Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
   4 * Author: Joerg Roedel <jroedel@suse.de>
   5 *         Leo Duran <leo.duran@amd.com>
   6 */
   7
   8#define pr_fmt(fmt)     "AMD-Vi: " fmt
   9#define dev_fmt(fmt)    pr_fmt(fmt)
  10
  11#include <linux/pci.h>
  12#include <linux/acpi.h>
  13#include <linux/list.h>
  14#include <linux/bitmap.h>
  15#include <linux/slab.h>
  16#include <linux/syscore_ops.h>
  17#include <linux/interrupt.h>
  18#include <linux/msi.h>
 
  19#include <linux/amd-iommu.h>
  20#include <linux/export.h>
  21#include <linux/kmemleak.h>
  22#include <linux/mem_encrypt.h>
 
  23#include <asm/pci-direct.h>
  24#include <asm/iommu.h>
  25#include <asm/apic.h>
  26#include <asm/msidef.h>
  27#include <asm/gart.h>
  28#include <asm/x86_init.h>
  29#include <asm/iommu_table.h>
  30#include <asm/io_apic.h>
  31#include <asm/irq_remapping.h>
 
  32
  33#include <linux/crash_dump.h>
  34
  35#include "amd_iommu.h"
  36#include "../irq_remapping.h"
  37
  38/*
  39 * definitions for the ACPI scanning code
  40 */
  41#define IVRS_HEADER_LENGTH 48
  42
  43#define ACPI_IVHD_TYPE_MAX_SUPPORTED	0x40
  44#define ACPI_IVMD_TYPE_ALL              0x20
  45#define ACPI_IVMD_TYPE                  0x21
  46#define ACPI_IVMD_TYPE_RANGE            0x22
  47
  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
  56#define IVHD_DEV_SPECIAL		0x48
  57#define IVHD_DEV_ACPI_HID		0xf0
  58
  59#define UID_NOT_PRESENT                 0
  60#define UID_IS_INTEGER                  1
  61#define UID_IS_CHARACTER                2
  62
  63#define IVHD_SPECIAL_IOAPIC		1
  64#define IVHD_SPECIAL_HPET		2
  65
  66#define IVHD_FLAG_HT_TUN_EN_MASK        0x01
  67#define IVHD_FLAG_PASSPW_EN_MASK        0x02
  68#define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
  69#define IVHD_FLAG_ISOC_EN_MASK          0x08
  70
  71#define IVMD_FLAG_EXCL_RANGE            0x08
  72#define IVMD_FLAG_IW                    0x04
  73#define IVMD_FLAG_IR                    0x02
  74#define IVMD_FLAG_UNITY_MAP             0x01
  75
  76#define ACPI_DEVFLAG_INITPASS           0x01
  77#define ACPI_DEVFLAG_EXTINT             0x02
  78#define ACPI_DEVFLAG_NMI                0x04
  79#define ACPI_DEVFLAG_SYSMGT1            0x10
  80#define ACPI_DEVFLAG_SYSMGT2            0x20
  81#define ACPI_DEVFLAG_LINT0              0x40
  82#define ACPI_DEVFLAG_LINT1              0x80
  83#define ACPI_DEVFLAG_ATSDIS             0x10000000
  84
  85#define LOOP_TIMEOUT	100000
 
 
 
 
  86/*
  87 * ACPI table definitions
  88 *
  89 * These data structures are laid over the table to parse the important values
  90 * out of it.
  91 */
  92
  93extern const struct iommu_ops amd_iommu_ops;
  94
  95/*
  96 * structure describing one IOMMU in the ACPI table. Typically followed by one
  97 * or more ivhd_entrys.
  98 */
  99struct ivhd_header {
 100	u8 type;
 101	u8 flags;
 102	u16 length;
 103	u16 devid;
 104	u16 cap_ptr;
 105	u64 mmio_phys;
 106	u16 pci_seg;
 107	u16 info;
 108	u32 efr_attr;
 109
 110	/* Following only valid on IVHD type 11h and 40h */
 111	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
 112	u64 res;
 113} __attribute__((packed));
 114
 115/*
 116 * A device entry describing which devices a specific IOMMU translates and
 117 * which requestor ids they use.
 118 */
 119struct ivhd_entry {
 120	u8 type;
 121	u16 devid;
 122	u8 flags;
 123	u32 ext;
 124	u32 hidh;
 
 
 125	u64 cid;
 126	u8 uidf;
 127	u8 uidl;
 128	u8 uid;
 129} __attribute__((packed));
 130
 131/*
 132 * An AMD IOMMU memory definition structure. It defines things like exclusion
 133 * ranges for devices and regions that should be unity mapped.
 134 */
 135struct ivmd_header {
 136	u8 type;
 137	u8 flags;
 138	u16 length;
 139	u16 devid;
 140	u16 aux;
 141	u64 resv;
 
 142	u64 range_start;
 143	u64 range_length;
 144} __attribute__((packed));
 145
 146bool amd_iommu_dump;
 147bool amd_iommu_irq_remap __read_mostly;
 148
 
 
 149int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
 150static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
 151
 152static bool amd_iommu_detected;
 153static bool __initdata amd_iommu_disabled;
 
 154static int amd_iommu_target_ivhd_type;
 155
 156u16 amd_iommu_last_bdf;			/* largest PCI device id we have
 157					   to handle */
 158LIST_HEAD(amd_iommu_unity_map);		/* a list of required unity mappings
 159					   we find in ACPI */
 160bool amd_iommu_unmap_flush;		/* if true, flush on every unmap */
 
 
 161
 
 162LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
 163					   system */
 164
 165/* Array to assign indices to IOMMUs*/
 166struct amd_iommu *amd_iommus[MAX_IOMMUS];
 167
 168/* Number of IOMMUs present in the system */
 169static int amd_iommus_present;
 170
 171/* IOMMUs have a non-present cache? */
 172bool amd_iommu_np_cache __read_mostly;
 173bool amd_iommu_iotlb_sup __read_mostly = true;
 174
 175u32 amd_iommu_max_pasid __read_mostly = ~0;
 176
 177bool amd_iommu_v2_present __read_mostly;
 178static bool amd_iommu_pc_present __read_mostly;
 
 179
 180bool amd_iommu_force_isolation __read_mostly;
 181
 182/*
 183 * Pointer to the device table which is shared by all AMD IOMMUs
 184 * it is indexed by the PCI device id or the HT unit id and contains
 185 * information about the domain the device belongs to as well as the
 186 * page table root pointer.
 187 */
 188struct dev_table_entry *amd_iommu_dev_table;
 189/*
 190 * Pointer to a device table which the content of old device table
 191 * will be copied to. It's only be used in kdump kernel.
 192 */
 193static struct dev_table_entry *old_dev_tbl_cpy;
 194
 195/*
 196 * The alias table is a driver specific data structure which contains the
 197 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
 198 * More than one device can share the same requestor id.
 199 */
 200u16 *amd_iommu_alias_table;
 201
 202/*
 203 * The rlookup table is used to find the IOMMU which is responsible
 204 * for a specific device. It is also indexed by the PCI device id.
 205 */
 206struct amd_iommu **amd_iommu_rlookup_table;
 207EXPORT_SYMBOL(amd_iommu_rlookup_table);
 208
 209/*
 210 * This table is used to find the irq remapping table for a given device id
 211 * quickly.
 212 */
 213struct irq_remap_table **irq_lookup_table;
 214
 215/*
 216 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
 217 * to know which ones are already in use.
 218 */
 219unsigned long *amd_iommu_pd_alloc_bitmap;
 220
 221static u32 dev_table_size;	/* size of the device table */
 222static u32 alias_table_size;	/* size of the alias table */
 223static u32 rlookup_table_size;	/* size if the rlookup table */
 224
 225enum iommu_init_state {
 226	IOMMU_START_STATE,
 227	IOMMU_IVRS_DETECTED,
 228	IOMMU_ACPI_FINISHED,
 229	IOMMU_ENABLED,
 230	IOMMU_PCI_INIT,
 231	IOMMU_INTERRUPTS_EN,
 232	IOMMU_DMA_OPS,
 233	IOMMU_INITIALIZED,
 234	IOMMU_NOT_FOUND,
 235	IOMMU_INIT_ERROR,
 236	IOMMU_CMDLINE_DISABLED,
 237};
 238
 239/* Early ioapic and hpet maps from kernel command line */
 240#define EARLY_MAP_SIZE		4
 241static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
 242static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
 243static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
 244
 245static int __initdata early_ioapic_map_size;
 246static int __initdata early_hpet_map_size;
 247static int __initdata early_acpihid_map_size;
 248
 249static bool __initdata cmdline_maps;
 250
 251static enum iommu_init_state init_state = IOMMU_START_STATE;
 252
 253static int amd_iommu_enable_interrupts(void);
 254static int __init iommu_go_to_state(enum iommu_init_state state);
 255static void init_device_table_dma(void);
 256
 257static bool amd_iommu_pre_enabled = true;
 258
 
 
 259bool translation_pre_enabled(struct amd_iommu *iommu)
 260{
 261	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
 262}
 263EXPORT_SYMBOL(translation_pre_enabled);
 264
 265static void clear_translation_pre_enabled(struct amd_iommu *iommu)
 266{
 267	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
 268}
 269
 270static void init_translation_status(struct amd_iommu *iommu)
 271{
 272	u64 ctrl;
 273
 274	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
 275	if (ctrl & (1<<CONTROL_IOMMU_EN))
 276		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
 277}
 278
 279static inline void update_last_devid(u16 devid)
 280{
 281	if (devid > amd_iommu_last_bdf)
 282		amd_iommu_last_bdf = devid;
 283}
 284
 285static inline unsigned long tbl_size(int entry_size)
 286{
 287	unsigned shift = PAGE_SHIFT +
 288			 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
 289
 290	return 1UL << shift;
 291}
 292
 293int amd_iommu_get_num_iommus(void)
 294{
 295	return amd_iommus_present;
 296}
 297
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 298/* Access to l1 and l2 indexed register spaces */
 299
 300static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
 301{
 302	u32 val;
 303
 304	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
 305	pci_read_config_dword(iommu->dev, 0xfc, &val);
 306	return val;
 307}
 308
 309static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
 310{
 311	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
 312	pci_write_config_dword(iommu->dev, 0xfc, val);
 313	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
 314}
 315
 316static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
 317{
 318	u32 val;
 319
 320	pci_write_config_dword(iommu->dev, 0xf0, address);
 321	pci_read_config_dword(iommu->dev, 0xf4, &val);
 322	return val;
 323}
 324
 325static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
 326{
 327	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
 328	pci_write_config_dword(iommu->dev, 0xf4, val);
 329}
 330
 331/****************************************************************************
 332 *
 333 * AMD IOMMU MMIO register space handling functions
 334 *
 335 * These functions are used to program the IOMMU device registers in
 336 * MMIO space required for that driver.
 337 *
 338 ****************************************************************************/
 339
 340/*
 341 * This function set the exclusion range in the IOMMU. DMA accesses to the
 342 * exclusion range are passed through untranslated
 343 */
 344static void iommu_set_exclusion_range(struct amd_iommu *iommu)
 345{
 346	u64 start = iommu->exclusion_start & PAGE_MASK;
 347	u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
 348	u64 entry;
 349
 350	if (!iommu->exclusion_start)
 351		return;
 352
 353	entry = start | MMIO_EXCL_ENABLE_MASK;
 354	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
 355			&entry, sizeof(entry));
 356
 357	entry = limit;
 358	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
 359			&entry, sizeof(entry));
 360}
 361
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 362/* Programs the physical address of the device table into the IOMMU hardware */
 363static void iommu_set_device_table(struct amd_iommu *iommu)
 364{
 365	u64 entry;
 
 
 366
 367	BUG_ON(iommu->mmio_base == NULL);
 368
 369	entry = iommu_virt_to_phys(amd_iommu_dev_table);
 370	entry |= (dev_table_size >> 12) - 1;
 371	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
 372			&entry, sizeof(entry));
 373}
 374
 375/* Generic functions to enable/disable certain features of the IOMMU. */
 376static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
 377{
 378	u64 ctrl;
 379
 380	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
 381	ctrl |= (1ULL << bit);
 382	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
 383}
 384
 385static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
 386{
 387	u64 ctrl;
 388
 389	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
 390	ctrl &= ~(1ULL << bit);
 391	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
 392}
 393
 394static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
 395{
 396	u64 ctrl;
 397
 398	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
 399	ctrl &= ~CTRL_INV_TO_MASK;
 400	ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
 401	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
 402}
 403
 404/* Function to enable the hardware */
 405static void iommu_enable(struct amd_iommu *iommu)
 406{
 407	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
 408}
 409
 410static void iommu_disable(struct amd_iommu *iommu)
 411{
 412	if (!iommu->mmio_base)
 413		return;
 414
 415	/* Disable command buffer */
 416	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
 417
 418	/* Disable event logging and event interrupts */
 419	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
 420	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
 421
 422	/* Disable IOMMU GA_LOG */
 423	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
 424	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
 425
 426	/* Disable IOMMU hardware itself */
 427	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
 428}
 429
 430/*
 431 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
 432 * the system has one.
 433 */
 434static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
 435{
 436	if (!request_mem_region(address, end, "amd_iommu")) {
 437		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
 438			address, end);
 439		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
 440		return NULL;
 441	}
 442
 443	return (u8 __iomem *)ioremap(address, end);
 444}
 445
 446static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
 447{
 448	if (iommu->mmio_base)
 449		iounmap(iommu->mmio_base);
 450	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
 451}
 452
 453static inline u32 get_ivhd_header_size(struct ivhd_header *h)
 454{
 455	u32 size = 0;
 456
 457	switch (h->type) {
 458	case 0x10:
 459		size = 24;
 460		break;
 461	case 0x11:
 462	case 0x40:
 463		size = 40;
 464		break;
 465	}
 466	return size;
 467}
 468
 469/****************************************************************************
 470 *
 471 * The functions below belong to the first pass of AMD IOMMU ACPI table
 472 * parsing. In this pass we try to find out the highest device id this
 473 * code has to handle. Upon this information the size of the shared data
 474 * structures is determined later.
 475 *
 476 ****************************************************************************/
 477
 478/*
 479 * This function calculates the length of a given IVHD entry
 480 */
 481static inline int ivhd_entry_length(u8 *ivhd)
 482{
 483	u32 type = ((struct ivhd_entry *)ivhd)->type;
 484
 485	if (type < 0x80) {
 486		return 0x04 << (*ivhd >> 6);
 487	} else if (type == IVHD_DEV_ACPI_HID) {
 488		/* For ACPI_HID, offset 21 is uid len */
 489		return *((u8 *)ivhd + 21) + 22;
 490	}
 491	return 0;
 492}
 493
 494/*
 495 * After reading the highest device id from the IOMMU PCI capability header
 496 * this function looks if there is a higher device id defined in the ACPI table
 497 */
 498static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
 499{
 500	u8 *p = (void *)h, *end = (void *)h;
 501	struct ivhd_entry *dev;
 
 502
 503	u32 ivhd_size = get_ivhd_header_size(h);
 504
 505	if (!ivhd_size) {
 506		pr_err("Unsupported IVHD type %#x\n", h->type);
 507		return -EINVAL;
 508	}
 509
 510	p += ivhd_size;
 511	end += h->length;
 512
 513	while (p < end) {
 514		dev = (struct ivhd_entry *)p;
 515		switch (dev->type) {
 516		case IVHD_DEV_ALL:
 517			/* Use maximum BDF value for DEV_ALL */
 518			update_last_devid(0xffff);
 519			break;
 520		case IVHD_DEV_SELECT:
 521		case IVHD_DEV_RANGE_END:
 522		case IVHD_DEV_ALIAS:
 523		case IVHD_DEV_EXT_SELECT:
 524			/* all the above subfield types refer to device ids */
 525			update_last_devid(dev->devid);
 
 526			break;
 527		default:
 528			break;
 529		}
 530		p += ivhd_entry_length(p);
 531	}
 532
 533	WARN_ON(p != end);
 534
 535	return 0;
 536}
 537
 538static int __init check_ivrs_checksum(struct acpi_table_header *table)
 539{
 540	int i;
 541	u8 checksum = 0, *p = (u8 *)table;
 542
 543	for (i = 0; i < table->length; ++i)
 544		checksum += p[i];
 545	if (checksum != 0) {
 546		/* ACPI table corrupt */
 547		pr_err(FW_BUG "IVRS invalid checksum\n");
 548		return -ENODEV;
 549	}
 550
 551	return 0;
 552}
 553
 554/*
 555 * Iterate over all IVHD entries in the ACPI table and find the highest device
 556 * id which we need to handle. This is the first of three functions which parse
 557 * the ACPI table. So we check the checksum here.
 558 */
 559static int __init find_last_devid_acpi(struct acpi_table_header *table)
 560{
 561	u8 *p = (u8 *)table, *end = (u8 *)table;
 562	struct ivhd_header *h;
 
 563
 564	p += IVRS_HEADER_LENGTH;
 565
 566	end += table->length;
 567	while (p < end) {
 568		h = (struct ivhd_header *)p;
 569		if (h->type == amd_iommu_target_ivhd_type) {
 570			int ret = find_last_devid_from_ivhd(h);
 571
 572			if (ret)
 573				return ret;
 
 
 
 574		}
 575		p += h->length;
 576	}
 577	WARN_ON(p != end);
 578
 579	return 0;
 580}
 581
 582/****************************************************************************
 583 *
 584 * The following functions belong to the code path which parses the ACPI table
 585 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
 586 * data structures, initialize the device/alias/rlookup table and also
 587 * basically initialize the hardware.
 588 *
 589 ****************************************************************************/
 590
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 591/*
 592 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
 593 * write commands to that buffer later and the IOMMU will execute them
 594 * asynchronously
 595 */
 596static int __init alloc_command_buffer(struct amd_iommu *iommu)
 597{
 598	iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 599						  get_order(CMD_BUFFER_SIZE));
 600
 601	return iommu->cmd_buf ? 0 : -ENOMEM;
 602}
 603
 604/*
 
 
 
 
 
 
 
 
 
 
 605 * This function resets the command buffer if the IOMMU stopped fetching
 606 * commands from it.
 607 */
 608void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
 609{
 610	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
 611
 612	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
 613	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
 614	iommu->cmd_buf_head = 0;
 615	iommu->cmd_buf_tail = 0;
 616
 617	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
 618}
 619
 620/*
 621 * This function writes the command buffer address to the hardware and
 622 * enables it.
 623 */
 624static void iommu_enable_command_buffer(struct amd_iommu *iommu)
 625{
 626	u64 entry;
 627
 628	BUG_ON(iommu->cmd_buf == NULL);
 629
 630	entry = iommu_virt_to_phys(iommu->cmd_buf);
 631	entry |= MMIO_CMD_SIZE_512;
 632
 633	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
 634		    &entry, sizeof(entry));
 635
 636	amd_iommu_reset_cmd_buffer(iommu);
 637}
 638
 639/*
 640 * This function disables the command buffer
 641 */
 642static void iommu_disable_command_buffer(struct amd_iommu *iommu)
 643{
 644	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
 645}
 646
 647static void __init free_command_buffer(struct amd_iommu *iommu)
 648{
 649	free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
 650}
 651
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 652/* allocates the memory where the IOMMU will log its events to */
 653static int __init alloc_event_buffer(struct amd_iommu *iommu)
 654{
 655	iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 656						  get_order(EVT_BUFFER_SIZE));
 657
 658	return iommu->evt_buf ? 0 : -ENOMEM;
 659}
 660
 661static void iommu_enable_event_buffer(struct amd_iommu *iommu)
 662{
 663	u64 entry;
 664
 665	BUG_ON(iommu->evt_buf == NULL);
 666
 667	entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
 668
 669	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
 670		    &entry, sizeof(entry));
 671
 672	/* set head and tail to zero manually */
 673	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
 674	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
 675
 676	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
 677}
 678
 679/*
 680 * This function disables the event log buffer
 681 */
 682static void iommu_disable_event_buffer(struct amd_iommu *iommu)
 683{
 684	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
 685}
 686
 687static void __init free_event_buffer(struct amd_iommu *iommu)
 688{
 689	free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
 690}
 691
 692/* allocates the memory where the IOMMU will log its events to */
 693static int __init alloc_ppr_log(struct amd_iommu *iommu)
 694{
 695	iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 696						  get_order(PPR_LOG_SIZE));
 697
 698	return iommu->ppr_log ? 0 : -ENOMEM;
 699}
 700
 701static void iommu_enable_ppr_log(struct amd_iommu *iommu)
 702{
 703	u64 entry;
 704
 705	if (iommu->ppr_log == NULL)
 706		return;
 707
 708	entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
 709
 710	memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
 711		    &entry, sizeof(entry));
 712
 713	/* set head and tail to zero manually */
 714	writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
 715	writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
 716
 717	iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
 718	iommu_feature_enable(iommu, CONTROL_PPR_EN);
 719}
 720
 721static void __init free_ppr_log(struct amd_iommu *iommu)
 722{
 723	free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
 724}
 725
 726static void free_ga_log(struct amd_iommu *iommu)
 727{
 728#ifdef CONFIG_IRQ_REMAP
 729	free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE));
 730	free_pages((unsigned long)iommu->ga_log_tail, get_order(8));
 731#endif
 732}
 733
 
 734static int iommu_ga_log_enable(struct amd_iommu *iommu)
 735{
 736#ifdef CONFIG_IRQ_REMAP
 737	u32 status, i;
 
 738
 739	if (!iommu->ga_log)
 740		return -EINVAL;
 741
 742	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
 
 
 
 
 
 
 
 
 743
 744	/* Check if already running */
 745	if (status & (MMIO_STATUS_GALOG_RUN_MASK))
 746		return 0;
 747
 748	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
 749	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
 750
 751	for (i = 0; i < LOOP_TIMEOUT; ++i) {
 752		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
 753		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
 754			break;
 
 755	}
 756
 757	if (i >= LOOP_TIMEOUT)
 758		return -EINVAL;
 759#endif /* CONFIG_IRQ_REMAP */
 760	return 0;
 761}
 762
 763#ifdef CONFIG_IRQ_REMAP
 764static int iommu_init_ga_log(struct amd_iommu *iommu)
 765{
 766	u64 entry;
 767
 768	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
 769		return 0;
 770
 771	iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 772					get_order(GA_LOG_SIZE));
 773	if (!iommu->ga_log)
 774		goto err_out;
 775
 776	iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 777					get_order(8));
 778	if (!iommu->ga_log_tail)
 779		goto err_out;
 780
 781	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
 782	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
 783		    &entry, sizeof(entry));
 784	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
 785		 (BIT_ULL(52)-1)) & ~7ULL;
 786	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
 787		    &entry, sizeof(entry));
 788	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
 789	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
 790
 791	return 0;
 792err_out:
 793	free_ga_log(iommu);
 794	return -EINVAL;
 795}
 796#endif /* CONFIG_IRQ_REMAP */
 797
 798static int iommu_init_ga(struct amd_iommu *iommu)
 799{
 800	int ret = 0;
 801
 802#ifdef CONFIG_IRQ_REMAP
 803	/* Note: We have already checked GASup from IVRS table.
 804	 *       Now, we need to make sure that GAMSup is set.
 805	 */
 806	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
 807	    !iommu_feature(iommu, FEATURE_GAM_VAPIC))
 808		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
 809
 810	ret = iommu_init_ga_log(iommu);
 811#endif /* CONFIG_IRQ_REMAP */
 812
 813	return ret;
 
 
 
 814}
 815
 816static void iommu_enable_xt(struct amd_iommu *iommu)
 817{
 818#ifdef CONFIG_IRQ_REMAP
 819	/*
 820	 * XT mode (32-bit APIC destination ID) requires
 821	 * GA mode (128-bit IRTE support) as a prerequisite.
 822	 */
 823	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
 824	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
 825		iommu_feature_enable(iommu, CONTROL_XT_EN);
 826#endif /* CONFIG_IRQ_REMAP */
 827}
 828
 829static void iommu_enable_gt(struct amd_iommu *iommu)
 830{
 831	if (!iommu_feature(iommu, FEATURE_GT))
 832		return;
 833
 834	iommu_feature_enable(iommu, CONTROL_GT_EN);
 835}
 836
 837/* sets a specific bit in the device table entry. */
 838static void set_dev_entry_bit(u16 devid, u8 bit)
 
 839{
 840	int i = (bit >> 6) & 0x03;
 841	int _bit = bit & 0x3f;
 842
 843	amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
 844}
 845
 846static int get_dev_entry_bit(u16 devid, u8 bit)
 
 
 
 
 
 
 
 
 847{
 848	int i = (bit >> 6) & 0x03;
 849	int _bit = bit & 0x3f;
 850
 851	return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
 852}
 853
 
 
 
 854
 855static bool copy_device_table(void)
 
 
 
 856{
 857	u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
 
 858	struct dev_table_entry *old_devtb = NULL;
 859	u32 lo, hi, devid, old_devtb_size;
 860	phys_addr_t old_devtb_phys;
 861	struct amd_iommu *iommu;
 862	u16 dom_id, dte_v, irq_v;
 863	gfp_t gfp_flag;
 864	u64 tmp;
 865
 866	if (!amd_iommu_pre_enabled)
 
 
 
 
 
 
 
 
 867		return false;
 868
 869	pr_warn("Translation is already enabled - trying to copy translation structures\n");
 870	for_each_iommu(iommu) {
 871		/* All IOMMUs should use the same device table with the same size */
 872		lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
 873		hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
 874		entry = (((u64) hi) << 32) + lo;
 875		if (last_entry && last_entry != entry) {
 876			pr_err("IOMMU:%d should use the same dev table as others!\n",
 877				iommu->index);
 878			return false;
 879		}
 880		last_entry = entry;
 881
 882		old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
 883		if (old_devtb_size != dev_table_size) {
 884			pr_err("The device table size of IOMMU:%d is not expected!\n",
 885				iommu->index);
 886			return false;
 887		}
 888	}
 889
 890	/*
 891	 * When SME is enabled in the first kernel, the entry includes the
 892	 * memory encryption mask(sme_me_mask), we must remove the memory
 893	 * encryption mask to obtain the true physical address in kdump kernel.
 894	 */
 895	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
 896
 897	if (old_devtb_phys >= 0x100000000ULL) {
 898		pr_err("The address of old device table is above 4G, not trustworthy!\n");
 899		return false;
 900	}
 901	old_devtb = (sme_active() && is_kdump_kernel())
 902		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
 903							dev_table_size)
 904		    : memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
 905
 906	if (!old_devtb)
 907		return false;
 908
 909	gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
 910	old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
 911				get_order(dev_table_size));
 912	if (old_dev_tbl_cpy == NULL) {
 913		pr_err("Failed to allocate memory for copying old device table!\n");
 
 914		return false;
 915	}
 916
 917	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
 918		old_dev_tbl_cpy[devid] = old_devtb[devid];
 919		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
 920		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
 921
 922		if (dte_v && dom_id) {
 923			old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
 924			old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
 925			__set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
 926			/* If gcr3 table existed, mask it out */
 927			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
 928				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
 929				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
 930				old_dev_tbl_cpy[devid].data[1] &= ~tmp;
 931				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
 932				tmp |= DTE_FLAG_GV;
 933				old_dev_tbl_cpy[devid].data[0] &= ~tmp;
 934			}
 935		}
 936
 937		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
 938		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
 939		int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
 940		if (irq_v && (int_ctl || int_tab_len)) {
 941			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
 942			    (int_tab_len != DTE_IRQ_TABLE_LEN)) {
 943				pr_err("Wrong old irq remapping flag: %#x\n", devid);
 
 944				return false;
 945			}
 946
 947		        old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
 948		}
 949	}
 950	memunmap(old_devtb);
 951
 952	return true;
 953}
 954
 955void amd_iommu_apply_erratum_63(u16 devid)
 956{
 957	int sysmgt;
 
 958
 959	sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
 960		 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
 961
 962	if (sysmgt == 0x01)
 963		set_dev_entry_bit(devid, DEV_ENTRY_IW);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 964}
 965
 966/* Writes the specific IOMMU for a device into the rlookup table */
 967static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
 968{
 969	amd_iommu_rlookup_table[devid] = iommu;
 
 
 
 
 
 
 970}
 971
 972/*
 973 * This function takes the device specific flags read from the ACPI
 974 * table and sets up the device table entry with that information
 975 */
 976static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
 977					   u16 devid, u32 flags, u32 ext_flags)
 978{
 979	if (flags & ACPI_DEVFLAG_INITPASS)
 980		set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
 981	if (flags & ACPI_DEVFLAG_EXTINT)
 982		set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
 983	if (flags & ACPI_DEVFLAG_NMI)
 984		set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
 985	if (flags & ACPI_DEVFLAG_SYSMGT1)
 986		set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
 987	if (flags & ACPI_DEVFLAG_SYSMGT2)
 988		set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
 989	if (flags & ACPI_DEVFLAG_LINT0)
 990		set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
 991	if (flags & ACPI_DEVFLAG_LINT1)
 992		set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
 993
 994	amd_iommu_apply_erratum_63(devid);
 995
 996	set_iommu_for_device(iommu, devid);
 997}
 998
 999int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1000{
1001	struct devid_map *entry;
1002	struct list_head *list;
1003
1004	if (type == IVHD_SPECIAL_IOAPIC)
1005		list = &ioapic_map;
1006	else if (type == IVHD_SPECIAL_HPET)
1007		list = &hpet_map;
1008	else
1009		return -EINVAL;
1010
1011	list_for_each_entry(entry, list, list) {
1012		if (!(entry->id == id && entry->cmd_line))
1013			continue;
1014
1015		pr_info("Command-line override present for %s id %d - ignoring\n",
1016			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1017
1018		*devid = entry->devid;
1019
1020		return 0;
1021	}
1022
1023	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1024	if (!entry)
1025		return -ENOMEM;
1026
1027	entry->id	= id;
1028	entry->devid	= *devid;
1029	entry->cmd_line	= cmd_line;
1030
1031	list_add_tail(&entry->list, list);
1032
1033	return 0;
1034}
1035
1036static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1037				      bool cmd_line)
1038{
1039	struct acpihid_map_entry *entry;
1040	struct list_head *list = &acpihid_map;
1041
1042	list_for_each_entry(entry, list, list) {
1043		if (strcmp(entry->hid, hid) ||
1044		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1045		    !entry->cmd_line)
1046			continue;
1047
1048		pr_info("Command-line override for hid:%s uid:%s\n",
1049			hid, uid);
1050		*devid = entry->devid;
1051		return 0;
1052	}
1053
1054	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1055	if (!entry)
1056		return -ENOMEM;
1057
1058	memcpy(entry->uid, uid, strlen(uid));
1059	memcpy(entry->hid, hid, strlen(hid));
1060	entry->devid = *devid;
1061	entry->cmd_line	= cmd_line;
1062	entry->root_devid = (entry->devid & (~0x7));
1063
1064	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1065		entry->cmd_line ? "cmd" : "ivrs",
1066		entry->hid, entry->uid, entry->root_devid);
1067
1068	list_add_tail(&entry->list, list);
1069	return 0;
1070}
1071
1072static int __init add_early_maps(void)
1073{
1074	int i, ret;
1075
1076	for (i = 0; i < early_ioapic_map_size; ++i) {
1077		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1078					 early_ioapic_map[i].id,
1079					 &early_ioapic_map[i].devid,
1080					 early_ioapic_map[i].cmd_line);
1081		if (ret)
1082			return ret;
1083	}
1084
1085	for (i = 0; i < early_hpet_map_size; ++i) {
1086		ret = add_special_device(IVHD_SPECIAL_HPET,
1087					 early_hpet_map[i].id,
1088					 &early_hpet_map[i].devid,
1089					 early_hpet_map[i].cmd_line);
1090		if (ret)
1091			return ret;
1092	}
1093
1094	for (i = 0; i < early_acpihid_map_size; ++i) {
1095		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1096					  early_acpihid_map[i].uid,
1097					  &early_acpihid_map[i].devid,
1098					  early_acpihid_map[i].cmd_line);
1099		if (ret)
1100			return ret;
1101	}
1102
1103	return 0;
1104}
1105
1106/*
1107 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1108 * initializes the hardware and our data structures with it.
1109 */
1110static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1111					struct ivhd_header *h)
1112{
1113	u8 *p = (u8 *)h;
1114	u8 *end = p, flags = 0;
1115	u16 devid = 0, devid_start = 0, devid_to = 0;
1116	u32 dev_i, ext_flags = 0;
1117	bool alias = false;
1118	struct ivhd_entry *e;
 
1119	u32 ivhd_size;
1120	int ret;
1121
1122
1123	ret = add_early_maps();
1124	if (ret)
1125		return ret;
1126
1127	amd_iommu_apply_ivrs_quirks();
1128
1129	/*
1130	 * First save the recommended feature enable bits from ACPI
1131	 */
1132	iommu->acpi_flags = h->flags;
1133
1134	/*
1135	 * Done. Now parse the device entries
1136	 */
1137	ivhd_size = get_ivhd_header_size(h);
1138	if (!ivhd_size) {
1139		pr_err("Unsupported IVHD type %#x\n", h->type);
1140		return -EINVAL;
1141	}
1142
1143	p += ivhd_size;
1144
1145	end += h->length;
1146
1147
1148	while (p < end) {
1149		e = (struct ivhd_entry *)p;
 
 
1150		switch (e->type) {
1151		case IVHD_DEV_ALL:
1152
1153			DUMP_printk("  DEV_ALL\t\t\tflags: %02x\n", e->flags);
1154
1155			for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1156				set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1157			break;
1158		case IVHD_DEV_SELECT:
1159
1160			DUMP_printk("  DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1161				    "flags: %02x\n",
1162				    PCI_BUS_NUM(e->devid),
1163				    PCI_SLOT(e->devid),
1164				    PCI_FUNC(e->devid),
1165				    e->flags);
1166
1167			devid = e->devid;
1168			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1169			break;
1170		case IVHD_DEV_SELECT_RANGE_START:
1171
1172			DUMP_printk("  DEV_SELECT_RANGE_START\t "
1173				    "devid: %02x:%02x.%x flags: %02x\n",
1174				    PCI_BUS_NUM(e->devid),
1175				    PCI_SLOT(e->devid),
1176				    PCI_FUNC(e->devid),
1177				    e->flags);
1178
1179			devid_start = e->devid;
1180			flags = e->flags;
1181			ext_flags = 0;
1182			alias = false;
1183			break;
1184		case IVHD_DEV_ALIAS:
1185
1186			DUMP_printk("  DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1187				    "flags: %02x devid_to: %02x:%02x.%x\n",
1188				    PCI_BUS_NUM(e->devid),
1189				    PCI_SLOT(e->devid),
1190				    PCI_FUNC(e->devid),
1191				    e->flags,
1192				    PCI_BUS_NUM(e->ext >> 8),
1193				    PCI_SLOT(e->ext >> 8),
1194				    PCI_FUNC(e->ext >> 8));
1195
1196			devid = e->devid;
1197			devid_to = e->ext >> 8;
1198			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1199			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1200			amd_iommu_alias_table[devid] = devid_to;
1201			break;
1202		case IVHD_DEV_ALIAS_RANGE:
1203
1204			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
1205				    "devid: %02x:%02x.%x flags: %02x "
1206				    "devid_to: %02x:%02x.%x\n",
1207				    PCI_BUS_NUM(e->devid),
1208				    PCI_SLOT(e->devid),
1209				    PCI_FUNC(e->devid),
1210				    e->flags,
1211				    PCI_BUS_NUM(e->ext >> 8),
1212				    PCI_SLOT(e->ext >> 8),
1213				    PCI_FUNC(e->ext >> 8));
1214
1215			devid_start = e->devid;
1216			flags = e->flags;
1217			devid_to = e->ext >> 8;
1218			ext_flags = 0;
1219			alias = true;
1220			break;
1221		case IVHD_DEV_EXT_SELECT:
1222
1223			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1224				    "flags: %02x ext: %08x\n",
1225				    PCI_BUS_NUM(e->devid),
1226				    PCI_SLOT(e->devid),
1227				    PCI_FUNC(e->devid),
1228				    e->flags, e->ext);
1229
1230			devid = e->devid;
1231			set_dev_entry_from_acpi(iommu, devid, e->flags,
1232						e->ext);
1233			break;
1234		case IVHD_DEV_EXT_SELECT_RANGE:
1235
1236			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
1237				    "%02x:%02x.%x flags: %02x ext: %08x\n",
1238				    PCI_BUS_NUM(e->devid),
1239				    PCI_SLOT(e->devid),
1240				    PCI_FUNC(e->devid),
1241				    e->flags, e->ext);
1242
1243			devid_start = e->devid;
1244			flags = e->flags;
1245			ext_flags = e->ext;
1246			alias = false;
1247			break;
1248		case IVHD_DEV_RANGE_END:
1249
1250			DUMP_printk("  DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1251				    PCI_BUS_NUM(e->devid),
1252				    PCI_SLOT(e->devid),
1253				    PCI_FUNC(e->devid));
1254
1255			devid = e->devid;
1256			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1257				if (alias) {
1258					amd_iommu_alias_table[dev_i] = devid_to;
1259					set_dev_entry_from_acpi(iommu,
1260						devid_to, flags, ext_flags);
1261				}
1262				set_dev_entry_from_acpi(iommu, dev_i,
1263							flags, ext_flags);
1264			}
1265			break;
1266		case IVHD_DEV_SPECIAL: {
1267			u8 handle, type;
1268			const char *var;
1269			u16 devid;
1270			int ret;
1271
1272			handle = e->ext & 0xff;
1273			devid  = (e->ext >>  8) & 0xffff;
1274			type   = (e->ext >> 24) & 0xff;
1275
1276			if (type == IVHD_SPECIAL_IOAPIC)
1277				var = "IOAPIC";
1278			else if (type == IVHD_SPECIAL_HPET)
1279				var = "HPET";
1280			else
1281				var = "UNKNOWN";
1282
1283			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1284				    var, (int)handle,
1285				    PCI_BUS_NUM(devid),
1286				    PCI_SLOT(devid),
1287				    PCI_FUNC(devid));
1288
1289			ret = add_special_device(type, handle, &devid, false);
1290			if (ret)
1291				return ret;
1292
1293			/*
1294			 * add_special_device might update the devid in case a
1295			 * command-line override is present. So call
1296			 * set_dev_entry_from_acpi after add_special_device.
1297			 */
1298			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1299
1300			break;
1301		}
1302		case IVHD_DEV_ACPI_HID: {
1303			u16 devid;
1304			u8 hid[ACPIHID_HID_LEN];
1305			u8 uid[ACPIHID_UID_LEN];
1306			int ret;
1307
1308			if (h->type != 0x40) {
1309				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1310				       e->type);
1311				break;
1312			}
1313
1314			memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
 
1315			hid[ACPIHID_HID_LEN - 1] = '\0';
1316
1317			if (!(*hid)) {
1318				pr_err(FW_BUG "Invalid HID.\n");
1319				break;
1320			}
1321
1322			uid[0] = '\0';
1323			switch (e->uidf) {
1324			case UID_NOT_PRESENT:
1325
1326				if (e->uidl != 0)
1327					pr_warn(FW_BUG "Invalid UID length.\n");
1328
1329				break;
1330			case UID_IS_INTEGER:
1331
1332				sprintf(uid, "%d", e->uid);
1333
1334				break;
1335			case UID_IS_CHARACTER:
1336
1337				memcpy(uid, &e->uid, e->uidl);
1338				uid[e->uidl] = '\0';
1339
1340				break;
1341			default:
1342				break;
1343			}
1344
1345			devid = e->devid;
1346			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1347				    hid, uid,
1348				    PCI_BUS_NUM(devid),
1349				    PCI_SLOT(devid),
1350				    PCI_FUNC(devid));
1351
1352			flags = e->flags;
1353
1354			ret = add_acpi_hid_device(hid, uid, &devid, false);
1355			if (ret)
1356				return ret;
1357
1358			/*
1359			 * add_special_device might update the devid in case a
1360			 * command-line override is present. So call
1361			 * set_dev_entry_from_acpi after add_special_device.
1362			 */
1363			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1364
1365			break;
1366		}
1367		default:
1368			break;
1369		}
1370
1371		p += ivhd_entry_length(p);
1372	}
1373
1374	return 0;
1375}
1376
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1377static void __init free_iommu_one(struct amd_iommu *iommu)
1378{
 
1379	free_command_buffer(iommu);
1380	free_event_buffer(iommu);
1381	free_ppr_log(iommu);
1382	free_ga_log(iommu);
1383	iommu_unmap_mmio_space(iommu);
1384}
1385
1386static void __init free_iommu_all(void)
1387{
1388	struct amd_iommu *iommu, *next;
1389
1390	for_each_iommu_safe(iommu, next) {
1391		list_del(&iommu->list);
1392		free_iommu_one(iommu);
1393		kfree(iommu);
1394	}
1395}
1396
1397/*
1398 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1399 * Workaround:
1400 *     BIOS should disable L2B micellaneous clock gating by setting
1401 *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1402 */
1403static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1404{
1405	u32 value;
1406
1407	if ((boot_cpu_data.x86 != 0x15) ||
1408	    (boot_cpu_data.x86_model < 0x10) ||
1409	    (boot_cpu_data.x86_model > 0x1f))
1410		return;
1411
1412	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1413	pci_read_config_dword(iommu->dev, 0xf4, &value);
1414
1415	if (value & BIT(2))
1416		return;
1417
1418	/* Select NB indirect register 0x90 and enable writing */
1419	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1420
1421	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1422	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1423
1424	/* Clear the enable writing bit */
1425	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1426}
1427
1428/*
1429 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1430 * Workaround:
1431 *     BIOS should enable ATS write permission check by setting
1432 *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1433 */
1434static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1435{
1436	u32 value;
1437
1438	if ((boot_cpu_data.x86 != 0x15) ||
1439	    (boot_cpu_data.x86_model < 0x30) ||
1440	    (boot_cpu_data.x86_model > 0x3f))
1441		return;
1442
1443	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1444	value = iommu_read_l2(iommu, 0x47);
1445
1446	if (value & BIT(0))
1447		return;
1448
1449	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1450	iommu_write_l2(iommu, 0x47, value | BIT(0));
1451
1452	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1453}
1454
1455/*
1456 * This function clues the initialization function for one IOMMU
1457 * together and also allocates the command buffer and programs the
1458 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1459 */
1460static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 
1461{
1462	int ret;
 
 
 
 
 
1463
1464	raw_spin_lock_init(&iommu->lock);
 
1465
1466	/* Add IOMMU to internal data structures */
1467	list_add_tail(&iommu->list, &amd_iommu_list);
1468	iommu->index = amd_iommus_present++;
1469
1470	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1471		WARN(1, "System has more IOMMUs than supported by this driver\n");
1472		return -ENOSYS;
1473	}
1474
1475	/* Index is fine - add IOMMU to the array */
1476	amd_iommus[iommu->index] = iommu;
1477
1478	/*
1479	 * Copy data from ACPI table entry to the iommu struct
1480	 */
1481	iommu->devid   = h->devid;
1482	iommu->cap_ptr = h->cap_ptr;
1483	iommu->pci_seg = h->pci_seg;
1484	iommu->mmio_phys = h->mmio_phys;
1485
1486	switch (h->type) {
1487	case 0x10:
1488		/* Check if IVHD EFR contains proper max banks/counters */
1489		if ((h->efr_attr != 0) &&
1490		    ((h->efr_attr & (0xF << 13)) != 0) &&
1491		    ((h->efr_attr & (0x3F << 17)) != 0))
1492			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1493		else
1494			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1495
1496		/*
1497		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1498		 * GAM also requires GA mode. Therefore, we need to
1499		 * check cmpxchg16b support before enabling it.
1500		 */
1501		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1502		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1503			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1504		break;
1505	case 0x11:
1506	case 0x40:
1507		if (h->efr_reg & (1 << 9))
1508			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1509		else
1510			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1511
1512		/*
1513		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1514		 * XT, GAM also requires GA mode. Therefore, we need to
1515		 * check cmpxchg16b support before enabling them.
1516		 */
1517		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1518		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1519			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1520			break;
1521		}
1522
1523		/*
1524		 * Note: Since iommu_update_intcapxt() leverages
1525		 * the IOMMU MMIO access to MSI capability block registers
1526		 * for MSI address lo/hi/data, we need to check both
1527		 * EFR[XtSup] and EFR[MsiCapMmioSup] for x2APIC support.
1528		 */
1529		if ((h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT)) &&
1530		    (h->efr_reg & BIT(IOMMU_EFR_MSICAPMMIOSUP_SHIFT)))
1531			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
 
 
 
1532		break;
1533	default:
1534		return -EINVAL;
1535	}
1536
1537	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1538						iommu->mmio_phys_end);
1539	if (!iommu->mmio_base)
1540		return -ENOMEM;
1541
 
 
 
 
 
 
 
 
 
 
1542	if (alloc_command_buffer(iommu))
1543		return -ENOMEM;
1544
1545	if (alloc_event_buffer(iommu))
1546		return -ENOMEM;
1547
1548	iommu->int_enabled = false;
1549
1550	init_translation_status(iommu);
1551	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1552		iommu_disable(iommu);
1553		clear_translation_pre_enabled(iommu);
1554		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1555			iommu->index);
1556	}
1557	if (amd_iommu_pre_enabled)
1558		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1559
1560	ret = init_iommu_from_acpi(iommu, h);
1561	if (ret)
1562		return ret;
1563
1564	ret = amd_iommu_create_irq_domain(iommu);
1565	if (ret)
1566		return ret;
1567
1568	/*
1569	 * Make sure IOMMU is not considered to translate itself. The IVRS
1570	 * table tells us so, but this is a lie!
1571	 */
1572	amd_iommu_rlookup_table[iommu->devid] = NULL;
1573
1574	return 0;
1575}
1576
1577/**
1578 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1579 * @ivrs          Pointer to the IVRS header
1580 *
1581 * This function search through all IVDB of the maximum supported IVHD
1582 */
1583static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1584{
1585	u8 *base = (u8 *)ivrs;
1586	struct ivhd_header *ivhd = (struct ivhd_header *)
1587					(base + IVRS_HEADER_LENGTH);
1588	u8 last_type = ivhd->type;
1589	u16 devid = ivhd->devid;
1590
1591	while (((u8 *)ivhd - base < ivrs->length) &&
1592	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1593		u8 *p = (u8 *) ivhd;
1594
1595		if (ivhd->devid == devid)
1596			last_type = ivhd->type;
1597		ivhd = (struct ivhd_header *)(p + ivhd->length);
1598	}
1599
1600	return last_type;
1601}
1602
1603/*
1604 * Iterates over all IOMMU entries in the ACPI table, allocates the
1605 * IOMMU structure and initializes it with init_iommu_one()
1606 */
1607static int __init init_iommu_all(struct acpi_table_header *table)
1608{
1609	u8 *p = (u8 *)table, *end = (u8 *)table;
1610	struct ivhd_header *h;
1611	struct amd_iommu *iommu;
1612	int ret;
1613
1614	end += table->length;
1615	p += IVRS_HEADER_LENGTH;
1616
 
1617	while (p < end) {
1618		h = (struct ivhd_header *)p;
1619		if (*p == amd_iommu_target_ivhd_type) {
1620
1621			DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1622				    "seg: %d flags: %01x info %04x\n",
1623				    PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1624				    PCI_FUNC(h->devid), h->cap_ptr,
1625				    h->pci_seg, h->flags, h->info);
1626			DUMP_printk("       mmio-addr: %016llx\n",
1627				    h->mmio_phys);
1628
1629			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1630			if (iommu == NULL)
1631				return -ENOMEM;
1632
1633			ret = init_iommu_one(iommu, h);
1634			if (ret)
1635				return ret;
1636		}
1637		p += h->length;
1638
1639	}
1640	WARN_ON(p != end);
1641
 
 
 
 
 
 
 
 
 
 
1642	return 0;
1643}
1644
1645static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1646				u8 fxn, u64 *value, bool is_write);
1647
1648static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1649{
 
1650	struct pci_dev *pdev = iommu->dev;
1651	u64 val = 0xabcd, val2 = 0, save_reg = 0;
1652
1653	if (!iommu_feature(iommu, FEATURE_PC))
1654		return;
1655
1656	amd_iommu_pc_present = true;
1657
1658	/* save the value to restore, if writable */
1659	if (iommu_pc_get_set_reg(iommu, 0, 0, 0, &save_reg, false))
1660		goto pc_false;
1661
1662	/* Check if the performance counters can be written to */
1663	if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1664	    (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1665	    (val != val2))
1666		goto pc_false;
1667
1668	/* restore */
1669	if (iommu_pc_get_set_reg(iommu, 0, 0, 0, &save_reg, true))
1670		goto pc_false;
1671
1672	pci_info(pdev, "IOMMU performance counters supported\n");
1673
1674	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1675	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1676	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1677
1678	return;
1679
1680pc_false:
1681	pci_err(pdev, "Unable to read/write to IOMMU perf counter.\n");
1682	amd_iommu_pc_present = false;
1683	return;
1684}
1685
1686static ssize_t amd_iommu_show_cap(struct device *dev,
1687				  struct device_attribute *attr,
1688				  char *buf)
1689{
1690	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1691	return sprintf(buf, "%x\n", iommu->cap);
1692}
1693static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1694
1695static ssize_t amd_iommu_show_features(struct device *dev,
1696				       struct device_attribute *attr,
1697				       char *buf)
1698{
1699	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1700	return sprintf(buf, "%llx\n", iommu->features);
1701}
1702static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1703
1704static struct attribute *amd_iommu_attrs[] = {
1705	&dev_attr_cap.attr,
1706	&dev_attr_features.attr,
1707	NULL,
1708};
1709
1710static struct attribute_group amd_iommu_group = {
1711	.name = "amd-iommu",
1712	.attrs = amd_iommu_attrs,
1713};
1714
1715static const struct attribute_group *amd_iommu_groups[] = {
1716	&amd_iommu_group,
1717	NULL,
1718};
1719
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1720static int __init iommu_init_pci(struct amd_iommu *iommu)
1721{
1722	int cap_ptr = iommu->cap_ptr;
1723	int ret;
1724
1725	iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
 
1726						 iommu->devid & 0xff);
1727	if (!iommu->dev)
1728		return -ENODEV;
1729
1730	/* Prevent binding other PCI device drivers to IOMMU devices */
1731	iommu->dev->match_driver = false;
1732
1733	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1734			      &iommu->cap);
1735
1736	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1737		amd_iommu_iotlb_sup = false;
1738
1739	/* read extended feature bits */
1740	iommu->features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1741
1742	if (iommu_feature(iommu, FEATURE_GT)) {
1743		int glxval;
1744		u32 max_pasid;
1745		u64 pasmax;
1746
1747		pasmax = iommu->features & FEATURE_PASID_MASK;
1748		pasmax >>= FEATURE_PASID_SHIFT;
1749		max_pasid  = (1 << (pasmax + 1)) - 1;
1750
1751		amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1752
1753		BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1754
1755		glxval   = iommu->features & FEATURE_GLXVAL_MASK;
1756		glxval >>= FEATURE_GLXVAL_SHIFT;
1757
1758		if (amd_iommu_max_glx_val == -1)
1759			amd_iommu_max_glx_val = glxval;
1760		else
1761			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1762	}
1763
1764	if (iommu_feature(iommu, FEATURE_GT) &&
1765	    iommu_feature(iommu, FEATURE_PPR)) {
1766		iommu->is_iommu_v2   = true;
1767		amd_iommu_v2_present = true;
1768	}
1769
1770	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1771		return -ENOMEM;
1772
1773	ret = iommu_init_ga(iommu);
1774	if (ret)
1775		return ret;
1776
1777	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1778		amd_iommu_np_cache = true;
 
1779
1780	init_iommu_perf_ctr(iommu);
1781
 
 
 
 
 
 
 
 
 
 
 
1782	if (is_rd890_iommu(iommu->dev)) {
1783		int i, j;
1784
1785		iommu->root_pdev =
1786			pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
 
1787						    PCI_DEVFN(0, 0));
1788
1789		/*
1790		 * Some rd890 systems may not be fully reconfigured by the
1791		 * BIOS, so it's necessary for us to store this information so
1792		 * it can be reprogrammed on resume
1793		 */
1794		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1795				&iommu->stored_addr_lo);
1796		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1797				&iommu->stored_addr_hi);
1798
1799		/* Low bit locks writes to configuration space */
1800		iommu->stored_addr_lo &= ~1;
1801
1802		for (i = 0; i < 6; i++)
1803			for (j = 0; j < 0x12; j++)
1804				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1805
1806		for (i = 0; i < 0x83; i++)
1807			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1808	}
1809
1810	amd_iommu_erratum_746_workaround(iommu);
1811	amd_iommu_ats_write_check_workaround(iommu);
1812
1813	iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1814			       amd_iommu_groups, "ivhd%d", iommu->index);
1815	iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1816	iommu_device_register(&iommu->iommu);
 
 
1817
1818	return pci_enable_device(iommu->dev);
1819}
1820
1821static void print_iommu_info(void)
1822{
1823	static const char * const feat_str[] = {
1824		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1825		"IA", "GA", "HE", "PC"
1826	};
1827	struct amd_iommu *iommu;
1828
1829	for_each_iommu(iommu) {
1830		struct pci_dev *pdev = iommu->dev;
1831		int i;
1832
1833		pci_info(pdev, "Found IOMMU cap 0x%hx\n", iommu->cap_ptr);
1834
1835		if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1836			pci_info(pdev, "Extended features (%#llx):",
1837				 iommu->features);
1838			for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1839				if (iommu_feature(iommu, (1ULL << i)))
1840					pr_cont(" %s", feat_str[i]);
1841			}
1842
1843			if (iommu->features & FEATURE_GAM_VAPIC)
1844				pr_cont(" GA_vAPIC");
1845
 
 
 
1846			pr_cont("\n");
1847		}
1848	}
1849	if (irq_remapping_enabled) {
1850		pr_info("Interrupt remapping enabled\n");
1851		if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1852			pr_info("Virtual APIC enabled\n");
1853		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1854			pr_info("X2APIC enabled\n");
1855	}
 
 
1856}
1857
1858static int __init amd_iommu_init_pci(void)
1859{
1860	struct amd_iommu *iommu;
1861	int ret = 0;
 
1862
1863	for_each_iommu(iommu) {
1864		ret = iommu_init_pci(iommu);
1865		if (ret)
1866			break;
 
 
 
 
 
1867	}
1868
1869	/*
1870	 * Order is important here to make sure any unity map requirements are
1871	 * fulfilled. The unity mappings are created and written to the device
1872	 * table during the amd_iommu_init_api() call.
1873	 *
1874	 * After that we call init_device_table_dma() to make sure any
1875	 * uninitialized DTE will block DMA, and in the end we flush the caches
1876	 * of all IOMMUs to make sure the changes to the device table are
1877	 * active.
1878	 */
1879	ret = amd_iommu_init_api();
1880
1881	init_device_table_dma();
1882
1883	for_each_iommu(iommu)
1884		iommu_flush_all_caches(iommu);
1885
1886	if (!ret)
1887		print_iommu_info();
1888
 
1889	return ret;
1890}
1891
1892/****************************************************************************
1893 *
1894 * The following functions initialize the MSI interrupts for all IOMMUs
1895 * in the system. It's a bit challenging because there could be multiple
1896 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1897 * pci_dev.
1898 *
1899 ****************************************************************************/
1900
1901static int iommu_setup_msi(struct amd_iommu *iommu)
1902{
1903	int r;
1904
1905	r = pci_enable_msi(iommu->dev);
1906	if (r)
1907		return r;
1908
1909	r = request_threaded_irq(iommu->dev->irq,
1910				 amd_iommu_int_handler,
1911				 amd_iommu_int_thread,
1912				 0, "AMD-Vi",
1913				 iommu);
1914
1915	if (r) {
1916		pci_disable_msi(iommu->dev);
1917		return r;
1918	}
1919
1920	iommu->int_enabled = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1921
 
 
 
 
 
1922	return 0;
1923}
1924
1925#define XT_INT_DEST_MODE(x)	(((x) & 0x1ULL) << 2)
1926#define XT_INT_DEST_LO(x)	(((x) & 0xFFFFFFULL) << 8)
1927#define XT_INT_VEC(x)		(((x) & 0xFFULL) << 32)
1928#define XT_INT_DEST_HI(x)	((((x) >> 24) & 0xFFULL) << 56)
1929
1930/**
1931 * Setup the IntCapXT registers with interrupt routing information
1932 * based on the PCI MSI capability block registers, accessed via
1933 * MMIO MSI address low/hi and MSI data registers.
1934 */
1935static void iommu_update_intcapxt(struct amd_iommu *iommu)
1936{
1937	u64 val;
1938	u32 addr_lo = readl(iommu->mmio_base + MMIO_MSI_ADDR_LO_OFFSET);
1939	u32 addr_hi = readl(iommu->mmio_base + MMIO_MSI_ADDR_HI_OFFSET);
1940	u32 data    = readl(iommu->mmio_base + MMIO_MSI_DATA_OFFSET);
1941	bool dm     = (addr_lo >> MSI_ADDR_DEST_MODE_SHIFT) & 0x1;
1942	u32 dest    = ((addr_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xFF);
1943
1944	if (x2apic_enabled())
1945		dest |= MSI_ADDR_EXT_DEST_ID(addr_hi);
1946
1947	val = XT_INT_VEC(data & 0xFF) |
1948	      XT_INT_DEST_MODE(dm) |
1949	      XT_INT_DEST_LO(dest) |
1950	      XT_INT_DEST_HI(dest);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1951
1952	/**
1953	 * Current IOMMU implemtation uses the same IRQ for all
1954	 * 3 IOMMU interrupts.
1955	 */
1956	writeq(val, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
1957	writeq(val, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
1958	writeq(val, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
1959}
1960
1961static void _irq_notifier_notify(struct irq_affinity_notify *notify,
1962				 const cpumask_t *mask)
1963{
1964	struct amd_iommu *iommu;
1965
1966	for_each_iommu(iommu) {
1967		if (iommu->dev->irq == notify->irq) {
1968			iommu_update_intcapxt(iommu);
1969			break;
1970		}
1971	}
1972}
1973
1974static void _irq_notifier_release(struct kref *ref)
 
 
1975{
 
 
 
 
 
 
 
1976}
1977
1978static int iommu_init_intcapxt(struct amd_iommu *iommu)
1979{
1980	int ret;
1981	struct irq_affinity_notify *notify = &iommu->intcapxt_notify;
1982
1983	/**
1984	 * IntCapXT requires XTSup=1 and MsiCapMmioSup=1,
1985	 * which can be inferred from amd_iommu_xt_mode.
1986	 */
1987	if (amd_iommu_xt_mode != IRQ_REMAP_X2APIC_MODE)
1988		return 0;
 
 
 
 
1989
1990	/**
1991	 * Also, we need to setup notifier to update the IntCapXT registers
1992	 * whenever the irq affinity is changed from user-space.
1993	 */
1994	notify->irq = iommu->dev->irq;
1995	notify->notify = _irq_notifier_notify,
1996	notify->release = _irq_notifier_release,
1997	ret = irq_set_affinity_notifier(iommu->dev->irq, notify);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1998	if (ret) {
1999		pr_err("Failed to register irq affinity notifier (devid=%#x, irq %d)\n",
2000		       iommu->devid, iommu->dev->irq);
2001		return ret;
2002	}
2003
2004	iommu_update_intcapxt(iommu);
2005	iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2006	return ret;
2007}
2008
2009static int iommu_init_msi(struct amd_iommu *iommu)
2010{
2011	int ret;
2012
2013	if (iommu->int_enabled)
2014		goto enable_faults;
2015
2016	if (iommu->dev->msi_cap)
 
 
2017		ret = iommu_setup_msi(iommu);
2018	else
2019		ret = -ENODEV;
2020
2021	if (ret)
2022		return ret;
2023
 
2024enable_faults:
2025	ret = iommu_init_intcapxt(iommu);
2026	if (ret)
2027		return ret;
2028
2029	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2030
2031	if (iommu->ppr_log != NULL)
2032		iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
2033
2034	iommu_ga_log_enable(iommu);
2035
2036	return 0;
2037}
2038
2039/****************************************************************************
2040 *
2041 * The next functions belong to the third pass of parsing the ACPI
2042 * table. In this last pass the memory mapping requirements are
2043 * gathered (like exclusion and unity mapping ranges).
2044 *
2045 ****************************************************************************/
2046
2047static void __init free_unity_maps(void)
2048{
2049	struct unity_map_entry *entry, *next;
 
2050
2051	list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
2052		list_del(&entry->list);
2053		kfree(entry);
 
 
2054	}
2055}
2056
2057/* called for unity map ACPI definition */
2058static int __init init_unity_map_range(struct ivmd_header *m)
 
2059{
2060	struct unity_map_entry *e = NULL;
 
2061	char *s;
2062
 
 
 
 
2063	e = kzalloc(sizeof(*e), GFP_KERNEL);
2064	if (e == NULL)
2065		return -ENOMEM;
2066
2067	switch (m->type) {
2068	default:
2069		kfree(e);
2070		return 0;
2071	case ACPI_IVMD_TYPE:
2072		s = "IVMD_TYPEi\t\t\t";
2073		e->devid_start = e->devid_end = m->devid;
2074		break;
2075	case ACPI_IVMD_TYPE_ALL:
2076		s = "IVMD_TYPE_ALL\t\t";
2077		e->devid_start = 0;
2078		e->devid_end = amd_iommu_last_bdf;
2079		break;
2080	case ACPI_IVMD_TYPE_RANGE:
2081		s = "IVMD_TYPE_RANGE\t\t";
2082		e->devid_start = m->devid;
2083		e->devid_end = m->aux;
2084		break;
2085	}
2086	e->address_start = PAGE_ALIGN(m->range_start);
2087	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2088	e->prot = m->flags >> 1;
2089
2090	/*
2091	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2092	 * since some buggy BIOSes might lead to the overwritten exclusion
2093	 * range (exclusion_start and exclusion_length members). This
2094	 * happens when there are multiple exclusion ranges (IVMD entries)
2095	 * defined in ACPI table.
2096	 */
2097	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2098		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2099
2100	DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2101		    " range_start: %016llx range_end: %016llx flags: %x\n", s,
 
2102		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2103		    PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
 
2104		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2105		    e->address_start, e->address_end, m->flags);
2106
2107	list_add_tail(&e->list, &amd_iommu_unity_map);
2108
2109	return 0;
2110}
2111
2112/* iterates over all memory definitions we find in the ACPI table */
2113static int __init init_memory_definitions(struct acpi_table_header *table)
2114{
2115	u8 *p = (u8 *)table, *end = (u8 *)table;
2116	struct ivmd_header *m;
2117
2118	end += table->length;
2119	p += IVRS_HEADER_LENGTH;
2120
2121	while (p < end) {
2122		m = (struct ivmd_header *)p;
2123		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2124			init_unity_map_range(m);
2125
2126		p += m->length;
2127	}
2128
2129	return 0;
2130}
2131
2132/*
2133 * Init the device table to not allow DMA access for devices
2134 */
2135static void init_device_table_dma(void)
2136{
2137	u32 devid;
 
 
 
 
2138
2139	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2140		set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2141		set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
 
2142	}
2143}
2144
2145static void __init uninit_device_table_dma(void)
2146{
2147	u32 devid;
 
 
 
 
2148
2149	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2150		amd_iommu_dev_table[devid].data[0] = 0ULL;
2151		amd_iommu_dev_table[devid].data[1] = 0ULL;
2152	}
2153}
2154
2155static void init_device_table(void)
2156{
 
2157	u32 devid;
2158
2159	if (!amd_iommu_irq_remap)
2160		return;
2161
2162	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2163		set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
 
 
 
2164}
2165
2166static void iommu_init_flags(struct amd_iommu *iommu)
2167{
2168	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2169		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2170		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2171
2172	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2173		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2174		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2175
2176	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2177		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2178		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2179
2180	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2181		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2182		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2183
2184	/*
2185	 * make IOMMU memory accesses cache coherent
2186	 */
2187	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2188
2189	/* Set IOTLB invalidation timeout to 1s */
2190	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2191}
2192
2193static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2194{
2195	int i, j;
2196	u32 ioc_feature_control;
2197	struct pci_dev *pdev = iommu->root_pdev;
2198
2199	/* RD890 BIOSes may not have completely reconfigured the iommu */
2200	if (!is_rd890_iommu(iommu->dev) || !pdev)
2201		return;
2202
2203	/*
2204	 * First, we need to ensure that the iommu is enabled. This is
2205	 * controlled by a register in the northbridge
2206	 */
2207
2208	/* Select Northbridge indirect register 0x75 and enable writing */
2209	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2210	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2211
2212	/* Enable the iommu */
2213	if (!(ioc_feature_control & 0x1))
2214		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2215
2216	/* Restore the iommu BAR */
2217	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2218			       iommu->stored_addr_lo);
2219	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2220			       iommu->stored_addr_hi);
2221
2222	/* Restore the l1 indirect regs for each of the 6 l1s */
2223	for (i = 0; i < 6; i++)
2224		for (j = 0; j < 0x12; j++)
2225			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2226
2227	/* Restore the l2 indirect regs */
2228	for (i = 0; i < 0x83; i++)
2229		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2230
2231	/* Lock PCI setup registers */
2232	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2233			       iommu->stored_addr_lo | 1);
2234}
2235
2236static void iommu_enable_ga(struct amd_iommu *iommu)
2237{
2238#ifdef CONFIG_IRQ_REMAP
2239	switch (amd_iommu_guest_ir) {
2240	case AMD_IOMMU_GUEST_IR_VAPIC:
2241		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2242		fallthrough;
2243	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2244		iommu_feature_enable(iommu, CONTROL_GA_EN);
2245		iommu->irte_ops = &irte_128_ops;
2246		break;
2247	default:
2248		iommu->irte_ops = &irte_32_ops;
2249		break;
2250	}
2251#endif
2252}
2253
2254static void early_enable_iommu(struct amd_iommu *iommu)
2255{
2256	iommu_disable(iommu);
2257	iommu_init_flags(iommu);
2258	iommu_set_device_table(iommu);
2259	iommu_enable_command_buffer(iommu);
2260	iommu_enable_event_buffer(iommu);
2261	iommu_set_exclusion_range(iommu);
2262	iommu_enable_ga(iommu);
2263	iommu_enable_xt(iommu);
2264	iommu_enable(iommu);
2265	iommu_flush_all_caches(iommu);
2266}
2267
2268/*
2269 * This function finally enables all IOMMUs found in the system after
2270 * they have been initialized.
2271 *
2272 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2273 * the old content of device table entries. Not this case or copy failed,
2274 * just continue as normal kernel does.
2275 */
2276static void early_enable_iommus(void)
2277{
2278	struct amd_iommu *iommu;
2279
2280
2281	if (!copy_device_table()) {
2282		/*
2283		 * If come here because of failure in copying device table from old
2284		 * kernel with all IOMMUs enabled, print error message and try to
2285		 * free allocated old_dev_tbl_cpy.
2286		 */
2287		if (amd_iommu_pre_enabled)
2288			pr_err("Failed to copy DEV table from previous kernel.\n");
2289		if (old_dev_tbl_cpy != NULL)
2290			free_pages((unsigned long)old_dev_tbl_cpy,
2291					get_order(dev_table_size));
 
 
 
 
 
2292
2293		for_each_iommu(iommu) {
2294			clear_translation_pre_enabled(iommu);
2295			early_enable_iommu(iommu);
2296		}
2297	} else {
2298		pr_info("Copied DEV table from previous kernel.\n");
2299		free_pages((unsigned long)amd_iommu_dev_table,
2300				get_order(dev_table_size));
2301		amd_iommu_dev_table = old_dev_tbl_cpy;
 
 
 
 
2302		for_each_iommu(iommu) {
2303			iommu_disable_command_buffer(iommu);
2304			iommu_disable_event_buffer(iommu);
2305			iommu_enable_command_buffer(iommu);
2306			iommu_enable_event_buffer(iommu);
2307			iommu_enable_ga(iommu);
2308			iommu_enable_xt(iommu);
2309			iommu_set_device_table(iommu);
2310			iommu_flush_all_caches(iommu);
2311		}
2312	}
2313
2314#ifdef CONFIG_IRQ_REMAP
2315	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2316		amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2317#endif
2318}
2319
2320static void enable_iommus_v2(void)
2321{
2322	struct amd_iommu *iommu;
2323
2324	for_each_iommu(iommu) {
2325		iommu_enable_ppr_log(iommu);
2326		iommu_enable_gt(iommu);
2327	}
2328}
2329
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2330static void enable_iommus(void)
2331{
2332	early_enable_iommus();
2333
2334	enable_iommus_v2();
2335}
2336
2337static void disable_iommus(void)
2338{
2339	struct amd_iommu *iommu;
2340
2341	for_each_iommu(iommu)
2342		iommu_disable(iommu);
2343
2344#ifdef CONFIG_IRQ_REMAP
2345	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2346		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2347#endif
2348}
2349
2350/*
2351 * Suspend/Resume support
2352 * disable suspend until real resume implemented
2353 */
2354
2355static void amd_iommu_resume(void)
2356{
2357	struct amd_iommu *iommu;
2358
2359	for_each_iommu(iommu)
2360		iommu_apply_resume_quirks(iommu);
2361
2362	/* re-load the hardware */
2363	enable_iommus();
2364
2365	amd_iommu_enable_interrupts();
2366}
2367
2368static int amd_iommu_suspend(void)
2369{
2370	/* disable IOMMUs to go out of the way for BIOS */
2371	disable_iommus();
2372
2373	return 0;
2374}
2375
2376static struct syscore_ops amd_iommu_syscore_ops = {
2377	.suspend = amd_iommu_suspend,
2378	.resume = amd_iommu_resume,
2379};
2380
2381static void __init free_iommu_resources(void)
2382{
2383	kmemleak_free(irq_lookup_table);
2384	free_pages((unsigned long)irq_lookup_table,
2385		   get_order(rlookup_table_size));
2386	irq_lookup_table = NULL;
2387
2388	kmem_cache_destroy(amd_iommu_irq_cache);
2389	amd_iommu_irq_cache = NULL;
2390
2391	free_pages((unsigned long)amd_iommu_rlookup_table,
2392		   get_order(rlookup_table_size));
2393	amd_iommu_rlookup_table = NULL;
2394
2395	free_pages((unsigned long)amd_iommu_alias_table,
2396		   get_order(alias_table_size));
2397	amd_iommu_alias_table = NULL;
2398
2399	free_pages((unsigned long)amd_iommu_dev_table,
2400		   get_order(dev_table_size));
2401	amd_iommu_dev_table = NULL;
2402
2403	free_iommu_all();
 
2404}
2405
2406/* SB IOAPIC is always on this device in AMD systems */
2407#define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2408
2409static bool __init check_ioapic_information(void)
2410{
2411	const char *fw_bug = FW_BUG;
2412	bool ret, has_sb_ioapic;
2413	int idx;
2414
2415	has_sb_ioapic = false;
2416	ret           = false;
2417
2418	/*
2419	 * If we have map overrides on the kernel command line the
2420	 * messages in this function might not describe firmware bugs
2421	 * anymore - so be careful
2422	 */
2423	if (cmdline_maps)
2424		fw_bug = "";
2425
2426	for (idx = 0; idx < nr_ioapics; idx++) {
2427		int devid, id = mpc_ioapic_id(idx);
2428
2429		devid = get_ioapic_devid(id);
2430		if (devid < 0) {
2431			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2432				fw_bug, id);
2433			ret = false;
2434		} else if (devid == IOAPIC_SB_DEVID) {
2435			has_sb_ioapic = true;
2436			ret           = true;
2437		}
2438	}
2439
2440	if (!has_sb_ioapic) {
2441		/*
2442		 * We expect the SB IOAPIC to be listed in the IVRS
2443		 * table. The system timer is connected to the SB IOAPIC
2444		 * and if we don't have it in the list the system will
2445		 * panic at boot time.  This situation usually happens
2446		 * when the BIOS is buggy and provides us the wrong
2447		 * device id for the IOAPIC in the system.
2448		 */
2449		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2450	}
2451
2452	if (!ret)
2453		pr_err("Disabling interrupt remapping\n");
2454
2455	return ret;
2456}
2457
2458static void __init free_dma_resources(void)
2459{
2460	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2461		   get_order(MAX_DOMAIN_ID/8));
2462	amd_iommu_pd_alloc_bitmap = NULL;
2463
2464	free_unity_maps();
2465}
2466
 
 
 
 
 
2467/*
2468 * This is the hardware init function for AMD IOMMU in the system.
2469 * This function is called either from amd_iommu_init or from the interrupt
2470 * remapping setup code.
2471 *
2472 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2473 * four times:
2474 *
2475 *	1 pass) Discover the most comprehensive IVHD type to use.
2476 *
2477 *	2 pass) Find the highest PCI device id the driver has to handle.
2478 *		Upon this information the size of the data structures is
2479 *		determined that needs to be allocated.
2480 *
2481 *	3 pass) Initialize the data structures just allocated with the
2482 *		information in the ACPI table about available AMD IOMMUs
2483 *		in the system. It also maps the PCI devices in the
2484 *		system to specific IOMMUs
2485 *
2486 *	4 pass) After the basic data structures are allocated and
2487 *		initialized we update them with information about memory
2488 *		remapping requirements parsed out of the ACPI table in
2489 *		this last pass.
2490 *
2491 * After everything is set up the IOMMUs are enabled and the necessary
2492 * hotplug and suspend notifiers are registered.
2493 */
2494static int __init early_amd_iommu_init(void)
2495{
2496	struct acpi_table_header *ivrs_base;
 
2497	acpi_status status;
2498	int i, remap_cache_sz, ret = 0;
2499	u32 pci_id;
2500
2501	if (!amd_iommu_detected)
2502		return -ENODEV;
2503
2504	status = acpi_get_table("IVRS", 0, &ivrs_base);
2505	if (status == AE_NOT_FOUND)
2506		return -ENODEV;
2507	else if (ACPI_FAILURE(status)) {
2508		const char *err = acpi_format_exception(status);
2509		pr_err("IVRS table error: %s\n", err);
2510		return -EINVAL;
2511	}
2512
2513	/*
2514	 * Validate checksum here so we don't need to do it when
2515	 * we actually parse the table
2516	 */
2517	ret = check_ivrs_checksum(ivrs_base);
2518	if (ret)
2519		goto out;
2520
 
 
2521	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2522	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2523
2524	/*
2525	 * First parse ACPI tables to find the largest Bus/Dev/Func
2526	 * we need to handle. Upon this information the shared data
2527	 * structures for the IOMMUs in the system will be allocated
2528	 */
2529	ret = find_last_devid_acpi(ivrs_base);
2530	if (ret)
2531		goto out;
2532
2533	dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE);
2534	alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2535	rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2536
2537	/* Device table - directly used by all IOMMUs */
2538	ret = -ENOMEM;
2539	amd_iommu_dev_table = (void *)__get_free_pages(
2540				      GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2541				      get_order(dev_table_size));
2542	if (amd_iommu_dev_table == NULL)
2543		goto out;
2544
2545	/*
2546	 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2547	 * IOMMU see for that device
2548	 */
2549	amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2550			get_order(alias_table_size));
2551	if (amd_iommu_alias_table == NULL)
2552		goto out;
2553
2554	/* IOMMU rlookup table - find the IOMMU for a specific device */
2555	amd_iommu_rlookup_table = (void *)__get_free_pages(
2556			GFP_KERNEL | __GFP_ZERO,
2557			get_order(rlookup_table_size));
2558	if (amd_iommu_rlookup_table == NULL)
2559		goto out;
2560
2561	amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2562					    GFP_KERNEL | __GFP_ZERO,
2563					    get_order(MAX_DOMAIN_ID/8));
2564	if (amd_iommu_pd_alloc_bitmap == NULL)
2565		goto out;
2566
2567	/*
2568	 * let all alias entries point to itself
2569	 */
2570	for (i = 0; i <= amd_iommu_last_bdf; ++i)
2571		amd_iommu_alias_table[i] = i;
2572
2573	/*
2574	 * never allocate domain 0 because its used as the non-allocated and
2575	 * error value placeholder
2576	 */
2577	__set_bit(0, amd_iommu_pd_alloc_bitmap);
2578
2579	/*
2580	 * now the data structures are allocated and basically initialized
2581	 * start the real acpi table scan
2582	 */
2583	ret = init_iommu_all(ivrs_base);
2584	if (ret)
2585		goto out;
2586
2587	/* Disable IOMMU if there's Stoney Ridge graphics */
2588	for (i = 0; i < 32; i++) {
2589		pci_id = read_pci_config(0, i, 0, 0);
2590		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
2591			pr_info("Disable IOMMU on Stoney Ridge\n");
2592			amd_iommu_disabled = true;
2593			break;
2594		}
2595	}
2596
2597	/* Disable any previously enabled IOMMUs */
2598	if (!is_kdump_kernel() || amd_iommu_disabled)
2599		disable_iommus();
2600
2601	if (amd_iommu_irq_remap)
2602		amd_iommu_irq_remap = check_ioapic_information();
2603
2604	if (amd_iommu_irq_remap) {
 
2605		/*
2606		 * Interrupt remapping enabled, create kmem_cache for the
2607		 * remapping tables.
2608		 */
2609		ret = -ENOMEM;
2610		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2611			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2612		else
2613			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2614		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2615							remap_cache_sz,
2616							IRQ_TABLE_ALIGNMENT,
2617							0, NULL);
2618		if (!amd_iommu_irq_cache)
2619			goto out;
2620
2621		irq_lookup_table = (void *)__get_free_pages(
2622				GFP_KERNEL | __GFP_ZERO,
2623				get_order(rlookup_table_size));
2624		kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2625			       1, GFP_KERNEL);
2626		if (!irq_lookup_table)
2627			goto out;
2628	}
2629
2630	ret = init_memory_definitions(ivrs_base);
2631	if (ret)
2632		goto out;
2633
2634	/* init the device table */
2635	init_device_table();
2636
2637out:
2638	/* Don't leak any ACPI memory */
2639	acpi_put_table(ivrs_base);
2640	ivrs_base = NULL;
2641
2642	return ret;
2643}
2644
2645static int amd_iommu_enable_interrupts(void)
2646{
2647	struct amd_iommu *iommu;
2648	int ret = 0;
2649
2650	for_each_iommu(iommu) {
2651		ret = iommu_init_msi(iommu);
2652		if (ret)
2653			goto out;
2654	}
2655
2656out:
2657	return ret;
2658}
2659
2660static bool detect_ivrs(void)
2661{
2662	struct acpi_table_header *ivrs_base;
2663	acpi_status status;
 
2664
2665	status = acpi_get_table("IVRS", 0, &ivrs_base);
2666	if (status == AE_NOT_FOUND)
2667		return false;
2668	else if (ACPI_FAILURE(status)) {
2669		const char *err = acpi_format_exception(status);
2670		pr_err("IVRS table error: %s\n", err);
2671		return false;
2672	}
2673
2674	acpi_put_table(ivrs_base);
2675
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2676	/* Make sure ACS will be enabled during PCI probe */
2677	pci_request_acs();
2678
2679	return true;
2680}
2681
2682/****************************************************************************
2683 *
2684 * AMD IOMMU Initialization State Machine
2685 *
2686 ****************************************************************************/
2687
2688static int __init state_next(void)
2689{
2690	int ret = 0;
2691
2692	switch (init_state) {
2693	case IOMMU_START_STATE:
2694		if (!detect_ivrs()) {
2695			init_state	= IOMMU_NOT_FOUND;
2696			ret		= -ENODEV;
2697		} else {
2698			init_state	= IOMMU_IVRS_DETECTED;
2699		}
2700		break;
2701	case IOMMU_IVRS_DETECTED:
2702		ret = early_amd_iommu_init();
2703		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2704		if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2705			pr_info("AMD IOMMU disabled\n");
2706			init_state = IOMMU_CMDLINE_DISABLED;
2707			ret = -EINVAL;
 
 
 
2708		}
2709		break;
2710	case IOMMU_ACPI_FINISHED:
2711		early_enable_iommus();
2712		x86_platform.iommu_shutdown = disable_iommus;
2713		init_state = IOMMU_ENABLED;
2714		break;
2715	case IOMMU_ENABLED:
2716		register_syscore_ops(&amd_iommu_syscore_ops);
2717		ret = amd_iommu_init_pci();
2718		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
 
2719		enable_iommus_v2();
2720		break;
2721	case IOMMU_PCI_INIT:
2722		ret = amd_iommu_enable_interrupts();
2723		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2724		break;
2725	case IOMMU_INTERRUPTS_EN:
2726		ret = amd_iommu_init_dma_ops();
2727		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2728		break;
2729	case IOMMU_DMA_OPS:
2730		init_state = IOMMU_INITIALIZED;
2731		break;
2732	case IOMMU_INITIALIZED:
2733		/* Nothing to do */
2734		break;
2735	case IOMMU_NOT_FOUND:
2736	case IOMMU_INIT_ERROR:
2737	case IOMMU_CMDLINE_DISABLED:
2738		/* Error states => do nothing */
2739		ret = -EINVAL;
2740		break;
2741	default:
2742		/* Unknown state */
2743		BUG();
2744	}
2745
2746	if (ret) {
2747		free_dma_resources();
2748		if (!irq_remapping_enabled) {
2749			disable_iommus();
2750			free_iommu_resources();
2751		} else {
2752			struct amd_iommu *iommu;
 
 
 
 
2753
2754			uninit_device_table_dma();
2755			for_each_iommu(iommu)
2756				iommu_flush_all_caches(iommu);
2757		}
2758	}
2759	return ret;
2760}
2761
2762static int __init iommu_go_to_state(enum iommu_init_state state)
2763{
2764	int ret = -EINVAL;
2765
2766	while (init_state != state) {
2767		if (init_state == IOMMU_NOT_FOUND         ||
2768		    init_state == IOMMU_INIT_ERROR        ||
2769		    init_state == IOMMU_CMDLINE_DISABLED)
2770			break;
2771		ret = state_next();
2772	}
2773
2774	return ret;
2775}
2776
2777#ifdef CONFIG_IRQ_REMAP
2778int __init amd_iommu_prepare(void)
2779{
2780	int ret;
2781
2782	amd_iommu_irq_remap = true;
2783
2784	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2785	if (ret)
 
2786		return ret;
 
 
2787	return amd_iommu_irq_remap ? 0 : -ENODEV;
2788}
2789
2790int __init amd_iommu_enable(void)
2791{
2792	int ret;
2793
2794	ret = iommu_go_to_state(IOMMU_ENABLED);
2795	if (ret)
2796		return ret;
2797
2798	irq_remapping_enabled = 1;
2799	return amd_iommu_xt_mode;
2800}
2801
2802void amd_iommu_disable(void)
2803{
2804	amd_iommu_suspend();
2805}
2806
2807int amd_iommu_reenable(int mode)
2808{
2809	amd_iommu_resume();
2810
2811	return 0;
2812}
2813
2814int __init amd_iommu_enable_faulting(void)
2815{
2816	/* We enable MSI later when PCI is initialized */
2817	return 0;
2818}
2819#endif
2820
2821/*
2822 * This is the core init function for AMD IOMMU hardware in the system.
2823 * This function is called from the generic x86 DMA layer initialization
2824 * code.
2825 */
2826static int __init amd_iommu_init(void)
2827{
2828	struct amd_iommu *iommu;
2829	int ret;
2830
2831	ret = iommu_go_to_state(IOMMU_INITIALIZED);
2832#ifdef CONFIG_GART_IOMMU
2833	if (ret && list_empty(&amd_iommu_list)) {
2834		/*
2835		 * We failed to initialize the AMD IOMMU - try fallback
2836		 * to GART if possible.
2837		 */
2838		gart_iommu_init();
2839	}
2840#endif
2841
2842	for_each_iommu(iommu)
2843		amd_iommu_debugfs_setup(iommu);
2844
2845	return ret;
2846}
2847
2848static bool amd_iommu_sme_check(void)
2849{
2850	if (!sme_active() || (boot_cpu_data.x86 != 0x17))
 
2851		return true;
2852
2853	/* For Fam17h, a specific level of support is required */
2854	if (boot_cpu_data.microcode >= 0x08001205)
2855		return true;
2856
2857	if ((boot_cpu_data.microcode >= 0x08001126) &&
2858	    (boot_cpu_data.microcode <= 0x080011ff))
2859		return true;
2860
2861	pr_notice("IOMMU not currently supported when SME is active\n");
2862
2863	return false;
2864}
2865
2866/****************************************************************************
2867 *
2868 * Early detect code. This code runs at IOMMU detection time in the DMA
2869 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2870 * IOMMUs
2871 *
2872 ****************************************************************************/
2873int __init amd_iommu_detect(void)
2874{
2875	int ret;
2876
2877	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2878		return -ENODEV;
2879
2880	if (!amd_iommu_sme_check())
2881		return -ENODEV;
2882
2883	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2884	if (ret)
2885		return ret;
2886
2887	amd_iommu_detected = true;
2888	iommu_detected = 1;
2889	x86_init.iommu.iommu_init = amd_iommu_init;
2890
2891	return 1;
2892}
2893
2894/****************************************************************************
2895 *
2896 * Parsing functions for the AMD IOMMU specific kernel command line
2897 * options.
2898 *
2899 ****************************************************************************/
2900
2901static int __init parse_amd_iommu_dump(char *str)
2902{
2903	amd_iommu_dump = true;
2904
2905	return 1;
2906}
2907
2908static int __init parse_amd_iommu_intr(char *str)
2909{
2910	for (; *str; ++str) {
2911		if (strncmp(str, "legacy", 6) == 0) {
2912			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2913			break;
2914		}
2915		if (strncmp(str, "vapic", 5) == 0) {
2916			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2917			break;
2918		}
2919	}
2920	return 1;
2921}
2922
2923static int __init parse_amd_iommu_options(char *str)
2924{
2925	for (; *str; ++str) {
2926		if (strncmp(str, "fullflush", 9) == 0)
2927			amd_iommu_unmap_flush = true;
2928		if (strncmp(str, "off", 3) == 0)
 
 
 
 
 
 
2929			amd_iommu_disabled = true;
2930		if (strncmp(str, "force_isolation", 15) == 0)
2931			amd_iommu_force_isolation = true;
 
 
 
 
 
 
 
 
 
 
 
2932	}
2933
2934	return 1;
2935}
2936
2937static int __init parse_ivrs_ioapic(char *str)
2938{
2939	unsigned int bus, dev, fn;
2940	int ret, id, i;
2941	u16 devid;
2942
2943	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2944
2945	if (ret != 4) {
2946		pr_err("Invalid command line: ivrs_ioapic%s\n", str);
2947		return 1;
 
 
 
 
 
 
2948	}
2949
 
 
 
 
2950	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2951		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2952			str);
2953		return 1;
2954	}
2955
2956	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2957
2958	cmdline_maps			= true;
2959	i				= early_ioapic_map_size++;
2960	early_ioapic_map[i].id		= id;
2961	early_ioapic_map[i].devid	= devid;
2962	early_ioapic_map[i].cmd_line	= true;
2963
2964	return 1;
2965}
2966
2967static int __init parse_ivrs_hpet(char *str)
2968{
2969	unsigned int bus, dev, fn;
2970	int ret, id, i;
2971	u16 devid;
2972
2973	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2974
2975	if (ret != 4) {
2976		pr_err("Invalid command line: ivrs_hpet%s\n", str);
2977		return 1;
 
 
 
 
 
 
2978	}
2979
 
 
 
 
2980	if (early_hpet_map_size == EARLY_MAP_SIZE) {
2981		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
2982			str);
2983		return 1;
2984	}
2985
2986	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2987
2988	cmdline_maps			= true;
2989	i				= early_hpet_map_size++;
2990	early_hpet_map[i].id		= id;
2991	early_hpet_map[i].devid		= devid;
2992	early_hpet_map[i].cmd_line	= true;
2993
2994	return 1;
2995}
2996
2997static int __init parse_ivrs_acpihid(char *str)
2998{
2999	u32 bus, dev, fn;
3000	char *hid, *uid, *p;
3001	char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3002	int ret, i;
3003
3004	ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
3005	if (ret != 4) {
3006		pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
3007		return 1;
 
 
 
 
 
3008	}
3009
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3010	p = acpiid;
3011	hid = strsep(&p, ":");
3012	uid = p;
3013
3014	if (!hid || !(*hid) || !uid) {
3015		pr_err("Invalid command line: hid or uid\n");
3016		return 1;
3017	}
3018
 
 
 
 
 
 
 
3019	i = early_acpihid_map_size++;
3020	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3021	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3022	early_acpihid_map[i].devid =
3023		((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3024	early_acpihid_map[i].cmd_line	= true;
3025
3026	return 1;
3027}
3028
3029__setup("amd_iommu_dump",	parse_amd_iommu_dump);
3030__setup("amd_iommu=",		parse_amd_iommu_options);
3031__setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3032__setup("ivrs_ioapic",		parse_ivrs_ioapic);
3033__setup("ivrs_hpet",		parse_ivrs_hpet);
3034__setup("ivrs_acpihid",		parse_ivrs_acpihid);
3035
3036IOMMU_INIT_FINISH(amd_iommu_detect,
3037		  gart_iommu_hole_init,
3038		  NULL,
3039		  NULL);
3040
3041bool amd_iommu_v2_supported(void)
3042{
3043	return amd_iommu_v2_present;
 
 
 
 
 
3044}
3045EXPORT_SYMBOL(amd_iommu_v2_supported);
3046
3047struct amd_iommu *get_amd_iommu(unsigned int idx)
3048{
3049	unsigned int i = 0;
3050	struct amd_iommu *iommu;
3051
3052	for_each_iommu(iommu)
3053		if (i++ == idx)
3054			return iommu;
3055	return NULL;
3056}
3057EXPORT_SYMBOL(get_amd_iommu);
3058
3059/****************************************************************************
3060 *
3061 * IOMMU EFR Performance Counter support functionality. This code allows
3062 * access to the IOMMU PC functionality.
3063 *
3064 ****************************************************************************/
3065
3066u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3067{
3068	struct amd_iommu *iommu = get_amd_iommu(idx);
3069
3070	if (iommu)
3071		return iommu->max_banks;
3072
3073	return 0;
3074}
3075EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3076
3077bool amd_iommu_pc_supported(void)
3078{
3079	return amd_iommu_pc_present;
3080}
3081EXPORT_SYMBOL(amd_iommu_pc_supported);
3082
3083u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3084{
3085	struct amd_iommu *iommu = get_amd_iommu(idx);
3086
3087	if (iommu)
3088		return iommu->max_counters;
3089
3090	return 0;
3091}
3092EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3093
3094static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3095				u8 fxn, u64 *value, bool is_write)
3096{
3097	u32 offset;
3098	u32 max_offset_lim;
3099
3100	/* Make sure the IOMMU PC resource is available */
3101	if (!amd_iommu_pc_present)
3102		return -ENODEV;
3103
3104	/* Check for valid iommu and pc register indexing */
3105	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3106		return -ENODEV;
3107
3108	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3109
3110	/* Limit the offset to the hw defined mmio region aperture */
3111	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3112				(iommu->max_counters << 8) | 0x28);
3113	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3114	    (offset > max_offset_lim))
3115		return -EINVAL;
3116
3117	if (is_write) {
3118		u64 val = *value & GENMASK_ULL(47, 0);
3119
3120		writel((u32)val, iommu->mmio_base + offset);
3121		writel((val >> 32), iommu->mmio_base + offset + 4);
3122	} else {
3123		*value = readl(iommu->mmio_base + offset + 4);
3124		*value <<= 32;
3125		*value |= readl(iommu->mmio_base + offset);
3126		*value &= GENMASK_ULL(47, 0);
3127	}
3128
3129	return 0;
3130}
3131
3132int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3133{
3134	if (!iommu)
3135		return -EINVAL;
3136
3137	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3138}
3139EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3140
3141int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3142{
3143	if (!iommu)
3144		return -EINVAL;
3145
3146	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3147}
3148EXPORT_SYMBOL(amd_iommu_pc_set_reg);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
   4 * Author: Joerg Roedel <jroedel@suse.de>
   5 *         Leo Duran <leo.duran@amd.com>
   6 */
   7
   8#define pr_fmt(fmt)     "AMD-Vi: " fmt
   9#define dev_fmt(fmt)    pr_fmt(fmt)
  10
  11#include <linux/pci.h>
  12#include <linux/acpi.h>
  13#include <linux/list.h>
  14#include <linux/bitmap.h>
  15#include <linux/slab.h>
  16#include <linux/syscore_ops.h>
  17#include <linux/interrupt.h>
  18#include <linux/msi.h>
  19#include <linux/irq.h>
  20#include <linux/amd-iommu.h>
  21#include <linux/export.h>
  22#include <linux/kmemleak.h>
  23#include <linux/cc_platform.h>
  24#include <linux/iopoll.h>
  25#include <asm/pci-direct.h>
  26#include <asm/iommu.h>
  27#include <asm/apic.h>
 
  28#include <asm/gart.h>
  29#include <asm/x86_init.h>
 
  30#include <asm/io_apic.h>
  31#include <asm/irq_remapping.h>
  32#include <asm/set_memory.h>
  33
  34#include <linux/crash_dump.h>
  35
  36#include "amd_iommu.h"
  37#include "../irq_remapping.h"
  38
  39/*
  40 * definitions for the ACPI scanning code
  41 */
  42#define IVRS_HEADER_LENGTH 48
  43
  44#define ACPI_IVHD_TYPE_MAX_SUPPORTED	0x40
  45#define ACPI_IVMD_TYPE_ALL              0x20
  46#define ACPI_IVMD_TYPE                  0x21
  47#define ACPI_IVMD_TYPE_RANGE            0x22
  48
  49#define IVHD_DEV_ALL                    0x01
  50#define IVHD_DEV_SELECT                 0x02
  51#define IVHD_DEV_SELECT_RANGE_START     0x03
  52#define IVHD_DEV_RANGE_END              0x04
  53#define IVHD_DEV_ALIAS                  0x42
  54#define IVHD_DEV_ALIAS_RANGE            0x43
  55#define IVHD_DEV_EXT_SELECT             0x46
  56#define IVHD_DEV_EXT_SELECT_RANGE       0x47
  57#define IVHD_DEV_SPECIAL		0x48
  58#define IVHD_DEV_ACPI_HID		0xf0
  59
  60#define UID_NOT_PRESENT                 0
  61#define UID_IS_INTEGER                  1
  62#define UID_IS_CHARACTER                2
  63
  64#define IVHD_SPECIAL_IOAPIC		1
  65#define IVHD_SPECIAL_HPET		2
  66
  67#define IVHD_FLAG_HT_TUN_EN_MASK        0x01
  68#define IVHD_FLAG_PASSPW_EN_MASK        0x02
  69#define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
  70#define IVHD_FLAG_ISOC_EN_MASK          0x08
  71
  72#define IVMD_FLAG_EXCL_RANGE            0x08
  73#define IVMD_FLAG_IW                    0x04
  74#define IVMD_FLAG_IR                    0x02
  75#define IVMD_FLAG_UNITY_MAP             0x01
  76
  77#define ACPI_DEVFLAG_INITPASS           0x01
  78#define ACPI_DEVFLAG_EXTINT             0x02
  79#define ACPI_DEVFLAG_NMI                0x04
  80#define ACPI_DEVFLAG_SYSMGT1            0x10
  81#define ACPI_DEVFLAG_SYSMGT2            0x20
  82#define ACPI_DEVFLAG_LINT0              0x40
  83#define ACPI_DEVFLAG_LINT1              0x80
  84#define ACPI_DEVFLAG_ATSDIS             0x10000000
  85
  86#define LOOP_TIMEOUT	2000000
  87
  88#define IVRS_GET_SBDF_ID(seg, bus, dev, fn)	(((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \
  89						 | ((dev & 0x1f) << 3) | (fn & 0x7))
  90
  91/*
  92 * ACPI table definitions
  93 *
  94 * These data structures are laid over the table to parse the important values
  95 * out of it.
  96 */
  97
 
 
  98/*
  99 * structure describing one IOMMU in the ACPI table. Typically followed by one
 100 * or more ivhd_entrys.
 101 */
 102struct ivhd_header {
 103	u8 type;
 104	u8 flags;
 105	u16 length;
 106	u16 devid;
 107	u16 cap_ptr;
 108	u64 mmio_phys;
 109	u16 pci_seg;
 110	u16 info;
 111	u32 efr_attr;
 112
 113	/* Following only valid on IVHD type 11h and 40h */
 114	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
 115	u64 efr_reg2;
 116} __attribute__((packed));
 117
 118/*
 119 * A device entry describing which devices a specific IOMMU translates and
 120 * which requestor ids they use.
 121 */
 122struct ivhd_entry {
 123	u8 type;
 124	u16 devid;
 125	u8 flags;
 126	struct_group(ext_hid,
 127		u32 ext;
 128		u32 hidh;
 129	);
 130	u64 cid;
 131	u8 uidf;
 132	u8 uidl;
 133	u8 uid;
 134} __attribute__((packed));
 135
 136/*
 137 * An AMD IOMMU memory definition structure. It defines things like exclusion
 138 * ranges for devices and regions that should be unity mapped.
 139 */
 140struct ivmd_header {
 141	u8 type;
 142	u8 flags;
 143	u16 length;
 144	u16 devid;
 145	u16 aux;
 146	u16 pci_seg;
 147	u8  resv[6];
 148	u64 range_start;
 149	u64 range_length;
 150} __attribute__((packed));
 151
 152bool amd_iommu_dump;
 153bool amd_iommu_irq_remap __read_mostly;
 154
 155enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1;
 156
 157int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
 158static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
 159
 160static bool amd_iommu_detected;
 161static bool amd_iommu_disabled __initdata;
 162static bool amd_iommu_force_enable __initdata;
 163static int amd_iommu_target_ivhd_type;
 164
 165/* Global EFR and EFR2 registers */
 166u64 amd_iommu_efr;
 167u64 amd_iommu_efr2;
 168
 169/* SNP is enabled on the system? */
 170bool amd_iommu_snp_en;
 171EXPORT_SYMBOL(amd_iommu_snp_en);
 172
 173LIST_HEAD(amd_iommu_pci_seg_list);	/* list of all PCI segments */
 174LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
 175					   system */
 176
 177/* Array to assign indices to IOMMUs*/
 178struct amd_iommu *amd_iommus[MAX_IOMMUS];
 179
 180/* Number of IOMMUs present in the system */
 181static int amd_iommus_present;
 182
 183/* IOMMUs have a non-present cache? */
 184bool amd_iommu_np_cache __read_mostly;
 185bool amd_iommu_iotlb_sup __read_mostly = true;
 186
 187u32 amd_iommu_max_pasid __read_mostly = ~0;
 188
 189bool amd_iommu_v2_present __read_mostly;
 190static bool amd_iommu_pc_present __read_mostly;
 191bool amdr_ivrs_remap_support __read_mostly;
 192
 193bool amd_iommu_force_isolation __read_mostly;
 194
 195/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 196 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
 197 * to know which ones are already in use.
 198 */
 199unsigned long *amd_iommu_pd_alloc_bitmap;
 200
 
 
 
 
 201enum iommu_init_state {
 202	IOMMU_START_STATE,
 203	IOMMU_IVRS_DETECTED,
 204	IOMMU_ACPI_FINISHED,
 205	IOMMU_ENABLED,
 206	IOMMU_PCI_INIT,
 207	IOMMU_INTERRUPTS_EN,
 
 208	IOMMU_INITIALIZED,
 209	IOMMU_NOT_FOUND,
 210	IOMMU_INIT_ERROR,
 211	IOMMU_CMDLINE_DISABLED,
 212};
 213
 214/* Early ioapic and hpet maps from kernel command line */
 215#define EARLY_MAP_SIZE		4
 216static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
 217static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
 218static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
 219
 220static int __initdata early_ioapic_map_size;
 221static int __initdata early_hpet_map_size;
 222static int __initdata early_acpihid_map_size;
 223
 224static bool __initdata cmdline_maps;
 225
 226static enum iommu_init_state init_state = IOMMU_START_STATE;
 227
 228static int amd_iommu_enable_interrupts(void);
 229static int __init iommu_go_to_state(enum iommu_init_state state);
 230static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
 231
 232static bool amd_iommu_pre_enabled = true;
 233
 234static u32 amd_iommu_ivinfo __initdata;
 235
 236bool translation_pre_enabled(struct amd_iommu *iommu)
 237{
 238	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
 239}
 
 240
 241static void clear_translation_pre_enabled(struct amd_iommu *iommu)
 242{
 243	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
 244}
 245
 246static void init_translation_status(struct amd_iommu *iommu)
 247{
 248	u64 ctrl;
 249
 250	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
 251	if (ctrl & (1<<CONTROL_IOMMU_EN))
 252		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
 253}
 254
 255static inline unsigned long tbl_size(int entry_size, int last_bdf)
 
 
 
 
 
 
 256{
 257	unsigned shift = PAGE_SHIFT +
 258			 get_order((last_bdf + 1) * entry_size);
 259
 260	return 1UL << shift;
 261}
 262
 263int amd_iommu_get_num_iommus(void)
 264{
 265	return amd_iommus_present;
 266}
 267
 268/*
 269 * Iterate through all the IOMMUs to get common EFR
 270 * masks among all IOMMUs and warn if found inconsistency.
 271 */
 272static void get_global_efr(void)
 273{
 274	struct amd_iommu *iommu;
 275
 276	for_each_iommu(iommu) {
 277		u64 tmp = iommu->features;
 278		u64 tmp2 = iommu->features2;
 279
 280		if (list_is_first(&iommu->list, &amd_iommu_list)) {
 281			amd_iommu_efr = tmp;
 282			amd_iommu_efr2 = tmp2;
 283			continue;
 284		}
 285
 286		if (amd_iommu_efr == tmp &&
 287		    amd_iommu_efr2 == tmp2)
 288			continue;
 289
 290		pr_err(FW_BUG
 291		       "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n",
 292		       tmp, tmp2, amd_iommu_efr, amd_iommu_efr2,
 293		       iommu->index, iommu->pci_seg->id,
 294		       PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid),
 295		       PCI_FUNC(iommu->devid));
 296
 297		amd_iommu_efr &= tmp;
 298		amd_iommu_efr2 &= tmp2;
 299	}
 300
 301	pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2);
 302}
 303
 304static bool check_feature_on_all_iommus(u64 mask)
 305{
 306	return !!(amd_iommu_efr & mask);
 307}
 308
 309/*
 310 * For IVHD type 0x11/0x40, EFR is also available via IVHD.
 311 * Default to IVHD EFR since it is available sooner
 312 * (i.e. before PCI init).
 313 */
 314static void __init early_iommu_features_init(struct amd_iommu *iommu,
 315					     struct ivhd_header *h)
 316{
 317	if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) {
 318		iommu->features = h->efr_reg;
 319		iommu->features2 = h->efr_reg2;
 320	}
 321	if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
 322		amdr_ivrs_remap_support = true;
 323}
 324
 325/* Access to l1 and l2 indexed register spaces */
 326
 327static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
 328{
 329	u32 val;
 330
 331	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
 332	pci_read_config_dword(iommu->dev, 0xfc, &val);
 333	return val;
 334}
 335
 336static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
 337{
 338	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
 339	pci_write_config_dword(iommu->dev, 0xfc, val);
 340	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
 341}
 342
 343static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
 344{
 345	u32 val;
 346
 347	pci_write_config_dword(iommu->dev, 0xf0, address);
 348	pci_read_config_dword(iommu->dev, 0xf4, &val);
 349	return val;
 350}
 351
 352static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
 353{
 354	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
 355	pci_write_config_dword(iommu->dev, 0xf4, val);
 356}
 357
 358/****************************************************************************
 359 *
 360 * AMD IOMMU MMIO register space handling functions
 361 *
 362 * These functions are used to program the IOMMU device registers in
 363 * MMIO space required for that driver.
 364 *
 365 ****************************************************************************/
 366
 367/*
 368 * This function set the exclusion range in the IOMMU. DMA accesses to the
 369 * exclusion range are passed through untranslated
 370 */
 371static void iommu_set_exclusion_range(struct amd_iommu *iommu)
 372{
 373	u64 start = iommu->exclusion_start & PAGE_MASK;
 374	u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
 375	u64 entry;
 376
 377	if (!iommu->exclusion_start)
 378		return;
 379
 380	entry = start | MMIO_EXCL_ENABLE_MASK;
 381	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
 382			&entry, sizeof(entry));
 383
 384	entry = limit;
 385	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
 386			&entry, sizeof(entry));
 387}
 388
 389static void iommu_set_cwwb_range(struct amd_iommu *iommu)
 390{
 391	u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
 392	u64 entry = start & PM_ADDR_MASK;
 393
 394	if (!check_feature_on_all_iommus(FEATURE_SNP))
 395		return;
 396
 397	/* Note:
 398	 * Re-purpose Exclusion base/limit registers for Completion wait
 399	 * write-back base/limit.
 400	 */
 401	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
 402		    &entry, sizeof(entry));
 403
 404	/* Note:
 405	 * Default to 4 Kbytes, which can be specified by setting base
 406	 * address equal to the limit address.
 407	 */
 408	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
 409		    &entry, sizeof(entry));
 410}
 411
 412/* Programs the physical address of the device table into the IOMMU hardware */
 413static void iommu_set_device_table(struct amd_iommu *iommu)
 414{
 415	u64 entry;
 416	u32 dev_table_size = iommu->pci_seg->dev_table_size;
 417	void *dev_table = (void *)get_dev_table(iommu);
 418
 419	BUG_ON(iommu->mmio_base == NULL);
 420
 421	entry = iommu_virt_to_phys(dev_table);
 422	entry |= (dev_table_size >> 12) - 1;
 423	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
 424			&entry, sizeof(entry));
 425}
 426
 427/* Generic functions to enable/disable certain features of the IOMMU. */
 428static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
 429{
 430	u64 ctrl;
 431
 432	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
 433	ctrl |= (1ULL << bit);
 434	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
 435}
 436
 437static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
 438{
 439	u64 ctrl;
 440
 441	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
 442	ctrl &= ~(1ULL << bit);
 443	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
 444}
 445
 446static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
 447{
 448	u64 ctrl;
 449
 450	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
 451	ctrl &= ~CTRL_INV_TO_MASK;
 452	ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
 453	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
 454}
 455
 456/* Function to enable the hardware */
 457static void iommu_enable(struct amd_iommu *iommu)
 458{
 459	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
 460}
 461
 462static void iommu_disable(struct amd_iommu *iommu)
 463{
 464	if (!iommu->mmio_base)
 465		return;
 466
 467	/* Disable command buffer */
 468	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
 469
 470	/* Disable event logging and event interrupts */
 471	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
 472	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
 473
 474	/* Disable IOMMU GA_LOG */
 475	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
 476	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
 477
 478	/* Disable IOMMU hardware itself */
 479	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
 480}
 481
 482/*
 483 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
 484 * the system has one.
 485 */
 486static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
 487{
 488	if (!request_mem_region(address, end, "amd_iommu")) {
 489		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
 490			address, end);
 491		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
 492		return NULL;
 493	}
 494
 495	return (u8 __iomem *)ioremap(address, end);
 496}
 497
 498static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
 499{
 500	if (iommu->mmio_base)
 501		iounmap(iommu->mmio_base);
 502	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
 503}
 504
 505static inline u32 get_ivhd_header_size(struct ivhd_header *h)
 506{
 507	u32 size = 0;
 508
 509	switch (h->type) {
 510	case 0x10:
 511		size = 24;
 512		break;
 513	case 0x11:
 514	case 0x40:
 515		size = 40;
 516		break;
 517	}
 518	return size;
 519}
 520
 521/****************************************************************************
 522 *
 523 * The functions below belong to the first pass of AMD IOMMU ACPI table
 524 * parsing. In this pass we try to find out the highest device id this
 525 * code has to handle. Upon this information the size of the shared data
 526 * structures is determined later.
 527 *
 528 ****************************************************************************/
 529
 530/*
 531 * This function calculates the length of a given IVHD entry
 532 */
 533static inline int ivhd_entry_length(u8 *ivhd)
 534{
 535	u32 type = ((struct ivhd_entry *)ivhd)->type;
 536
 537	if (type < 0x80) {
 538		return 0x04 << (*ivhd >> 6);
 539	} else if (type == IVHD_DEV_ACPI_HID) {
 540		/* For ACPI_HID, offset 21 is uid len */
 541		return *((u8 *)ivhd + 21) + 22;
 542	}
 543	return 0;
 544}
 545
 546/*
 547 * After reading the highest device id from the IOMMU PCI capability header
 548 * this function looks if there is a higher device id defined in the ACPI table
 549 */
 550static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
 551{
 552	u8 *p = (void *)h, *end = (void *)h;
 553	struct ivhd_entry *dev;
 554	int last_devid = -EINVAL;
 555
 556	u32 ivhd_size = get_ivhd_header_size(h);
 557
 558	if (!ivhd_size) {
 559		pr_err("Unsupported IVHD type %#x\n", h->type);
 560		return -EINVAL;
 561	}
 562
 563	p += ivhd_size;
 564	end += h->length;
 565
 566	while (p < end) {
 567		dev = (struct ivhd_entry *)p;
 568		switch (dev->type) {
 569		case IVHD_DEV_ALL:
 570			/* Use maximum BDF value for DEV_ALL */
 571			return 0xffff;
 
 572		case IVHD_DEV_SELECT:
 573		case IVHD_DEV_RANGE_END:
 574		case IVHD_DEV_ALIAS:
 575		case IVHD_DEV_EXT_SELECT:
 576			/* all the above subfield types refer to device ids */
 577			if (dev->devid > last_devid)
 578				last_devid = dev->devid;
 579			break;
 580		default:
 581			break;
 582		}
 583		p += ivhd_entry_length(p);
 584	}
 585
 586	WARN_ON(p != end);
 587
 588	return last_devid;
 589}
 590
 591static int __init check_ivrs_checksum(struct acpi_table_header *table)
 592{
 593	int i;
 594	u8 checksum = 0, *p = (u8 *)table;
 595
 596	for (i = 0; i < table->length; ++i)
 597		checksum += p[i];
 598	if (checksum != 0) {
 599		/* ACPI table corrupt */
 600		pr_err(FW_BUG "IVRS invalid checksum\n");
 601		return -ENODEV;
 602	}
 603
 604	return 0;
 605}
 606
 607/*
 608 * Iterate over all IVHD entries in the ACPI table and find the highest device
 609 * id which we need to handle. This is the first of three functions which parse
 610 * the ACPI table. So we check the checksum here.
 611 */
 612static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
 613{
 614	u8 *p = (u8 *)table, *end = (u8 *)table;
 615	struct ivhd_header *h;
 616	int last_devid, last_bdf = 0;
 617
 618	p += IVRS_HEADER_LENGTH;
 619
 620	end += table->length;
 621	while (p < end) {
 622		h = (struct ivhd_header *)p;
 623		if (h->pci_seg == pci_seg &&
 624		    h->type == amd_iommu_target_ivhd_type) {
 625			last_devid = find_last_devid_from_ivhd(h);
 626
 627			if (last_devid < 0)
 628				return -EINVAL;
 629			if (last_devid > last_bdf)
 630				last_bdf = last_devid;
 631		}
 632		p += h->length;
 633	}
 634	WARN_ON(p != end);
 635
 636	return last_bdf;
 637}
 638
 639/****************************************************************************
 640 *
 641 * The following functions belong to the code path which parses the ACPI table
 642 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
 643 * data structures, initialize the per PCI segment device/alias/rlookup table
 644 * and also basically initialize the hardware.
 645 *
 646 ****************************************************************************/
 647
 648/* Allocate per PCI segment device table */
 649static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
 650{
 651	pci_seg->dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
 652						      get_order(pci_seg->dev_table_size));
 653	if (!pci_seg->dev_table)
 654		return -ENOMEM;
 655
 656	return 0;
 657}
 658
 659static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
 660{
 661	free_pages((unsigned long)pci_seg->dev_table,
 662		    get_order(pci_seg->dev_table_size));
 663	pci_seg->dev_table = NULL;
 664}
 665
 666/* Allocate per PCI segment IOMMU rlookup table. */
 667static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
 668{
 669	pci_seg->rlookup_table = (void *)__get_free_pages(
 670						GFP_KERNEL | __GFP_ZERO,
 671						get_order(pci_seg->rlookup_table_size));
 672	if (pci_seg->rlookup_table == NULL)
 673		return -ENOMEM;
 674
 675	return 0;
 676}
 677
 678static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
 679{
 680	free_pages((unsigned long)pci_seg->rlookup_table,
 681		   get_order(pci_seg->rlookup_table_size));
 682	pci_seg->rlookup_table = NULL;
 683}
 684
 685static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
 686{
 687	pci_seg->irq_lookup_table = (void *)__get_free_pages(
 688					     GFP_KERNEL | __GFP_ZERO,
 689					     get_order(pci_seg->rlookup_table_size));
 690	kmemleak_alloc(pci_seg->irq_lookup_table,
 691		       pci_seg->rlookup_table_size, 1, GFP_KERNEL);
 692	if (pci_seg->irq_lookup_table == NULL)
 693		return -ENOMEM;
 694
 695	return 0;
 696}
 697
 698static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
 699{
 700	kmemleak_free(pci_seg->irq_lookup_table);
 701	free_pages((unsigned long)pci_seg->irq_lookup_table,
 702		   get_order(pci_seg->rlookup_table_size));
 703	pci_seg->irq_lookup_table = NULL;
 704}
 705
 706static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
 707{
 708	int i;
 709
 710	pci_seg->alias_table = (void *)__get_free_pages(GFP_KERNEL,
 711					get_order(pci_seg->alias_table_size));
 712	if (!pci_seg->alias_table)
 713		return -ENOMEM;
 714
 715	/*
 716	 * let all alias entries point to itself
 717	 */
 718	for (i = 0; i <= pci_seg->last_bdf; ++i)
 719		pci_seg->alias_table[i] = i;
 720
 721	return 0;
 722}
 723
 724static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
 725{
 726	free_pages((unsigned long)pci_seg->alias_table,
 727		   get_order(pci_seg->alias_table_size));
 728	pci_seg->alias_table = NULL;
 729}
 730
 731/*
 732 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
 733 * write commands to that buffer later and the IOMMU will execute them
 734 * asynchronously
 735 */
 736static int __init alloc_command_buffer(struct amd_iommu *iommu)
 737{
 738	iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 739						  get_order(CMD_BUFFER_SIZE));
 740
 741	return iommu->cmd_buf ? 0 : -ENOMEM;
 742}
 743
 744/*
 745 * This function restarts event logging in case the IOMMU experienced
 746 * an event log buffer overflow.
 747 */
 748void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
 749{
 750	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
 751	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
 752}
 753
 754/*
 755 * This function resets the command buffer if the IOMMU stopped fetching
 756 * commands from it.
 757 */
 758static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
 759{
 760	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
 761
 762	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
 763	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
 764	iommu->cmd_buf_head = 0;
 765	iommu->cmd_buf_tail = 0;
 766
 767	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
 768}
 769
 770/*
 771 * This function writes the command buffer address to the hardware and
 772 * enables it.
 773 */
 774static void iommu_enable_command_buffer(struct amd_iommu *iommu)
 775{
 776	u64 entry;
 777
 778	BUG_ON(iommu->cmd_buf == NULL);
 779
 780	entry = iommu_virt_to_phys(iommu->cmd_buf);
 781	entry |= MMIO_CMD_SIZE_512;
 782
 783	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
 784		    &entry, sizeof(entry));
 785
 786	amd_iommu_reset_cmd_buffer(iommu);
 787}
 788
 789/*
 790 * This function disables the command buffer
 791 */
 792static void iommu_disable_command_buffer(struct amd_iommu *iommu)
 793{
 794	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
 795}
 796
 797static void __init free_command_buffer(struct amd_iommu *iommu)
 798{
 799	free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
 800}
 801
 802static void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu,
 803					 gfp_t gfp, size_t size)
 804{
 805	int order = get_order(size);
 806	void *buf = (void *)__get_free_pages(gfp, order);
 807
 808	if (buf &&
 809	    check_feature_on_all_iommus(FEATURE_SNP) &&
 810	    set_memory_4k((unsigned long)buf, (1 << order))) {
 811		free_pages((unsigned long)buf, order);
 812		buf = NULL;
 813	}
 814
 815	return buf;
 816}
 817
 818/* allocates the memory where the IOMMU will log its events to */
 819static int __init alloc_event_buffer(struct amd_iommu *iommu)
 820{
 821	iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
 822					      EVT_BUFFER_SIZE);
 823
 824	return iommu->evt_buf ? 0 : -ENOMEM;
 825}
 826
 827static void iommu_enable_event_buffer(struct amd_iommu *iommu)
 828{
 829	u64 entry;
 830
 831	BUG_ON(iommu->evt_buf == NULL);
 832
 833	entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
 834
 835	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
 836		    &entry, sizeof(entry));
 837
 838	/* set head and tail to zero manually */
 839	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
 840	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
 841
 842	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
 843}
 844
 845/*
 846 * This function disables the event log buffer
 847 */
 848static void iommu_disable_event_buffer(struct amd_iommu *iommu)
 849{
 850	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
 851}
 852
 853static void __init free_event_buffer(struct amd_iommu *iommu)
 854{
 855	free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
 856}
 857
 858/* allocates the memory where the IOMMU will log its events to */
 859static int __init alloc_ppr_log(struct amd_iommu *iommu)
 860{
 861	iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
 862					      PPR_LOG_SIZE);
 863
 864	return iommu->ppr_log ? 0 : -ENOMEM;
 865}
 866
 867static void iommu_enable_ppr_log(struct amd_iommu *iommu)
 868{
 869	u64 entry;
 870
 871	if (iommu->ppr_log == NULL)
 872		return;
 873
 874	entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
 875
 876	memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
 877		    &entry, sizeof(entry));
 878
 879	/* set head and tail to zero manually */
 880	writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
 881	writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
 882
 883	iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
 884	iommu_feature_enable(iommu, CONTROL_PPR_EN);
 885}
 886
 887static void __init free_ppr_log(struct amd_iommu *iommu)
 888{
 889	free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
 890}
 891
 892static void free_ga_log(struct amd_iommu *iommu)
 893{
 894#ifdef CONFIG_IRQ_REMAP
 895	free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE));
 896	free_pages((unsigned long)iommu->ga_log_tail, get_order(8));
 897#endif
 898}
 899
 900#ifdef CONFIG_IRQ_REMAP
 901static int iommu_ga_log_enable(struct amd_iommu *iommu)
 902{
 
 903	u32 status, i;
 904	u64 entry;
 905
 906	if (!iommu->ga_log)
 907		return -EINVAL;
 908
 909	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
 910	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
 911		    &entry, sizeof(entry));
 912	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
 913		 (BIT_ULL(52)-1)) & ~7ULL;
 914	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
 915		    &entry, sizeof(entry));
 916	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
 917	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
 918
 
 
 
 919
 920	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
 921	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
 922
 923	for (i = 0; i < LOOP_TIMEOUT; ++i) {
 924		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
 925		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
 926			break;
 927		udelay(10);
 928	}
 929
 930	if (WARN_ON(i >= LOOP_TIMEOUT))
 931		return -EINVAL;
 932
 933	return 0;
 934}
 935
 
 936static int iommu_init_ga_log(struct amd_iommu *iommu)
 937{
 
 
 938	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
 939		return 0;
 940
 941	iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 942					get_order(GA_LOG_SIZE));
 943	if (!iommu->ga_log)
 944		goto err_out;
 945
 946	iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 947					get_order(8));
 948	if (!iommu->ga_log_tail)
 949		goto err_out;
 950
 
 
 
 
 
 
 
 
 
 
 951	return 0;
 952err_out:
 953	free_ga_log(iommu);
 954	return -EINVAL;
 955}
 956#endif /* CONFIG_IRQ_REMAP */
 957
 958static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
 959{
 960	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 1);
 961
 962	return iommu->cmd_sem ? 0 : -ENOMEM;
 963}
 
 
 
 
 
 
 
 
 964
 965static void __init free_cwwb_sem(struct amd_iommu *iommu)
 966{
 967	if (iommu->cmd_sem)
 968		free_page((unsigned long)iommu->cmd_sem);
 969}
 970
 971static void iommu_enable_xt(struct amd_iommu *iommu)
 972{
 973#ifdef CONFIG_IRQ_REMAP
 974	/*
 975	 * XT mode (32-bit APIC destination ID) requires
 976	 * GA mode (128-bit IRTE support) as a prerequisite.
 977	 */
 978	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
 979	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
 980		iommu_feature_enable(iommu, CONTROL_XT_EN);
 981#endif /* CONFIG_IRQ_REMAP */
 982}
 983
 984static void iommu_enable_gt(struct amd_iommu *iommu)
 985{
 986	if (!iommu_feature(iommu, FEATURE_GT))
 987		return;
 988
 989	iommu_feature_enable(iommu, CONTROL_GT_EN);
 990}
 991
 992/* sets a specific bit in the device table entry. */
 993static void __set_dev_entry_bit(struct dev_table_entry *dev_table,
 994				u16 devid, u8 bit)
 995{
 996	int i = (bit >> 6) & 0x03;
 997	int _bit = bit & 0x3f;
 998
 999	dev_table[devid].data[i] |= (1UL << _bit);
1000}
1001
1002static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1003{
1004	struct dev_table_entry *dev_table = get_dev_table(iommu);
1005
1006	return __set_dev_entry_bit(dev_table, devid, bit);
1007}
1008
1009static int __get_dev_entry_bit(struct dev_table_entry *dev_table,
1010			       u16 devid, u8 bit)
1011{
1012	int i = (bit >> 6) & 0x03;
1013	int _bit = bit & 0x3f;
1014
1015	return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1016}
1017
1018static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1019{
1020	struct dev_table_entry *dev_table = get_dev_table(iommu);
1021
1022	return __get_dev_entry_bit(dev_table, devid, bit);
1023}
1024
1025static bool __copy_device_table(struct amd_iommu *iommu)
1026{
1027	u64 int_ctl, int_tab_len, entry = 0;
1028	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1029	struct dev_table_entry *old_devtb = NULL;
1030	u32 lo, hi, devid, old_devtb_size;
1031	phys_addr_t old_devtb_phys;
 
1032	u16 dom_id, dte_v, irq_v;
1033	gfp_t gfp_flag;
1034	u64 tmp;
1035
1036	/* Each IOMMU use separate device table with the same size */
1037	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1038	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1039	entry = (((u64) hi) << 32) + lo;
1040
1041	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1042	if (old_devtb_size != pci_seg->dev_table_size) {
1043		pr_err("The device table size of IOMMU:%d is not expected!\n",
1044			iommu->index);
1045		return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1046	}
1047
1048	/*
1049	 * When SME is enabled in the first kernel, the entry includes the
1050	 * memory encryption mask(sme_me_mask), we must remove the memory
1051	 * encryption mask to obtain the true physical address in kdump kernel.
1052	 */
1053	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1054
1055	if (old_devtb_phys >= 0x100000000ULL) {
1056		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1057		return false;
1058	}
1059	old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1060		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
1061							pci_seg->dev_table_size)
1062		    : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB);
1063
1064	if (!old_devtb)
1065		return false;
1066
1067	gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
1068	pci_seg->old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
1069						    get_order(pci_seg->dev_table_size));
1070	if (pci_seg->old_dev_tbl_cpy == NULL) {
1071		pr_err("Failed to allocate memory for copying old device table!\n");
1072		memunmap(old_devtb);
1073		return false;
1074	}
1075
1076	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
1077		pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1078		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1079		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1080
1081		if (dte_v && dom_id) {
1082			pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1083			pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1084			__set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
1085			/* If gcr3 table existed, mask it out */
1086			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1087				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1088				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1089				pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1090				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1091				tmp |= DTE_FLAG_GV;
1092				pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1093			}
1094		}
1095
1096		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1097		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1098		int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1099		if (irq_v && (int_ctl || int_tab_len)) {
1100			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1101			    (int_tab_len != DTE_INTTABLEN)) {
1102				pr_err("Wrong old irq remapping flag: %#x\n", devid);
1103				memunmap(old_devtb);
1104				return false;
1105			}
1106
1107			pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1108		}
1109	}
1110	memunmap(old_devtb);
1111
1112	return true;
1113}
1114
1115static bool copy_device_table(void)
1116{
1117	struct amd_iommu *iommu;
1118	struct amd_iommu_pci_seg *pci_seg;
1119
1120	if (!amd_iommu_pre_enabled)
1121		return false;
1122
1123	pr_warn("Translation is already enabled - trying to copy translation structures\n");
1124
1125	/*
1126	 * All IOMMUs within PCI segment shares common device table.
1127	 * Hence copy device table only once per PCI segment.
1128	 */
1129	for_each_pci_segment(pci_seg) {
1130		for_each_iommu(iommu) {
1131			if (pci_seg->id != iommu->pci_seg->id)
1132				continue;
1133			if (!__copy_device_table(iommu))
1134				return false;
1135			break;
1136		}
1137	}
1138
1139	return true;
1140}
1141
1142void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid)
 
1143{
1144	int sysmgt;
1145
1146	sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) |
1147		 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1);
1148
1149	if (sysmgt == 0x01)
1150		set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW);
1151}
1152
1153/*
1154 * This function takes the device specific flags read from the ACPI
1155 * table and sets up the device table entry with that information
1156 */
1157static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1158					   u16 devid, u32 flags, u32 ext_flags)
1159{
1160	if (flags & ACPI_DEVFLAG_INITPASS)
1161		set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS);
1162	if (flags & ACPI_DEVFLAG_EXTINT)
1163		set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS);
1164	if (flags & ACPI_DEVFLAG_NMI)
1165		set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS);
1166	if (flags & ACPI_DEVFLAG_SYSMGT1)
1167		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1);
1168	if (flags & ACPI_DEVFLAG_SYSMGT2)
1169		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2);
1170	if (flags & ACPI_DEVFLAG_LINT0)
1171		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS);
1172	if (flags & ACPI_DEVFLAG_LINT1)
1173		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS);
1174
1175	amd_iommu_apply_erratum_63(iommu, devid);
1176
1177	amd_iommu_set_rlookup_table(iommu, devid);
1178}
1179
1180int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line)
1181{
1182	struct devid_map *entry;
1183	struct list_head *list;
1184
1185	if (type == IVHD_SPECIAL_IOAPIC)
1186		list = &ioapic_map;
1187	else if (type == IVHD_SPECIAL_HPET)
1188		list = &hpet_map;
1189	else
1190		return -EINVAL;
1191
1192	list_for_each_entry(entry, list, list) {
1193		if (!(entry->id == id && entry->cmd_line))
1194			continue;
1195
1196		pr_info("Command-line override present for %s id %d - ignoring\n",
1197			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1198
1199		*devid = entry->devid;
1200
1201		return 0;
1202	}
1203
1204	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1205	if (!entry)
1206		return -ENOMEM;
1207
1208	entry->id	= id;
1209	entry->devid	= *devid;
1210	entry->cmd_line	= cmd_line;
1211
1212	list_add_tail(&entry->list, list);
1213
1214	return 0;
1215}
1216
1217static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid,
1218				      bool cmd_line)
1219{
1220	struct acpihid_map_entry *entry;
1221	struct list_head *list = &acpihid_map;
1222
1223	list_for_each_entry(entry, list, list) {
1224		if (strcmp(entry->hid, hid) ||
1225		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1226		    !entry->cmd_line)
1227			continue;
1228
1229		pr_info("Command-line override for hid:%s uid:%s\n",
1230			hid, uid);
1231		*devid = entry->devid;
1232		return 0;
1233	}
1234
1235	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1236	if (!entry)
1237		return -ENOMEM;
1238
1239	memcpy(entry->uid, uid, strlen(uid));
1240	memcpy(entry->hid, hid, strlen(hid));
1241	entry->devid = *devid;
1242	entry->cmd_line	= cmd_line;
1243	entry->root_devid = (entry->devid & (~0x7));
1244
1245	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1246		entry->cmd_line ? "cmd" : "ivrs",
1247		entry->hid, entry->uid, entry->root_devid);
1248
1249	list_add_tail(&entry->list, list);
1250	return 0;
1251}
1252
1253static int __init add_early_maps(void)
1254{
1255	int i, ret;
1256
1257	for (i = 0; i < early_ioapic_map_size; ++i) {
1258		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1259					 early_ioapic_map[i].id,
1260					 &early_ioapic_map[i].devid,
1261					 early_ioapic_map[i].cmd_line);
1262		if (ret)
1263			return ret;
1264	}
1265
1266	for (i = 0; i < early_hpet_map_size; ++i) {
1267		ret = add_special_device(IVHD_SPECIAL_HPET,
1268					 early_hpet_map[i].id,
1269					 &early_hpet_map[i].devid,
1270					 early_hpet_map[i].cmd_line);
1271		if (ret)
1272			return ret;
1273	}
1274
1275	for (i = 0; i < early_acpihid_map_size; ++i) {
1276		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1277					  early_acpihid_map[i].uid,
1278					  &early_acpihid_map[i].devid,
1279					  early_acpihid_map[i].cmd_line);
1280		if (ret)
1281			return ret;
1282	}
1283
1284	return 0;
1285}
1286
1287/*
1288 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1289 * initializes the hardware and our data structures with it.
1290 */
1291static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1292					struct ivhd_header *h)
1293{
1294	u8 *p = (u8 *)h;
1295	u8 *end = p, flags = 0;
1296	u16 devid = 0, devid_start = 0, devid_to = 0, seg_id;
1297	u32 dev_i, ext_flags = 0;
1298	bool alias = false;
1299	struct ivhd_entry *e;
1300	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1301	u32 ivhd_size;
1302	int ret;
1303
1304
1305	ret = add_early_maps();
1306	if (ret)
1307		return ret;
1308
1309	amd_iommu_apply_ivrs_quirks();
1310
1311	/*
1312	 * First save the recommended feature enable bits from ACPI
1313	 */
1314	iommu->acpi_flags = h->flags;
1315
1316	/*
1317	 * Done. Now parse the device entries
1318	 */
1319	ivhd_size = get_ivhd_header_size(h);
1320	if (!ivhd_size) {
1321		pr_err("Unsupported IVHD type %#x\n", h->type);
1322		return -EINVAL;
1323	}
1324
1325	p += ivhd_size;
1326
1327	end += h->length;
1328
1329
1330	while (p < end) {
1331		e = (struct ivhd_entry *)p;
1332		seg_id = pci_seg->id;
1333
1334		switch (e->type) {
1335		case IVHD_DEV_ALL:
1336
1337			DUMP_printk("  DEV_ALL\t\t\tflags: %02x\n", e->flags);
1338
1339			for (dev_i = 0; dev_i <= pci_seg->last_bdf; ++dev_i)
1340				set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1341			break;
1342		case IVHD_DEV_SELECT:
1343
1344			DUMP_printk("  DEV_SELECT\t\t\t devid: %04x:%02x:%02x.%x "
1345				    "flags: %02x\n",
1346				    seg_id, PCI_BUS_NUM(e->devid),
1347				    PCI_SLOT(e->devid),
1348				    PCI_FUNC(e->devid),
1349				    e->flags);
1350
1351			devid = e->devid;
1352			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1353			break;
1354		case IVHD_DEV_SELECT_RANGE_START:
1355
1356			DUMP_printk("  DEV_SELECT_RANGE_START\t "
1357				    "devid: %04x:%02x:%02x.%x flags: %02x\n",
1358				    seg_id, PCI_BUS_NUM(e->devid),
1359				    PCI_SLOT(e->devid),
1360				    PCI_FUNC(e->devid),
1361				    e->flags);
1362
1363			devid_start = e->devid;
1364			flags = e->flags;
1365			ext_flags = 0;
1366			alias = false;
1367			break;
1368		case IVHD_DEV_ALIAS:
1369
1370			DUMP_printk("  DEV_ALIAS\t\t\t devid: %04x:%02x:%02x.%x "
1371				    "flags: %02x devid_to: %02x:%02x.%x\n",
1372				    seg_id, PCI_BUS_NUM(e->devid),
1373				    PCI_SLOT(e->devid),
1374				    PCI_FUNC(e->devid),
1375				    e->flags,
1376				    PCI_BUS_NUM(e->ext >> 8),
1377				    PCI_SLOT(e->ext >> 8),
1378				    PCI_FUNC(e->ext >> 8));
1379
1380			devid = e->devid;
1381			devid_to = e->ext >> 8;
1382			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1383			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1384			pci_seg->alias_table[devid] = devid_to;
1385			break;
1386		case IVHD_DEV_ALIAS_RANGE:
1387
1388			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
1389				    "devid: %04x:%02x:%02x.%x flags: %02x "
1390				    "devid_to: %04x:%02x:%02x.%x\n",
1391				    seg_id, PCI_BUS_NUM(e->devid),
1392				    PCI_SLOT(e->devid),
1393				    PCI_FUNC(e->devid),
1394				    e->flags,
1395				    seg_id, PCI_BUS_NUM(e->ext >> 8),
1396				    PCI_SLOT(e->ext >> 8),
1397				    PCI_FUNC(e->ext >> 8));
1398
1399			devid_start = e->devid;
1400			flags = e->flags;
1401			devid_to = e->ext >> 8;
1402			ext_flags = 0;
1403			alias = true;
1404			break;
1405		case IVHD_DEV_EXT_SELECT:
1406
1407			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %04x:%02x:%02x.%x "
1408				    "flags: %02x ext: %08x\n",
1409				    seg_id, PCI_BUS_NUM(e->devid),
1410				    PCI_SLOT(e->devid),
1411				    PCI_FUNC(e->devid),
1412				    e->flags, e->ext);
1413
1414			devid = e->devid;
1415			set_dev_entry_from_acpi(iommu, devid, e->flags,
1416						e->ext);
1417			break;
1418		case IVHD_DEV_EXT_SELECT_RANGE:
1419
1420			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
1421				    "%04x:%02x:%02x.%x flags: %02x ext: %08x\n",
1422				    seg_id, PCI_BUS_NUM(e->devid),
1423				    PCI_SLOT(e->devid),
1424				    PCI_FUNC(e->devid),
1425				    e->flags, e->ext);
1426
1427			devid_start = e->devid;
1428			flags = e->flags;
1429			ext_flags = e->ext;
1430			alias = false;
1431			break;
1432		case IVHD_DEV_RANGE_END:
1433
1434			DUMP_printk("  DEV_RANGE_END\t\t devid: %04x:%02x:%02x.%x\n",
1435				    seg_id, PCI_BUS_NUM(e->devid),
1436				    PCI_SLOT(e->devid),
1437				    PCI_FUNC(e->devid));
1438
1439			devid = e->devid;
1440			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1441				if (alias) {
1442					pci_seg->alias_table[dev_i] = devid_to;
1443					set_dev_entry_from_acpi(iommu,
1444						devid_to, flags, ext_flags);
1445				}
1446				set_dev_entry_from_acpi(iommu, dev_i,
1447							flags, ext_flags);
1448			}
1449			break;
1450		case IVHD_DEV_SPECIAL: {
1451			u8 handle, type;
1452			const char *var;
1453			u32 devid;
1454			int ret;
1455
1456			handle = e->ext & 0xff;
1457			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8));
1458			type   = (e->ext >> 24) & 0xff;
1459
1460			if (type == IVHD_SPECIAL_IOAPIC)
1461				var = "IOAPIC";
1462			else if (type == IVHD_SPECIAL_HPET)
1463				var = "HPET";
1464			else
1465				var = "UNKNOWN";
1466
1467			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x\n",
1468				    var, (int)handle,
1469				    seg_id, PCI_BUS_NUM(devid),
1470				    PCI_SLOT(devid),
1471				    PCI_FUNC(devid));
1472
1473			ret = add_special_device(type, handle, &devid, false);
1474			if (ret)
1475				return ret;
1476
1477			/*
1478			 * add_special_device might update the devid in case a
1479			 * command-line override is present. So call
1480			 * set_dev_entry_from_acpi after add_special_device.
1481			 */
1482			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1483
1484			break;
1485		}
1486		case IVHD_DEV_ACPI_HID: {
1487			u32 devid;
1488			u8 hid[ACPIHID_HID_LEN];
1489			u8 uid[ACPIHID_UID_LEN];
1490			int ret;
1491
1492			if (h->type != 0x40) {
1493				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1494				       e->type);
1495				break;
1496			}
1497
1498			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1499			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1500			hid[ACPIHID_HID_LEN - 1] = '\0';
1501
1502			if (!(*hid)) {
1503				pr_err(FW_BUG "Invalid HID.\n");
1504				break;
1505			}
1506
1507			uid[0] = '\0';
1508			switch (e->uidf) {
1509			case UID_NOT_PRESENT:
1510
1511				if (e->uidl != 0)
1512					pr_warn(FW_BUG "Invalid UID length.\n");
1513
1514				break;
1515			case UID_IS_INTEGER:
1516
1517				sprintf(uid, "%d", e->uid);
1518
1519				break;
1520			case UID_IS_CHARACTER:
1521
1522				memcpy(uid, &e->uid, e->uidl);
1523				uid[e->uidl] = '\0';
1524
1525				break;
1526			default:
1527				break;
1528			}
1529
1530			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid);
1531			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x\n",
1532				    hid, uid, seg_id,
1533				    PCI_BUS_NUM(devid),
1534				    PCI_SLOT(devid),
1535				    PCI_FUNC(devid));
1536
1537			flags = e->flags;
1538
1539			ret = add_acpi_hid_device(hid, uid, &devid, false);
1540			if (ret)
1541				return ret;
1542
1543			/*
1544			 * add_special_device might update the devid in case a
1545			 * command-line override is present. So call
1546			 * set_dev_entry_from_acpi after add_special_device.
1547			 */
1548			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1549
1550			break;
1551		}
1552		default:
1553			break;
1554		}
1555
1556		p += ivhd_entry_length(p);
1557	}
1558
1559	return 0;
1560}
1561
1562/* Allocate PCI segment data structure */
1563static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1564					  struct acpi_table_header *ivrs_base)
1565{
1566	struct amd_iommu_pci_seg *pci_seg;
1567	int last_bdf;
1568
1569	/*
1570	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1571	 * handle in this PCI segment. Upon this information the shared data
1572	 * structures for the PCI segments in the system will be allocated.
1573	 */
1574	last_bdf = find_last_devid_acpi(ivrs_base, id);
1575	if (last_bdf < 0)
1576		return NULL;
1577
1578	pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1579	if (pci_seg == NULL)
1580		return NULL;
1581
1582	pci_seg->last_bdf = last_bdf;
1583	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1584	pci_seg->dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE, last_bdf);
1585	pci_seg->alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE, last_bdf);
1586	pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE, last_bdf);
1587
1588	pci_seg->id = id;
1589	init_llist_head(&pci_seg->dev_data_list);
1590	INIT_LIST_HEAD(&pci_seg->unity_map);
1591	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1592
1593	if (alloc_dev_table(pci_seg))
1594		return NULL;
1595	if (alloc_alias_table(pci_seg))
1596		return NULL;
1597	if (alloc_rlookup_table(pci_seg))
1598		return NULL;
1599
1600	return pci_seg;
1601}
1602
1603static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1604					struct acpi_table_header *ivrs_base)
1605{
1606	struct amd_iommu_pci_seg *pci_seg;
1607
1608	for_each_pci_segment(pci_seg) {
1609		if (pci_seg->id == id)
1610			return pci_seg;
1611	}
1612
1613	return alloc_pci_segment(id, ivrs_base);
1614}
1615
1616static void __init free_pci_segments(void)
1617{
1618	struct amd_iommu_pci_seg *pci_seg, *next;
1619
1620	for_each_pci_segment_safe(pci_seg, next) {
1621		list_del(&pci_seg->list);
1622		free_irq_lookup_table(pci_seg);
1623		free_rlookup_table(pci_seg);
1624		free_alias_table(pci_seg);
1625		free_dev_table(pci_seg);
1626		kfree(pci_seg);
1627	}
1628}
1629
1630static void __init free_iommu_one(struct amd_iommu *iommu)
1631{
1632	free_cwwb_sem(iommu);
1633	free_command_buffer(iommu);
1634	free_event_buffer(iommu);
1635	free_ppr_log(iommu);
1636	free_ga_log(iommu);
1637	iommu_unmap_mmio_space(iommu);
1638}
1639
1640static void __init free_iommu_all(void)
1641{
1642	struct amd_iommu *iommu, *next;
1643
1644	for_each_iommu_safe(iommu, next) {
1645		list_del(&iommu->list);
1646		free_iommu_one(iommu);
1647		kfree(iommu);
1648	}
1649}
1650
1651/*
1652 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1653 * Workaround:
1654 *     BIOS should disable L2B micellaneous clock gating by setting
1655 *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1656 */
1657static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1658{
1659	u32 value;
1660
1661	if ((boot_cpu_data.x86 != 0x15) ||
1662	    (boot_cpu_data.x86_model < 0x10) ||
1663	    (boot_cpu_data.x86_model > 0x1f))
1664		return;
1665
1666	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1667	pci_read_config_dword(iommu->dev, 0xf4, &value);
1668
1669	if (value & BIT(2))
1670		return;
1671
1672	/* Select NB indirect register 0x90 and enable writing */
1673	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1674
1675	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1676	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1677
1678	/* Clear the enable writing bit */
1679	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1680}
1681
1682/*
1683 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1684 * Workaround:
1685 *     BIOS should enable ATS write permission check by setting
1686 *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1687 */
1688static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1689{
1690	u32 value;
1691
1692	if ((boot_cpu_data.x86 != 0x15) ||
1693	    (boot_cpu_data.x86_model < 0x30) ||
1694	    (boot_cpu_data.x86_model > 0x3f))
1695		return;
1696
1697	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1698	value = iommu_read_l2(iommu, 0x47);
1699
1700	if (value & BIT(0))
1701		return;
1702
1703	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1704	iommu_write_l2(iommu, 0x47, value | BIT(0));
1705
1706	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1707}
1708
1709/*
1710 * This function glues the initialization function for one IOMMU
1711 * together and also allocates the command buffer and programs the
1712 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1713 */
1714static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1715				 struct acpi_table_header *ivrs_base)
1716{
1717	struct amd_iommu_pci_seg *pci_seg;
1718
1719	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1720	if (pci_seg == NULL)
1721		return -ENOMEM;
1722	iommu->pci_seg = pci_seg;
1723
1724	raw_spin_lock_init(&iommu->lock);
1725	iommu->cmd_sem_val = 0;
1726
1727	/* Add IOMMU to internal data structures */
1728	list_add_tail(&iommu->list, &amd_iommu_list);
1729	iommu->index = amd_iommus_present++;
1730
1731	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1732		WARN(1, "System has more IOMMUs than supported by this driver\n");
1733		return -ENOSYS;
1734	}
1735
1736	/* Index is fine - add IOMMU to the array */
1737	amd_iommus[iommu->index] = iommu;
1738
1739	/*
1740	 * Copy data from ACPI table entry to the iommu struct
1741	 */
1742	iommu->devid   = h->devid;
1743	iommu->cap_ptr = h->cap_ptr;
 
1744	iommu->mmio_phys = h->mmio_phys;
1745
1746	switch (h->type) {
1747	case 0x10:
1748		/* Check if IVHD EFR contains proper max banks/counters */
1749		if ((h->efr_attr != 0) &&
1750		    ((h->efr_attr & (0xF << 13)) != 0) &&
1751		    ((h->efr_attr & (0x3F << 17)) != 0))
1752			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1753		else
1754			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1755
1756		/*
1757		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1758		 * GAM also requires GA mode. Therefore, we need to
1759		 * check cmpxchg16b support before enabling it.
1760		 */
1761		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1762		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1763			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1764		break;
1765	case 0x11:
1766	case 0x40:
1767		if (h->efr_reg & (1 << 9))
1768			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1769		else
1770			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1771
1772		/*
1773		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1774		 * XT, GAM also requires GA mode. Therefore, we need to
1775		 * check cmpxchg16b support before enabling them.
1776		 */
1777		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1778		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1779			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1780			break;
1781		}
1782
1783		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
 
 
 
 
 
 
 
1784			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1785
1786		early_iommu_features_init(iommu, h);
1787
1788		break;
1789	default:
1790		return -EINVAL;
1791	}
1792
1793	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1794						iommu->mmio_phys_end);
1795	if (!iommu->mmio_base)
1796		return -ENOMEM;
1797
1798	return init_iommu_from_acpi(iommu, h);
1799}
1800
1801static int __init init_iommu_one_late(struct amd_iommu *iommu)
1802{
1803	int ret;
1804
1805	if (alloc_cwwb_sem(iommu))
1806		return -ENOMEM;
1807
1808	if (alloc_command_buffer(iommu))
1809		return -ENOMEM;
1810
1811	if (alloc_event_buffer(iommu))
1812		return -ENOMEM;
1813
1814	iommu->int_enabled = false;
1815
1816	init_translation_status(iommu);
1817	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1818		iommu_disable(iommu);
1819		clear_translation_pre_enabled(iommu);
1820		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1821			iommu->index);
1822	}
1823	if (amd_iommu_pre_enabled)
1824		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1825
1826	if (amd_iommu_irq_remap) {
1827		ret = amd_iommu_create_irq_domain(iommu);
1828		if (ret)
1829			return ret;
1830	}
 
 
1831
1832	/*
1833	 * Make sure IOMMU is not considered to translate itself. The IVRS
1834	 * table tells us so, but this is a lie!
1835	 */
1836	iommu->pci_seg->rlookup_table[iommu->devid] = NULL;
1837
1838	return 0;
1839}
1840
1841/**
1842 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1843 * @ivrs: Pointer to the IVRS header
1844 *
1845 * This function search through all IVDB of the maximum supported IVHD
1846 */
1847static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1848{
1849	u8 *base = (u8 *)ivrs;
1850	struct ivhd_header *ivhd = (struct ivhd_header *)
1851					(base + IVRS_HEADER_LENGTH);
1852	u8 last_type = ivhd->type;
1853	u16 devid = ivhd->devid;
1854
1855	while (((u8 *)ivhd - base < ivrs->length) &&
1856	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1857		u8 *p = (u8 *) ivhd;
1858
1859		if (ivhd->devid == devid)
1860			last_type = ivhd->type;
1861		ivhd = (struct ivhd_header *)(p + ivhd->length);
1862	}
1863
1864	return last_type;
1865}
1866
1867/*
1868 * Iterates over all IOMMU entries in the ACPI table, allocates the
1869 * IOMMU structure and initializes it with init_iommu_one()
1870 */
1871static int __init init_iommu_all(struct acpi_table_header *table)
1872{
1873	u8 *p = (u8 *)table, *end = (u8 *)table;
1874	struct ivhd_header *h;
1875	struct amd_iommu *iommu;
1876	int ret;
1877
1878	end += table->length;
1879	p += IVRS_HEADER_LENGTH;
1880
1881	/* Phase 1: Process all IVHD blocks */
1882	while (p < end) {
1883		h = (struct ivhd_header *)p;
1884		if (*p == amd_iommu_target_ivhd_type) {
1885
1886			DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x "
1887				    "flags: %01x info %04x\n",
1888				    h->pci_seg, PCI_BUS_NUM(h->devid),
1889				    PCI_SLOT(h->devid), PCI_FUNC(h->devid),
1890				    h->cap_ptr, h->flags, h->info);
1891			DUMP_printk("       mmio-addr: %016llx\n",
1892				    h->mmio_phys);
1893
1894			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1895			if (iommu == NULL)
1896				return -ENOMEM;
1897
1898			ret = init_iommu_one(iommu, h, table);
1899			if (ret)
1900				return ret;
1901		}
1902		p += h->length;
1903
1904	}
1905	WARN_ON(p != end);
1906
1907	/* Phase 2 : Early feature support check */
1908	get_global_efr();
1909
1910	/* Phase 3 : Enabling IOMMU features */
1911	for_each_iommu(iommu) {
1912		ret = init_iommu_one_late(iommu);
1913		if (ret)
1914			return ret;
1915	}
1916
1917	return 0;
1918}
1919
 
 
 
1920static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1921{
1922	u64 val;
1923	struct pci_dev *pdev = iommu->dev;
 
1924
1925	if (!iommu_feature(iommu, FEATURE_PC))
1926		return;
1927
1928	amd_iommu_pc_present = true;
1929
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1930	pci_info(pdev, "IOMMU performance counters supported\n");
1931
1932	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1933	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1934	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1935
1936	return;
 
 
 
 
 
1937}
1938
1939static ssize_t amd_iommu_show_cap(struct device *dev,
1940				  struct device_attribute *attr,
1941				  char *buf)
1942{
1943	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1944	return sprintf(buf, "%x\n", iommu->cap);
1945}
1946static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1947
1948static ssize_t amd_iommu_show_features(struct device *dev,
1949				       struct device_attribute *attr,
1950				       char *buf)
1951{
1952	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1953	return sprintf(buf, "%llx:%llx\n", iommu->features2, iommu->features);
1954}
1955static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1956
1957static struct attribute *amd_iommu_attrs[] = {
1958	&dev_attr_cap.attr,
1959	&dev_attr_features.attr,
1960	NULL,
1961};
1962
1963static struct attribute_group amd_iommu_group = {
1964	.name = "amd-iommu",
1965	.attrs = amd_iommu_attrs,
1966};
1967
1968static const struct attribute_group *amd_iommu_groups[] = {
1969	&amd_iommu_group,
1970	NULL,
1971};
1972
1973/*
1974 * Note: IVHD 0x11 and 0x40 also contains exact copy
1975 * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1976 * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1977 */
1978static void __init late_iommu_features_init(struct amd_iommu *iommu)
1979{
1980	u64 features, features2;
1981
1982	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1983		return;
1984
1985	/* read extended feature bits */
1986	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1987	features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2);
1988
1989	if (!iommu->features) {
1990		iommu->features = features;
1991		iommu->features2 = features2;
1992		return;
1993	}
1994
1995	/*
1996	 * Sanity check and warn if EFR values from
1997	 * IVHD and MMIO conflict.
1998	 */
1999	if (features != iommu->features ||
2000	    features2 != iommu->features2) {
2001		pr_warn(FW_WARN
2002			"EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n",
2003			features, iommu->features,
2004			features2, iommu->features2);
2005	}
2006}
2007
2008static int __init iommu_init_pci(struct amd_iommu *iommu)
2009{
2010	int cap_ptr = iommu->cap_ptr;
2011	int ret;
2012
2013	iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2014						 PCI_BUS_NUM(iommu->devid),
2015						 iommu->devid & 0xff);
2016	if (!iommu->dev)
2017		return -ENODEV;
2018
2019	/* Prevent binding other PCI device drivers to IOMMU devices */
2020	iommu->dev->match_driver = false;
2021
2022	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2023			      &iommu->cap);
2024
2025	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2026		amd_iommu_iotlb_sup = false;
2027
2028	late_iommu_features_init(iommu);
 
2029
2030	if (iommu_feature(iommu, FEATURE_GT)) {
2031		int glxval;
2032		u32 max_pasid;
2033		u64 pasmax;
2034
2035		pasmax = iommu->features & FEATURE_PASID_MASK;
2036		pasmax >>= FEATURE_PASID_SHIFT;
2037		max_pasid  = (1 << (pasmax + 1)) - 1;
2038
2039		amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
2040
2041		BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
2042
2043		glxval   = iommu->features & FEATURE_GLXVAL_MASK;
2044		glxval >>= FEATURE_GLXVAL_SHIFT;
2045
2046		if (amd_iommu_max_glx_val == -1)
2047			amd_iommu_max_glx_val = glxval;
2048		else
2049			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2050	}
2051
2052	if (iommu_feature(iommu, FEATURE_GT) &&
2053	    iommu_feature(iommu, FEATURE_PPR)) {
2054		iommu->is_iommu_v2   = true;
2055		amd_iommu_v2_present = true;
2056	}
2057
2058	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
2059		return -ENOMEM;
2060
2061	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2062		pr_info("Using strict mode due to virtualization\n");
2063		iommu_set_dma_strict();
 
 
2064		amd_iommu_np_cache = true;
2065	}
2066
2067	init_iommu_perf_ctr(iommu);
2068
2069	if (amd_iommu_pgtable == AMD_IOMMU_V2) {
2070		if (!iommu_feature(iommu, FEATURE_GIOSUP) ||
2071		    !iommu_feature(iommu, FEATURE_GT)) {
2072			pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n");
2073			amd_iommu_pgtable = AMD_IOMMU_V1;
2074		} else if (iommu_default_passthrough()) {
2075			pr_warn("V2 page table doesn't support passthrough mode. Fallback to v1.\n");
2076			amd_iommu_pgtable = AMD_IOMMU_V1;
2077		}
2078	}
2079
2080	if (is_rd890_iommu(iommu->dev)) {
2081		int i, j;
2082
2083		iommu->root_pdev =
2084			pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2085						    iommu->dev->bus->number,
2086						    PCI_DEVFN(0, 0));
2087
2088		/*
2089		 * Some rd890 systems may not be fully reconfigured by the
2090		 * BIOS, so it's necessary for us to store this information so
2091		 * it can be reprogrammed on resume
2092		 */
2093		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2094				&iommu->stored_addr_lo);
2095		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2096				&iommu->stored_addr_hi);
2097
2098		/* Low bit locks writes to configuration space */
2099		iommu->stored_addr_lo &= ~1;
2100
2101		for (i = 0; i < 6; i++)
2102			for (j = 0; j < 0x12; j++)
2103				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2104
2105		for (i = 0; i < 0x83; i++)
2106			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2107	}
2108
2109	amd_iommu_erratum_746_workaround(iommu);
2110	amd_iommu_ats_write_check_workaround(iommu);
2111
2112	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2113			       amd_iommu_groups, "ivhd%d", iommu->index);
2114	if (ret)
2115		return ret;
2116
2117	iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2118
2119	return pci_enable_device(iommu->dev);
2120}
2121
2122static void print_iommu_info(void)
2123{
2124	static const char * const feat_str[] = {
2125		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2126		"IA", "GA", "HE", "PC"
2127	};
2128	struct amd_iommu *iommu;
2129
2130	for_each_iommu(iommu) {
2131		struct pci_dev *pdev = iommu->dev;
2132		int i;
2133
2134		pci_info(pdev, "Found IOMMU cap 0x%x\n", iommu->cap_ptr);
2135
2136		if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
2137			pr_info("Extended features (%#llx, %#llx):", iommu->features, iommu->features2);
2138
2139			for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2140				if (iommu_feature(iommu, (1ULL << i)))
2141					pr_cont(" %s", feat_str[i]);
2142			}
2143
2144			if (iommu->features & FEATURE_GAM_VAPIC)
2145				pr_cont(" GA_vAPIC");
2146
2147			if (iommu->features & FEATURE_SNP)
2148				pr_cont(" SNP");
2149
2150			pr_cont("\n");
2151		}
2152	}
2153	if (irq_remapping_enabled) {
2154		pr_info("Interrupt remapping enabled\n");
 
 
2155		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2156			pr_info("X2APIC enabled\n");
2157	}
2158	if (amd_iommu_pgtable == AMD_IOMMU_V2)
2159		pr_info("V2 page table enabled\n");
2160}
2161
2162static int __init amd_iommu_init_pci(void)
2163{
2164	struct amd_iommu *iommu;
2165	struct amd_iommu_pci_seg *pci_seg;
2166	int ret;
2167
2168	for_each_iommu(iommu) {
2169		ret = iommu_init_pci(iommu);
2170		if (ret) {
2171			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2172			       iommu->index, ret);
2173			goto out;
2174		}
2175		/* Need to setup range after PCI init */
2176		iommu_set_cwwb_range(iommu);
2177	}
2178
2179	/*
2180	 * Order is important here to make sure any unity map requirements are
2181	 * fulfilled. The unity mappings are created and written to the device
2182	 * table during the iommu_init_pci() call.
2183	 *
2184	 * After that we call init_device_table_dma() to make sure any
2185	 * uninitialized DTE will block DMA, and in the end we flush the caches
2186	 * of all IOMMUs to make sure the changes to the device table are
2187	 * active.
2188	 */
2189	for_each_pci_segment(pci_seg)
2190		init_device_table_dma(pci_seg);
 
2191
2192	for_each_iommu(iommu)
2193		iommu_flush_all_caches(iommu);
2194
2195	print_iommu_info();
 
2196
2197out:
2198	return ret;
2199}
2200
2201/****************************************************************************
2202 *
2203 * The following functions initialize the MSI interrupts for all IOMMUs
2204 * in the system. It's a bit challenging because there could be multiple
2205 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2206 * pci_dev.
2207 *
2208 ****************************************************************************/
2209
2210static int iommu_setup_msi(struct amd_iommu *iommu)
2211{
2212	int r;
2213
2214	r = pci_enable_msi(iommu->dev);
2215	if (r)
2216		return r;
2217
2218	r = request_threaded_irq(iommu->dev->irq,
2219				 amd_iommu_int_handler,
2220				 amd_iommu_int_thread,
2221				 0, "AMD-Vi",
2222				 iommu);
2223
2224	if (r) {
2225		pci_disable_msi(iommu->dev);
2226		return r;
2227	}
2228
2229	return 0;
2230}
2231
2232union intcapxt {
2233	u64	capxt;
2234	struct {
2235		u64	reserved_0		:  2,
2236			dest_mode_logical	:  1,
2237			reserved_1		:  5,
2238			destid_0_23		: 24,
2239			vector			:  8,
2240			reserved_2		: 16,
2241			destid_24_31		:  8;
2242	};
2243} __attribute__ ((packed));
2244
2245
2246static struct irq_chip intcapxt_controller;
2247
2248static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2249				       struct irq_data *irqd, bool reserve)
2250{
2251	return 0;
2252}
2253
2254static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2255					  struct irq_data *irqd)
2256{
2257}
2258
2259
2260static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2261				    unsigned int nr_irqs, void *arg)
 
 
 
2262{
2263	struct irq_alloc_info *info = arg;
2264	int i, ret;
2265
2266	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2267		return -EINVAL;
2268
2269	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2270	if (ret < 0)
2271		return ret;
2272
2273	for (i = virq; i < virq + nr_irqs; i++) {
2274		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2275
2276		irqd->chip = &intcapxt_controller;
2277		irqd->chip_data = info->data;
2278		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2279	}
2280
2281	return ret;
2282}
2283
2284static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2285				    unsigned int nr_irqs)
2286{
2287	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2288}
2289
2290
2291static void intcapxt_unmask_irq(struct irq_data *irqd)
2292{
2293	struct amd_iommu *iommu = irqd->chip_data;
2294	struct irq_cfg *cfg = irqd_cfg(irqd);
2295	union intcapxt xt;
2296
2297	xt.capxt = 0ULL;
2298	xt.dest_mode_logical = apic->dest_mode_logical;
2299	xt.vector = cfg->vector;
2300	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2301	xt.destid_24_31 = cfg->dest_apicid >> 24;
2302
2303	/**
2304	 * Current IOMMU implementation uses the same IRQ for all
2305	 * 3 IOMMU interrupts.
2306	 */
2307	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2308	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2309	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2310}
2311
2312static void intcapxt_mask_irq(struct irq_data *irqd)
 
2313{
2314	struct amd_iommu *iommu = irqd->chip_data;
2315
2316	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2317	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2318	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
 
 
 
2319}
2320
2321
2322static int intcapxt_set_affinity(struct irq_data *irqd,
2323				 const struct cpumask *mask, bool force)
2324{
2325	struct irq_data *parent = irqd->parent_data;
2326	int ret;
2327
2328	ret = parent->chip->irq_set_affinity(parent, mask, force);
2329	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2330		return ret;
2331	return 0;
2332}
2333
2334static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2335{
2336	return on ? -EOPNOTSUPP : 0;
2337}
2338
2339static struct irq_chip intcapxt_controller = {
2340	.name			= "IOMMU-MSI",
2341	.irq_unmask		= intcapxt_unmask_irq,
2342	.irq_mask		= intcapxt_mask_irq,
2343	.irq_ack		= irq_chip_ack_parent,
2344	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2345	.irq_set_affinity       = intcapxt_set_affinity,
2346	.irq_set_wake		= intcapxt_set_wake,
2347	.flags			= IRQCHIP_MASK_ON_SUSPEND,
2348};
2349
2350static const struct irq_domain_ops intcapxt_domain_ops = {
2351	.alloc			= intcapxt_irqdomain_alloc,
2352	.free			= intcapxt_irqdomain_free,
2353	.activate		= intcapxt_irqdomain_activate,
2354	.deactivate		= intcapxt_irqdomain_deactivate,
2355};
2356
2357
2358static struct irq_domain *iommu_irqdomain;
2359
2360static struct irq_domain *iommu_get_irqdomain(void)
2361{
2362	struct fwnode_handle *fn;
2363
2364	/* No need for locking here (yet) as the init is single-threaded */
2365	if (iommu_irqdomain)
2366		return iommu_irqdomain;
2367
2368	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2369	if (!fn)
2370		return NULL;
2371
2372	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2373						      fn, &intcapxt_domain_ops,
2374						      NULL);
2375	if (!iommu_irqdomain)
2376		irq_domain_free_fwnode(fn);
2377
2378	return iommu_irqdomain;
2379}
2380
2381static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2382{
2383	struct irq_domain *domain;
2384	struct irq_alloc_info info;
2385	int irq, ret;
2386
2387	domain = iommu_get_irqdomain();
2388	if (!domain)
2389		return -ENXIO;
2390
2391	init_irq_alloc_info(&info, NULL);
2392	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2393	info.data = iommu;
2394
2395	irq = irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
2396	if (irq < 0) {
2397		irq_domain_remove(domain);
2398		return irq;
2399	}
2400
2401	ret = request_threaded_irq(irq, amd_iommu_int_handler,
2402				   amd_iommu_int_thread, 0, "AMD-Vi", iommu);
2403	if (ret) {
2404		irq_domain_free_irqs(irq, 1);
2405		irq_domain_remove(domain);
2406		return ret;
2407	}
2408
2409	return 0;
 
 
2410}
2411
2412static int iommu_init_irq(struct amd_iommu *iommu)
2413{
2414	int ret;
2415
2416	if (iommu->int_enabled)
2417		goto enable_faults;
2418
2419	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2420		ret = iommu_setup_intcapxt(iommu);
2421	else if (iommu->dev->msi_cap)
2422		ret = iommu_setup_msi(iommu);
2423	else
2424		ret = -ENODEV;
2425
2426	if (ret)
2427		return ret;
2428
2429	iommu->int_enabled = true;
2430enable_faults:
2431
2432	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2433		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2434
2435	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2436
2437	if (iommu->ppr_log != NULL)
2438		iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
 
 
 
2439	return 0;
2440}
2441
2442/****************************************************************************
2443 *
2444 * The next functions belong to the third pass of parsing the ACPI
2445 * table. In this last pass the memory mapping requirements are
2446 * gathered (like exclusion and unity mapping ranges).
2447 *
2448 ****************************************************************************/
2449
2450static void __init free_unity_maps(void)
2451{
2452	struct unity_map_entry *entry, *next;
2453	struct amd_iommu_pci_seg *p, *pci_seg;
2454
2455	for_each_pci_segment_safe(pci_seg, p) {
2456		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2457			list_del(&entry->list);
2458			kfree(entry);
2459		}
2460	}
2461}
2462
2463/* called for unity map ACPI definition */
2464static int __init init_unity_map_range(struct ivmd_header *m,
2465				       struct acpi_table_header *ivrs_base)
2466{
2467	struct unity_map_entry *e = NULL;
2468	struct amd_iommu_pci_seg *pci_seg;
2469	char *s;
2470
2471	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2472	if (pci_seg == NULL)
2473		return -ENOMEM;
2474
2475	e = kzalloc(sizeof(*e), GFP_KERNEL);
2476	if (e == NULL)
2477		return -ENOMEM;
2478
2479	switch (m->type) {
2480	default:
2481		kfree(e);
2482		return 0;
2483	case ACPI_IVMD_TYPE:
2484		s = "IVMD_TYPEi\t\t\t";
2485		e->devid_start = e->devid_end = m->devid;
2486		break;
2487	case ACPI_IVMD_TYPE_ALL:
2488		s = "IVMD_TYPE_ALL\t\t";
2489		e->devid_start = 0;
2490		e->devid_end = pci_seg->last_bdf;
2491		break;
2492	case ACPI_IVMD_TYPE_RANGE:
2493		s = "IVMD_TYPE_RANGE\t\t";
2494		e->devid_start = m->devid;
2495		e->devid_end = m->aux;
2496		break;
2497	}
2498	e->address_start = PAGE_ALIGN(m->range_start);
2499	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2500	e->prot = m->flags >> 1;
2501
2502	/*
2503	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2504	 * since some buggy BIOSes might lead to the overwritten exclusion
2505	 * range (exclusion_start and exclusion_length members). This
2506	 * happens when there are multiple exclusion ranges (IVMD entries)
2507	 * defined in ACPI table.
2508	 */
2509	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2510		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2511
2512	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2513		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2514		    " flags: %x\n", s, m->pci_seg,
2515		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2516		    PCI_FUNC(e->devid_start), m->pci_seg,
2517		    PCI_BUS_NUM(e->devid_end),
2518		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2519		    e->address_start, e->address_end, m->flags);
2520
2521	list_add_tail(&e->list, &pci_seg->unity_map);
2522
2523	return 0;
2524}
2525
2526/* iterates over all memory definitions we find in the ACPI table */
2527static int __init init_memory_definitions(struct acpi_table_header *table)
2528{
2529	u8 *p = (u8 *)table, *end = (u8 *)table;
2530	struct ivmd_header *m;
2531
2532	end += table->length;
2533	p += IVRS_HEADER_LENGTH;
2534
2535	while (p < end) {
2536		m = (struct ivmd_header *)p;
2537		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2538			init_unity_map_range(m, table);
2539
2540		p += m->length;
2541	}
2542
2543	return 0;
2544}
2545
2546/*
2547 * Init the device table to not allow DMA access for devices
2548 */
2549static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2550{
2551	u32 devid;
2552	struct dev_table_entry *dev_table = pci_seg->dev_table;
2553
2554	if (dev_table == NULL)
2555		return;
2556
2557	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2558		__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID);
2559		if (!amd_iommu_snp_en)
2560			__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION);
2561	}
2562}
2563
2564static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2565{
2566	u32 devid;
2567	struct dev_table_entry *dev_table = pci_seg->dev_table;
2568
2569	if (dev_table == NULL)
2570		return;
2571
2572	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2573		dev_table[devid].data[0] = 0ULL;
2574		dev_table[devid].data[1] = 0ULL;
2575	}
2576}
2577
2578static void init_device_table(void)
2579{
2580	struct amd_iommu_pci_seg *pci_seg;
2581	u32 devid;
2582
2583	if (!amd_iommu_irq_remap)
2584		return;
2585
2586	for_each_pci_segment(pci_seg) {
2587		for (devid = 0; devid <= pci_seg->last_bdf; ++devid)
2588			__set_dev_entry_bit(pci_seg->dev_table,
2589					    devid, DEV_ENTRY_IRQ_TBL_EN);
2590	}
2591}
2592
2593static void iommu_init_flags(struct amd_iommu *iommu)
2594{
2595	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2596		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2597		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2598
2599	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2600		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2601		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2602
2603	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2604		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2605		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2606
2607	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2608		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2609		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2610
2611	/*
2612	 * make IOMMU memory accesses cache coherent
2613	 */
2614	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2615
2616	/* Set IOTLB invalidation timeout to 1s */
2617	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2618}
2619
2620static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2621{
2622	int i, j;
2623	u32 ioc_feature_control;
2624	struct pci_dev *pdev = iommu->root_pdev;
2625
2626	/* RD890 BIOSes may not have completely reconfigured the iommu */
2627	if (!is_rd890_iommu(iommu->dev) || !pdev)
2628		return;
2629
2630	/*
2631	 * First, we need to ensure that the iommu is enabled. This is
2632	 * controlled by a register in the northbridge
2633	 */
2634
2635	/* Select Northbridge indirect register 0x75 and enable writing */
2636	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2637	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2638
2639	/* Enable the iommu */
2640	if (!(ioc_feature_control & 0x1))
2641		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2642
2643	/* Restore the iommu BAR */
2644	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2645			       iommu->stored_addr_lo);
2646	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2647			       iommu->stored_addr_hi);
2648
2649	/* Restore the l1 indirect regs for each of the 6 l1s */
2650	for (i = 0; i < 6; i++)
2651		for (j = 0; j < 0x12; j++)
2652			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2653
2654	/* Restore the l2 indirect regs */
2655	for (i = 0; i < 0x83; i++)
2656		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2657
2658	/* Lock PCI setup registers */
2659	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2660			       iommu->stored_addr_lo | 1);
2661}
2662
2663static void iommu_enable_ga(struct amd_iommu *iommu)
2664{
2665#ifdef CONFIG_IRQ_REMAP
2666	switch (amd_iommu_guest_ir) {
2667	case AMD_IOMMU_GUEST_IR_VAPIC:
 
 
2668	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2669		iommu_feature_enable(iommu, CONTROL_GA_EN);
2670		iommu->irte_ops = &irte_128_ops;
2671		break;
2672	default:
2673		iommu->irte_ops = &irte_32_ops;
2674		break;
2675	}
2676#endif
2677}
2678
2679static void early_enable_iommu(struct amd_iommu *iommu)
2680{
2681	iommu_disable(iommu);
2682	iommu_init_flags(iommu);
2683	iommu_set_device_table(iommu);
2684	iommu_enable_command_buffer(iommu);
2685	iommu_enable_event_buffer(iommu);
2686	iommu_set_exclusion_range(iommu);
2687	iommu_enable_ga(iommu);
2688	iommu_enable_xt(iommu);
2689	iommu_enable(iommu);
2690	iommu_flush_all_caches(iommu);
2691}
2692
2693/*
2694 * This function finally enables all IOMMUs found in the system after
2695 * they have been initialized.
2696 *
2697 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2698 * the old content of device table entries. Not this case or copy failed,
2699 * just continue as normal kernel does.
2700 */
2701static void early_enable_iommus(void)
2702{
2703	struct amd_iommu *iommu;
2704	struct amd_iommu_pci_seg *pci_seg;
2705
2706	if (!copy_device_table()) {
2707		/*
2708		 * If come here because of failure in copying device table from old
2709		 * kernel with all IOMMUs enabled, print error message and try to
2710		 * free allocated old_dev_tbl_cpy.
2711		 */
2712		if (amd_iommu_pre_enabled)
2713			pr_err("Failed to copy DEV table from previous kernel.\n");
2714
2715		for_each_pci_segment(pci_seg) {
2716			if (pci_seg->old_dev_tbl_cpy != NULL) {
2717				free_pages((unsigned long)pci_seg->old_dev_tbl_cpy,
2718						get_order(pci_seg->dev_table_size));
2719				pci_seg->old_dev_tbl_cpy = NULL;
2720			}
2721		}
2722
2723		for_each_iommu(iommu) {
2724			clear_translation_pre_enabled(iommu);
2725			early_enable_iommu(iommu);
2726		}
2727	} else {
2728		pr_info("Copied DEV table from previous kernel.\n");
2729
2730		for_each_pci_segment(pci_seg) {
2731			free_pages((unsigned long)pci_seg->dev_table,
2732				   get_order(pci_seg->dev_table_size));
2733			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2734		}
2735
2736		for_each_iommu(iommu) {
2737			iommu_disable_command_buffer(iommu);
2738			iommu_disable_event_buffer(iommu);
2739			iommu_enable_command_buffer(iommu);
2740			iommu_enable_event_buffer(iommu);
2741			iommu_enable_ga(iommu);
2742			iommu_enable_xt(iommu);
2743			iommu_set_device_table(iommu);
2744			iommu_flush_all_caches(iommu);
2745		}
2746	}
 
 
 
 
 
2747}
2748
2749static void enable_iommus_v2(void)
2750{
2751	struct amd_iommu *iommu;
2752
2753	for_each_iommu(iommu) {
2754		iommu_enable_ppr_log(iommu);
2755		iommu_enable_gt(iommu);
2756	}
2757}
2758
2759static void enable_iommus_vapic(void)
2760{
2761#ifdef CONFIG_IRQ_REMAP
2762	u32 status, i;
2763	struct amd_iommu *iommu;
2764
2765	for_each_iommu(iommu) {
2766		/*
2767		 * Disable GALog if already running. It could have been enabled
2768		 * in the previous boot before kdump.
2769		 */
2770		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2771		if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2772			continue;
2773
2774		iommu_feature_disable(iommu, CONTROL_GALOG_EN);
2775		iommu_feature_disable(iommu, CONTROL_GAINT_EN);
2776
2777		/*
2778		 * Need to set and poll check the GALOGRun bit to zero before
2779		 * we can set/ modify GA Log registers safely.
2780		 */
2781		for (i = 0; i < LOOP_TIMEOUT; ++i) {
2782			status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2783			if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2784				break;
2785			udelay(10);
2786		}
2787
2788		if (WARN_ON(i >= LOOP_TIMEOUT))
2789			return;
2790	}
2791
2792	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2793	    !check_feature_on_all_iommus(FEATURE_GAM_VAPIC)) {
2794		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2795		return;
2796	}
2797
2798	if (amd_iommu_snp_en &&
2799	    !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) {
2800		pr_warn("Force to disable Virtual APIC due to SNP\n");
2801		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2802		return;
2803	}
2804
2805	/* Enabling GAM and SNPAVIC support */
2806	for_each_iommu(iommu) {
2807		if (iommu_init_ga_log(iommu) ||
2808		    iommu_ga_log_enable(iommu))
2809			return;
2810
2811		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2812		if (amd_iommu_snp_en)
2813			iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN);
2814	}
2815
2816	amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2817	pr_info("Virtual APIC enabled\n");
2818#endif
2819}
2820
2821static void enable_iommus(void)
2822{
2823	early_enable_iommus();
2824	enable_iommus_vapic();
2825	enable_iommus_v2();
2826}
2827
2828static void disable_iommus(void)
2829{
2830	struct amd_iommu *iommu;
2831
2832	for_each_iommu(iommu)
2833		iommu_disable(iommu);
2834
2835#ifdef CONFIG_IRQ_REMAP
2836	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2837		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2838#endif
2839}
2840
2841/*
2842 * Suspend/Resume support
2843 * disable suspend until real resume implemented
2844 */
2845
2846static void amd_iommu_resume(void)
2847{
2848	struct amd_iommu *iommu;
2849
2850	for_each_iommu(iommu)
2851		iommu_apply_resume_quirks(iommu);
2852
2853	/* re-load the hardware */
2854	enable_iommus();
2855
2856	amd_iommu_enable_interrupts();
2857}
2858
2859static int amd_iommu_suspend(void)
2860{
2861	/* disable IOMMUs to go out of the way for BIOS */
2862	disable_iommus();
2863
2864	return 0;
2865}
2866
2867static struct syscore_ops amd_iommu_syscore_ops = {
2868	.suspend = amd_iommu_suspend,
2869	.resume = amd_iommu_resume,
2870};
2871
2872static void __init free_iommu_resources(void)
2873{
 
 
 
 
 
2874	kmem_cache_destroy(amd_iommu_irq_cache);
2875	amd_iommu_irq_cache = NULL;
2876
 
 
 
 
 
 
 
 
 
 
 
 
2877	free_iommu_all();
2878	free_pci_segments();
2879}
2880
2881/* SB IOAPIC is always on this device in AMD systems */
2882#define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2883
2884static bool __init check_ioapic_information(void)
2885{
2886	const char *fw_bug = FW_BUG;
2887	bool ret, has_sb_ioapic;
2888	int idx;
2889
2890	has_sb_ioapic = false;
2891	ret           = false;
2892
2893	/*
2894	 * If we have map overrides on the kernel command line the
2895	 * messages in this function might not describe firmware bugs
2896	 * anymore - so be careful
2897	 */
2898	if (cmdline_maps)
2899		fw_bug = "";
2900
2901	for (idx = 0; idx < nr_ioapics; idx++) {
2902		int devid, id = mpc_ioapic_id(idx);
2903
2904		devid = get_ioapic_devid(id);
2905		if (devid < 0) {
2906			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2907				fw_bug, id);
2908			ret = false;
2909		} else if (devid == IOAPIC_SB_DEVID) {
2910			has_sb_ioapic = true;
2911			ret           = true;
2912		}
2913	}
2914
2915	if (!has_sb_ioapic) {
2916		/*
2917		 * We expect the SB IOAPIC to be listed in the IVRS
2918		 * table. The system timer is connected to the SB IOAPIC
2919		 * and if we don't have it in the list the system will
2920		 * panic at boot time.  This situation usually happens
2921		 * when the BIOS is buggy and provides us the wrong
2922		 * device id for the IOAPIC in the system.
2923		 */
2924		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2925	}
2926
2927	if (!ret)
2928		pr_err("Disabling interrupt remapping\n");
2929
2930	return ret;
2931}
2932
2933static void __init free_dma_resources(void)
2934{
2935	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2936		   get_order(MAX_DOMAIN_ID/8));
2937	amd_iommu_pd_alloc_bitmap = NULL;
2938
2939	free_unity_maps();
2940}
2941
2942static void __init ivinfo_init(void *ivrs)
2943{
2944	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
2945}
2946
2947/*
2948 * This is the hardware init function for AMD IOMMU in the system.
2949 * This function is called either from amd_iommu_init or from the interrupt
2950 * remapping setup code.
2951 *
2952 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2953 * four times:
2954 *
2955 *	1 pass) Discover the most comprehensive IVHD type to use.
2956 *
2957 *	2 pass) Find the highest PCI device id the driver has to handle.
2958 *		Upon this information the size of the data structures is
2959 *		determined that needs to be allocated.
2960 *
2961 *	3 pass) Initialize the data structures just allocated with the
2962 *		information in the ACPI table about available AMD IOMMUs
2963 *		in the system. It also maps the PCI devices in the
2964 *		system to specific IOMMUs
2965 *
2966 *	4 pass) After the basic data structures are allocated and
2967 *		initialized we update them with information about memory
2968 *		remapping requirements parsed out of the ACPI table in
2969 *		this last pass.
2970 *
2971 * After everything is set up the IOMMUs are enabled and the necessary
2972 * hotplug and suspend notifiers are registered.
2973 */
2974static int __init early_amd_iommu_init(void)
2975{
2976	struct acpi_table_header *ivrs_base;
2977	int remap_cache_sz, ret;
2978	acpi_status status;
 
 
2979
2980	if (!amd_iommu_detected)
2981		return -ENODEV;
2982
2983	status = acpi_get_table("IVRS", 0, &ivrs_base);
2984	if (status == AE_NOT_FOUND)
2985		return -ENODEV;
2986	else if (ACPI_FAILURE(status)) {
2987		const char *err = acpi_format_exception(status);
2988		pr_err("IVRS table error: %s\n", err);
2989		return -EINVAL;
2990	}
2991
2992	/*
2993	 * Validate checksum here so we don't need to do it when
2994	 * we actually parse the table
2995	 */
2996	ret = check_ivrs_checksum(ivrs_base);
2997	if (ret)
2998		goto out;
2999
3000	ivinfo_init(ivrs_base);
3001
3002	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
3003	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
3004
 
 
 
 
 
 
 
 
 
 
 
 
 
3005	/* Device table - directly used by all IOMMUs */
3006	ret = -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3007
3008	amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
3009					    GFP_KERNEL | __GFP_ZERO,
3010					    get_order(MAX_DOMAIN_ID/8));
3011	if (amd_iommu_pd_alloc_bitmap == NULL)
3012		goto out;
3013
3014	/*
 
 
 
 
 
 
3015	 * never allocate domain 0 because its used as the non-allocated and
3016	 * error value placeholder
3017	 */
3018	__set_bit(0, amd_iommu_pd_alloc_bitmap);
3019
3020	/*
3021	 * now the data structures are allocated and basically initialized
3022	 * start the real acpi table scan
3023	 */
3024	ret = init_iommu_all(ivrs_base);
3025	if (ret)
3026		goto out;
3027
 
 
 
 
 
 
 
 
 
 
3028	/* Disable any previously enabled IOMMUs */
3029	if (!is_kdump_kernel() || amd_iommu_disabled)
3030		disable_iommus();
3031
3032	if (amd_iommu_irq_remap)
3033		amd_iommu_irq_remap = check_ioapic_information();
3034
3035	if (amd_iommu_irq_remap) {
3036		struct amd_iommu_pci_seg *pci_seg;
3037		/*
3038		 * Interrupt remapping enabled, create kmem_cache for the
3039		 * remapping tables.
3040		 */
3041		ret = -ENOMEM;
3042		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
3043			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
3044		else
3045			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
3046		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
3047							remap_cache_sz,
3048							DTE_INTTAB_ALIGNMENT,
3049							0, NULL);
3050		if (!amd_iommu_irq_cache)
3051			goto out;
3052
3053		for_each_pci_segment(pci_seg) {
3054			if (alloc_irq_lookup_table(pci_seg))
3055				goto out;
3056		}
 
 
 
3057	}
3058
3059	ret = init_memory_definitions(ivrs_base);
3060	if (ret)
3061		goto out;
3062
3063	/* init the device table */
3064	init_device_table();
3065
3066out:
3067	/* Don't leak any ACPI memory */
3068	acpi_put_table(ivrs_base);
 
3069
3070	return ret;
3071}
3072
3073static int amd_iommu_enable_interrupts(void)
3074{
3075	struct amd_iommu *iommu;
3076	int ret = 0;
3077
3078	for_each_iommu(iommu) {
3079		ret = iommu_init_irq(iommu);
3080		if (ret)
3081			goto out;
3082	}
3083
3084out:
3085	return ret;
3086}
3087
3088static bool __init detect_ivrs(void)
3089{
3090	struct acpi_table_header *ivrs_base;
3091	acpi_status status;
3092	int i;
3093
3094	status = acpi_get_table("IVRS", 0, &ivrs_base);
3095	if (status == AE_NOT_FOUND)
3096		return false;
3097	else if (ACPI_FAILURE(status)) {
3098		const char *err = acpi_format_exception(status);
3099		pr_err("IVRS table error: %s\n", err);
3100		return false;
3101	}
3102
3103	acpi_put_table(ivrs_base);
3104
3105	if (amd_iommu_force_enable)
3106		goto out;
3107
3108	/* Don't use IOMMU if there is Stoney Ridge graphics */
3109	for (i = 0; i < 32; i++) {
3110		u32 pci_id;
3111
3112		pci_id = read_pci_config(0, i, 0, 0);
3113		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3114			pr_info("Disable IOMMU on Stoney Ridge\n");
3115			return false;
3116		}
3117	}
3118
3119out:
3120	/* Make sure ACS will be enabled during PCI probe */
3121	pci_request_acs();
3122
3123	return true;
3124}
3125
3126/****************************************************************************
3127 *
3128 * AMD IOMMU Initialization State Machine
3129 *
3130 ****************************************************************************/
3131
3132static int __init state_next(void)
3133{
3134	int ret = 0;
3135
3136	switch (init_state) {
3137	case IOMMU_START_STATE:
3138		if (!detect_ivrs()) {
3139			init_state	= IOMMU_NOT_FOUND;
3140			ret		= -ENODEV;
3141		} else {
3142			init_state	= IOMMU_IVRS_DETECTED;
3143		}
3144		break;
3145	case IOMMU_IVRS_DETECTED:
3146		if (amd_iommu_disabled) {
 
 
 
3147			init_state = IOMMU_CMDLINE_DISABLED;
3148			ret = -EINVAL;
3149		} else {
3150			ret = early_amd_iommu_init();
3151			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3152		}
3153		break;
3154	case IOMMU_ACPI_FINISHED:
3155		early_enable_iommus();
3156		x86_platform.iommu_shutdown = disable_iommus;
3157		init_state = IOMMU_ENABLED;
3158		break;
3159	case IOMMU_ENABLED:
3160		register_syscore_ops(&amd_iommu_syscore_ops);
3161		ret = amd_iommu_init_pci();
3162		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3163		enable_iommus_vapic();
3164		enable_iommus_v2();
3165		break;
3166	case IOMMU_PCI_INIT:
3167		ret = amd_iommu_enable_interrupts();
3168		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3169		break;
3170	case IOMMU_INTERRUPTS_EN:
 
 
 
 
3171		init_state = IOMMU_INITIALIZED;
3172		break;
3173	case IOMMU_INITIALIZED:
3174		/* Nothing to do */
3175		break;
3176	case IOMMU_NOT_FOUND:
3177	case IOMMU_INIT_ERROR:
3178	case IOMMU_CMDLINE_DISABLED:
3179		/* Error states => do nothing */
3180		ret = -EINVAL;
3181		break;
3182	default:
3183		/* Unknown state */
3184		BUG();
3185	}
3186
3187	if (ret) {
3188		free_dma_resources();
3189		if (!irq_remapping_enabled) {
3190			disable_iommus();
3191			free_iommu_resources();
3192		} else {
3193			struct amd_iommu *iommu;
3194			struct amd_iommu_pci_seg *pci_seg;
3195
3196			for_each_pci_segment(pci_seg)
3197				uninit_device_table_dma(pci_seg);
3198
 
3199			for_each_iommu(iommu)
3200				iommu_flush_all_caches(iommu);
3201		}
3202	}
3203	return ret;
3204}
3205
3206static int __init iommu_go_to_state(enum iommu_init_state state)
3207{
3208	int ret = -EINVAL;
3209
3210	while (init_state != state) {
3211		if (init_state == IOMMU_NOT_FOUND         ||
3212		    init_state == IOMMU_INIT_ERROR        ||
3213		    init_state == IOMMU_CMDLINE_DISABLED)
3214			break;
3215		ret = state_next();
3216	}
3217
3218	return ret;
3219}
3220
3221#ifdef CONFIG_IRQ_REMAP
3222int __init amd_iommu_prepare(void)
3223{
3224	int ret;
3225
3226	amd_iommu_irq_remap = true;
3227
3228	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3229	if (ret) {
3230		amd_iommu_irq_remap = false;
3231		return ret;
3232	}
3233
3234	return amd_iommu_irq_remap ? 0 : -ENODEV;
3235}
3236
3237int __init amd_iommu_enable(void)
3238{
3239	int ret;
3240
3241	ret = iommu_go_to_state(IOMMU_ENABLED);
3242	if (ret)
3243		return ret;
3244
3245	irq_remapping_enabled = 1;
3246	return amd_iommu_xt_mode;
3247}
3248
3249void amd_iommu_disable(void)
3250{
3251	amd_iommu_suspend();
3252}
3253
3254int amd_iommu_reenable(int mode)
3255{
3256	amd_iommu_resume();
3257
3258	return 0;
3259}
3260
3261int __init amd_iommu_enable_faulting(void)
3262{
3263	/* We enable MSI later when PCI is initialized */
3264	return 0;
3265}
3266#endif
3267
3268/*
3269 * This is the core init function for AMD IOMMU hardware in the system.
3270 * This function is called from the generic x86 DMA layer initialization
3271 * code.
3272 */
3273static int __init amd_iommu_init(void)
3274{
3275	struct amd_iommu *iommu;
3276	int ret;
3277
3278	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3279#ifdef CONFIG_GART_IOMMU
3280	if (ret && list_empty(&amd_iommu_list)) {
3281		/*
3282		 * We failed to initialize the AMD IOMMU - try fallback
3283		 * to GART if possible.
3284		 */
3285		gart_iommu_init();
3286	}
3287#endif
3288
3289	for_each_iommu(iommu)
3290		amd_iommu_debugfs_setup(iommu);
3291
3292	return ret;
3293}
3294
3295static bool amd_iommu_sme_check(void)
3296{
3297	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3298	    (boot_cpu_data.x86 != 0x17))
3299		return true;
3300
3301	/* For Fam17h, a specific level of support is required */
3302	if (boot_cpu_data.microcode >= 0x08001205)
3303		return true;
3304
3305	if ((boot_cpu_data.microcode >= 0x08001126) &&
3306	    (boot_cpu_data.microcode <= 0x080011ff))
3307		return true;
3308
3309	pr_notice("IOMMU not currently supported when SME is active\n");
3310
3311	return false;
3312}
3313
3314/****************************************************************************
3315 *
3316 * Early detect code. This code runs at IOMMU detection time in the DMA
3317 * layer. It just looks if there is an IVRS ACPI table to detect AMD
3318 * IOMMUs
3319 *
3320 ****************************************************************************/
3321int __init amd_iommu_detect(void)
3322{
3323	int ret;
3324
3325	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3326		return -ENODEV;
3327
3328	if (!amd_iommu_sme_check())
3329		return -ENODEV;
3330
3331	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3332	if (ret)
3333		return ret;
3334
3335	amd_iommu_detected = true;
3336	iommu_detected = 1;
3337	x86_init.iommu.iommu_init = amd_iommu_init;
3338
3339	return 1;
3340}
3341
3342/****************************************************************************
3343 *
3344 * Parsing functions for the AMD IOMMU specific kernel command line
3345 * options.
3346 *
3347 ****************************************************************************/
3348
3349static int __init parse_amd_iommu_dump(char *str)
3350{
3351	amd_iommu_dump = true;
3352
3353	return 1;
3354}
3355
3356static int __init parse_amd_iommu_intr(char *str)
3357{
3358	for (; *str; ++str) {
3359		if (strncmp(str, "legacy", 6) == 0) {
3360			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3361			break;
3362		}
3363		if (strncmp(str, "vapic", 5) == 0) {
3364			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3365			break;
3366		}
3367	}
3368	return 1;
3369}
3370
3371static int __init parse_amd_iommu_options(char *str)
3372{
3373	if (!str)
3374		return -EINVAL;
3375
3376	while (*str) {
3377		if (strncmp(str, "fullflush", 9) == 0) {
3378			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3379			iommu_set_dma_strict();
3380		} else if (strncmp(str, "force_enable", 12) == 0) {
3381			amd_iommu_force_enable = true;
3382		} else if (strncmp(str, "off", 3) == 0) {
3383			amd_iommu_disabled = true;
3384		} else if (strncmp(str, "force_isolation", 15) == 0) {
3385			amd_iommu_force_isolation = true;
3386		} else if (strncmp(str, "pgtbl_v1", 8) == 0) {
3387			amd_iommu_pgtable = AMD_IOMMU_V1;
3388		} else if (strncmp(str, "pgtbl_v2", 8) == 0) {
3389			amd_iommu_pgtable = AMD_IOMMU_V2;
3390		} else {
3391			pr_notice("Unknown option - '%s'\n", str);
3392		}
3393
3394		str += strcspn(str, ",");
3395		while (*str == ',')
3396			str++;
3397	}
3398
3399	return 1;
3400}
3401
3402static int __init parse_ivrs_ioapic(char *str)
3403{
3404	u32 seg = 0, bus, dev, fn;
3405	int id, i;
3406	u32 devid;
 
 
3407
3408	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3409	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3410		goto found;
3411
3412	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3413	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3414		pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n",
3415			str, id, seg, bus, dev, fn);
3416		goto found;
3417	}
3418
3419	pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3420	return 1;
3421
3422found:
3423	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3424		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3425			str);
3426		return 1;
3427	}
3428
3429	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3430
3431	cmdline_maps			= true;
3432	i				= early_ioapic_map_size++;
3433	early_ioapic_map[i].id		= id;
3434	early_ioapic_map[i].devid	= devid;
3435	early_ioapic_map[i].cmd_line	= true;
3436
3437	return 1;
3438}
3439
3440static int __init parse_ivrs_hpet(char *str)
3441{
3442	u32 seg = 0, bus, dev, fn;
3443	int id, i;
3444	u32 devid;
 
 
3445
3446	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3447	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3448		goto found;
3449
3450	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3451	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3452		pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n",
3453			str, id, seg, bus, dev, fn);
3454		goto found;
3455	}
3456
3457	pr_err("Invalid command line: ivrs_hpet%s\n", str);
3458	return 1;
3459
3460found:
3461	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3462		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3463			str);
3464		return 1;
3465	}
3466
3467	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3468
3469	cmdline_maps			= true;
3470	i				= early_hpet_map_size++;
3471	early_hpet_map[i].id		= id;
3472	early_hpet_map[i].devid		= devid;
3473	early_hpet_map[i].cmd_line	= true;
3474
3475	return 1;
3476}
3477
3478static int __init parse_ivrs_acpihid(char *str)
3479{
3480	u32 seg = 0, bus, dev, fn;
3481	char *hid, *uid, *p, *addr;
3482	char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3483	int i;
3484
3485	addr = strchr(str, '@');
3486	if (!addr) {
3487		if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 ||
3488		    sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) {
3489			pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n",
3490				str, acpiid, seg, bus, dev, fn);
3491			goto found;
3492		}
3493		goto not_found;
3494	}
3495
3496	/* We have the '@', make it the terminator to get just the acpiid */
3497	*addr++ = 0;
3498
3499	if (sscanf(str, "=%s", acpiid) != 1)
3500		goto not_found;
3501
3502	if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 ||
3503	    sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4)
3504		goto found;
3505
3506not_found:
3507	pr_err("Invalid command line: ivrs_acpihid%s\n", str);
3508	return 1;
3509
3510found:
3511	p = acpiid;
3512	hid = strsep(&p, ":");
3513	uid = p;
3514
3515	if (!hid || !(*hid) || !uid) {
3516		pr_err("Invalid command line: hid or uid\n");
3517		return 1;
3518	}
3519
3520	/*
3521	 * Ignore leading zeroes after ':', so e.g., AMDI0095:00
3522	 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match
3523	 */
3524	while (*uid == '0' && *(uid + 1))
3525		uid++;
3526
3527	i = early_acpihid_map_size++;
3528	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3529	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3530	early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
 
3531	early_acpihid_map[i].cmd_line	= true;
3532
3533	return 1;
3534}
3535
3536__setup("amd_iommu_dump",	parse_amd_iommu_dump);
3537__setup("amd_iommu=",		parse_amd_iommu_options);
3538__setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3539__setup("ivrs_ioapic",		parse_ivrs_ioapic);
3540__setup("ivrs_hpet",		parse_ivrs_hpet);
3541__setup("ivrs_acpihid",		parse_ivrs_acpihid);
3542
 
 
 
 
 
3543bool amd_iommu_v2_supported(void)
3544{
3545	/*
3546	 * Since DTE[Mode]=0 is prohibited on SNP-enabled system
3547	 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without
3548	 * setting up IOMMUv1 page table.
3549	 */
3550	return amd_iommu_v2_present && !amd_iommu_snp_en;
3551}
3552EXPORT_SYMBOL(amd_iommu_v2_supported);
3553
3554struct amd_iommu *get_amd_iommu(unsigned int idx)
3555{
3556	unsigned int i = 0;
3557	struct amd_iommu *iommu;
3558
3559	for_each_iommu(iommu)
3560		if (i++ == idx)
3561			return iommu;
3562	return NULL;
3563}
 
3564
3565/****************************************************************************
3566 *
3567 * IOMMU EFR Performance Counter support functionality. This code allows
3568 * access to the IOMMU PC functionality.
3569 *
3570 ****************************************************************************/
3571
3572u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3573{
3574	struct amd_iommu *iommu = get_amd_iommu(idx);
3575
3576	if (iommu)
3577		return iommu->max_banks;
3578
3579	return 0;
3580}
3581EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3582
3583bool amd_iommu_pc_supported(void)
3584{
3585	return amd_iommu_pc_present;
3586}
3587EXPORT_SYMBOL(amd_iommu_pc_supported);
3588
3589u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3590{
3591	struct amd_iommu *iommu = get_amd_iommu(idx);
3592
3593	if (iommu)
3594		return iommu->max_counters;
3595
3596	return 0;
3597}
3598EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3599
3600static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3601				u8 fxn, u64 *value, bool is_write)
3602{
3603	u32 offset;
3604	u32 max_offset_lim;
3605
3606	/* Make sure the IOMMU PC resource is available */
3607	if (!amd_iommu_pc_present)
3608		return -ENODEV;
3609
3610	/* Check for valid iommu and pc register indexing */
3611	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3612		return -ENODEV;
3613
3614	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3615
3616	/* Limit the offset to the hw defined mmio region aperture */
3617	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3618				(iommu->max_counters << 8) | 0x28);
3619	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3620	    (offset > max_offset_lim))
3621		return -EINVAL;
3622
3623	if (is_write) {
3624		u64 val = *value & GENMASK_ULL(47, 0);
3625
3626		writel((u32)val, iommu->mmio_base + offset);
3627		writel((val >> 32), iommu->mmio_base + offset + 4);
3628	} else {
3629		*value = readl(iommu->mmio_base + offset + 4);
3630		*value <<= 32;
3631		*value |= readl(iommu->mmio_base + offset);
3632		*value &= GENMASK_ULL(47, 0);
3633	}
3634
3635	return 0;
3636}
3637
3638int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3639{
3640	if (!iommu)
3641		return -EINVAL;
3642
3643	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3644}
 
3645
3646int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3647{
3648	if (!iommu)
3649		return -EINVAL;
3650
3651	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3652}
3653
3654#ifdef CONFIG_AMD_MEM_ENCRYPT
3655int amd_iommu_snp_enable(void)
3656{
3657	/*
3658	 * The SNP support requires that IOMMU must be enabled, and is
3659	 * not configured in the passthrough mode.
3660	 */
3661	if (no_iommu || iommu_default_passthrough()) {
3662		pr_err("SNP: IOMMU is disabled or configured in passthrough mode, SNP cannot be supported");
3663		return -EINVAL;
3664	}
3665
3666	/*
3667	 * Prevent enabling SNP after IOMMU_ENABLED state because this process
3668	 * affect how IOMMU driver sets up data structures and configures
3669	 * IOMMU hardware.
3670	 */
3671	if (init_state > IOMMU_ENABLED) {
3672		pr_err("SNP: Too late to enable SNP for IOMMU.\n");
3673		return -EINVAL;
3674	}
3675
3676	amd_iommu_snp_en = check_feature_on_all_iommus(FEATURE_SNP);
3677	if (!amd_iommu_snp_en)
3678		return -EINVAL;
3679
3680	pr_info("SNP enabled\n");
3681
3682	/* Enforce IOMMU v1 pagetable when SNP is enabled. */
3683	if (amd_iommu_pgtable != AMD_IOMMU_V1) {
3684		pr_warn("Force to using AMD IOMMU v1 page table due to SNP\n");
3685		amd_iommu_pgtable = AMD_IOMMU_V1;
3686	}
3687
3688	return 0;
3689}
3690#endif