xref: /openbmc/linux/drivers/iommu/amd/init.c (revision 307959008d80cdfd67fb19a146932582bb68a399)
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  * ACPI table definitions
89  *
90  * These data structures are laid over the table to parse the important values
91  * out of it.
92  */
93 
94 extern const struct iommu_ops amd_iommu_ops;
95 
96 /*
97  * structure describing one IOMMU in the ACPI table. Typically followed by one
98  * or more ivhd_entrys.
99  */
100 struct ivhd_header {
101 	u8 type;
102 	u8 flags;
103 	u16 length;
104 	u16 devid;
105 	u16 cap_ptr;
106 	u64 mmio_phys;
107 	u16 pci_seg;
108 	u16 info;
109 	u32 efr_attr;
110 
111 	/* Following only valid on IVHD type 11h and 40h */
112 	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
113 	u64 res;
114 } __attribute__((packed));
115 
116 /*
117  * A device entry describing which devices a specific IOMMU translates and
118  * which requestor ids they use.
119  */
120 struct ivhd_entry {
121 	u8 type;
122 	u16 devid;
123 	u8 flags;
124 	struct_group(ext_hid,
125 		u32 ext;
126 		u32 hidh;
127 	);
128 	u64 cid;
129 	u8 uidf;
130 	u8 uidl;
131 	u8 uid;
132 } __attribute__((packed));
133 
134 /*
135  * An AMD IOMMU memory definition structure. It defines things like exclusion
136  * ranges for devices and regions that should be unity mapped.
137  */
138 struct ivmd_header {
139 	u8 type;
140 	u8 flags;
141 	u16 length;
142 	u16 devid;
143 	u16 aux;
144 	u16 pci_seg;
145 	u8  resv[6];
146 	u64 range_start;
147 	u64 range_length;
148 } __attribute__((packed));
149 
150 bool amd_iommu_dump;
151 bool amd_iommu_irq_remap __read_mostly;
152 
153 enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1;
154 
155 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
156 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
157 
158 static bool amd_iommu_detected;
159 static bool amd_iommu_disabled __initdata;
160 static bool amd_iommu_force_enable __initdata;
161 static int amd_iommu_target_ivhd_type;
162 
163 u16 amd_iommu_last_bdf;			/* largest PCI device id we have
164 					   to handle */
165 
166 LIST_HEAD(amd_iommu_pci_seg_list);	/* list of all PCI segments */
167 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
168 					   system */
169 
170 /* Array to assign indices to IOMMUs*/
171 struct amd_iommu *amd_iommus[MAX_IOMMUS];
172 
173 /* Number of IOMMUs present in the system */
174 static int amd_iommus_present;
175 
176 /* IOMMUs have a non-present cache? */
177 bool amd_iommu_np_cache __read_mostly;
178 bool amd_iommu_iotlb_sup __read_mostly = true;
179 
180 u32 amd_iommu_max_pasid __read_mostly = ~0;
181 
182 bool amd_iommu_v2_present __read_mostly;
183 static bool amd_iommu_pc_present __read_mostly;
184 bool amdr_ivrs_remap_support __read_mostly;
185 
186 bool amd_iommu_force_isolation __read_mostly;
187 
188 /*
189  * Pointer to the device table which is shared by all AMD IOMMUs
190  * it is indexed by the PCI device id or the HT unit id and contains
191  * information about the domain the device belongs to as well as the
192  * page table root pointer.
193  */
194 struct dev_table_entry *amd_iommu_dev_table;
195 
196 /*
197  * The alias table is a driver specific data structure which contains the
198  * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
199  * More than one device can share the same requestor id.
200  */
201 u16 *amd_iommu_alias_table;
202 
203 /*
204  * The rlookup table is used to find the IOMMU which is responsible
205  * for a specific device. It is also indexed by the PCI device id.
206  */
207 struct amd_iommu **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  */
213 struct 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  */
219 unsigned long *amd_iommu_pd_alloc_bitmap;
220 
221 static u32 dev_table_size;	/* size of the device table */
222 static u32 alias_table_size;	/* size of the alias table */
223 static u32 rlookup_table_size;	/* size if the rlookup table */
224 
225 enum 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_INITIALIZED,
233 	IOMMU_NOT_FOUND,
234 	IOMMU_INIT_ERROR,
235 	IOMMU_CMDLINE_DISABLED,
236 };
237 
238 /* Early ioapic and hpet maps from kernel command line */
239 #define EARLY_MAP_SIZE		4
240 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
241 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
242 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
243 
244 static int __initdata early_ioapic_map_size;
245 static int __initdata early_hpet_map_size;
246 static int __initdata early_acpihid_map_size;
247 
248 static bool __initdata cmdline_maps;
249 
250 static enum iommu_init_state init_state = IOMMU_START_STATE;
251 
252 static int amd_iommu_enable_interrupts(void);
253 static int __init iommu_go_to_state(enum iommu_init_state state);
254 static void init_device_table_dma(void);
255 
256 static bool amd_iommu_pre_enabled = true;
257 
258 static u32 amd_iommu_ivinfo __initdata;
259 
260 bool translation_pre_enabled(struct amd_iommu *iommu)
261 {
262 	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
263 }
264 
265 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
266 {
267 	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
268 }
269 
270 static 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 
279 static inline void update_last_devid(u16 devid)
280 {
281 	if (devid > amd_iommu_last_bdf)
282 		amd_iommu_last_bdf = devid;
283 }
284 
285 static 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 
293 int amd_iommu_get_num_iommus(void)
294 {
295 	return amd_iommus_present;
296 }
297 
298 #ifdef CONFIG_IRQ_REMAP
299 static bool check_feature_on_all_iommus(u64 mask)
300 {
301 	bool ret = false;
302 	struct amd_iommu *iommu;
303 
304 	for_each_iommu(iommu) {
305 		ret = iommu_feature(iommu, mask);
306 		if (!ret)
307 			return false;
308 	}
309 
310 	return true;
311 }
312 #endif
313 
314 /*
315  * For IVHD type 0x11/0x40, EFR is also available via IVHD.
316  * Default to IVHD EFR since it is available sooner
317  * (i.e. before PCI init).
318  */
319 static void __init early_iommu_features_init(struct amd_iommu *iommu,
320 					     struct ivhd_header *h)
321 {
322 	if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP)
323 		iommu->features = h->efr_reg;
324 	if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
325 		amdr_ivrs_remap_support = true;
326 }
327 
328 /* Access to l1 and l2 indexed register spaces */
329 
330 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
331 {
332 	u32 val;
333 
334 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
335 	pci_read_config_dword(iommu->dev, 0xfc, &val);
336 	return val;
337 }
338 
339 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
340 {
341 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
342 	pci_write_config_dword(iommu->dev, 0xfc, val);
343 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
344 }
345 
346 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
347 {
348 	u32 val;
349 
350 	pci_write_config_dword(iommu->dev, 0xf0, address);
351 	pci_read_config_dword(iommu->dev, 0xf4, &val);
352 	return val;
353 }
354 
355 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
356 {
357 	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
358 	pci_write_config_dword(iommu->dev, 0xf4, val);
359 }
360 
361 /****************************************************************************
362  *
363  * AMD IOMMU MMIO register space handling functions
364  *
365  * These functions are used to program the IOMMU device registers in
366  * MMIO space required for that driver.
367  *
368  ****************************************************************************/
369 
370 /*
371  * This function set the exclusion range in the IOMMU. DMA accesses to the
372  * exclusion range are passed through untranslated
373  */
374 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
375 {
376 	u64 start = iommu->exclusion_start & PAGE_MASK;
377 	u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
378 	u64 entry;
379 
380 	if (!iommu->exclusion_start)
381 		return;
382 
383 	entry = start | MMIO_EXCL_ENABLE_MASK;
384 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
385 			&entry, sizeof(entry));
386 
387 	entry = limit;
388 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
389 			&entry, sizeof(entry));
390 }
391 
392 static void iommu_set_cwwb_range(struct amd_iommu *iommu)
393 {
394 	u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
395 	u64 entry = start & PM_ADDR_MASK;
396 
397 	if (!iommu_feature(iommu, FEATURE_SNP))
398 		return;
399 
400 	/* Note:
401 	 * Re-purpose Exclusion base/limit registers for Completion wait
402 	 * write-back base/limit.
403 	 */
404 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
405 		    &entry, sizeof(entry));
406 
407 	/* Note:
408 	 * Default to 4 Kbytes, which can be specified by setting base
409 	 * address equal to the limit address.
410 	 */
411 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
412 		    &entry, sizeof(entry));
413 }
414 
415 /* Programs the physical address of the device table into the IOMMU hardware */
416 static void iommu_set_device_table(struct amd_iommu *iommu)
417 {
418 	u64 entry;
419 
420 	BUG_ON(iommu->mmio_base == NULL);
421 
422 	entry = iommu_virt_to_phys(amd_iommu_dev_table);
423 	entry |= (dev_table_size >> 12) - 1;
424 	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
425 			&entry, sizeof(entry));
426 }
427 
428 /* Generic functions to enable/disable certain features of the IOMMU. */
429 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
430 {
431 	u64 ctrl;
432 
433 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
434 	ctrl |= (1ULL << bit);
435 	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
436 }
437 
438 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
439 {
440 	u64 ctrl;
441 
442 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
443 	ctrl &= ~(1ULL << bit);
444 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
445 }
446 
447 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
448 {
449 	u64 ctrl;
450 
451 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
452 	ctrl &= ~CTRL_INV_TO_MASK;
453 	ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
454 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
455 }
456 
457 /* Function to enable the hardware */
458 static void iommu_enable(struct amd_iommu *iommu)
459 {
460 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
461 }
462 
463 static void iommu_disable(struct amd_iommu *iommu)
464 {
465 	if (!iommu->mmio_base)
466 		return;
467 
468 	/* Disable command buffer */
469 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
470 
471 	/* Disable event logging and event interrupts */
472 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
473 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
474 
475 	/* Disable IOMMU GA_LOG */
476 	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
477 	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
478 
479 	/* Disable IOMMU hardware itself */
480 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
481 }
482 
483 /*
484  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
485  * the system has one.
486  */
487 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
488 {
489 	if (!request_mem_region(address, end, "amd_iommu")) {
490 		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
491 			address, end);
492 		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
493 		return NULL;
494 	}
495 
496 	return (u8 __iomem *)ioremap(address, end);
497 }
498 
499 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
500 {
501 	if (iommu->mmio_base)
502 		iounmap(iommu->mmio_base);
503 	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
504 }
505 
506 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
507 {
508 	u32 size = 0;
509 
510 	switch (h->type) {
511 	case 0x10:
512 		size = 24;
513 		break;
514 	case 0x11:
515 	case 0x40:
516 		size = 40;
517 		break;
518 	}
519 	return size;
520 }
521 
522 /****************************************************************************
523  *
524  * The functions below belong to the first pass of AMD IOMMU ACPI table
525  * parsing. In this pass we try to find out the highest device id this
526  * code has to handle. Upon this information the size of the shared data
527  * structures is determined later.
528  *
529  ****************************************************************************/
530 
531 /*
532  * This function calculates the length of a given IVHD entry
533  */
534 static inline int ivhd_entry_length(u8 *ivhd)
535 {
536 	u32 type = ((struct ivhd_entry *)ivhd)->type;
537 
538 	if (type < 0x80) {
539 		return 0x04 << (*ivhd >> 6);
540 	} else if (type == IVHD_DEV_ACPI_HID) {
541 		/* For ACPI_HID, offset 21 is uid len */
542 		return *((u8 *)ivhd + 21) + 22;
543 	}
544 	return 0;
545 }
546 
547 /*
548  * After reading the highest device id from the IOMMU PCI capability header
549  * this function looks if there is a higher device id defined in the ACPI table
550  */
551 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
552 {
553 	u8 *p = (void *)h, *end = (void *)h;
554 	struct ivhd_entry *dev;
555 	int last_devid = -EINVAL;
556 
557 	u32 ivhd_size = get_ivhd_header_size(h);
558 
559 	if (!ivhd_size) {
560 		pr_err("Unsupported IVHD type %#x\n", h->type);
561 		return -EINVAL;
562 	}
563 
564 	p += ivhd_size;
565 	end += h->length;
566 
567 	while (p < end) {
568 		dev = (struct ivhd_entry *)p;
569 		switch (dev->type) {
570 		case IVHD_DEV_ALL:
571 			/* Use maximum BDF value for DEV_ALL */
572 			update_last_devid(0xffff);
573 			return 0xffff;
574 		case IVHD_DEV_SELECT:
575 		case IVHD_DEV_RANGE_END:
576 		case IVHD_DEV_ALIAS:
577 		case IVHD_DEV_EXT_SELECT:
578 			/* all the above subfield types refer to device ids */
579 			update_last_devid(dev->devid);
580 			if (dev->devid > last_devid)
581 				last_devid = dev->devid;
582 			break;
583 		default:
584 			break;
585 		}
586 		p += ivhd_entry_length(p);
587 	}
588 
589 	WARN_ON(p != end);
590 
591 	return last_devid;
592 }
593 
594 static int __init check_ivrs_checksum(struct acpi_table_header *table)
595 {
596 	int i;
597 	u8 checksum = 0, *p = (u8 *)table;
598 
599 	for (i = 0; i < table->length; ++i)
600 		checksum += p[i];
601 	if (checksum != 0) {
602 		/* ACPI table corrupt */
603 		pr_err(FW_BUG "IVRS invalid checksum\n");
604 		return -ENODEV;
605 	}
606 
607 	return 0;
608 }
609 
610 /*
611  * Iterate over all IVHD entries in the ACPI table and find the highest device
612  * id which we need to handle. This is the first of three functions which parse
613  * the ACPI table. So we check the checksum here.
614  */
615 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
616 {
617 	u8 *p = (u8 *)table, *end = (u8 *)table;
618 	struct ivhd_header *h;
619 	int last_devid, last_bdf = 0;
620 
621 	p += IVRS_HEADER_LENGTH;
622 
623 	end += table->length;
624 	while (p < end) {
625 		h = (struct ivhd_header *)p;
626 		if (h->pci_seg == pci_seg &&
627 		    h->type == amd_iommu_target_ivhd_type) {
628 			last_devid = find_last_devid_from_ivhd(h);
629 
630 			if (last_devid < 0)
631 				return -EINVAL;
632 			if (last_devid > last_bdf)
633 				last_bdf = last_devid;
634 		}
635 		p += h->length;
636 	}
637 	WARN_ON(p != end);
638 
639 	return last_bdf;
640 }
641 
642 /****************************************************************************
643  *
644  * The following functions belong to the code path which parses the ACPI table
645  * the second time. In this ACPI parsing iteration we allocate IOMMU specific
646  * data structures, initialize the per PCI segment device/alias/rlookup table
647  * and also basically initialize the hardware.
648  *
649  ****************************************************************************/
650 
651 /* Allocate per PCI segment device table */
652 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
653 {
654 	pci_seg->dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
655 						      get_order(dev_table_size));
656 	if (!pci_seg->dev_table)
657 		return -ENOMEM;
658 
659 	return 0;
660 }
661 
662 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
663 {
664 	free_pages((unsigned long)pci_seg->dev_table,
665 		    get_order(dev_table_size));
666 	pci_seg->dev_table = NULL;
667 }
668 
669 /* Allocate per PCI segment IOMMU rlookup table. */
670 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
671 {
672 	pci_seg->rlookup_table = (void *)__get_free_pages(
673 						GFP_KERNEL | __GFP_ZERO,
674 						get_order(rlookup_table_size));
675 	if (pci_seg->rlookup_table == NULL)
676 		return -ENOMEM;
677 
678 	return 0;
679 }
680 
681 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
682 {
683 	free_pages((unsigned long)pci_seg->rlookup_table,
684 		   get_order(rlookup_table_size));
685 	pci_seg->rlookup_table = NULL;
686 }
687 
688 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
689 {
690 	pci_seg->irq_lookup_table = (void *)__get_free_pages(
691 					     GFP_KERNEL | __GFP_ZERO,
692 					     get_order(rlookup_table_size));
693 	kmemleak_alloc(pci_seg->irq_lookup_table,
694 		       rlookup_table_size, 1, GFP_KERNEL);
695 	if (pci_seg->irq_lookup_table == NULL)
696 		return -ENOMEM;
697 
698 	return 0;
699 }
700 
701 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
702 {
703 	kmemleak_free(pci_seg->irq_lookup_table);
704 	free_pages((unsigned long)pci_seg->irq_lookup_table,
705 		   get_order(rlookup_table_size));
706 	pci_seg->irq_lookup_table = NULL;
707 }
708 
709 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
710 {
711 	int i;
712 
713 	pci_seg->alias_table = (void *)__get_free_pages(GFP_KERNEL,
714 						    get_order(alias_table_size));
715 	if (!pci_seg->alias_table)
716 		return -ENOMEM;
717 
718 	/*
719 	 * let all alias entries point to itself
720 	 */
721 	for (i = 0; i <= amd_iommu_last_bdf; ++i)
722 		pci_seg->alias_table[i] = i;
723 
724 	return 0;
725 }
726 
727 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
728 {
729 	free_pages((unsigned long)pci_seg->alias_table,
730 		   get_order(alias_table_size));
731 	pci_seg->alias_table = NULL;
732 }
733 
734 /*
735  * Allocates the command buffer. This buffer is per AMD IOMMU. We can
736  * write commands to that buffer later and the IOMMU will execute them
737  * asynchronously
738  */
739 static int __init alloc_command_buffer(struct amd_iommu *iommu)
740 {
741 	iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
742 						  get_order(CMD_BUFFER_SIZE));
743 
744 	return iommu->cmd_buf ? 0 : -ENOMEM;
745 }
746 
747 /*
748  * This function restarts event logging in case the IOMMU experienced
749  * an event log buffer overflow.
750  */
751 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
752 {
753 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
754 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
755 }
756 
757 /*
758  * This function resets the command buffer if the IOMMU stopped fetching
759  * commands from it.
760  */
761 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
762 {
763 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
764 
765 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
766 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
767 	iommu->cmd_buf_head = 0;
768 	iommu->cmd_buf_tail = 0;
769 
770 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
771 }
772 
773 /*
774  * This function writes the command buffer address to the hardware and
775  * enables it.
776  */
777 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
778 {
779 	u64 entry;
780 
781 	BUG_ON(iommu->cmd_buf == NULL);
782 
783 	entry = iommu_virt_to_phys(iommu->cmd_buf);
784 	entry |= MMIO_CMD_SIZE_512;
785 
786 	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
787 		    &entry, sizeof(entry));
788 
789 	amd_iommu_reset_cmd_buffer(iommu);
790 }
791 
792 /*
793  * This function disables the command buffer
794  */
795 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
796 {
797 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
798 }
799 
800 static void __init free_command_buffer(struct amd_iommu *iommu)
801 {
802 	free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
803 }
804 
805 static void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu,
806 					 gfp_t gfp, size_t size)
807 {
808 	int order = get_order(size);
809 	void *buf = (void *)__get_free_pages(gfp, order);
810 
811 	if (buf &&
812 	    iommu_feature(iommu, FEATURE_SNP) &&
813 	    set_memory_4k((unsigned long)buf, (1 << order))) {
814 		free_pages((unsigned long)buf, order);
815 		buf = NULL;
816 	}
817 
818 	return buf;
819 }
820 
821 /* allocates the memory where the IOMMU will log its events to */
822 static int __init alloc_event_buffer(struct amd_iommu *iommu)
823 {
824 	iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
825 					      EVT_BUFFER_SIZE);
826 
827 	return iommu->evt_buf ? 0 : -ENOMEM;
828 }
829 
830 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
831 {
832 	u64 entry;
833 
834 	BUG_ON(iommu->evt_buf == NULL);
835 
836 	entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
837 
838 	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
839 		    &entry, sizeof(entry));
840 
841 	/* set head and tail to zero manually */
842 	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
843 	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
844 
845 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
846 }
847 
848 /*
849  * This function disables the event log buffer
850  */
851 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
852 {
853 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
854 }
855 
856 static void __init free_event_buffer(struct amd_iommu *iommu)
857 {
858 	free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
859 }
860 
861 /* allocates the memory where the IOMMU will log its events to */
862 static int __init alloc_ppr_log(struct amd_iommu *iommu)
863 {
864 	iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
865 					      PPR_LOG_SIZE);
866 
867 	return iommu->ppr_log ? 0 : -ENOMEM;
868 }
869 
870 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
871 {
872 	u64 entry;
873 
874 	if (iommu->ppr_log == NULL)
875 		return;
876 
877 	entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
878 
879 	memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
880 		    &entry, sizeof(entry));
881 
882 	/* set head and tail to zero manually */
883 	writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
884 	writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
885 
886 	iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
887 	iommu_feature_enable(iommu, CONTROL_PPR_EN);
888 }
889 
890 static void __init free_ppr_log(struct amd_iommu *iommu)
891 {
892 	free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
893 }
894 
895 static void free_ga_log(struct amd_iommu *iommu)
896 {
897 #ifdef CONFIG_IRQ_REMAP
898 	free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE));
899 	free_pages((unsigned long)iommu->ga_log_tail, get_order(8));
900 #endif
901 }
902 
903 static int iommu_ga_log_enable(struct amd_iommu *iommu)
904 {
905 #ifdef CONFIG_IRQ_REMAP
906 	u32 status, i;
907 	u64 entry;
908 
909 	if (!iommu->ga_log)
910 		return -EINVAL;
911 
912 	/* Check if already running */
913 	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
914 	if (WARN_ON(status & (MMIO_STATUS_GALOG_RUN_MASK)))
915 		return 0;
916 
917 	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
918 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
919 		    &entry, sizeof(entry));
920 	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
921 		 (BIT_ULL(52)-1)) & ~7ULL;
922 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
923 		    &entry, sizeof(entry));
924 	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
925 	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
926 
927 
928 	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
929 	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
930 
931 	for (i = 0; i < LOOP_TIMEOUT; ++i) {
932 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
933 		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
934 			break;
935 		udelay(10);
936 	}
937 
938 	if (WARN_ON(i >= LOOP_TIMEOUT))
939 		return -EINVAL;
940 #endif /* CONFIG_IRQ_REMAP */
941 	return 0;
942 }
943 
944 static int iommu_init_ga_log(struct amd_iommu *iommu)
945 {
946 #ifdef CONFIG_IRQ_REMAP
947 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
948 		return 0;
949 
950 	iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
951 					get_order(GA_LOG_SIZE));
952 	if (!iommu->ga_log)
953 		goto err_out;
954 
955 	iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
956 					get_order(8));
957 	if (!iommu->ga_log_tail)
958 		goto err_out;
959 
960 	return 0;
961 err_out:
962 	free_ga_log(iommu);
963 	return -EINVAL;
964 #else
965 	return 0;
966 #endif /* CONFIG_IRQ_REMAP */
967 }
968 
969 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
970 {
971 	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 1);
972 
973 	return iommu->cmd_sem ? 0 : -ENOMEM;
974 }
975 
976 static void __init free_cwwb_sem(struct amd_iommu *iommu)
977 {
978 	if (iommu->cmd_sem)
979 		free_page((unsigned long)iommu->cmd_sem);
980 }
981 
982 static void iommu_enable_xt(struct amd_iommu *iommu)
983 {
984 #ifdef CONFIG_IRQ_REMAP
985 	/*
986 	 * XT mode (32-bit APIC destination ID) requires
987 	 * GA mode (128-bit IRTE support) as a prerequisite.
988 	 */
989 	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
990 	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
991 		iommu_feature_enable(iommu, CONTROL_XT_EN);
992 #endif /* CONFIG_IRQ_REMAP */
993 }
994 
995 static void iommu_enable_gt(struct amd_iommu *iommu)
996 {
997 	if (!iommu_feature(iommu, FEATURE_GT))
998 		return;
999 
1000 	iommu_feature_enable(iommu, CONTROL_GT_EN);
1001 }
1002 
1003 /* sets a specific bit in the device table entry. */
1004 static void set_dev_entry_bit(u16 devid, u8 bit)
1005 {
1006 	int i = (bit >> 6) & 0x03;
1007 	int _bit = bit & 0x3f;
1008 
1009 	amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
1010 }
1011 
1012 static int get_dev_entry_bit(u16 devid, u8 bit)
1013 {
1014 	int i = (bit >> 6) & 0x03;
1015 	int _bit = bit & 0x3f;
1016 
1017 	return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1018 }
1019 
1020 
1021 static bool __copy_device_table(struct amd_iommu *iommu)
1022 {
1023 	u64 int_ctl, int_tab_len, entry = 0;
1024 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1025 	struct dev_table_entry *old_devtb = NULL;
1026 	u32 lo, hi, devid, old_devtb_size;
1027 	phys_addr_t old_devtb_phys;
1028 	u16 dom_id, dte_v, irq_v;
1029 	gfp_t gfp_flag;
1030 	u64 tmp;
1031 
1032 	/* Each IOMMU use separate device table with the same size */
1033 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1034 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1035 	entry = (((u64) hi) << 32) + lo;
1036 
1037 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1038 	if (old_devtb_size != dev_table_size) {
1039 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1040 			iommu->index);
1041 		return false;
1042 	}
1043 
1044 	/*
1045 	 * When SME is enabled in the first kernel, the entry includes the
1046 	 * memory encryption mask(sme_me_mask), we must remove the memory
1047 	 * encryption mask to obtain the true physical address in kdump kernel.
1048 	 */
1049 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1050 
1051 	if (old_devtb_phys >= 0x100000000ULL) {
1052 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1053 		return false;
1054 	}
1055 	old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1056 		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
1057 							dev_table_size)
1058 		    : memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
1059 
1060 	if (!old_devtb)
1061 		return false;
1062 
1063 	gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
1064 	pci_seg->old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
1065 						    get_order(dev_table_size));
1066 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1067 		pr_err("Failed to allocate memory for copying old device table!\n");
1068 		memunmap(old_devtb);
1069 		return false;
1070 	}
1071 
1072 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1073 		pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1074 		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1075 		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1076 
1077 		if (dte_v && dom_id) {
1078 			pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1079 			pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1080 			__set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
1081 			/* If gcr3 table existed, mask it out */
1082 			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1083 				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1084 				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1085 				pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1086 				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1087 				tmp |= DTE_FLAG_GV;
1088 				pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1089 			}
1090 		}
1091 
1092 		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1093 		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1094 		int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1095 		if (irq_v && (int_ctl || int_tab_len)) {
1096 			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1097 			    (int_tab_len != DTE_INTTABLEN)) {
1098 				pr_err("Wrong old irq remapping flag: %#x\n", devid);
1099 				memunmap(old_devtb);
1100 				return false;
1101 			}
1102 
1103 			pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1104 		}
1105 	}
1106 	memunmap(old_devtb);
1107 
1108 	return true;
1109 }
1110 
1111 static bool copy_device_table(void)
1112 {
1113 	struct amd_iommu *iommu;
1114 	struct amd_iommu_pci_seg *pci_seg;
1115 
1116 	if (!amd_iommu_pre_enabled)
1117 		return false;
1118 
1119 	pr_warn("Translation is already enabled - trying to copy translation structures\n");
1120 
1121 	/*
1122 	 * All IOMMUs within PCI segment shares common device table.
1123 	 * Hence copy device table only once per PCI segment.
1124 	 */
1125 	for_each_pci_segment(pci_seg) {
1126 		for_each_iommu(iommu) {
1127 			if (pci_seg->id != iommu->pci_seg->id)
1128 				continue;
1129 			if (!__copy_device_table(iommu))
1130 				return false;
1131 			break;
1132 		}
1133 	}
1134 
1135 	return true;
1136 }
1137 
1138 void amd_iommu_apply_erratum_63(u16 devid)
1139 {
1140 	int sysmgt;
1141 
1142 	sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
1143 		 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
1144 
1145 	if (sysmgt == 0x01)
1146 		set_dev_entry_bit(devid, DEV_ENTRY_IW);
1147 }
1148 
1149 /* Writes the specific IOMMU for a device into the rlookup table */
1150 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
1151 {
1152 	amd_iommu_rlookup_table[devid] = iommu;
1153 }
1154 
1155 /*
1156  * This function takes the device specific flags read from the ACPI
1157  * table and sets up the device table entry with that information
1158  */
1159 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1160 					   u16 devid, u32 flags, u32 ext_flags)
1161 {
1162 	if (flags & ACPI_DEVFLAG_INITPASS)
1163 		set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
1164 	if (flags & ACPI_DEVFLAG_EXTINT)
1165 		set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
1166 	if (flags & ACPI_DEVFLAG_NMI)
1167 		set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
1168 	if (flags & ACPI_DEVFLAG_SYSMGT1)
1169 		set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
1170 	if (flags & ACPI_DEVFLAG_SYSMGT2)
1171 		set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
1172 	if (flags & ACPI_DEVFLAG_LINT0)
1173 		set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
1174 	if (flags & ACPI_DEVFLAG_LINT1)
1175 		set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
1176 
1177 	amd_iommu_apply_erratum_63(devid);
1178 
1179 	set_iommu_for_device(iommu, devid);
1180 }
1181 
1182 int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1183 {
1184 	struct devid_map *entry;
1185 	struct list_head *list;
1186 
1187 	if (type == IVHD_SPECIAL_IOAPIC)
1188 		list = &ioapic_map;
1189 	else if (type == IVHD_SPECIAL_HPET)
1190 		list = &hpet_map;
1191 	else
1192 		return -EINVAL;
1193 
1194 	list_for_each_entry(entry, list, list) {
1195 		if (!(entry->id == id && entry->cmd_line))
1196 			continue;
1197 
1198 		pr_info("Command-line override present for %s id %d - ignoring\n",
1199 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1200 
1201 		*devid = entry->devid;
1202 
1203 		return 0;
1204 	}
1205 
1206 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1207 	if (!entry)
1208 		return -ENOMEM;
1209 
1210 	entry->id	= id;
1211 	entry->devid	= *devid;
1212 	entry->cmd_line	= cmd_line;
1213 
1214 	list_add_tail(&entry->list, list);
1215 
1216 	return 0;
1217 }
1218 
1219 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1220 				      bool cmd_line)
1221 {
1222 	struct acpihid_map_entry *entry;
1223 	struct list_head *list = &acpihid_map;
1224 
1225 	list_for_each_entry(entry, list, list) {
1226 		if (strcmp(entry->hid, hid) ||
1227 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1228 		    !entry->cmd_line)
1229 			continue;
1230 
1231 		pr_info("Command-line override for hid:%s uid:%s\n",
1232 			hid, uid);
1233 		*devid = entry->devid;
1234 		return 0;
1235 	}
1236 
1237 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1238 	if (!entry)
1239 		return -ENOMEM;
1240 
1241 	memcpy(entry->uid, uid, strlen(uid));
1242 	memcpy(entry->hid, hid, strlen(hid));
1243 	entry->devid = *devid;
1244 	entry->cmd_line	= cmd_line;
1245 	entry->root_devid = (entry->devid & (~0x7));
1246 
1247 	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1248 		entry->cmd_line ? "cmd" : "ivrs",
1249 		entry->hid, entry->uid, entry->root_devid);
1250 
1251 	list_add_tail(&entry->list, list);
1252 	return 0;
1253 }
1254 
1255 static int __init add_early_maps(void)
1256 {
1257 	int i, ret;
1258 
1259 	for (i = 0; i < early_ioapic_map_size; ++i) {
1260 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1261 					 early_ioapic_map[i].id,
1262 					 &early_ioapic_map[i].devid,
1263 					 early_ioapic_map[i].cmd_line);
1264 		if (ret)
1265 			return ret;
1266 	}
1267 
1268 	for (i = 0; i < early_hpet_map_size; ++i) {
1269 		ret = add_special_device(IVHD_SPECIAL_HPET,
1270 					 early_hpet_map[i].id,
1271 					 &early_hpet_map[i].devid,
1272 					 early_hpet_map[i].cmd_line);
1273 		if (ret)
1274 			return ret;
1275 	}
1276 
1277 	for (i = 0; i < early_acpihid_map_size; ++i) {
1278 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1279 					  early_acpihid_map[i].uid,
1280 					  &early_acpihid_map[i].devid,
1281 					  early_acpihid_map[i].cmd_line);
1282 		if (ret)
1283 			return ret;
1284 	}
1285 
1286 	return 0;
1287 }
1288 
1289 /*
1290  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1291  * initializes the hardware and our data structures with it.
1292  */
1293 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1294 					struct ivhd_header *h)
1295 {
1296 	u8 *p = (u8 *)h;
1297 	u8 *end = p, flags = 0;
1298 	u16 devid = 0, devid_start = 0, devid_to = 0;
1299 	u32 dev_i, ext_flags = 0;
1300 	bool alias = false;
1301 	struct ivhd_entry *e;
1302 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1303 	u32 ivhd_size;
1304 	int ret;
1305 
1306 
1307 	ret = add_early_maps();
1308 	if (ret)
1309 		return ret;
1310 
1311 	amd_iommu_apply_ivrs_quirks();
1312 
1313 	/*
1314 	 * First save the recommended feature enable bits from ACPI
1315 	 */
1316 	iommu->acpi_flags = h->flags;
1317 
1318 	/*
1319 	 * Done. Now parse the device entries
1320 	 */
1321 	ivhd_size = get_ivhd_header_size(h);
1322 	if (!ivhd_size) {
1323 		pr_err("Unsupported IVHD type %#x\n", h->type);
1324 		return -EINVAL;
1325 	}
1326 
1327 	p += ivhd_size;
1328 
1329 	end += h->length;
1330 
1331 
1332 	while (p < end) {
1333 		e = (struct ivhd_entry *)p;
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 <= amd_iommu_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: %02x:%02x.%x "
1345 				    "flags: %02x\n",
1346 				    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: %02x:%02x.%x flags: %02x\n",
1358 				    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: %02x:%02x.%x "
1371 				    "flags: %02x devid_to: %02x:%02x.%x\n",
1372 				    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: %02x:%02x.%x flags: %02x "
1390 				    "devid_to: %02x:%02x.%x\n",
1391 				    PCI_BUS_NUM(e->devid),
1392 				    PCI_SLOT(e->devid),
1393 				    PCI_FUNC(e->devid),
1394 				    e->flags,
1395 				    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: %02x:%02x.%x "
1408 				    "flags: %02x ext: %08x\n",
1409 				    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 				    "%02x:%02x.%x flags: %02x ext: %08x\n",
1422 				    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: %02x:%02x.%x\n",
1435 				    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 			u16 devid;
1454 			int ret;
1455 
1456 			handle = e->ext & 0xff;
1457 			devid  = (e->ext >>  8) & 0xffff;
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: %02x:%02x.%x\n",
1468 				    var, (int)handle,
1469 				    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 			u16 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 = e->devid;
1531 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1532 				    hid, uid,
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 */
1563 static 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 
1585 	pci_seg->id = id;
1586 	init_llist_head(&pci_seg->dev_data_list);
1587 	INIT_LIST_HEAD(&pci_seg->unity_map);
1588 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1589 
1590 	if (alloc_dev_table(pci_seg))
1591 		return NULL;
1592 	if (alloc_alias_table(pci_seg))
1593 		return NULL;
1594 	if (alloc_rlookup_table(pci_seg))
1595 		return NULL;
1596 
1597 	return pci_seg;
1598 }
1599 
1600 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1601 					struct acpi_table_header *ivrs_base)
1602 {
1603 	struct amd_iommu_pci_seg *pci_seg;
1604 
1605 	for_each_pci_segment(pci_seg) {
1606 		if (pci_seg->id == id)
1607 			return pci_seg;
1608 	}
1609 
1610 	return alloc_pci_segment(id, ivrs_base);
1611 }
1612 
1613 static void __init free_pci_segments(void)
1614 {
1615 	struct amd_iommu_pci_seg *pci_seg, *next;
1616 
1617 	for_each_pci_segment_safe(pci_seg, next) {
1618 		list_del(&pci_seg->list);
1619 		free_irq_lookup_table(pci_seg);
1620 		free_rlookup_table(pci_seg);
1621 		free_alias_table(pci_seg);
1622 		free_dev_table(pci_seg);
1623 		kfree(pci_seg);
1624 	}
1625 }
1626 
1627 static void __init free_iommu_one(struct amd_iommu *iommu)
1628 {
1629 	free_cwwb_sem(iommu);
1630 	free_command_buffer(iommu);
1631 	free_event_buffer(iommu);
1632 	free_ppr_log(iommu);
1633 	free_ga_log(iommu);
1634 	iommu_unmap_mmio_space(iommu);
1635 }
1636 
1637 static void __init free_iommu_all(void)
1638 {
1639 	struct amd_iommu *iommu, *next;
1640 
1641 	for_each_iommu_safe(iommu, next) {
1642 		list_del(&iommu->list);
1643 		free_iommu_one(iommu);
1644 		kfree(iommu);
1645 	}
1646 }
1647 
1648 /*
1649  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1650  * Workaround:
1651  *     BIOS should disable L2B micellaneous clock gating by setting
1652  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1653  */
1654 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1655 {
1656 	u32 value;
1657 
1658 	if ((boot_cpu_data.x86 != 0x15) ||
1659 	    (boot_cpu_data.x86_model < 0x10) ||
1660 	    (boot_cpu_data.x86_model > 0x1f))
1661 		return;
1662 
1663 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1664 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1665 
1666 	if (value & BIT(2))
1667 		return;
1668 
1669 	/* Select NB indirect register 0x90 and enable writing */
1670 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1671 
1672 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1673 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1674 
1675 	/* Clear the enable writing bit */
1676 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1677 }
1678 
1679 /*
1680  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1681  * Workaround:
1682  *     BIOS should enable ATS write permission check by setting
1683  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1684  */
1685 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1686 {
1687 	u32 value;
1688 
1689 	if ((boot_cpu_data.x86 != 0x15) ||
1690 	    (boot_cpu_data.x86_model < 0x30) ||
1691 	    (boot_cpu_data.x86_model > 0x3f))
1692 		return;
1693 
1694 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1695 	value = iommu_read_l2(iommu, 0x47);
1696 
1697 	if (value & BIT(0))
1698 		return;
1699 
1700 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1701 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1702 
1703 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1704 }
1705 
1706 /*
1707  * This function glues the initialization function for one IOMMU
1708  * together and also allocates the command buffer and programs the
1709  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1710  */
1711 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1712 				 struct acpi_table_header *ivrs_base)
1713 {
1714 	struct amd_iommu_pci_seg *pci_seg;
1715 	int ret;
1716 
1717 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1718 	if (pci_seg == NULL)
1719 		return -ENOMEM;
1720 	iommu->pci_seg = pci_seg;
1721 
1722 	raw_spin_lock_init(&iommu->lock);
1723 	iommu->cmd_sem_val = 0;
1724 
1725 	/* Add IOMMU to internal data structures */
1726 	list_add_tail(&iommu->list, &amd_iommu_list);
1727 	iommu->index = amd_iommus_present++;
1728 
1729 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1730 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1731 		return -ENOSYS;
1732 	}
1733 
1734 	/* Index is fine - add IOMMU to the array */
1735 	amd_iommus[iommu->index] = iommu;
1736 
1737 	/*
1738 	 * Copy data from ACPI table entry to the iommu struct
1739 	 */
1740 	iommu->devid   = h->devid;
1741 	iommu->cap_ptr = h->cap_ptr;
1742 	iommu->mmio_phys = h->mmio_phys;
1743 
1744 	switch (h->type) {
1745 	case 0x10:
1746 		/* Check if IVHD EFR contains proper max banks/counters */
1747 		if ((h->efr_attr != 0) &&
1748 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1749 		    ((h->efr_attr & (0x3F << 17)) != 0))
1750 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1751 		else
1752 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1753 
1754 		/*
1755 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1756 		 * GAM also requires GA mode. Therefore, we need to
1757 		 * check cmpxchg16b support before enabling it.
1758 		 */
1759 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1760 		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1761 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1762 		break;
1763 	case 0x11:
1764 	case 0x40:
1765 		if (h->efr_reg & (1 << 9))
1766 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1767 		else
1768 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1769 
1770 		/*
1771 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1772 		 * XT, GAM also requires GA mode. Therefore, we need to
1773 		 * check cmpxchg16b support before enabling them.
1774 		 */
1775 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1776 		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1777 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1778 			break;
1779 		}
1780 
1781 		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1782 			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1783 
1784 		early_iommu_features_init(iommu, h);
1785 
1786 		break;
1787 	default:
1788 		return -EINVAL;
1789 	}
1790 
1791 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1792 						iommu->mmio_phys_end);
1793 	if (!iommu->mmio_base)
1794 		return -ENOMEM;
1795 
1796 	if (alloc_cwwb_sem(iommu))
1797 		return -ENOMEM;
1798 
1799 	if (alloc_command_buffer(iommu))
1800 		return -ENOMEM;
1801 
1802 	if (alloc_event_buffer(iommu))
1803 		return -ENOMEM;
1804 
1805 	iommu->int_enabled = false;
1806 
1807 	init_translation_status(iommu);
1808 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1809 		iommu_disable(iommu);
1810 		clear_translation_pre_enabled(iommu);
1811 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1812 			iommu->index);
1813 	}
1814 	if (amd_iommu_pre_enabled)
1815 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1816 
1817 	ret = init_iommu_from_acpi(iommu, h);
1818 	if (ret)
1819 		return ret;
1820 
1821 	if (amd_iommu_irq_remap) {
1822 		ret = amd_iommu_create_irq_domain(iommu);
1823 		if (ret)
1824 			return ret;
1825 	}
1826 
1827 	/*
1828 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1829 	 * table tells us so, but this is a lie!
1830 	 */
1831 	amd_iommu_rlookup_table[iommu->devid] = NULL;
1832 
1833 	return 0;
1834 }
1835 
1836 /**
1837  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1838  * @ivrs: Pointer to the IVRS header
1839  *
1840  * This function search through all IVDB of the maximum supported IVHD
1841  */
1842 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1843 {
1844 	u8 *base = (u8 *)ivrs;
1845 	struct ivhd_header *ivhd = (struct ivhd_header *)
1846 					(base + IVRS_HEADER_LENGTH);
1847 	u8 last_type = ivhd->type;
1848 	u16 devid = ivhd->devid;
1849 
1850 	while (((u8 *)ivhd - base < ivrs->length) &&
1851 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1852 		u8 *p = (u8 *) ivhd;
1853 
1854 		if (ivhd->devid == devid)
1855 			last_type = ivhd->type;
1856 		ivhd = (struct ivhd_header *)(p + ivhd->length);
1857 	}
1858 
1859 	return last_type;
1860 }
1861 
1862 /*
1863  * Iterates over all IOMMU entries in the ACPI table, allocates the
1864  * IOMMU structure and initializes it with init_iommu_one()
1865  */
1866 static int __init init_iommu_all(struct acpi_table_header *table)
1867 {
1868 	u8 *p = (u8 *)table, *end = (u8 *)table;
1869 	struct ivhd_header *h;
1870 	struct amd_iommu *iommu;
1871 	int ret;
1872 
1873 	end += table->length;
1874 	p += IVRS_HEADER_LENGTH;
1875 
1876 	while (p < end) {
1877 		h = (struct ivhd_header *)p;
1878 		if (*p == amd_iommu_target_ivhd_type) {
1879 
1880 			DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1881 				    "seg: %d flags: %01x info %04x\n",
1882 				    PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1883 				    PCI_FUNC(h->devid), h->cap_ptr,
1884 				    h->pci_seg, h->flags, h->info);
1885 			DUMP_printk("       mmio-addr: %016llx\n",
1886 				    h->mmio_phys);
1887 
1888 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1889 			if (iommu == NULL)
1890 				return -ENOMEM;
1891 
1892 			ret = init_iommu_one(iommu, h, table);
1893 			if (ret)
1894 				return ret;
1895 		}
1896 		p += h->length;
1897 
1898 	}
1899 	WARN_ON(p != end);
1900 
1901 	return 0;
1902 }
1903 
1904 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1905 {
1906 	u64 val;
1907 	struct pci_dev *pdev = iommu->dev;
1908 
1909 	if (!iommu_feature(iommu, FEATURE_PC))
1910 		return;
1911 
1912 	amd_iommu_pc_present = true;
1913 
1914 	pci_info(pdev, "IOMMU performance counters supported\n");
1915 
1916 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1917 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1918 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1919 
1920 	return;
1921 }
1922 
1923 static ssize_t amd_iommu_show_cap(struct device *dev,
1924 				  struct device_attribute *attr,
1925 				  char *buf)
1926 {
1927 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1928 	return sprintf(buf, "%x\n", iommu->cap);
1929 }
1930 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1931 
1932 static ssize_t amd_iommu_show_features(struct device *dev,
1933 				       struct device_attribute *attr,
1934 				       char *buf)
1935 {
1936 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1937 	return sprintf(buf, "%llx\n", iommu->features);
1938 }
1939 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1940 
1941 static struct attribute *amd_iommu_attrs[] = {
1942 	&dev_attr_cap.attr,
1943 	&dev_attr_features.attr,
1944 	NULL,
1945 };
1946 
1947 static struct attribute_group amd_iommu_group = {
1948 	.name = "amd-iommu",
1949 	.attrs = amd_iommu_attrs,
1950 };
1951 
1952 static const struct attribute_group *amd_iommu_groups[] = {
1953 	&amd_iommu_group,
1954 	NULL,
1955 };
1956 
1957 /*
1958  * Note: IVHD 0x11 and 0x40 also contains exact copy
1959  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1960  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1961  */
1962 static void __init late_iommu_features_init(struct amd_iommu *iommu)
1963 {
1964 	u64 features;
1965 
1966 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1967 		return;
1968 
1969 	/* read extended feature bits */
1970 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1971 
1972 	if (!iommu->features) {
1973 		iommu->features = features;
1974 		return;
1975 	}
1976 
1977 	/*
1978 	 * Sanity check and warn if EFR values from
1979 	 * IVHD and MMIO conflict.
1980 	 */
1981 	if (features != iommu->features)
1982 		pr_warn(FW_WARN "EFR mismatch. Use IVHD EFR (%#llx : %#llx).\n",
1983 			features, iommu->features);
1984 }
1985 
1986 static int __init iommu_init_pci(struct amd_iommu *iommu)
1987 {
1988 	int cap_ptr = iommu->cap_ptr;
1989 	int ret;
1990 
1991 	iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1992 						 iommu->devid & 0xff);
1993 	if (!iommu->dev)
1994 		return -ENODEV;
1995 
1996 	/* Prevent binding other PCI device drivers to IOMMU devices */
1997 	iommu->dev->match_driver = false;
1998 
1999 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2000 			      &iommu->cap);
2001 
2002 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2003 		amd_iommu_iotlb_sup = false;
2004 
2005 	late_iommu_features_init(iommu);
2006 
2007 	if (iommu_feature(iommu, FEATURE_GT)) {
2008 		int glxval;
2009 		u32 max_pasid;
2010 		u64 pasmax;
2011 
2012 		pasmax = iommu->features & FEATURE_PASID_MASK;
2013 		pasmax >>= FEATURE_PASID_SHIFT;
2014 		max_pasid  = (1 << (pasmax + 1)) - 1;
2015 
2016 		amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
2017 
2018 		BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
2019 
2020 		glxval   = iommu->features & FEATURE_GLXVAL_MASK;
2021 		glxval >>= FEATURE_GLXVAL_SHIFT;
2022 
2023 		if (amd_iommu_max_glx_val == -1)
2024 			amd_iommu_max_glx_val = glxval;
2025 		else
2026 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2027 	}
2028 
2029 	if (iommu_feature(iommu, FEATURE_GT) &&
2030 	    iommu_feature(iommu, FEATURE_PPR)) {
2031 		iommu->is_iommu_v2   = true;
2032 		amd_iommu_v2_present = true;
2033 	}
2034 
2035 	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
2036 		return -ENOMEM;
2037 
2038 	ret = iommu_init_ga_log(iommu);
2039 	if (ret)
2040 		return ret;
2041 
2042 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2043 		pr_info("Using strict mode due to virtualization\n");
2044 		iommu_set_dma_strict();
2045 		amd_iommu_np_cache = true;
2046 	}
2047 
2048 	init_iommu_perf_ctr(iommu);
2049 
2050 	if (is_rd890_iommu(iommu->dev)) {
2051 		int i, j;
2052 
2053 		iommu->root_pdev =
2054 			pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
2055 						    PCI_DEVFN(0, 0));
2056 
2057 		/*
2058 		 * Some rd890 systems may not be fully reconfigured by the
2059 		 * BIOS, so it's necessary for us to store this information so
2060 		 * it can be reprogrammed on resume
2061 		 */
2062 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2063 				&iommu->stored_addr_lo);
2064 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2065 				&iommu->stored_addr_hi);
2066 
2067 		/* Low bit locks writes to configuration space */
2068 		iommu->stored_addr_lo &= ~1;
2069 
2070 		for (i = 0; i < 6; i++)
2071 			for (j = 0; j < 0x12; j++)
2072 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2073 
2074 		for (i = 0; i < 0x83; i++)
2075 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2076 	}
2077 
2078 	amd_iommu_erratum_746_workaround(iommu);
2079 	amd_iommu_ats_write_check_workaround(iommu);
2080 
2081 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2082 			       amd_iommu_groups, "ivhd%d", iommu->index);
2083 	if (ret)
2084 		return ret;
2085 
2086 	iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2087 
2088 	return pci_enable_device(iommu->dev);
2089 }
2090 
2091 static void print_iommu_info(void)
2092 {
2093 	static const char * const feat_str[] = {
2094 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2095 		"IA", "GA", "HE", "PC"
2096 	};
2097 	struct amd_iommu *iommu;
2098 
2099 	for_each_iommu(iommu) {
2100 		struct pci_dev *pdev = iommu->dev;
2101 		int i;
2102 
2103 		pci_info(pdev, "Found IOMMU cap 0x%x\n", iommu->cap_ptr);
2104 
2105 		if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
2106 			pr_info("Extended features (%#llx):", iommu->features);
2107 
2108 			for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2109 				if (iommu_feature(iommu, (1ULL << i)))
2110 					pr_cont(" %s", feat_str[i]);
2111 			}
2112 
2113 			if (iommu->features & FEATURE_GAM_VAPIC)
2114 				pr_cont(" GA_vAPIC");
2115 
2116 			pr_cont("\n");
2117 		}
2118 	}
2119 	if (irq_remapping_enabled) {
2120 		pr_info("Interrupt remapping enabled\n");
2121 		if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2122 			pr_info("Virtual APIC enabled\n");
2123 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2124 			pr_info("X2APIC enabled\n");
2125 	}
2126 }
2127 
2128 static int __init amd_iommu_init_pci(void)
2129 {
2130 	struct amd_iommu *iommu;
2131 	int ret;
2132 
2133 	for_each_iommu(iommu) {
2134 		ret = iommu_init_pci(iommu);
2135 		if (ret) {
2136 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2137 			       iommu->index, ret);
2138 			goto out;
2139 		}
2140 		/* Need to setup range after PCI init */
2141 		iommu_set_cwwb_range(iommu);
2142 	}
2143 
2144 	/*
2145 	 * Order is important here to make sure any unity map requirements are
2146 	 * fulfilled. The unity mappings are created and written to the device
2147 	 * table during the amd_iommu_init_api() call.
2148 	 *
2149 	 * After that we call init_device_table_dma() to make sure any
2150 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2151 	 * of all IOMMUs to make sure the changes to the device table are
2152 	 * active.
2153 	 */
2154 	ret = amd_iommu_init_api();
2155 	if (ret) {
2156 		pr_err("IOMMU: Failed to initialize IOMMU-API interface (error=%d)!\n",
2157 		       ret);
2158 		goto out;
2159 	}
2160 
2161 	init_device_table_dma();
2162 
2163 	for_each_iommu(iommu)
2164 		iommu_flush_all_caches(iommu);
2165 
2166 	print_iommu_info();
2167 
2168 out:
2169 	return ret;
2170 }
2171 
2172 /****************************************************************************
2173  *
2174  * The following functions initialize the MSI interrupts for all IOMMUs
2175  * in the system. It's a bit challenging because there could be multiple
2176  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2177  * pci_dev.
2178  *
2179  ****************************************************************************/
2180 
2181 static int iommu_setup_msi(struct amd_iommu *iommu)
2182 {
2183 	int r;
2184 
2185 	r = pci_enable_msi(iommu->dev);
2186 	if (r)
2187 		return r;
2188 
2189 	r = request_threaded_irq(iommu->dev->irq,
2190 				 amd_iommu_int_handler,
2191 				 amd_iommu_int_thread,
2192 				 0, "AMD-Vi",
2193 				 iommu);
2194 
2195 	if (r) {
2196 		pci_disable_msi(iommu->dev);
2197 		return r;
2198 	}
2199 
2200 	return 0;
2201 }
2202 
2203 union intcapxt {
2204 	u64	capxt;
2205 	struct {
2206 		u64	reserved_0		:  2,
2207 			dest_mode_logical	:  1,
2208 			reserved_1		:  5,
2209 			destid_0_23		: 24,
2210 			vector			:  8,
2211 			reserved_2		: 16,
2212 			destid_24_31		:  8;
2213 	};
2214 } __attribute__ ((packed));
2215 
2216 
2217 static struct irq_chip intcapxt_controller;
2218 
2219 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2220 				       struct irq_data *irqd, bool reserve)
2221 {
2222 	return 0;
2223 }
2224 
2225 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2226 					  struct irq_data *irqd)
2227 {
2228 }
2229 
2230 
2231 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2232 				    unsigned int nr_irqs, void *arg)
2233 {
2234 	struct irq_alloc_info *info = arg;
2235 	int i, ret;
2236 
2237 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2238 		return -EINVAL;
2239 
2240 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2241 	if (ret < 0)
2242 		return ret;
2243 
2244 	for (i = virq; i < virq + nr_irqs; i++) {
2245 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2246 
2247 		irqd->chip = &intcapxt_controller;
2248 		irqd->chip_data = info->data;
2249 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2250 	}
2251 
2252 	return ret;
2253 }
2254 
2255 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2256 				    unsigned int nr_irqs)
2257 {
2258 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2259 }
2260 
2261 
2262 static void intcapxt_unmask_irq(struct irq_data *irqd)
2263 {
2264 	struct amd_iommu *iommu = irqd->chip_data;
2265 	struct irq_cfg *cfg = irqd_cfg(irqd);
2266 	union intcapxt xt;
2267 
2268 	xt.capxt = 0ULL;
2269 	xt.dest_mode_logical = apic->dest_mode_logical;
2270 	xt.vector = cfg->vector;
2271 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2272 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2273 
2274 	/**
2275 	 * Current IOMMU implementation uses the same IRQ for all
2276 	 * 3 IOMMU interrupts.
2277 	 */
2278 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2279 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2280 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2281 }
2282 
2283 static void intcapxt_mask_irq(struct irq_data *irqd)
2284 {
2285 	struct amd_iommu *iommu = irqd->chip_data;
2286 
2287 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2288 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2289 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2290 }
2291 
2292 
2293 static int intcapxt_set_affinity(struct irq_data *irqd,
2294 				 const struct cpumask *mask, bool force)
2295 {
2296 	struct irq_data *parent = irqd->parent_data;
2297 	int ret;
2298 
2299 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2300 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2301 		return ret;
2302 	return 0;
2303 }
2304 
2305 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2306 {
2307 	return on ? -EOPNOTSUPP : 0;
2308 }
2309 
2310 static struct irq_chip intcapxt_controller = {
2311 	.name			= "IOMMU-MSI",
2312 	.irq_unmask		= intcapxt_unmask_irq,
2313 	.irq_mask		= intcapxt_mask_irq,
2314 	.irq_ack		= irq_chip_ack_parent,
2315 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2316 	.irq_set_affinity       = intcapxt_set_affinity,
2317 	.irq_set_wake		= intcapxt_set_wake,
2318 	.flags			= IRQCHIP_MASK_ON_SUSPEND,
2319 };
2320 
2321 static const struct irq_domain_ops intcapxt_domain_ops = {
2322 	.alloc			= intcapxt_irqdomain_alloc,
2323 	.free			= intcapxt_irqdomain_free,
2324 	.activate		= intcapxt_irqdomain_activate,
2325 	.deactivate		= intcapxt_irqdomain_deactivate,
2326 };
2327 
2328 
2329 static struct irq_domain *iommu_irqdomain;
2330 
2331 static struct irq_domain *iommu_get_irqdomain(void)
2332 {
2333 	struct fwnode_handle *fn;
2334 
2335 	/* No need for locking here (yet) as the init is single-threaded */
2336 	if (iommu_irqdomain)
2337 		return iommu_irqdomain;
2338 
2339 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2340 	if (!fn)
2341 		return NULL;
2342 
2343 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2344 						      fn, &intcapxt_domain_ops,
2345 						      NULL);
2346 	if (!iommu_irqdomain)
2347 		irq_domain_free_fwnode(fn);
2348 
2349 	return iommu_irqdomain;
2350 }
2351 
2352 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2353 {
2354 	struct irq_domain *domain;
2355 	struct irq_alloc_info info;
2356 	int irq, ret;
2357 
2358 	domain = iommu_get_irqdomain();
2359 	if (!domain)
2360 		return -ENXIO;
2361 
2362 	init_irq_alloc_info(&info, NULL);
2363 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2364 	info.data = iommu;
2365 
2366 	irq = irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
2367 	if (irq < 0) {
2368 		irq_domain_remove(domain);
2369 		return irq;
2370 	}
2371 
2372 	ret = request_threaded_irq(irq, amd_iommu_int_handler,
2373 				   amd_iommu_int_thread, 0, "AMD-Vi", iommu);
2374 	if (ret) {
2375 		irq_domain_free_irqs(irq, 1);
2376 		irq_domain_remove(domain);
2377 		return ret;
2378 	}
2379 
2380 	return 0;
2381 }
2382 
2383 static int iommu_init_irq(struct amd_iommu *iommu)
2384 {
2385 	int ret;
2386 
2387 	if (iommu->int_enabled)
2388 		goto enable_faults;
2389 
2390 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2391 		ret = iommu_setup_intcapxt(iommu);
2392 	else if (iommu->dev->msi_cap)
2393 		ret = iommu_setup_msi(iommu);
2394 	else
2395 		ret = -ENODEV;
2396 
2397 	if (ret)
2398 		return ret;
2399 
2400 	iommu->int_enabled = true;
2401 enable_faults:
2402 
2403 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2404 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2405 
2406 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2407 
2408 	if (iommu->ppr_log != NULL)
2409 		iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
2410 
2411 	iommu_ga_log_enable(iommu);
2412 
2413 	return 0;
2414 }
2415 
2416 /****************************************************************************
2417  *
2418  * The next functions belong to the third pass of parsing the ACPI
2419  * table. In this last pass the memory mapping requirements are
2420  * gathered (like exclusion and unity mapping ranges).
2421  *
2422  ****************************************************************************/
2423 
2424 static void __init free_unity_maps(void)
2425 {
2426 	struct unity_map_entry *entry, *next;
2427 	struct amd_iommu_pci_seg *p, *pci_seg;
2428 
2429 	for_each_pci_segment_safe(pci_seg, p) {
2430 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2431 			list_del(&entry->list);
2432 			kfree(entry);
2433 		}
2434 	}
2435 }
2436 
2437 /* called for unity map ACPI definition */
2438 static int __init init_unity_map_range(struct ivmd_header *m,
2439 				       struct acpi_table_header *ivrs_base)
2440 {
2441 	struct unity_map_entry *e = NULL;
2442 	struct amd_iommu_pci_seg *pci_seg;
2443 	char *s;
2444 
2445 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2446 	if (pci_seg == NULL)
2447 		return -ENOMEM;
2448 
2449 	e = kzalloc(sizeof(*e), GFP_KERNEL);
2450 	if (e == NULL)
2451 		return -ENOMEM;
2452 
2453 	switch (m->type) {
2454 	default:
2455 		kfree(e);
2456 		return 0;
2457 	case ACPI_IVMD_TYPE:
2458 		s = "IVMD_TYPEi\t\t\t";
2459 		e->devid_start = e->devid_end = m->devid;
2460 		break;
2461 	case ACPI_IVMD_TYPE_ALL:
2462 		s = "IVMD_TYPE_ALL\t\t";
2463 		e->devid_start = 0;
2464 		e->devid_end = amd_iommu_last_bdf;
2465 		break;
2466 	case ACPI_IVMD_TYPE_RANGE:
2467 		s = "IVMD_TYPE_RANGE\t\t";
2468 		e->devid_start = m->devid;
2469 		e->devid_end = m->aux;
2470 		break;
2471 	}
2472 	e->address_start = PAGE_ALIGN(m->range_start);
2473 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2474 	e->prot = m->flags >> 1;
2475 
2476 	/*
2477 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2478 	 * since some buggy BIOSes might lead to the overwritten exclusion
2479 	 * range (exclusion_start and exclusion_length members). This
2480 	 * happens when there are multiple exclusion ranges (IVMD entries)
2481 	 * defined in ACPI table.
2482 	 */
2483 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2484 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2485 
2486 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2487 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2488 		    " flags: %x\n", s, m->pci_seg,
2489 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2490 		    PCI_FUNC(e->devid_start), m->pci_seg,
2491 		    PCI_BUS_NUM(e->devid_end),
2492 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2493 		    e->address_start, e->address_end, m->flags);
2494 
2495 	list_add_tail(&e->list, &pci_seg->unity_map);
2496 
2497 	return 0;
2498 }
2499 
2500 /* iterates over all memory definitions we find in the ACPI table */
2501 static int __init init_memory_definitions(struct acpi_table_header *table)
2502 {
2503 	u8 *p = (u8 *)table, *end = (u8 *)table;
2504 	struct ivmd_header *m;
2505 
2506 	end += table->length;
2507 	p += IVRS_HEADER_LENGTH;
2508 
2509 	while (p < end) {
2510 		m = (struct ivmd_header *)p;
2511 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2512 			init_unity_map_range(m, table);
2513 
2514 		p += m->length;
2515 	}
2516 
2517 	return 0;
2518 }
2519 
2520 /*
2521  * Init the device table to not allow DMA access for devices
2522  */
2523 static void init_device_table_dma(void)
2524 {
2525 	u32 devid;
2526 
2527 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2528 		set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2529 		set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2530 	}
2531 }
2532 
2533 static void __init uninit_device_table_dma(void)
2534 {
2535 	u32 devid;
2536 
2537 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2538 		amd_iommu_dev_table[devid].data[0] = 0ULL;
2539 		amd_iommu_dev_table[devid].data[1] = 0ULL;
2540 	}
2541 }
2542 
2543 static void init_device_table(void)
2544 {
2545 	u32 devid;
2546 
2547 	if (!amd_iommu_irq_remap)
2548 		return;
2549 
2550 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2551 		set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2552 }
2553 
2554 static void iommu_init_flags(struct amd_iommu *iommu)
2555 {
2556 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2557 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2558 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2559 
2560 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2561 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2562 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2563 
2564 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2565 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2566 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2567 
2568 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2569 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2570 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2571 
2572 	/*
2573 	 * make IOMMU memory accesses cache coherent
2574 	 */
2575 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2576 
2577 	/* Set IOTLB invalidation timeout to 1s */
2578 	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2579 }
2580 
2581 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2582 {
2583 	int i, j;
2584 	u32 ioc_feature_control;
2585 	struct pci_dev *pdev = iommu->root_pdev;
2586 
2587 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2588 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2589 		return;
2590 
2591 	/*
2592 	 * First, we need to ensure that the iommu is enabled. This is
2593 	 * controlled by a register in the northbridge
2594 	 */
2595 
2596 	/* Select Northbridge indirect register 0x75 and enable writing */
2597 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2598 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2599 
2600 	/* Enable the iommu */
2601 	if (!(ioc_feature_control & 0x1))
2602 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2603 
2604 	/* Restore the iommu BAR */
2605 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2606 			       iommu->stored_addr_lo);
2607 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2608 			       iommu->stored_addr_hi);
2609 
2610 	/* Restore the l1 indirect regs for each of the 6 l1s */
2611 	for (i = 0; i < 6; i++)
2612 		for (j = 0; j < 0x12; j++)
2613 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2614 
2615 	/* Restore the l2 indirect regs */
2616 	for (i = 0; i < 0x83; i++)
2617 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2618 
2619 	/* Lock PCI setup registers */
2620 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2621 			       iommu->stored_addr_lo | 1);
2622 }
2623 
2624 static void iommu_enable_ga(struct amd_iommu *iommu)
2625 {
2626 #ifdef CONFIG_IRQ_REMAP
2627 	switch (amd_iommu_guest_ir) {
2628 	case AMD_IOMMU_GUEST_IR_VAPIC:
2629 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2630 		fallthrough;
2631 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2632 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2633 		iommu->irte_ops = &irte_128_ops;
2634 		break;
2635 	default:
2636 		iommu->irte_ops = &irte_32_ops;
2637 		break;
2638 	}
2639 #endif
2640 }
2641 
2642 static void early_enable_iommu(struct amd_iommu *iommu)
2643 {
2644 	iommu_disable(iommu);
2645 	iommu_init_flags(iommu);
2646 	iommu_set_device_table(iommu);
2647 	iommu_enable_command_buffer(iommu);
2648 	iommu_enable_event_buffer(iommu);
2649 	iommu_set_exclusion_range(iommu);
2650 	iommu_enable_ga(iommu);
2651 	iommu_enable_xt(iommu);
2652 	iommu_enable(iommu);
2653 	iommu_flush_all_caches(iommu);
2654 }
2655 
2656 /*
2657  * This function finally enables all IOMMUs found in the system after
2658  * they have been initialized.
2659  *
2660  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2661  * the old content of device table entries. Not this case or copy failed,
2662  * just continue as normal kernel does.
2663  */
2664 static void early_enable_iommus(void)
2665 {
2666 	struct amd_iommu *iommu;
2667 	struct amd_iommu_pci_seg *pci_seg;
2668 
2669 	if (!copy_device_table()) {
2670 		/*
2671 		 * If come here because of failure in copying device table from old
2672 		 * kernel with all IOMMUs enabled, print error message and try to
2673 		 * free allocated old_dev_tbl_cpy.
2674 		 */
2675 		if (amd_iommu_pre_enabled)
2676 			pr_err("Failed to copy DEV table from previous kernel.\n");
2677 
2678 		for_each_pci_segment(pci_seg) {
2679 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2680 				free_pages((unsigned long)pci_seg->old_dev_tbl_cpy,
2681 						get_order(dev_table_size));
2682 				pci_seg->old_dev_tbl_cpy = NULL;
2683 			}
2684 		}
2685 
2686 		for_each_iommu(iommu) {
2687 			clear_translation_pre_enabled(iommu);
2688 			early_enable_iommu(iommu);
2689 		}
2690 	} else {
2691 		pr_info("Copied DEV table from previous kernel.\n");
2692 
2693 		for_each_pci_segment(pci_seg) {
2694 			free_pages((unsigned long)pci_seg->dev_table,
2695 				   get_order(dev_table_size));
2696 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2697 		}
2698 
2699 		for_each_iommu(iommu) {
2700 			iommu_disable_command_buffer(iommu);
2701 			iommu_disable_event_buffer(iommu);
2702 			iommu_enable_command_buffer(iommu);
2703 			iommu_enable_event_buffer(iommu);
2704 			iommu_enable_ga(iommu);
2705 			iommu_enable_xt(iommu);
2706 			iommu_set_device_table(iommu);
2707 			iommu_flush_all_caches(iommu);
2708 		}
2709 	}
2710 
2711 #ifdef CONFIG_IRQ_REMAP
2712 	/*
2713 	 * Note: We have already checked GASup from IVRS table.
2714 	 *       Now, we need to make sure that GAMSup is set.
2715 	 */
2716 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2717 	    !check_feature_on_all_iommus(FEATURE_GAM_VAPIC))
2718 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2719 
2720 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2721 		amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2722 #endif
2723 }
2724 
2725 static void enable_iommus_v2(void)
2726 {
2727 	struct amd_iommu *iommu;
2728 
2729 	for_each_iommu(iommu) {
2730 		iommu_enable_ppr_log(iommu);
2731 		iommu_enable_gt(iommu);
2732 	}
2733 }
2734 
2735 static void enable_iommus(void)
2736 {
2737 	early_enable_iommus();
2738 
2739 	enable_iommus_v2();
2740 }
2741 
2742 static void disable_iommus(void)
2743 {
2744 	struct amd_iommu *iommu;
2745 
2746 	for_each_iommu(iommu)
2747 		iommu_disable(iommu);
2748 
2749 #ifdef CONFIG_IRQ_REMAP
2750 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2751 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2752 #endif
2753 }
2754 
2755 /*
2756  * Suspend/Resume support
2757  * disable suspend until real resume implemented
2758  */
2759 
2760 static void amd_iommu_resume(void)
2761 {
2762 	struct amd_iommu *iommu;
2763 
2764 	for_each_iommu(iommu)
2765 		iommu_apply_resume_quirks(iommu);
2766 
2767 	/* re-load the hardware */
2768 	enable_iommus();
2769 
2770 	amd_iommu_enable_interrupts();
2771 }
2772 
2773 static int amd_iommu_suspend(void)
2774 {
2775 	/* disable IOMMUs to go out of the way for BIOS */
2776 	disable_iommus();
2777 
2778 	return 0;
2779 }
2780 
2781 static struct syscore_ops amd_iommu_syscore_ops = {
2782 	.suspend = amd_iommu_suspend,
2783 	.resume = amd_iommu_resume,
2784 };
2785 
2786 static void __init free_iommu_resources(void)
2787 {
2788 	kmemleak_free(irq_lookup_table);
2789 	free_pages((unsigned long)irq_lookup_table,
2790 		   get_order(rlookup_table_size));
2791 	irq_lookup_table = NULL;
2792 
2793 	kmem_cache_destroy(amd_iommu_irq_cache);
2794 	amd_iommu_irq_cache = NULL;
2795 
2796 	free_pages((unsigned long)amd_iommu_rlookup_table,
2797 		   get_order(rlookup_table_size));
2798 	amd_iommu_rlookup_table = NULL;
2799 
2800 	free_pages((unsigned long)amd_iommu_alias_table,
2801 		   get_order(alias_table_size));
2802 	amd_iommu_alias_table = NULL;
2803 
2804 	free_pages((unsigned long)amd_iommu_dev_table,
2805 		   get_order(dev_table_size));
2806 	amd_iommu_dev_table = NULL;
2807 
2808 	free_iommu_all();
2809 	free_pci_segments();
2810 }
2811 
2812 /* SB IOAPIC is always on this device in AMD systems */
2813 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2814 
2815 static bool __init check_ioapic_information(void)
2816 {
2817 	const char *fw_bug = FW_BUG;
2818 	bool ret, has_sb_ioapic;
2819 	int idx;
2820 
2821 	has_sb_ioapic = false;
2822 	ret           = false;
2823 
2824 	/*
2825 	 * If we have map overrides on the kernel command line the
2826 	 * messages in this function might not describe firmware bugs
2827 	 * anymore - so be careful
2828 	 */
2829 	if (cmdline_maps)
2830 		fw_bug = "";
2831 
2832 	for (idx = 0; idx < nr_ioapics; idx++) {
2833 		int devid, id = mpc_ioapic_id(idx);
2834 
2835 		devid = get_ioapic_devid(id);
2836 		if (devid < 0) {
2837 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2838 				fw_bug, id);
2839 			ret = false;
2840 		} else if (devid == IOAPIC_SB_DEVID) {
2841 			has_sb_ioapic = true;
2842 			ret           = true;
2843 		}
2844 	}
2845 
2846 	if (!has_sb_ioapic) {
2847 		/*
2848 		 * We expect the SB IOAPIC to be listed in the IVRS
2849 		 * table. The system timer is connected to the SB IOAPIC
2850 		 * and if we don't have it in the list the system will
2851 		 * panic at boot time.  This situation usually happens
2852 		 * when the BIOS is buggy and provides us the wrong
2853 		 * device id for the IOAPIC in the system.
2854 		 */
2855 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2856 	}
2857 
2858 	if (!ret)
2859 		pr_err("Disabling interrupt remapping\n");
2860 
2861 	return ret;
2862 }
2863 
2864 static void __init free_dma_resources(void)
2865 {
2866 	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2867 		   get_order(MAX_DOMAIN_ID/8));
2868 	amd_iommu_pd_alloc_bitmap = NULL;
2869 
2870 	free_unity_maps();
2871 }
2872 
2873 static void __init ivinfo_init(void *ivrs)
2874 {
2875 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
2876 }
2877 
2878 /*
2879  * This is the hardware init function for AMD IOMMU in the system.
2880  * This function is called either from amd_iommu_init or from the interrupt
2881  * remapping setup code.
2882  *
2883  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2884  * four times:
2885  *
2886  *	1 pass) Discover the most comprehensive IVHD type to use.
2887  *
2888  *	2 pass) Find the highest PCI device id the driver has to handle.
2889  *		Upon this information the size of the data structures is
2890  *		determined that needs to be allocated.
2891  *
2892  *	3 pass) Initialize the data structures just allocated with the
2893  *		information in the ACPI table about available AMD IOMMUs
2894  *		in the system. It also maps the PCI devices in the
2895  *		system to specific IOMMUs
2896  *
2897  *	4 pass) After the basic data structures are allocated and
2898  *		initialized we update them with information about memory
2899  *		remapping requirements parsed out of the ACPI table in
2900  *		this last pass.
2901  *
2902  * After everything is set up the IOMMUs are enabled and the necessary
2903  * hotplug and suspend notifiers are registered.
2904  */
2905 static int __init early_amd_iommu_init(void)
2906 {
2907 	struct acpi_table_header *ivrs_base;
2908 	int remap_cache_sz, ret;
2909 	acpi_status status;
2910 
2911 	if (!amd_iommu_detected)
2912 		return -ENODEV;
2913 
2914 	status = acpi_get_table("IVRS", 0, &ivrs_base);
2915 	if (status == AE_NOT_FOUND)
2916 		return -ENODEV;
2917 	else if (ACPI_FAILURE(status)) {
2918 		const char *err = acpi_format_exception(status);
2919 		pr_err("IVRS table error: %s\n", err);
2920 		return -EINVAL;
2921 	}
2922 
2923 	/*
2924 	 * Validate checksum here so we don't need to do it when
2925 	 * we actually parse the table
2926 	 */
2927 	ret = check_ivrs_checksum(ivrs_base);
2928 	if (ret)
2929 		goto out;
2930 
2931 	ivinfo_init(ivrs_base);
2932 
2933 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2934 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2935 
2936 	dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE);
2937 	alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2938 	rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2939 
2940 	/* Device table - directly used by all IOMMUs */
2941 	ret = -ENOMEM;
2942 	amd_iommu_dev_table = (void *)__get_free_pages(
2943 				      GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2944 				      get_order(dev_table_size));
2945 	if (amd_iommu_dev_table == NULL)
2946 		goto out;
2947 
2948 	/*
2949 	 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2950 	 * IOMMU see for that device
2951 	 */
2952 	amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2953 			get_order(alias_table_size));
2954 	if (amd_iommu_alias_table == NULL)
2955 		goto out;
2956 
2957 	/* IOMMU rlookup table - find the IOMMU for a specific device */
2958 	amd_iommu_rlookup_table = (void *)__get_free_pages(
2959 			GFP_KERNEL | __GFP_ZERO,
2960 			get_order(rlookup_table_size));
2961 	if (amd_iommu_rlookup_table == NULL)
2962 		goto out;
2963 
2964 	amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2965 					    GFP_KERNEL | __GFP_ZERO,
2966 					    get_order(MAX_DOMAIN_ID/8));
2967 	if (amd_iommu_pd_alloc_bitmap == NULL)
2968 		goto out;
2969 
2970 	/*
2971 	 * never allocate domain 0 because its used as the non-allocated and
2972 	 * error value placeholder
2973 	 */
2974 	__set_bit(0, amd_iommu_pd_alloc_bitmap);
2975 
2976 	/*
2977 	 * now the data structures are allocated and basically initialized
2978 	 * start the real acpi table scan
2979 	 */
2980 	ret = init_iommu_all(ivrs_base);
2981 	if (ret)
2982 		goto out;
2983 
2984 	/* Disable any previously enabled IOMMUs */
2985 	if (!is_kdump_kernel() || amd_iommu_disabled)
2986 		disable_iommus();
2987 
2988 	if (amd_iommu_irq_remap)
2989 		amd_iommu_irq_remap = check_ioapic_information();
2990 
2991 	if (amd_iommu_irq_remap) {
2992 		struct amd_iommu_pci_seg *pci_seg;
2993 		/*
2994 		 * Interrupt remapping enabled, create kmem_cache for the
2995 		 * remapping tables.
2996 		 */
2997 		ret = -ENOMEM;
2998 		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2999 			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
3000 		else
3001 			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
3002 		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
3003 							remap_cache_sz,
3004 							DTE_INTTAB_ALIGNMENT,
3005 							0, NULL);
3006 		if (!amd_iommu_irq_cache)
3007 			goto out;
3008 
3009 		for_each_pci_segment(pci_seg) {
3010 			if (alloc_irq_lookup_table(pci_seg))
3011 				goto out;
3012 		}
3013 
3014 		irq_lookup_table = (void *)__get_free_pages(
3015 				GFP_KERNEL | __GFP_ZERO,
3016 				get_order(rlookup_table_size));
3017 		kmemleak_alloc(irq_lookup_table, rlookup_table_size,
3018 			       1, GFP_KERNEL);
3019 		if (!irq_lookup_table)
3020 			goto out;
3021 	}
3022 
3023 	ret = init_memory_definitions(ivrs_base);
3024 	if (ret)
3025 		goto out;
3026 
3027 	/* init the device table */
3028 	init_device_table();
3029 
3030 out:
3031 	/* Don't leak any ACPI memory */
3032 	acpi_put_table(ivrs_base);
3033 
3034 	return ret;
3035 }
3036 
3037 static int amd_iommu_enable_interrupts(void)
3038 {
3039 	struct amd_iommu *iommu;
3040 	int ret = 0;
3041 
3042 	for_each_iommu(iommu) {
3043 		ret = iommu_init_irq(iommu);
3044 		if (ret)
3045 			goto out;
3046 	}
3047 
3048 out:
3049 	return ret;
3050 }
3051 
3052 static bool __init detect_ivrs(void)
3053 {
3054 	struct acpi_table_header *ivrs_base;
3055 	acpi_status status;
3056 	int i;
3057 
3058 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3059 	if (status == AE_NOT_FOUND)
3060 		return false;
3061 	else if (ACPI_FAILURE(status)) {
3062 		const char *err = acpi_format_exception(status);
3063 		pr_err("IVRS table error: %s\n", err);
3064 		return false;
3065 	}
3066 
3067 	acpi_put_table(ivrs_base);
3068 
3069 	if (amd_iommu_force_enable)
3070 		goto out;
3071 
3072 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3073 	for (i = 0; i < 32; i++) {
3074 		u32 pci_id;
3075 
3076 		pci_id = read_pci_config(0, i, 0, 0);
3077 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3078 			pr_info("Disable IOMMU on Stoney Ridge\n");
3079 			return false;
3080 		}
3081 	}
3082 
3083 out:
3084 	/* Make sure ACS will be enabled during PCI probe */
3085 	pci_request_acs();
3086 
3087 	return true;
3088 }
3089 
3090 /****************************************************************************
3091  *
3092  * AMD IOMMU Initialization State Machine
3093  *
3094  ****************************************************************************/
3095 
3096 static int __init state_next(void)
3097 {
3098 	int ret = 0;
3099 
3100 	switch (init_state) {
3101 	case IOMMU_START_STATE:
3102 		if (!detect_ivrs()) {
3103 			init_state	= IOMMU_NOT_FOUND;
3104 			ret		= -ENODEV;
3105 		} else {
3106 			init_state	= IOMMU_IVRS_DETECTED;
3107 		}
3108 		break;
3109 	case IOMMU_IVRS_DETECTED:
3110 		if (amd_iommu_disabled) {
3111 			init_state = IOMMU_CMDLINE_DISABLED;
3112 			ret = -EINVAL;
3113 		} else {
3114 			ret = early_amd_iommu_init();
3115 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3116 		}
3117 		break;
3118 	case IOMMU_ACPI_FINISHED:
3119 		early_enable_iommus();
3120 		x86_platform.iommu_shutdown = disable_iommus;
3121 		init_state = IOMMU_ENABLED;
3122 		break;
3123 	case IOMMU_ENABLED:
3124 		register_syscore_ops(&amd_iommu_syscore_ops);
3125 		ret = amd_iommu_init_pci();
3126 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3127 		enable_iommus_v2();
3128 		break;
3129 	case IOMMU_PCI_INIT:
3130 		ret = amd_iommu_enable_interrupts();
3131 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3132 		break;
3133 	case IOMMU_INTERRUPTS_EN:
3134 		init_state = IOMMU_INITIALIZED;
3135 		break;
3136 	case IOMMU_INITIALIZED:
3137 		/* Nothing to do */
3138 		break;
3139 	case IOMMU_NOT_FOUND:
3140 	case IOMMU_INIT_ERROR:
3141 	case IOMMU_CMDLINE_DISABLED:
3142 		/* Error states => do nothing */
3143 		ret = -EINVAL;
3144 		break;
3145 	default:
3146 		/* Unknown state */
3147 		BUG();
3148 	}
3149 
3150 	if (ret) {
3151 		free_dma_resources();
3152 		if (!irq_remapping_enabled) {
3153 			disable_iommus();
3154 			free_iommu_resources();
3155 		} else {
3156 			struct amd_iommu *iommu;
3157 
3158 			uninit_device_table_dma();
3159 			for_each_iommu(iommu)
3160 				iommu_flush_all_caches(iommu);
3161 		}
3162 	}
3163 	return ret;
3164 }
3165 
3166 static int __init iommu_go_to_state(enum iommu_init_state state)
3167 {
3168 	int ret = -EINVAL;
3169 
3170 	while (init_state != state) {
3171 		if (init_state == IOMMU_NOT_FOUND         ||
3172 		    init_state == IOMMU_INIT_ERROR        ||
3173 		    init_state == IOMMU_CMDLINE_DISABLED)
3174 			break;
3175 		ret = state_next();
3176 	}
3177 
3178 	return ret;
3179 }
3180 
3181 #ifdef CONFIG_IRQ_REMAP
3182 int __init amd_iommu_prepare(void)
3183 {
3184 	int ret;
3185 
3186 	amd_iommu_irq_remap = true;
3187 
3188 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3189 	if (ret) {
3190 		amd_iommu_irq_remap = false;
3191 		return ret;
3192 	}
3193 
3194 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3195 }
3196 
3197 int __init amd_iommu_enable(void)
3198 {
3199 	int ret;
3200 
3201 	ret = iommu_go_to_state(IOMMU_ENABLED);
3202 	if (ret)
3203 		return ret;
3204 
3205 	irq_remapping_enabled = 1;
3206 	return amd_iommu_xt_mode;
3207 }
3208 
3209 void amd_iommu_disable(void)
3210 {
3211 	amd_iommu_suspend();
3212 }
3213 
3214 int amd_iommu_reenable(int mode)
3215 {
3216 	amd_iommu_resume();
3217 
3218 	return 0;
3219 }
3220 
3221 int __init amd_iommu_enable_faulting(void)
3222 {
3223 	/* We enable MSI later when PCI is initialized */
3224 	return 0;
3225 }
3226 #endif
3227 
3228 /*
3229  * This is the core init function for AMD IOMMU hardware in the system.
3230  * This function is called from the generic x86 DMA layer initialization
3231  * code.
3232  */
3233 static int __init amd_iommu_init(void)
3234 {
3235 	struct amd_iommu *iommu;
3236 	int ret;
3237 
3238 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3239 #ifdef CONFIG_GART_IOMMU
3240 	if (ret && list_empty(&amd_iommu_list)) {
3241 		/*
3242 		 * We failed to initialize the AMD IOMMU - try fallback
3243 		 * to GART if possible.
3244 		 */
3245 		gart_iommu_init();
3246 	}
3247 #endif
3248 
3249 	for_each_iommu(iommu)
3250 		amd_iommu_debugfs_setup(iommu);
3251 
3252 	return ret;
3253 }
3254 
3255 static bool amd_iommu_sme_check(void)
3256 {
3257 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3258 	    (boot_cpu_data.x86 != 0x17))
3259 		return true;
3260 
3261 	/* For Fam17h, a specific level of support is required */
3262 	if (boot_cpu_data.microcode >= 0x08001205)
3263 		return true;
3264 
3265 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3266 	    (boot_cpu_data.microcode <= 0x080011ff))
3267 		return true;
3268 
3269 	pr_notice("IOMMU not currently supported when SME is active\n");
3270 
3271 	return false;
3272 }
3273 
3274 /****************************************************************************
3275  *
3276  * Early detect code. This code runs at IOMMU detection time in the DMA
3277  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3278  * IOMMUs
3279  *
3280  ****************************************************************************/
3281 int __init amd_iommu_detect(void)
3282 {
3283 	int ret;
3284 
3285 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3286 		return -ENODEV;
3287 
3288 	if (!amd_iommu_sme_check())
3289 		return -ENODEV;
3290 
3291 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3292 	if (ret)
3293 		return ret;
3294 
3295 	amd_iommu_detected = true;
3296 	iommu_detected = 1;
3297 	x86_init.iommu.iommu_init = amd_iommu_init;
3298 
3299 	return 1;
3300 }
3301 
3302 /****************************************************************************
3303  *
3304  * Parsing functions for the AMD IOMMU specific kernel command line
3305  * options.
3306  *
3307  ****************************************************************************/
3308 
3309 static int __init parse_amd_iommu_dump(char *str)
3310 {
3311 	amd_iommu_dump = true;
3312 
3313 	return 1;
3314 }
3315 
3316 static int __init parse_amd_iommu_intr(char *str)
3317 {
3318 	for (; *str; ++str) {
3319 		if (strncmp(str, "legacy", 6) == 0) {
3320 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3321 			break;
3322 		}
3323 		if (strncmp(str, "vapic", 5) == 0) {
3324 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3325 			break;
3326 		}
3327 	}
3328 	return 1;
3329 }
3330 
3331 static int __init parse_amd_iommu_options(char *str)
3332 {
3333 	for (; *str; ++str) {
3334 		if (strncmp(str, "fullflush", 9) == 0) {
3335 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3336 			iommu_set_dma_strict();
3337 		}
3338 		if (strncmp(str, "force_enable", 12) == 0)
3339 			amd_iommu_force_enable = true;
3340 		if (strncmp(str, "off", 3) == 0)
3341 			amd_iommu_disabled = true;
3342 		if (strncmp(str, "force_isolation", 15) == 0)
3343 			amd_iommu_force_isolation = true;
3344 	}
3345 
3346 	return 1;
3347 }
3348 
3349 static int __init parse_ivrs_ioapic(char *str)
3350 {
3351 	unsigned int bus, dev, fn;
3352 	int ret, id, i;
3353 	u16 devid;
3354 
3355 	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
3356 
3357 	if (ret != 4) {
3358 		pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3359 		return 1;
3360 	}
3361 
3362 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3363 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3364 			str);
3365 		return 1;
3366 	}
3367 
3368 	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3369 
3370 	cmdline_maps			= true;
3371 	i				= early_ioapic_map_size++;
3372 	early_ioapic_map[i].id		= id;
3373 	early_ioapic_map[i].devid	= devid;
3374 	early_ioapic_map[i].cmd_line	= true;
3375 
3376 	return 1;
3377 }
3378 
3379 static int __init parse_ivrs_hpet(char *str)
3380 {
3381 	unsigned int bus, dev, fn;
3382 	int ret, id, i;
3383 	u16 devid;
3384 
3385 	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
3386 
3387 	if (ret != 4) {
3388 		pr_err("Invalid command line: ivrs_hpet%s\n", str);
3389 		return 1;
3390 	}
3391 
3392 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3393 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3394 			str);
3395 		return 1;
3396 	}
3397 
3398 	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3399 
3400 	cmdline_maps			= true;
3401 	i				= early_hpet_map_size++;
3402 	early_hpet_map[i].id		= id;
3403 	early_hpet_map[i].devid		= devid;
3404 	early_hpet_map[i].cmd_line	= true;
3405 
3406 	return 1;
3407 }
3408 
3409 static int __init parse_ivrs_acpihid(char *str)
3410 {
3411 	u32 bus, dev, fn;
3412 	char *hid, *uid, *p;
3413 	char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3414 	int ret, i;
3415 
3416 	ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
3417 	if (ret != 4) {
3418 		pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
3419 		return 1;
3420 	}
3421 
3422 	p = acpiid;
3423 	hid = strsep(&p, ":");
3424 	uid = p;
3425 
3426 	if (!hid || !(*hid) || !uid) {
3427 		pr_err("Invalid command line: hid or uid\n");
3428 		return 1;
3429 	}
3430 
3431 	i = early_acpihid_map_size++;
3432 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3433 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3434 	early_acpihid_map[i].devid =
3435 		((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3436 	early_acpihid_map[i].cmd_line	= true;
3437 
3438 	return 1;
3439 }
3440 
3441 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3442 __setup("amd_iommu=",		parse_amd_iommu_options);
3443 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3444 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3445 __setup("ivrs_hpet",		parse_ivrs_hpet);
3446 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3447 
3448 bool amd_iommu_v2_supported(void)
3449 {
3450 	return amd_iommu_v2_present;
3451 }
3452 EXPORT_SYMBOL(amd_iommu_v2_supported);
3453 
3454 struct amd_iommu *get_amd_iommu(unsigned int idx)
3455 {
3456 	unsigned int i = 0;
3457 	struct amd_iommu *iommu;
3458 
3459 	for_each_iommu(iommu)
3460 		if (i++ == idx)
3461 			return iommu;
3462 	return NULL;
3463 }
3464 
3465 /****************************************************************************
3466  *
3467  * IOMMU EFR Performance Counter support functionality. This code allows
3468  * access to the IOMMU PC functionality.
3469  *
3470  ****************************************************************************/
3471 
3472 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3473 {
3474 	struct amd_iommu *iommu = get_amd_iommu(idx);
3475 
3476 	if (iommu)
3477 		return iommu->max_banks;
3478 
3479 	return 0;
3480 }
3481 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3482 
3483 bool amd_iommu_pc_supported(void)
3484 {
3485 	return amd_iommu_pc_present;
3486 }
3487 EXPORT_SYMBOL(amd_iommu_pc_supported);
3488 
3489 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3490 {
3491 	struct amd_iommu *iommu = get_amd_iommu(idx);
3492 
3493 	if (iommu)
3494 		return iommu->max_counters;
3495 
3496 	return 0;
3497 }
3498 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3499 
3500 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3501 				u8 fxn, u64 *value, bool is_write)
3502 {
3503 	u32 offset;
3504 	u32 max_offset_lim;
3505 
3506 	/* Make sure the IOMMU PC resource is available */
3507 	if (!amd_iommu_pc_present)
3508 		return -ENODEV;
3509 
3510 	/* Check for valid iommu and pc register indexing */
3511 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3512 		return -ENODEV;
3513 
3514 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3515 
3516 	/* Limit the offset to the hw defined mmio region aperture */
3517 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3518 				(iommu->max_counters << 8) | 0x28);
3519 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3520 	    (offset > max_offset_lim))
3521 		return -EINVAL;
3522 
3523 	if (is_write) {
3524 		u64 val = *value & GENMASK_ULL(47, 0);
3525 
3526 		writel((u32)val, iommu->mmio_base + offset);
3527 		writel((val >> 32), iommu->mmio_base + offset + 4);
3528 	} else {
3529 		*value = readl(iommu->mmio_base + offset + 4);
3530 		*value <<= 32;
3531 		*value |= readl(iommu->mmio_base + offset);
3532 		*value &= GENMASK_ULL(47, 0);
3533 	}
3534 
3535 	return 0;
3536 }
3537 
3538 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3539 {
3540 	if (!iommu)
3541 		return -EINVAL;
3542 
3543 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3544 }
3545 
3546 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3547 {
3548 	if (!iommu)
3549 		return -EINVAL;
3550 
3551 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3552 }
3553