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1523 if (amd_iommu_init_err) {
1524 ret = amd_iommu_init_err;
1528 ret = amd_iommu_init_devices();
1534 if (iommu_pass_through)
1535 ret = amd_iommu_init_passthrough();
1537 ret = amd_iommu_init_dma_ops();
1542 amd_iommu_init_api();
1544 amd_iommu_init_notifier();
1546 register_syscore_ops(&amd_iommu_syscore_ops);
1548 if (iommu_pass_through)
1551 if (amd_iommu_unmap_flush)
1552 printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
1554 printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n");
1556 x86_platform.iommu_shutdown = disable_iommus;
1564 amd_iommu_uninit_devices();
1566 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1567 get_order(MAX_DOMAIN_ID/8));
1569 free_pages((unsigned long)amd_iommu_rlookup_table,
1570 get_order(rlookup_table_size));
1572 free_pages((unsigned long)amd_iommu_alias_table,
1573 get_order(alias_table_size));
1575 free_pages((unsigned long)amd_iommu_dev_table,
1576 get_order(dev_table_size));
1582 #ifdef CONFIG_GART_IOMMU
1584 * We failed to initialize the AMD IOMMU - try fallback to GART
1594 /****************************************************************************
1596 * Early detect code. This code runs at IOMMU detection time in the DMA
1597 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1600 ****************************************************************************/
1601 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1606 int __init amd_iommu_detect(void)
1608 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
1611 if (amd_iommu_disabled)
1614 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1616 amd_iommu_detected = 1;
1617 x86_init.iommu.iommu_init = amd_iommu_init;
1619 /* Make sure ACS will be enabled */
1626 /****************************************************************************
1628 * Parsing functions for the AMD IOMMU specific kernel command line
1631 ****************************************************************************/
1633 static int __init parse_amd_iommu_dump(char *str)
1635 amd_iommu_dump = true;
1640 static int __init parse_amd_iommu_options(char *str)
1642 for (; *str; ++str) {
1643 if (strncmp(str, "fullflush", 9) == 0)
1644 amd_iommu_unmap_flush = true;
1645 if (strncmp(str, "off", 3) == 0)
1646 amd_iommu_disabled = true;
1647 if (strncmp(str, "force_isolation", 15) == 0)
1648 amd_iommu_force_isolation = true;
1654 __setup("amd_iommu_dump", parse_amd_iommu_dump);
1655 __setup("amd_iommu=", parse_amd_iommu_options);
1657 IOMMU_INIT_FINISH(amd_iommu_detect,
1658 gart_iommu_hole_init,
1662 bool amd_iommu_v2_supported(void)
1664 return amd_iommu_v2_present;
1666 EXPORT_SYMBOL(amd_iommu_v2_supported);