xref: /openbmc/linux/drivers/iommu/amd/init.c (revision 56fb79514c52947107001ff9313870d76c4c8008)
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  * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
205  * to know which ones are already in use.
206  */
207 unsigned long *amd_iommu_pd_alloc_bitmap;
208 
209 static u32 dev_table_size;	/* size of the device table */
210 static u32 alias_table_size;	/* size of the alias table */
211 
212 enum iommu_init_state {
213 	IOMMU_START_STATE,
214 	IOMMU_IVRS_DETECTED,
215 	IOMMU_ACPI_FINISHED,
216 	IOMMU_ENABLED,
217 	IOMMU_PCI_INIT,
218 	IOMMU_INTERRUPTS_EN,
219 	IOMMU_INITIALIZED,
220 	IOMMU_NOT_FOUND,
221 	IOMMU_INIT_ERROR,
222 	IOMMU_CMDLINE_DISABLED,
223 };
224 
225 /* Early ioapic and hpet maps from kernel command line */
226 #define EARLY_MAP_SIZE		4
227 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
228 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
229 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
230 
231 static int __initdata early_ioapic_map_size;
232 static int __initdata early_hpet_map_size;
233 static int __initdata early_acpihid_map_size;
234 
235 static bool __initdata cmdline_maps;
236 
237 static enum iommu_init_state init_state = IOMMU_START_STATE;
238 
239 static int amd_iommu_enable_interrupts(void);
240 static int __init iommu_go_to_state(enum iommu_init_state state);
241 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
242 
243 static bool amd_iommu_pre_enabled = true;
244 
245 static u32 amd_iommu_ivinfo __initdata;
246 
247 bool translation_pre_enabled(struct amd_iommu *iommu)
248 {
249 	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
250 }
251 
252 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
253 {
254 	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
255 }
256 
257 static void init_translation_status(struct amd_iommu *iommu)
258 {
259 	u64 ctrl;
260 
261 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
262 	if (ctrl & (1<<CONTROL_IOMMU_EN))
263 		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
264 }
265 
266 static inline void update_last_devid(u16 devid)
267 {
268 	if (devid > amd_iommu_last_bdf)
269 		amd_iommu_last_bdf = devid;
270 }
271 
272 static inline unsigned long tbl_size(int entry_size)
273 {
274 	unsigned shift = PAGE_SHIFT +
275 			 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
276 
277 	return 1UL << shift;
278 }
279 
280 int amd_iommu_get_num_iommus(void)
281 {
282 	return amd_iommus_present;
283 }
284 
285 #ifdef CONFIG_IRQ_REMAP
286 static bool check_feature_on_all_iommus(u64 mask)
287 {
288 	bool ret = false;
289 	struct amd_iommu *iommu;
290 
291 	for_each_iommu(iommu) {
292 		ret = iommu_feature(iommu, mask);
293 		if (!ret)
294 			return false;
295 	}
296 
297 	return true;
298 }
299 #endif
300 
301 /*
302  * For IVHD type 0x11/0x40, EFR is also available via IVHD.
303  * Default to IVHD EFR since it is available sooner
304  * (i.e. before PCI init).
305  */
306 static void __init early_iommu_features_init(struct amd_iommu *iommu,
307 					     struct ivhd_header *h)
308 {
309 	if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP)
310 		iommu->features = h->efr_reg;
311 	if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
312 		amdr_ivrs_remap_support = true;
313 }
314 
315 /* Access to l1 and l2 indexed register spaces */
316 
317 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
318 {
319 	u32 val;
320 
321 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
322 	pci_read_config_dword(iommu->dev, 0xfc, &val);
323 	return val;
324 }
325 
326 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
327 {
328 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
329 	pci_write_config_dword(iommu->dev, 0xfc, val);
330 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
331 }
332 
333 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
334 {
335 	u32 val;
336 
337 	pci_write_config_dword(iommu->dev, 0xf0, address);
338 	pci_read_config_dword(iommu->dev, 0xf4, &val);
339 	return val;
340 }
341 
342 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
343 {
344 	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
345 	pci_write_config_dword(iommu->dev, 0xf4, val);
346 }
347 
348 /****************************************************************************
349  *
350  * AMD IOMMU MMIO register space handling functions
351  *
352  * These functions are used to program the IOMMU device registers in
353  * MMIO space required for that driver.
354  *
355  ****************************************************************************/
356 
357 /*
358  * This function set the exclusion range in the IOMMU. DMA accesses to the
359  * exclusion range are passed through untranslated
360  */
361 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
362 {
363 	u64 start = iommu->exclusion_start & PAGE_MASK;
364 	u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
365 	u64 entry;
366 
367 	if (!iommu->exclusion_start)
368 		return;
369 
370 	entry = start | MMIO_EXCL_ENABLE_MASK;
371 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
372 			&entry, sizeof(entry));
373 
374 	entry = limit;
375 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
376 			&entry, sizeof(entry));
377 }
378 
379 static void iommu_set_cwwb_range(struct amd_iommu *iommu)
380 {
381 	u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
382 	u64 entry = start & PM_ADDR_MASK;
383 
384 	if (!iommu_feature(iommu, FEATURE_SNP))
385 		return;
386 
387 	/* Note:
388 	 * Re-purpose Exclusion base/limit registers for Completion wait
389 	 * write-back base/limit.
390 	 */
391 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
392 		    &entry, sizeof(entry));
393 
394 	/* Note:
395 	 * Default to 4 Kbytes, which can be specified by setting base
396 	 * address equal to the limit address.
397 	 */
398 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
399 		    &entry, sizeof(entry));
400 }
401 
402 /* Programs the physical address of the device table into the IOMMU hardware */
403 static void iommu_set_device_table(struct amd_iommu *iommu)
404 {
405 	u64 entry;
406 	u32 dev_table_size = iommu->pci_seg->dev_table_size;
407 
408 	BUG_ON(iommu->mmio_base == NULL);
409 
410 	entry = iommu_virt_to_phys(amd_iommu_dev_table);
411 	entry |= (dev_table_size >> 12) - 1;
412 	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
413 			&entry, sizeof(entry));
414 }
415 
416 /* Generic functions to enable/disable certain features of the IOMMU. */
417 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
418 {
419 	u64 ctrl;
420 
421 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
422 	ctrl |= (1ULL << bit);
423 	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
424 }
425 
426 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
427 {
428 	u64 ctrl;
429 
430 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
431 	ctrl &= ~(1ULL << bit);
432 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
433 }
434 
435 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
436 {
437 	u64 ctrl;
438 
439 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
440 	ctrl &= ~CTRL_INV_TO_MASK;
441 	ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
442 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
443 }
444 
445 /* Function to enable the hardware */
446 static void iommu_enable(struct amd_iommu *iommu)
447 {
448 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
449 }
450 
451 static void iommu_disable(struct amd_iommu *iommu)
452 {
453 	if (!iommu->mmio_base)
454 		return;
455 
456 	/* Disable command buffer */
457 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
458 
459 	/* Disable event logging and event interrupts */
460 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
461 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
462 
463 	/* Disable IOMMU GA_LOG */
464 	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
465 	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
466 
467 	/* Disable IOMMU hardware itself */
468 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
469 }
470 
471 /*
472  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
473  * the system has one.
474  */
475 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
476 {
477 	if (!request_mem_region(address, end, "amd_iommu")) {
478 		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
479 			address, end);
480 		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
481 		return NULL;
482 	}
483 
484 	return (u8 __iomem *)ioremap(address, end);
485 }
486 
487 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
488 {
489 	if (iommu->mmio_base)
490 		iounmap(iommu->mmio_base);
491 	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
492 }
493 
494 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
495 {
496 	u32 size = 0;
497 
498 	switch (h->type) {
499 	case 0x10:
500 		size = 24;
501 		break;
502 	case 0x11:
503 	case 0x40:
504 		size = 40;
505 		break;
506 	}
507 	return size;
508 }
509 
510 /****************************************************************************
511  *
512  * The functions below belong to the first pass of AMD IOMMU ACPI table
513  * parsing. In this pass we try to find out the highest device id this
514  * code has to handle. Upon this information the size of the shared data
515  * structures is determined later.
516  *
517  ****************************************************************************/
518 
519 /*
520  * This function calculates the length of a given IVHD entry
521  */
522 static inline int ivhd_entry_length(u8 *ivhd)
523 {
524 	u32 type = ((struct ivhd_entry *)ivhd)->type;
525 
526 	if (type < 0x80) {
527 		return 0x04 << (*ivhd >> 6);
528 	} else if (type == IVHD_DEV_ACPI_HID) {
529 		/* For ACPI_HID, offset 21 is uid len */
530 		return *((u8 *)ivhd + 21) + 22;
531 	}
532 	return 0;
533 }
534 
535 /*
536  * After reading the highest device id from the IOMMU PCI capability header
537  * this function looks if there is a higher device id defined in the ACPI table
538  */
539 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
540 {
541 	u8 *p = (void *)h, *end = (void *)h;
542 	struct ivhd_entry *dev;
543 	int last_devid = -EINVAL;
544 
545 	u32 ivhd_size = get_ivhd_header_size(h);
546 
547 	if (!ivhd_size) {
548 		pr_err("Unsupported IVHD type %#x\n", h->type);
549 		return -EINVAL;
550 	}
551 
552 	p += ivhd_size;
553 	end += h->length;
554 
555 	while (p < end) {
556 		dev = (struct ivhd_entry *)p;
557 		switch (dev->type) {
558 		case IVHD_DEV_ALL:
559 			/* Use maximum BDF value for DEV_ALL */
560 			update_last_devid(0xffff);
561 			return 0xffff;
562 		case IVHD_DEV_SELECT:
563 		case IVHD_DEV_RANGE_END:
564 		case IVHD_DEV_ALIAS:
565 		case IVHD_DEV_EXT_SELECT:
566 			/* all the above subfield types refer to device ids */
567 			update_last_devid(dev->devid);
568 			if (dev->devid > last_devid)
569 				last_devid = dev->devid;
570 			break;
571 		default:
572 			break;
573 		}
574 		p += ivhd_entry_length(p);
575 	}
576 
577 	WARN_ON(p != end);
578 
579 	return last_devid;
580 }
581 
582 static int __init check_ivrs_checksum(struct acpi_table_header *table)
583 {
584 	int i;
585 	u8 checksum = 0, *p = (u8 *)table;
586 
587 	for (i = 0; i < table->length; ++i)
588 		checksum += p[i];
589 	if (checksum != 0) {
590 		/* ACPI table corrupt */
591 		pr_err(FW_BUG "IVRS invalid checksum\n");
592 		return -ENODEV;
593 	}
594 
595 	return 0;
596 }
597 
598 /*
599  * Iterate over all IVHD entries in the ACPI table and find the highest device
600  * id which we need to handle. This is the first of three functions which parse
601  * the ACPI table. So we check the checksum here.
602  */
603 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
604 {
605 	u8 *p = (u8 *)table, *end = (u8 *)table;
606 	struct ivhd_header *h;
607 	int last_devid, last_bdf = 0;
608 
609 	p += IVRS_HEADER_LENGTH;
610 
611 	end += table->length;
612 	while (p < end) {
613 		h = (struct ivhd_header *)p;
614 		if (h->pci_seg == pci_seg &&
615 		    h->type == amd_iommu_target_ivhd_type) {
616 			last_devid = find_last_devid_from_ivhd(h);
617 
618 			if (last_devid < 0)
619 				return -EINVAL;
620 			if (last_devid > last_bdf)
621 				last_bdf = last_devid;
622 		}
623 		p += h->length;
624 	}
625 	WARN_ON(p != end);
626 
627 	return last_bdf;
628 }
629 
630 /****************************************************************************
631  *
632  * The following functions belong to the code path which parses the ACPI table
633  * the second time. In this ACPI parsing iteration we allocate IOMMU specific
634  * data structures, initialize the per PCI segment device/alias/rlookup table
635  * and also basically initialize the hardware.
636  *
637  ****************************************************************************/
638 
639 /* Allocate per PCI segment device table */
640 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
641 {
642 	pci_seg->dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
643 						      get_order(pci_seg->dev_table_size));
644 	if (!pci_seg->dev_table)
645 		return -ENOMEM;
646 
647 	return 0;
648 }
649 
650 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
651 {
652 	free_pages((unsigned long)pci_seg->dev_table,
653 		    get_order(pci_seg->dev_table_size));
654 	pci_seg->dev_table = NULL;
655 }
656 
657 /* Allocate per PCI segment IOMMU rlookup table. */
658 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
659 {
660 	pci_seg->rlookup_table = (void *)__get_free_pages(
661 						GFP_KERNEL | __GFP_ZERO,
662 						get_order(pci_seg->rlookup_table_size));
663 	if (pci_seg->rlookup_table == NULL)
664 		return -ENOMEM;
665 
666 	return 0;
667 }
668 
669 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
670 {
671 	free_pages((unsigned long)pci_seg->rlookup_table,
672 		   get_order(pci_seg->rlookup_table_size));
673 	pci_seg->rlookup_table = NULL;
674 }
675 
676 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
677 {
678 	pci_seg->irq_lookup_table = (void *)__get_free_pages(
679 					     GFP_KERNEL | __GFP_ZERO,
680 					     get_order(pci_seg->rlookup_table_size));
681 	kmemleak_alloc(pci_seg->irq_lookup_table,
682 		       pci_seg->rlookup_table_size, 1, GFP_KERNEL);
683 	if (pci_seg->irq_lookup_table == NULL)
684 		return -ENOMEM;
685 
686 	return 0;
687 }
688 
689 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
690 {
691 	kmemleak_free(pci_seg->irq_lookup_table);
692 	free_pages((unsigned long)pci_seg->irq_lookup_table,
693 		   get_order(pci_seg->rlookup_table_size));
694 	pci_seg->irq_lookup_table = NULL;
695 }
696 
697 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
698 {
699 	int i;
700 
701 	pci_seg->alias_table = (void *)__get_free_pages(GFP_KERNEL,
702 					get_order(pci_seg->alias_table_size));
703 	if (!pci_seg->alias_table)
704 		return -ENOMEM;
705 
706 	/*
707 	 * let all alias entries point to itself
708 	 */
709 	for (i = 0; i <= amd_iommu_last_bdf; ++i)
710 		pci_seg->alias_table[i] = i;
711 
712 	return 0;
713 }
714 
715 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
716 {
717 	free_pages((unsigned long)pci_seg->alias_table,
718 		   get_order(pci_seg->alias_table_size));
719 	pci_seg->alias_table = NULL;
720 }
721 
722 /*
723  * Allocates the command buffer. This buffer is per AMD IOMMU. We can
724  * write commands to that buffer later and the IOMMU will execute them
725  * asynchronously
726  */
727 static int __init alloc_command_buffer(struct amd_iommu *iommu)
728 {
729 	iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
730 						  get_order(CMD_BUFFER_SIZE));
731 
732 	return iommu->cmd_buf ? 0 : -ENOMEM;
733 }
734 
735 /*
736  * This function restarts event logging in case the IOMMU experienced
737  * an event log buffer overflow.
738  */
739 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
740 {
741 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
742 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
743 }
744 
745 /*
746  * This function resets the command buffer if the IOMMU stopped fetching
747  * commands from it.
748  */
749 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
750 {
751 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
752 
753 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
754 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
755 	iommu->cmd_buf_head = 0;
756 	iommu->cmd_buf_tail = 0;
757 
758 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
759 }
760 
761 /*
762  * This function writes the command buffer address to the hardware and
763  * enables it.
764  */
765 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
766 {
767 	u64 entry;
768 
769 	BUG_ON(iommu->cmd_buf == NULL);
770 
771 	entry = iommu_virt_to_phys(iommu->cmd_buf);
772 	entry |= MMIO_CMD_SIZE_512;
773 
774 	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
775 		    &entry, sizeof(entry));
776 
777 	amd_iommu_reset_cmd_buffer(iommu);
778 }
779 
780 /*
781  * This function disables the command buffer
782  */
783 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
784 {
785 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
786 }
787 
788 static void __init free_command_buffer(struct amd_iommu *iommu)
789 {
790 	free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
791 }
792 
793 static void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu,
794 					 gfp_t gfp, size_t size)
795 {
796 	int order = get_order(size);
797 	void *buf = (void *)__get_free_pages(gfp, order);
798 
799 	if (buf &&
800 	    iommu_feature(iommu, FEATURE_SNP) &&
801 	    set_memory_4k((unsigned long)buf, (1 << order))) {
802 		free_pages((unsigned long)buf, order);
803 		buf = NULL;
804 	}
805 
806 	return buf;
807 }
808 
809 /* allocates the memory where the IOMMU will log its events to */
810 static int __init alloc_event_buffer(struct amd_iommu *iommu)
811 {
812 	iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
813 					      EVT_BUFFER_SIZE);
814 
815 	return iommu->evt_buf ? 0 : -ENOMEM;
816 }
817 
818 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
819 {
820 	u64 entry;
821 
822 	BUG_ON(iommu->evt_buf == NULL);
823 
824 	entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
825 
826 	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
827 		    &entry, sizeof(entry));
828 
829 	/* set head and tail to zero manually */
830 	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
831 	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
832 
833 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
834 }
835 
836 /*
837  * This function disables the event log buffer
838  */
839 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
840 {
841 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
842 }
843 
844 static void __init free_event_buffer(struct amd_iommu *iommu)
845 {
846 	free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
847 }
848 
849 /* allocates the memory where the IOMMU will log its events to */
850 static int __init alloc_ppr_log(struct amd_iommu *iommu)
851 {
852 	iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
853 					      PPR_LOG_SIZE);
854 
855 	return iommu->ppr_log ? 0 : -ENOMEM;
856 }
857 
858 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
859 {
860 	u64 entry;
861 
862 	if (iommu->ppr_log == NULL)
863 		return;
864 
865 	entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
866 
867 	memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
868 		    &entry, sizeof(entry));
869 
870 	/* set head and tail to zero manually */
871 	writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
872 	writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
873 
874 	iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
875 	iommu_feature_enable(iommu, CONTROL_PPR_EN);
876 }
877 
878 static void __init free_ppr_log(struct amd_iommu *iommu)
879 {
880 	free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
881 }
882 
883 static void free_ga_log(struct amd_iommu *iommu)
884 {
885 #ifdef CONFIG_IRQ_REMAP
886 	free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE));
887 	free_pages((unsigned long)iommu->ga_log_tail, get_order(8));
888 #endif
889 }
890 
891 static int iommu_ga_log_enable(struct amd_iommu *iommu)
892 {
893 #ifdef CONFIG_IRQ_REMAP
894 	u32 status, i;
895 	u64 entry;
896 
897 	if (!iommu->ga_log)
898 		return -EINVAL;
899 
900 	/* Check if already running */
901 	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
902 	if (WARN_ON(status & (MMIO_STATUS_GALOG_RUN_MASK)))
903 		return 0;
904 
905 	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
906 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
907 		    &entry, sizeof(entry));
908 	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
909 		 (BIT_ULL(52)-1)) & ~7ULL;
910 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
911 		    &entry, sizeof(entry));
912 	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
913 	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
914 
915 
916 	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
917 	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
918 
919 	for (i = 0; i < LOOP_TIMEOUT; ++i) {
920 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
921 		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
922 			break;
923 		udelay(10);
924 	}
925 
926 	if (WARN_ON(i >= LOOP_TIMEOUT))
927 		return -EINVAL;
928 #endif /* CONFIG_IRQ_REMAP */
929 	return 0;
930 }
931 
932 static int iommu_init_ga_log(struct amd_iommu *iommu)
933 {
934 #ifdef CONFIG_IRQ_REMAP
935 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
936 		return 0;
937 
938 	iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
939 					get_order(GA_LOG_SIZE));
940 	if (!iommu->ga_log)
941 		goto err_out;
942 
943 	iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
944 					get_order(8));
945 	if (!iommu->ga_log_tail)
946 		goto err_out;
947 
948 	return 0;
949 err_out:
950 	free_ga_log(iommu);
951 	return -EINVAL;
952 #else
953 	return 0;
954 #endif /* CONFIG_IRQ_REMAP */
955 }
956 
957 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
958 {
959 	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 1);
960 
961 	return iommu->cmd_sem ? 0 : -ENOMEM;
962 }
963 
964 static void __init free_cwwb_sem(struct amd_iommu *iommu)
965 {
966 	if (iommu->cmd_sem)
967 		free_page((unsigned long)iommu->cmd_sem);
968 }
969 
970 static void iommu_enable_xt(struct amd_iommu *iommu)
971 {
972 #ifdef CONFIG_IRQ_REMAP
973 	/*
974 	 * XT mode (32-bit APIC destination ID) requires
975 	 * GA mode (128-bit IRTE support) as a prerequisite.
976 	 */
977 	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
978 	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
979 		iommu_feature_enable(iommu, CONTROL_XT_EN);
980 #endif /* CONFIG_IRQ_REMAP */
981 }
982 
983 static void iommu_enable_gt(struct amd_iommu *iommu)
984 {
985 	if (!iommu_feature(iommu, FEATURE_GT))
986 		return;
987 
988 	iommu_feature_enable(iommu, CONTROL_GT_EN);
989 }
990 
991 /* sets a specific bit in the device table entry. */
992 static void __set_dev_entry_bit(struct dev_table_entry *dev_table,
993 				u16 devid, u8 bit)
994 {
995 	int i = (bit >> 6) & 0x03;
996 	int _bit = bit & 0x3f;
997 
998 	dev_table[devid].data[i] |= (1UL << _bit);
999 }
1000 
1001 static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1002 {
1003 	struct dev_table_entry *dev_table = get_dev_table(iommu);
1004 
1005 	return __set_dev_entry_bit(dev_table, devid, bit);
1006 }
1007 
1008 static int __get_dev_entry_bit(struct dev_table_entry *dev_table,
1009 			       u16 devid, u8 bit)
1010 {
1011 	int i = (bit >> 6) & 0x03;
1012 	int _bit = bit & 0x3f;
1013 
1014 	return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1015 }
1016 
1017 static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1018 {
1019 	struct dev_table_entry *dev_table = get_dev_table(iommu);
1020 
1021 	return __get_dev_entry_bit(dev_table, devid, bit);
1022 }
1023 
1024 static bool __copy_device_table(struct amd_iommu *iommu)
1025 {
1026 	u64 int_ctl, int_tab_len, entry = 0;
1027 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1028 	struct dev_table_entry *old_devtb = NULL;
1029 	u32 lo, hi, devid, old_devtb_size;
1030 	phys_addr_t old_devtb_phys;
1031 	u16 dom_id, dte_v, irq_v;
1032 	gfp_t gfp_flag;
1033 	u64 tmp;
1034 
1035 	/* Each IOMMU use separate device table with the same size */
1036 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1037 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1038 	entry = (((u64) hi) << 32) + lo;
1039 
1040 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1041 	if (old_devtb_size != pci_seg->dev_table_size) {
1042 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1043 			iommu->index);
1044 		return false;
1045 	}
1046 
1047 	/*
1048 	 * When SME is enabled in the first kernel, the entry includes the
1049 	 * memory encryption mask(sme_me_mask), we must remove the memory
1050 	 * encryption mask to obtain the true physical address in kdump kernel.
1051 	 */
1052 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1053 
1054 	if (old_devtb_phys >= 0x100000000ULL) {
1055 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1056 		return false;
1057 	}
1058 	old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1059 		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
1060 							pci_seg->dev_table_size)
1061 		    : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB);
1062 
1063 	if (!old_devtb)
1064 		return false;
1065 
1066 	gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
1067 	pci_seg->old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
1068 						    get_order(pci_seg->dev_table_size));
1069 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1070 		pr_err("Failed to allocate memory for copying old device table!\n");
1071 		memunmap(old_devtb);
1072 		return false;
1073 	}
1074 
1075 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1076 		pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1077 		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1078 		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1079 
1080 		if (dte_v && dom_id) {
1081 			pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1082 			pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1083 			__set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
1084 			/* If gcr3 table existed, mask it out */
1085 			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1086 				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1087 				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1088 				pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1089 				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1090 				tmp |= DTE_FLAG_GV;
1091 				pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1092 			}
1093 		}
1094 
1095 		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1096 		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1097 		int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1098 		if (irq_v && (int_ctl || int_tab_len)) {
1099 			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1100 			    (int_tab_len != DTE_INTTABLEN)) {
1101 				pr_err("Wrong old irq remapping flag: %#x\n", devid);
1102 				memunmap(old_devtb);
1103 				return false;
1104 			}
1105 
1106 			pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1107 		}
1108 	}
1109 	memunmap(old_devtb);
1110 
1111 	return true;
1112 }
1113 
1114 static bool copy_device_table(void)
1115 {
1116 	struct amd_iommu *iommu;
1117 	struct amd_iommu_pci_seg *pci_seg;
1118 
1119 	if (!amd_iommu_pre_enabled)
1120 		return false;
1121 
1122 	pr_warn("Translation is already enabled - trying to copy translation structures\n");
1123 
1124 	/*
1125 	 * All IOMMUs within PCI segment shares common device table.
1126 	 * Hence copy device table only once per PCI segment.
1127 	 */
1128 	for_each_pci_segment(pci_seg) {
1129 		for_each_iommu(iommu) {
1130 			if (pci_seg->id != iommu->pci_seg->id)
1131 				continue;
1132 			if (!__copy_device_table(iommu))
1133 				return false;
1134 			break;
1135 		}
1136 	}
1137 
1138 	return true;
1139 }
1140 
1141 void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid)
1142 {
1143 	int sysmgt;
1144 
1145 	sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) |
1146 		 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1);
1147 
1148 	if (sysmgt == 0x01)
1149 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW);
1150 }
1151 
1152 /* Writes the specific IOMMU for a device into the rlookup table */
1153 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
1154 {
1155 	iommu->pci_seg->rlookup_table[devid] = iommu;
1156 }
1157 
1158 /*
1159  * This function takes the device specific flags read from the ACPI
1160  * table and sets up the device table entry with that information
1161  */
1162 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1163 					   u16 devid, u32 flags, u32 ext_flags)
1164 {
1165 	if (flags & ACPI_DEVFLAG_INITPASS)
1166 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS);
1167 	if (flags & ACPI_DEVFLAG_EXTINT)
1168 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS);
1169 	if (flags & ACPI_DEVFLAG_NMI)
1170 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS);
1171 	if (flags & ACPI_DEVFLAG_SYSMGT1)
1172 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1);
1173 	if (flags & ACPI_DEVFLAG_SYSMGT2)
1174 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2);
1175 	if (flags & ACPI_DEVFLAG_LINT0)
1176 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS);
1177 	if (flags & ACPI_DEVFLAG_LINT1)
1178 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS);
1179 
1180 	amd_iommu_apply_erratum_63(iommu, devid);
1181 
1182 	set_iommu_for_device(iommu, devid);
1183 }
1184 
1185 int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1186 {
1187 	struct devid_map *entry;
1188 	struct list_head *list;
1189 
1190 	if (type == IVHD_SPECIAL_IOAPIC)
1191 		list = &ioapic_map;
1192 	else if (type == IVHD_SPECIAL_HPET)
1193 		list = &hpet_map;
1194 	else
1195 		return -EINVAL;
1196 
1197 	list_for_each_entry(entry, list, list) {
1198 		if (!(entry->id == id && entry->cmd_line))
1199 			continue;
1200 
1201 		pr_info("Command-line override present for %s id %d - ignoring\n",
1202 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1203 
1204 		*devid = entry->devid;
1205 
1206 		return 0;
1207 	}
1208 
1209 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1210 	if (!entry)
1211 		return -ENOMEM;
1212 
1213 	entry->id	= id;
1214 	entry->devid	= *devid;
1215 	entry->cmd_line	= cmd_line;
1216 
1217 	list_add_tail(&entry->list, list);
1218 
1219 	return 0;
1220 }
1221 
1222 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1223 				      bool cmd_line)
1224 {
1225 	struct acpihid_map_entry *entry;
1226 	struct list_head *list = &acpihid_map;
1227 
1228 	list_for_each_entry(entry, list, list) {
1229 		if (strcmp(entry->hid, hid) ||
1230 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1231 		    !entry->cmd_line)
1232 			continue;
1233 
1234 		pr_info("Command-line override for hid:%s uid:%s\n",
1235 			hid, uid);
1236 		*devid = entry->devid;
1237 		return 0;
1238 	}
1239 
1240 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1241 	if (!entry)
1242 		return -ENOMEM;
1243 
1244 	memcpy(entry->uid, uid, strlen(uid));
1245 	memcpy(entry->hid, hid, strlen(hid));
1246 	entry->devid = *devid;
1247 	entry->cmd_line	= cmd_line;
1248 	entry->root_devid = (entry->devid & (~0x7));
1249 
1250 	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1251 		entry->cmd_line ? "cmd" : "ivrs",
1252 		entry->hid, entry->uid, entry->root_devid);
1253 
1254 	list_add_tail(&entry->list, list);
1255 	return 0;
1256 }
1257 
1258 static int __init add_early_maps(void)
1259 {
1260 	int i, ret;
1261 
1262 	for (i = 0; i < early_ioapic_map_size; ++i) {
1263 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1264 					 early_ioapic_map[i].id,
1265 					 &early_ioapic_map[i].devid,
1266 					 early_ioapic_map[i].cmd_line);
1267 		if (ret)
1268 			return ret;
1269 	}
1270 
1271 	for (i = 0; i < early_hpet_map_size; ++i) {
1272 		ret = add_special_device(IVHD_SPECIAL_HPET,
1273 					 early_hpet_map[i].id,
1274 					 &early_hpet_map[i].devid,
1275 					 early_hpet_map[i].cmd_line);
1276 		if (ret)
1277 			return ret;
1278 	}
1279 
1280 	for (i = 0; i < early_acpihid_map_size; ++i) {
1281 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1282 					  early_acpihid_map[i].uid,
1283 					  &early_acpihid_map[i].devid,
1284 					  early_acpihid_map[i].cmd_line);
1285 		if (ret)
1286 			return ret;
1287 	}
1288 
1289 	return 0;
1290 }
1291 
1292 /*
1293  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1294  * initializes the hardware and our data structures with it.
1295  */
1296 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1297 					struct ivhd_header *h)
1298 {
1299 	u8 *p = (u8 *)h;
1300 	u8 *end = p, flags = 0;
1301 	u16 devid = 0, devid_start = 0, devid_to = 0;
1302 	u32 dev_i, ext_flags = 0;
1303 	bool alias = false;
1304 	struct ivhd_entry *e;
1305 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1306 	u32 ivhd_size;
1307 	int ret;
1308 
1309 
1310 	ret = add_early_maps();
1311 	if (ret)
1312 		return ret;
1313 
1314 	amd_iommu_apply_ivrs_quirks();
1315 
1316 	/*
1317 	 * First save the recommended feature enable bits from ACPI
1318 	 */
1319 	iommu->acpi_flags = h->flags;
1320 
1321 	/*
1322 	 * Done. Now parse the device entries
1323 	 */
1324 	ivhd_size = get_ivhd_header_size(h);
1325 	if (!ivhd_size) {
1326 		pr_err("Unsupported IVHD type %#x\n", h->type);
1327 		return -EINVAL;
1328 	}
1329 
1330 	p += ivhd_size;
1331 
1332 	end += h->length;
1333 
1334 
1335 	while (p < end) {
1336 		e = (struct ivhd_entry *)p;
1337 		switch (e->type) {
1338 		case IVHD_DEV_ALL:
1339 
1340 			DUMP_printk("  DEV_ALL\t\t\tflags: %02x\n", e->flags);
1341 
1342 			for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1343 				set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1344 			break;
1345 		case IVHD_DEV_SELECT:
1346 
1347 			DUMP_printk("  DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1348 				    "flags: %02x\n",
1349 				    PCI_BUS_NUM(e->devid),
1350 				    PCI_SLOT(e->devid),
1351 				    PCI_FUNC(e->devid),
1352 				    e->flags);
1353 
1354 			devid = e->devid;
1355 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1356 			break;
1357 		case IVHD_DEV_SELECT_RANGE_START:
1358 
1359 			DUMP_printk("  DEV_SELECT_RANGE_START\t "
1360 				    "devid: %02x:%02x.%x flags: %02x\n",
1361 				    PCI_BUS_NUM(e->devid),
1362 				    PCI_SLOT(e->devid),
1363 				    PCI_FUNC(e->devid),
1364 				    e->flags);
1365 
1366 			devid_start = e->devid;
1367 			flags = e->flags;
1368 			ext_flags = 0;
1369 			alias = false;
1370 			break;
1371 		case IVHD_DEV_ALIAS:
1372 
1373 			DUMP_printk("  DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1374 				    "flags: %02x devid_to: %02x:%02x.%x\n",
1375 				    PCI_BUS_NUM(e->devid),
1376 				    PCI_SLOT(e->devid),
1377 				    PCI_FUNC(e->devid),
1378 				    e->flags,
1379 				    PCI_BUS_NUM(e->ext >> 8),
1380 				    PCI_SLOT(e->ext >> 8),
1381 				    PCI_FUNC(e->ext >> 8));
1382 
1383 			devid = e->devid;
1384 			devid_to = e->ext >> 8;
1385 			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1386 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1387 			pci_seg->alias_table[devid] = devid_to;
1388 			break;
1389 		case IVHD_DEV_ALIAS_RANGE:
1390 
1391 			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
1392 				    "devid: %02x:%02x.%x flags: %02x "
1393 				    "devid_to: %02x:%02x.%x\n",
1394 				    PCI_BUS_NUM(e->devid),
1395 				    PCI_SLOT(e->devid),
1396 				    PCI_FUNC(e->devid),
1397 				    e->flags,
1398 				    PCI_BUS_NUM(e->ext >> 8),
1399 				    PCI_SLOT(e->ext >> 8),
1400 				    PCI_FUNC(e->ext >> 8));
1401 
1402 			devid_start = e->devid;
1403 			flags = e->flags;
1404 			devid_to = e->ext >> 8;
1405 			ext_flags = 0;
1406 			alias = true;
1407 			break;
1408 		case IVHD_DEV_EXT_SELECT:
1409 
1410 			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1411 				    "flags: %02x ext: %08x\n",
1412 				    PCI_BUS_NUM(e->devid),
1413 				    PCI_SLOT(e->devid),
1414 				    PCI_FUNC(e->devid),
1415 				    e->flags, e->ext);
1416 
1417 			devid = e->devid;
1418 			set_dev_entry_from_acpi(iommu, devid, e->flags,
1419 						e->ext);
1420 			break;
1421 		case IVHD_DEV_EXT_SELECT_RANGE:
1422 
1423 			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
1424 				    "%02x:%02x.%x flags: %02x ext: %08x\n",
1425 				    PCI_BUS_NUM(e->devid),
1426 				    PCI_SLOT(e->devid),
1427 				    PCI_FUNC(e->devid),
1428 				    e->flags, e->ext);
1429 
1430 			devid_start = e->devid;
1431 			flags = e->flags;
1432 			ext_flags = e->ext;
1433 			alias = false;
1434 			break;
1435 		case IVHD_DEV_RANGE_END:
1436 
1437 			DUMP_printk("  DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1438 				    PCI_BUS_NUM(e->devid),
1439 				    PCI_SLOT(e->devid),
1440 				    PCI_FUNC(e->devid));
1441 
1442 			devid = e->devid;
1443 			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1444 				if (alias) {
1445 					pci_seg->alias_table[dev_i] = devid_to;
1446 					set_dev_entry_from_acpi(iommu,
1447 						devid_to, flags, ext_flags);
1448 				}
1449 				set_dev_entry_from_acpi(iommu, dev_i,
1450 							flags, ext_flags);
1451 			}
1452 			break;
1453 		case IVHD_DEV_SPECIAL: {
1454 			u8 handle, type;
1455 			const char *var;
1456 			u16 devid;
1457 			int ret;
1458 
1459 			handle = e->ext & 0xff;
1460 			devid  = (e->ext >>  8) & 0xffff;
1461 			type   = (e->ext >> 24) & 0xff;
1462 
1463 			if (type == IVHD_SPECIAL_IOAPIC)
1464 				var = "IOAPIC";
1465 			else if (type == IVHD_SPECIAL_HPET)
1466 				var = "HPET";
1467 			else
1468 				var = "UNKNOWN";
1469 
1470 			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1471 				    var, (int)handle,
1472 				    PCI_BUS_NUM(devid),
1473 				    PCI_SLOT(devid),
1474 				    PCI_FUNC(devid));
1475 
1476 			ret = add_special_device(type, handle, &devid, false);
1477 			if (ret)
1478 				return ret;
1479 
1480 			/*
1481 			 * add_special_device might update the devid in case a
1482 			 * command-line override is present. So call
1483 			 * set_dev_entry_from_acpi after add_special_device.
1484 			 */
1485 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1486 
1487 			break;
1488 		}
1489 		case IVHD_DEV_ACPI_HID: {
1490 			u16 devid;
1491 			u8 hid[ACPIHID_HID_LEN];
1492 			u8 uid[ACPIHID_UID_LEN];
1493 			int ret;
1494 
1495 			if (h->type != 0x40) {
1496 				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1497 				       e->type);
1498 				break;
1499 			}
1500 
1501 			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1502 			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1503 			hid[ACPIHID_HID_LEN - 1] = '\0';
1504 
1505 			if (!(*hid)) {
1506 				pr_err(FW_BUG "Invalid HID.\n");
1507 				break;
1508 			}
1509 
1510 			uid[0] = '\0';
1511 			switch (e->uidf) {
1512 			case UID_NOT_PRESENT:
1513 
1514 				if (e->uidl != 0)
1515 					pr_warn(FW_BUG "Invalid UID length.\n");
1516 
1517 				break;
1518 			case UID_IS_INTEGER:
1519 
1520 				sprintf(uid, "%d", e->uid);
1521 
1522 				break;
1523 			case UID_IS_CHARACTER:
1524 
1525 				memcpy(uid, &e->uid, e->uidl);
1526 				uid[e->uidl] = '\0';
1527 
1528 				break;
1529 			default:
1530 				break;
1531 			}
1532 
1533 			devid = e->devid;
1534 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1535 				    hid, uid,
1536 				    PCI_BUS_NUM(devid),
1537 				    PCI_SLOT(devid),
1538 				    PCI_FUNC(devid));
1539 
1540 			flags = e->flags;
1541 
1542 			ret = add_acpi_hid_device(hid, uid, &devid, false);
1543 			if (ret)
1544 				return ret;
1545 
1546 			/*
1547 			 * add_special_device might update the devid in case a
1548 			 * command-line override is present. So call
1549 			 * set_dev_entry_from_acpi after add_special_device.
1550 			 */
1551 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1552 
1553 			break;
1554 		}
1555 		default:
1556 			break;
1557 		}
1558 
1559 		p += ivhd_entry_length(p);
1560 	}
1561 
1562 	return 0;
1563 }
1564 
1565 /* Allocate PCI segment data structure */
1566 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1567 					  struct acpi_table_header *ivrs_base)
1568 {
1569 	struct amd_iommu_pci_seg *pci_seg;
1570 	int last_bdf;
1571 
1572 	/*
1573 	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1574 	 * handle in this PCI segment. Upon this information the shared data
1575 	 * structures for the PCI segments in the system will be allocated.
1576 	 */
1577 	last_bdf = find_last_devid_acpi(ivrs_base, id);
1578 	if (last_bdf < 0)
1579 		return NULL;
1580 
1581 	pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1582 	if (pci_seg == NULL)
1583 		return NULL;
1584 
1585 	pci_seg->last_bdf = last_bdf;
1586 	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1587 	pci_seg->dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE);
1588 	pci_seg->alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1589 	pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1590 
1591 	pci_seg->id = id;
1592 	init_llist_head(&pci_seg->dev_data_list);
1593 	INIT_LIST_HEAD(&pci_seg->unity_map);
1594 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1595 
1596 	if (alloc_dev_table(pci_seg))
1597 		return NULL;
1598 	if (alloc_alias_table(pci_seg))
1599 		return NULL;
1600 	if (alloc_rlookup_table(pci_seg))
1601 		return NULL;
1602 
1603 	return pci_seg;
1604 }
1605 
1606 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1607 					struct acpi_table_header *ivrs_base)
1608 {
1609 	struct amd_iommu_pci_seg *pci_seg;
1610 
1611 	for_each_pci_segment(pci_seg) {
1612 		if (pci_seg->id == id)
1613 			return pci_seg;
1614 	}
1615 
1616 	return alloc_pci_segment(id, ivrs_base);
1617 }
1618 
1619 static void __init free_pci_segments(void)
1620 {
1621 	struct amd_iommu_pci_seg *pci_seg, *next;
1622 
1623 	for_each_pci_segment_safe(pci_seg, next) {
1624 		list_del(&pci_seg->list);
1625 		free_irq_lookup_table(pci_seg);
1626 		free_rlookup_table(pci_seg);
1627 		free_alias_table(pci_seg);
1628 		free_dev_table(pci_seg);
1629 		kfree(pci_seg);
1630 	}
1631 }
1632 
1633 static void __init free_iommu_one(struct amd_iommu *iommu)
1634 {
1635 	free_cwwb_sem(iommu);
1636 	free_command_buffer(iommu);
1637 	free_event_buffer(iommu);
1638 	free_ppr_log(iommu);
1639 	free_ga_log(iommu);
1640 	iommu_unmap_mmio_space(iommu);
1641 }
1642 
1643 static void __init free_iommu_all(void)
1644 {
1645 	struct amd_iommu *iommu, *next;
1646 
1647 	for_each_iommu_safe(iommu, next) {
1648 		list_del(&iommu->list);
1649 		free_iommu_one(iommu);
1650 		kfree(iommu);
1651 	}
1652 }
1653 
1654 /*
1655  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1656  * Workaround:
1657  *     BIOS should disable L2B micellaneous clock gating by setting
1658  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1659  */
1660 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1661 {
1662 	u32 value;
1663 
1664 	if ((boot_cpu_data.x86 != 0x15) ||
1665 	    (boot_cpu_data.x86_model < 0x10) ||
1666 	    (boot_cpu_data.x86_model > 0x1f))
1667 		return;
1668 
1669 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1670 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1671 
1672 	if (value & BIT(2))
1673 		return;
1674 
1675 	/* Select NB indirect register 0x90 and enable writing */
1676 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1677 
1678 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1679 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1680 
1681 	/* Clear the enable writing bit */
1682 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1683 }
1684 
1685 /*
1686  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1687  * Workaround:
1688  *     BIOS should enable ATS write permission check by setting
1689  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1690  */
1691 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1692 {
1693 	u32 value;
1694 
1695 	if ((boot_cpu_data.x86 != 0x15) ||
1696 	    (boot_cpu_data.x86_model < 0x30) ||
1697 	    (boot_cpu_data.x86_model > 0x3f))
1698 		return;
1699 
1700 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1701 	value = iommu_read_l2(iommu, 0x47);
1702 
1703 	if (value & BIT(0))
1704 		return;
1705 
1706 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1707 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1708 
1709 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1710 }
1711 
1712 /*
1713  * This function glues the initialization function for one IOMMU
1714  * together and also allocates the command buffer and programs the
1715  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1716  */
1717 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1718 				 struct acpi_table_header *ivrs_base)
1719 {
1720 	struct amd_iommu_pci_seg *pci_seg;
1721 	int ret;
1722 
1723 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1724 	if (pci_seg == NULL)
1725 		return -ENOMEM;
1726 	iommu->pci_seg = pci_seg;
1727 
1728 	raw_spin_lock_init(&iommu->lock);
1729 	iommu->cmd_sem_val = 0;
1730 
1731 	/* Add IOMMU to internal data structures */
1732 	list_add_tail(&iommu->list, &amd_iommu_list);
1733 	iommu->index = amd_iommus_present++;
1734 
1735 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1736 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1737 		return -ENOSYS;
1738 	}
1739 
1740 	/* Index is fine - add IOMMU to the array */
1741 	amd_iommus[iommu->index] = iommu;
1742 
1743 	/*
1744 	 * Copy data from ACPI table entry to the iommu struct
1745 	 */
1746 	iommu->devid   = h->devid;
1747 	iommu->cap_ptr = h->cap_ptr;
1748 	iommu->mmio_phys = h->mmio_phys;
1749 
1750 	switch (h->type) {
1751 	case 0x10:
1752 		/* Check if IVHD EFR contains proper max banks/counters */
1753 		if ((h->efr_attr != 0) &&
1754 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1755 		    ((h->efr_attr & (0x3F << 17)) != 0))
1756 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1757 		else
1758 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1759 
1760 		/*
1761 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1762 		 * GAM also requires GA mode. Therefore, we need to
1763 		 * check cmpxchg16b support before enabling it.
1764 		 */
1765 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1766 		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1767 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1768 		break;
1769 	case 0x11:
1770 	case 0x40:
1771 		if (h->efr_reg & (1 << 9))
1772 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1773 		else
1774 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1775 
1776 		/*
1777 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1778 		 * XT, GAM also requires GA mode. Therefore, we need to
1779 		 * check cmpxchg16b support before enabling them.
1780 		 */
1781 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1782 		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1783 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1784 			break;
1785 		}
1786 
1787 		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1788 			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1789 
1790 		early_iommu_features_init(iommu, h);
1791 
1792 		break;
1793 	default:
1794 		return -EINVAL;
1795 	}
1796 
1797 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1798 						iommu->mmio_phys_end);
1799 	if (!iommu->mmio_base)
1800 		return -ENOMEM;
1801 
1802 	if (alloc_cwwb_sem(iommu))
1803 		return -ENOMEM;
1804 
1805 	if (alloc_command_buffer(iommu))
1806 		return -ENOMEM;
1807 
1808 	if (alloc_event_buffer(iommu))
1809 		return -ENOMEM;
1810 
1811 	iommu->int_enabled = false;
1812 
1813 	init_translation_status(iommu);
1814 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1815 		iommu_disable(iommu);
1816 		clear_translation_pre_enabled(iommu);
1817 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1818 			iommu->index);
1819 	}
1820 	if (amd_iommu_pre_enabled)
1821 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1822 
1823 	ret = init_iommu_from_acpi(iommu, h);
1824 	if (ret)
1825 		return ret;
1826 
1827 	if (amd_iommu_irq_remap) {
1828 		ret = amd_iommu_create_irq_domain(iommu);
1829 		if (ret)
1830 			return ret;
1831 	}
1832 
1833 	/*
1834 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1835 	 * table tells us so, but this is a lie!
1836 	 */
1837 	pci_seg->rlookup_table[iommu->devid] = NULL;
1838 
1839 	return 0;
1840 }
1841 
1842 /**
1843  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1844  * @ivrs: Pointer to the IVRS header
1845  *
1846  * This function search through all IVDB of the maximum supported IVHD
1847  */
1848 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1849 {
1850 	u8 *base = (u8 *)ivrs;
1851 	struct ivhd_header *ivhd = (struct ivhd_header *)
1852 					(base + IVRS_HEADER_LENGTH);
1853 	u8 last_type = ivhd->type;
1854 	u16 devid = ivhd->devid;
1855 
1856 	while (((u8 *)ivhd - base < ivrs->length) &&
1857 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1858 		u8 *p = (u8 *) ivhd;
1859 
1860 		if (ivhd->devid == devid)
1861 			last_type = ivhd->type;
1862 		ivhd = (struct ivhd_header *)(p + ivhd->length);
1863 	}
1864 
1865 	return last_type;
1866 }
1867 
1868 /*
1869  * Iterates over all IOMMU entries in the ACPI table, allocates the
1870  * IOMMU structure and initializes it with init_iommu_one()
1871  */
1872 static int __init init_iommu_all(struct acpi_table_header *table)
1873 {
1874 	u8 *p = (u8 *)table, *end = (u8 *)table;
1875 	struct ivhd_header *h;
1876 	struct amd_iommu *iommu;
1877 	int ret;
1878 
1879 	end += table->length;
1880 	p += IVRS_HEADER_LENGTH;
1881 
1882 	while (p < end) {
1883 		h = (struct ivhd_header *)p;
1884 		if (*p == amd_iommu_target_ivhd_type) {
1885 
1886 			DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1887 				    "seg: %d flags: %01x info %04x\n",
1888 				    PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1889 				    PCI_FUNC(h->devid), h->cap_ptr,
1890 				    h->pci_seg, h->flags, h->info);
1891 			DUMP_printk("       mmio-addr: %016llx\n",
1892 				    h->mmio_phys);
1893 
1894 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1895 			if (iommu == NULL)
1896 				return -ENOMEM;
1897 
1898 			ret = init_iommu_one(iommu, h, table);
1899 			if (ret)
1900 				return ret;
1901 		}
1902 		p += h->length;
1903 
1904 	}
1905 	WARN_ON(p != end);
1906 
1907 	return 0;
1908 }
1909 
1910 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1911 {
1912 	u64 val;
1913 	struct pci_dev *pdev = iommu->dev;
1914 
1915 	if (!iommu_feature(iommu, FEATURE_PC))
1916 		return;
1917 
1918 	amd_iommu_pc_present = true;
1919 
1920 	pci_info(pdev, "IOMMU performance counters supported\n");
1921 
1922 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1923 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1924 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1925 
1926 	return;
1927 }
1928 
1929 static ssize_t amd_iommu_show_cap(struct device *dev,
1930 				  struct device_attribute *attr,
1931 				  char *buf)
1932 {
1933 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1934 	return sprintf(buf, "%x\n", iommu->cap);
1935 }
1936 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1937 
1938 static ssize_t amd_iommu_show_features(struct device *dev,
1939 				       struct device_attribute *attr,
1940 				       char *buf)
1941 {
1942 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1943 	return sprintf(buf, "%llx\n", iommu->features);
1944 }
1945 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1946 
1947 static struct attribute *amd_iommu_attrs[] = {
1948 	&dev_attr_cap.attr,
1949 	&dev_attr_features.attr,
1950 	NULL,
1951 };
1952 
1953 static struct attribute_group amd_iommu_group = {
1954 	.name = "amd-iommu",
1955 	.attrs = amd_iommu_attrs,
1956 };
1957 
1958 static const struct attribute_group *amd_iommu_groups[] = {
1959 	&amd_iommu_group,
1960 	NULL,
1961 };
1962 
1963 /*
1964  * Note: IVHD 0x11 and 0x40 also contains exact copy
1965  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1966  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1967  */
1968 static void __init late_iommu_features_init(struct amd_iommu *iommu)
1969 {
1970 	u64 features;
1971 
1972 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1973 		return;
1974 
1975 	/* read extended feature bits */
1976 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1977 
1978 	if (!iommu->features) {
1979 		iommu->features = features;
1980 		return;
1981 	}
1982 
1983 	/*
1984 	 * Sanity check and warn if EFR values from
1985 	 * IVHD and MMIO conflict.
1986 	 */
1987 	if (features != iommu->features)
1988 		pr_warn(FW_WARN "EFR mismatch. Use IVHD EFR (%#llx : %#llx).\n",
1989 			features, iommu->features);
1990 }
1991 
1992 static int __init iommu_init_pci(struct amd_iommu *iommu)
1993 {
1994 	int cap_ptr = iommu->cap_ptr;
1995 	int ret;
1996 
1997 	iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1998 						 iommu->devid & 0xff);
1999 	if (!iommu->dev)
2000 		return -ENODEV;
2001 
2002 	/* Prevent binding other PCI device drivers to IOMMU devices */
2003 	iommu->dev->match_driver = false;
2004 
2005 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2006 			      &iommu->cap);
2007 
2008 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2009 		amd_iommu_iotlb_sup = false;
2010 
2011 	late_iommu_features_init(iommu);
2012 
2013 	if (iommu_feature(iommu, FEATURE_GT)) {
2014 		int glxval;
2015 		u32 max_pasid;
2016 		u64 pasmax;
2017 
2018 		pasmax = iommu->features & FEATURE_PASID_MASK;
2019 		pasmax >>= FEATURE_PASID_SHIFT;
2020 		max_pasid  = (1 << (pasmax + 1)) - 1;
2021 
2022 		amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
2023 
2024 		BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
2025 
2026 		glxval   = iommu->features & FEATURE_GLXVAL_MASK;
2027 		glxval >>= FEATURE_GLXVAL_SHIFT;
2028 
2029 		if (amd_iommu_max_glx_val == -1)
2030 			amd_iommu_max_glx_val = glxval;
2031 		else
2032 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2033 	}
2034 
2035 	if (iommu_feature(iommu, FEATURE_GT) &&
2036 	    iommu_feature(iommu, FEATURE_PPR)) {
2037 		iommu->is_iommu_v2   = true;
2038 		amd_iommu_v2_present = true;
2039 	}
2040 
2041 	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
2042 		return -ENOMEM;
2043 
2044 	ret = iommu_init_ga_log(iommu);
2045 	if (ret)
2046 		return ret;
2047 
2048 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2049 		pr_info("Using strict mode due to virtualization\n");
2050 		iommu_set_dma_strict();
2051 		amd_iommu_np_cache = true;
2052 	}
2053 
2054 	init_iommu_perf_ctr(iommu);
2055 
2056 	if (is_rd890_iommu(iommu->dev)) {
2057 		int i, j;
2058 
2059 		iommu->root_pdev =
2060 			pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
2061 						    PCI_DEVFN(0, 0));
2062 
2063 		/*
2064 		 * Some rd890 systems may not be fully reconfigured by the
2065 		 * BIOS, so it's necessary for us to store this information so
2066 		 * it can be reprogrammed on resume
2067 		 */
2068 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2069 				&iommu->stored_addr_lo);
2070 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2071 				&iommu->stored_addr_hi);
2072 
2073 		/* Low bit locks writes to configuration space */
2074 		iommu->stored_addr_lo &= ~1;
2075 
2076 		for (i = 0; i < 6; i++)
2077 			for (j = 0; j < 0x12; j++)
2078 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2079 
2080 		for (i = 0; i < 0x83; i++)
2081 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2082 	}
2083 
2084 	amd_iommu_erratum_746_workaround(iommu);
2085 	amd_iommu_ats_write_check_workaround(iommu);
2086 
2087 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2088 			       amd_iommu_groups, "ivhd%d", iommu->index);
2089 	if (ret)
2090 		return ret;
2091 
2092 	iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2093 
2094 	return pci_enable_device(iommu->dev);
2095 }
2096 
2097 static void print_iommu_info(void)
2098 {
2099 	static const char * const feat_str[] = {
2100 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2101 		"IA", "GA", "HE", "PC"
2102 	};
2103 	struct amd_iommu *iommu;
2104 
2105 	for_each_iommu(iommu) {
2106 		struct pci_dev *pdev = iommu->dev;
2107 		int i;
2108 
2109 		pci_info(pdev, "Found IOMMU cap 0x%x\n", iommu->cap_ptr);
2110 
2111 		if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
2112 			pr_info("Extended features (%#llx):", iommu->features);
2113 
2114 			for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2115 				if (iommu_feature(iommu, (1ULL << i)))
2116 					pr_cont(" %s", feat_str[i]);
2117 			}
2118 
2119 			if (iommu->features & FEATURE_GAM_VAPIC)
2120 				pr_cont(" GA_vAPIC");
2121 
2122 			pr_cont("\n");
2123 		}
2124 	}
2125 	if (irq_remapping_enabled) {
2126 		pr_info("Interrupt remapping enabled\n");
2127 		if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2128 			pr_info("Virtual APIC enabled\n");
2129 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2130 			pr_info("X2APIC enabled\n");
2131 	}
2132 }
2133 
2134 static int __init amd_iommu_init_pci(void)
2135 {
2136 	struct amd_iommu *iommu;
2137 	struct amd_iommu_pci_seg *pci_seg;
2138 	int ret;
2139 
2140 	for_each_iommu(iommu) {
2141 		ret = iommu_init_pci(iommu);
2142 		if (ret) {
2143 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2144 			       iommu->index, ret);
2145 			goto out;
2146 		}
2147 		/* Need to setup range after PCI init */
2148 		iommu_set_cwwb_range(iommu);
2149 	}
2150 
2151 	/*
2152 	 * Order is important here to make sure any unity map requirements are
2153 	 * fulfilled. The unity mappings are created and written to the device
2154 	 * table during the amd_iommu_init_api() call.
2155 	 *
2156 	 * After that we call init_device_table_dma() to make sure any
2157 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2158 	 * of all IOMMUs to make sure the changes to the device table are
2159 	 * active.
2160 	 */
2161 	ret = amd_iommu_init_api();
2162 	if (ret) {
2163 		pr_err("IOMMU: Failed to initialize IOMMU-API interface (error=%d)!\n",
2164 		       ret);
2165 		goto out;
2166 	}
2167 
2168 	for_each_pci_segment(pci_seg)
2169 		init_device_table_dma(pci_seg);
2170 
2171 	for_each_iommu(iommu)
2172 		iommu_flush_all_caches(iommu);
2173 
2174 	print_iommu_info();
2175 
2176 out:
2177 	return ret;
2178 }
2179 
2180 /****************************************************************************
2181  *
2182  * The following functions initialize the MSI interrupts for all IOMMUs
2183  * in the system. It's a bit challenging because there could be multiple
2184  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2185  * pci_dev.
2186  *
2187  ****************************************************************************/
2188 
2189 static int iommu_setup_msi(struct amd_iommu *iommu)
2190 {
2191 	int r;
2192 
2193 	r = pci_enable_msi(iommu->dev);
2194 	if (r)
2195 		return r;
2196 
2197 	r = request_threaded_irq(iommu->dev->irq,
2198 				 amd_iommu_int_handler,
2199 				 amd_iommu_int_thread,
2200 				 0, "AMD-Vi",
2201 				 iommu);
2202 
2203 	if (r) {
2204 		pci_disable_msi(iommu->dev);
2205 		return r;
2206 	}
2207 
2208 	return 0;
2209 }
2210 
2211 union intcapxt {
2212 	u64	capxt;
2213 	struct {
2214 		u64	reserved_0		:  2,
2215 			dest_mode_logical	:  1,
2216 			reserved_1		:  5,
2217 			destid_0_23		: 24,
2218 			vector			:  8,
2219 			reserved_2		: 16,
2220 			destid_24_31		:  8;
2221 	};
2222 } __attribute__ ((packed));
2223 
2224 
2225 static struct irq_chip intcapxt_controller;
2226 
2227 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2228 				       struct irq_data *irqd, bool reserve)
2229 {
2230 	return 0;
2231 }
2232 
2233 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2234 					  struct irq_data *irqd)
2235 {
2236 }
2237 
2238 
2239 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2240 				    unsigned int nr_irqs, void *arg)
2241 {
2242 	struct irq_alloc_info *info = arg;
2243 	int i, ret;
2244 
2245 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2246 		return -EINVAL;
2247 
2248 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2249 	if (ret < 0)
2250 		return ret;
2251 
2252 	for (i = virq; i < virq + nr_irqs; i++) {
2253 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2254 
2255 		irqd->chip = &intcapxt_controller;
2256 		irqd->chip_data = info->data;
2257 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2258 	}
2259 
2260 	return ret;
2261 }
2262 
2263 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2264 				    unsigned int nr_irqs)
2265 {
2266 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2267 }
2268 
2269 
2270 static void intcapxt_unmask_irq(struct irq_data *irqd)
2271 {
2272 	struct amd_iommu *iommu = irqd->chip_data;
2273 	struct irq_cfg *cfg = irqd_cfg(irqd);
2274 	union intcapxt xt;
2275 
2276 	xt.capxt = 0ULL;
2277 	xt.dest_mode_logical = apic->dest_mode_logical;
2278 	xt.vector = cfg->vector;
2279 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2280 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2281 
2282 	/**
2283 	 * Current IOMMU implementation uses the same IRQ for all
2284 	 * 3 IOMMU interrupts.
2285 	 */
2286 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2287 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2288 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2289 }
2290 
2291 static void intcapxt_mask_irq(struct irq_data *irqd)
2292 {
2293 	struct amd_iommu *iommu = irqd->chip_data;
2294 
2295 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2296 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2297 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2298 }
2299 
2300 
2301 static int intcapxt_set_affinity(struct irq_data *irqd,
2302 				 const struct cpumask *mask, bool force)
2303 {
2304 	struct irq_data *parent = irqd->parent_data;
2305 	int ret;
2306 
2307 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2308 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2309 		return ret;
2310 	return 0;
2311 }
2312 
2313 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2314 {
2315 	return on ? -EOPNOTSUPP : 0;
2316 }
2317 
2318 static struct irq_chip intcapxt_controller = {
2319 	.name			= "IOMMU-MSI",
2320 	.irq_unmask		= intcapxt_unmask_irq,
2321 	.irq_mask		= intcapxt_mask_irq,
2322 	.irq_ack		= irq_chip_ack_parent,
2323 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2324 	.irq_set_affinity       = intcapxt_set_affinity,
2325 	.irq_set_wake		= intcapxt_set_wake,
2326 	.flags			= IRQCHIP_MASK_ON_SUSPEND,
2327 };
2328 
2329 static const struct irq_domain_ops intcapxt_domain_ops = {
2330 	.alloc			= intcapxt_irqdomain_alloc,
2331 	.free			= intcapxt_irqdomain_free,
2332 	.activate		= intcapxt_irqdomain_activate,
2333 	.deactivate		= intcapxt_irqdomain_deactivate,
2334 };
2335 
2336 
2337 static struct irq_domain *iommu_irqdomain;
2338 
2339 static struct irq_domain *iommu_get_irqdomain(void)
2340 {
2341 	struct fwnode_handle *fn;
2342 
2343 	/* No need for locking here (yet) as the init is single-threaded */
2344 	if (iommu_irqdomain)
2345 		return iommu_irqdomain;
2346 
2347 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2348 	if (!fn)
2349 		return NULL;
2350 
2351 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2352 						      fn, &intcapxt_domain_ops,
2353 						      NULL);
2354 	if (!iommu_irqdomain)
2355 		irq_domain_free_fwnode(fn);
2356 
2357 	return iommu_irqdomain;
2358 }
2359 
2360 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2361 {
2362 	struct irq_domain *domain;
2363 	struct irq_alloc_info info;
2364 	int irq, ret;
2365 
2366 	domain = iommu_get_irqdomain();
2367 	if (!domain)
2368 		return -ENXIO;
2369 
2370 	init_irq_alloc_info(&info, NULL);
2371 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2372 	info.data = iommu;
2373 
2374 	irq = irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
2375 	if (irq < 0) {
2376 		irq_domain_remove(domain);
2377 		return irq;
2378 	}
2379 
2380 	ret = request_threaded_irq(irq, amd_iommu_int_handler,
2381 				   amd_iommu_int_thread, 0, "AMD-Vi", iommu);
2382 	if (ret) {
2383 		irq_domain_free_irqs(irq, 1);
2384 		irq_domain_remove(domain);
2385 		return ret;
2386 	}
2387 
2388 	return 0;
2389 }
2390 
2391 static int iommu_init_irq(struct amd_iommu *iommu)
2392 {
2393 	int ret;
2394 
2395 	if (iommu->int_enabled)
2396 		goto enable_faults;
2397 
2398 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2399 		ret = iommu_setup_intcapxt(iommu);
2400 	else if (iommu->dev->msi_cap)
2401 		ret = iommu_setup_msi(iommu);
2402 	else
2403 		ret = -ENODEV;
2404 
2405 	if (ret)
2406 		return ret;
2407 
2408 	iommu->int_enabled = true;
2409 enable_faults:
2410 
2411 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2412 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2413 
2414 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2415 
2416 	if (iommu->ppr_log != NULL)
2417 		iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
2418 
2419 	iommu_ga_log_enable(iommu);
2420 
2421 	return 0;
2422 }
2423 
2424 /****************************************************************************
2425  *
2426  * The next functions belong to the third pass of parsing the ACPI
2427  * table. In this last pass the memory mapping requirements are
2428  * gathered (like exclusion and unity mapping ranges).
2429  *
2430  ****************************************************************************/
2431 
2432 static void __init free_unity_maps(void)
2433 {
2434 	struct unity_map_entry *entry, *next;
2435 	struct amd_iommu_pci_seg *p, *pci_seg;
2436 
2437 	for_each_pci_segment_safe(pci_seg, p) {
2438 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2439 			list_del(&entry->list);
2440 			kfree(entry);
2441 		}
2442 	}
2443 }
2444 
2445 /* called for unity map ACPI definition */
2446 static int __init init_unity_map_range(struct ivmd_header *m,
2447 				       struct acpi_table_header *ivrs_base)
2448 {
2449 	struct unity_map_entry *e = NULL;
2450 	struct amd_iommu_pci_seg *pci_seg;
2451 	char *s;
2452 
2453 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2454 	if (pci_seg == NULL)
2455 		return -ENOMEM;
2456 
2457 	e = kzalloc(sizeof(*e), GFP_KERNEL);
2458 	if (e == NULL)
2459 		return -ENOMEM;
2460 
2461 	switch (m->type) {
2462 	default:
2463 		kfree(e);
2464 		return 0;
2465 	case ACPI_IVMD_TYPE:
2466 		s = "IVMD_TYPEi\t\t\t";
2467 		e->devid_start = e->devid_end = m->devid;
2468 		break;
2469 	case ACPI_IVMD_TYPE_ALL:
2470 		s = "IVMD_TYPE_ALL\t\t";
2471 		e->devid_start = 0;
2472 		e->devid_end = amd_iommu_last_bdf;
2473 		break;
2474 	case ACPI_IVMD_TYPE_RANGE:
2475 		s = "IVMD_TYPE_RANGE\t\t";
2476 		e->devid_start = m->devid;
2477 		e->devid_end = m->aux;
2478 		break;
2479 	}
2480 	e->address_start = PAGE_ALIGN(m->range_start);
2481 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2482 	e->prot = m->flags >> 1;
2483 
2484 	/*
2485 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2486 	 * since some buggy BIOSes might lead to the overwritten exclusion
2487 	 * range (exclusion_start and exclusion_length members). This
2488 	 * happens when there are multiple exclusion ranges (IVMD entries)
2489 	 * defined in ACPI table.
2490 	 */
2491 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2492 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2493 
2494 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2495 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2496 		    " flags: %x\n", s, m->pci_seg,
2497 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2498 		    PCI_FUNC(e->devid_start), m->pci_seg,
2499 		    PCI_BUS_NUM(e->devid_end),
2500 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2501 		    e->address_start, e->address_end, m->flags);
2502 
2503 	list_add_tail(&e->list, &pci_seg->unity_map);
2504 
2505 	return 0;
2506 }
2507 
2508 /* iterates over all memory definitions we find in the ACPI table */
2509 static int __init init_memory_definitions(struct acpi_table_header *table)
2510 {
2511 	u8 *p = (u8 *)table, *end = (u8 *)table;
2512 	struct ivmd_header *m;
2513 
2514 	end += table->length;
2515 	p += IVRS_HEADER_LENGTH;
2516 
2517 	while (p < end) {
2518 		m = (struct ivmd_header *)p;
2519 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2520 			init_unity_map_range(m, table);
2521 
2522 		p += m->length;
2523 	}
2524 
2525 	return 0;
2526 }
2527 
2528 /*
2529  * Init the device table to not allow DMA access for devices
2530  */
2531 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2532 {
2533 	u32 devid;
2534 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2535 
2536 	if (dev_table == NULL)
2537 		return;
2538 
2539 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2540 		__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID);
2541 		__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION);
2542 	}
2543 }
2544 
2545 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2546 {
2547 	u32 devid;
2548 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2549 
2550 	if (dev_table == NULL)
2551 		return;
2552 
2553 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2554 		dev_table[devid].data[0] = 0ULL;
2555 		dev_table[devid].data[1] = 0ULL;
2556 	}
2557 }
2558 
2559 static void init_device_table(void)
2560 {
2561 	struct amd_iommu_pci_seg *pci_seg;
2562 	u32 devid;
2563 
2564 	if (!amd_iommu_irq_remap)
2565 		return;
2566 
2567 	for_each_pci_segment(pci_seg) {
2568 		for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2569 			__set_dev_entry_bit(pci_seg->dev_table,
2570 					    devid, DEV_ENTRY_IRQ_TBL_EN);
2571 	}
2572 }
2573 
2574 static void iommu_init_flags(struct amd_iommu *iommu)
2575 {
2576 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2577 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2578 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2579 
2580 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2581 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2582 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2583 
2584 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2585 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2586 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2587 
2588 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2589 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2590 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2591 
2592 	/*
2593 	 * make IOMMU memory accesses cache coherent
2594 	 */
2595 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2596 
2597 	/* Set IOTLB invalidation timeout to 1s */
2598 	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2599 }
2600 
2601 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2602 {
2603 	int i, j;
2604 	u32 ioc_feature_control;
2605 	struct pci_dev *pdev = iommu->root_pdev;
2606 
2607 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2608 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2609 		return;
2610 
2611 	/*
2612 	 * First, we need to ensure that the iommu is enabled. This is
2613 	 * controlled by a register in the northbridge
2614 	 */
2615 
2616 	/* Select Northbridge indirect register 0x75 and enable writing */
2617 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2618 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2619 
2620 	/* Enable the iommu */
2621 	if (!(ioc_feature_control & 0x1))
2622 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2623 
2624 	/* Restore the iommu BAR */
2625 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2626 			       iommu->stored_addr_lo);
2627 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2628 			       iommu->stored_addr_hi);
2629 
2630 	/* Restore the l1 indirect regs for each of the 6 l1s */
2631 	for (i = 0; i < 6; i++)
2632 		for (j = 0; j < 0x12; j++)
2633 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2634 
2635 	/* Restore the l2 indirect regs */
2636 	for (i = 0; i < 0x83; i++)
2637 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2638 
2639 	/* Lock PCI setup registers */
2640 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2641 			       iommu->stored_addr_lo | 1);
2642 }
2643 
2644 static void iommu_enable_ga(struct amd_iommu *iommu)
2645 {
2646 #ifdef CONFIG_IRQ_REMAP
2647 	switch (amd_iommu_guest_ir) {
2648 	case AMD_IOMMU_GUEST_IR_VAPIC:
2649 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2650 		fallthrough;
2651 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2652 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2653 		iommu->irte_ops = &irte_128_ops;
2654 		break;
2655 	default:
2656 		iommu->irte_ops = &irte_32_ops;
2657 		break;
2658 	}
2659 #endif
2660 }
2661 
2662 static void early_enable_iommu(struct amd_iommu *iommu)
2663 {
2664 	iommu_disable(iommu);
2665 	iommu_init_flags(iommu);
2666 	iommu_set_device_table(iommu);
2667 	iommu_enable_command_buffer(iommu);
2668 	iommu_enable_event_buffer(iommu);
2669 	iommu_set_exclusion_range(iommu);
2670 	iommu_enable_ga(iommu);
2671 	iommu_enable_xt(iommu);
2672 	iommu_enable(iommu);
2673 	iommu_flush_all_caches(iommu);
2674 }
2675 
2676 /*
2677  * This function finally enables all IOMMUs found in the system after
2678  * they have been initialized.
2679  *
2680  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2681  * the old content of device table entries. Not this case or copy failed,
2682  * just continue as normal kernel does.
2683  */
2684 static void early_enable_iommus(void)
2685 {
2686 	struct amd_iommu *iommu;
2687 	struct amd_iommu_pci_seg *pci_seg;
2688 
2689 	if (!copy_device_table()) {
2690 		/*
2691 		 * If come here because of failure in copying device table from old
2692 		 * kernel with all IOMMUs enabled, print error message and try to
2693 		 * free allocated old_dev_tbl_cpy.
2694 		 */
2695 		if (amd_iommu_pre_enabled)
2696 			pr_err("Failed to copy DEV table from previous kernel.\n");
2697 
2698 		for_each_pci_segment(pci_seg) {
2699 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2700 				free_pages((unsigned long)pci_seg->old_dev_tbl_cpy,
2701 						get_order(pci_seg->dev_table_size));
2702 				pci_seg->old_dev_tbl_cpy = NULL;
2703 			}
2704 		}
2705 
2706 		for_each_iommu(iommu) {
2707 			clear_translation_pre_enabled(iommu);
2708 			early_enable_iommu(iommu);
2709 		}
2710 	} else {
2711 		pr_info("Copied DEV table from previous kernel.\n");
2712 
2713 		for_each_pci_segment(pci_seg) {
2714 			free_pages((unsigned long)pci_seg->dev_table,
2715 				   get_order(pci_seg->dev_table_size));
2716 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2717 		}
2718 
2719 		for_each_iommu(iommu) {
2720 			iommu_disable_command_buffer(iommu);
2721 			iommu_disable_event_buffer(iommu);
2722 			iommu_enable_command_buffer(iommu);
2723 			iommu_enable_event_buffer(iommu);
2724 			iommu_enable_ga(iommu);
2725 			iommu_enable_xt(iommu);
2726 			iommu_set_device_table(iommu);
2727 			iommu_flush_all_caches(iommu);
2728 		}
2729 	}
2730 
2731 #ifdef CONFIG_IRQ_REMAP
2732 	/*
2733 	 * Note: We have already checked GASup from IVRS table.
2734 	 *       Now, we need to make sure that GAMSup is set.
2735 	 */
2736 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2737 	    !check_feature_on_all_iommus(FEATURE_GAM_VAPIC))
2738 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2739 
2740 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2741 		amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2742 #endif
2743 }
2744 
2745 static void enable_iommus_v2(void)
2746 {
2747 	struct amd_iommu *iommu;
2748 
2749 	for_each_iommu(iommu) {
2750 		iommu_enable_ppr_log(iommu);
2751 		iommu_enable_gt(iommu);
2752 	}
2753 }
2754 
2755 static void enable_iommus(void)
2756 {
2757 	early_enable_iommus();
2758 
2759 	enable_iommus_v2();
2760 }
2761 
2762 static void disable_iommus(void)
2763 {
2764 	struct amd_iommu *iommu;
2765 
2766 	for_each_iommu(iommu)
2767 		iommu_disable(iommu);
2768 
2769 #ifdef CONFIG_IRQ_REMAP
2770 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2771 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2772 #endif
2773 }
2774 
2775 /*
2776  * Suspend/Resume support
2777  * disable suspend until real resume implemented
2778  */
2779 
2780 static void amd_iommu_resume(void)
2781 {
2782 	struct amd_iommu *iommu;
2783 
2784 	for_each_iommu(iommu)
2785 		iommu_apply_resume_quirks(iommu);
2786 
2787 	/* re-load the hardware */
2788 	enable_iommus();
2789 
2790 	amd_iommu_enable_interrupts();
2791 }
2792 
2793 static int amd_iommu_suspend(void)
2794 {
2795 	/* disable IOMMUs to go out of the way for BIOS */
2796 	disable_iommus();
2797 
2798 	return 0;
2799 }
2800 
2801 static struct syscore_ops amd_iommu_syscore_ops = {
2802 	.suspend = amd_iommu_suspend,
2803 	.resume = amd_iommu_resume,
2804 };
2805 
2806 static void __init free_iommu_resources(void)
2807 {
2808 	kmem_cache_destroy(amd_iommu_irq_cache);
2809 	amd_iommu_irq_cache = NULL;
2810 
2811 	free_pages((unsigned long)amd_iommu_alias_table,
2812 		   get_order(alias_table_size));
2813 	amd_iommu_alias_table = NULL;
2814 
2815 	free_pages((unsigned long)amd_iommu_dev_table,
2816 		   get_order(dev_table_size));
2817 	amd_iommu_dev_table = NULL;
2818 
2819 	free_iommu_all();
2820 	free_pci_segments();
2821 }
2822 
2823 /* SB IOAPIC is always on this device in AMD systems */
2824 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2825 
2826 static bool __init check_ioapic_information(void)
2827 {
2828 	const char *fw_bug = FW_BUG;
2829 	bool ret, has_sb_ioapic;
2830 	int idx;
2831 
2832 	has_sb_ioapic = false;
2833 	ret           = false;
2834 
2835 	/*
2836 	 * If we have map overrides on the kernel command line the
2837 	 * messages in this function might not describe firmware bugs
2838 	 * anymore - so be careful
2839 	 */
2840 	if (cmdline_maps)
2841 		fw_bug = "";
2842 
2843 	for (idx = 0; idx < nr_ioapics; idx++) {
2844 		int devid, id = mpc_ioapic_id(idx);
2845 
2846 		devid = get_ioapic_devid(id);
2847 		if (devid < 0) {
2848 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2849 				fw_bug, id);
2850 			ret = false;
2851 		} else if (devid == IOAPIC_SB_DEVID) {
2852 			has_sb_ioapic = true;
2853 			ret           = true;
2854 		}
2855 	}
2856 
2857 	if (!has_sb_ioapic) {
2858 		/*
2859 		 * We expect the SB IOAPIC to be listed in the IVRS
2860 		 * table. The system timer is connected to the SB IOAPIC
2861 		 * and if we don't have it in the list the system will
2862 		 * panic at boot time.  This situation usually happens
2863 		 * when the BIOS is buggy and provides us the wrong
2864 		 * device id for the IOAPIC in the system.
2865 		 */
2866 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2867 	}
2868 
2869 	if (!ret)
2870 		pr_err("Disabling interrupt remapping\n");
2871 
2872 	return ret;
2873 }
2874 
2875 static void __init free_dma_resources(void)
2876 {
2877 	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2878 		   get_order(MAX_DOMAIN_ID/8));
2879 	amd_iommu_pd_alloc_bitmap = NULL;
2880 
2881 	free_unity_maps();
2882 }
2883 
2884 static void __init ivinfo_init(void *ivrs)
2885 {
2886 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
2887 }
2888 
2889 /*
2890  * This is the hardware init function for AMD IOMMU in the system.
2891  * This function is called either from amd_iommu_init or from the interrupt
2892  * remapping setup code.
2893  *
2894  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2895  * four times:
2896  *
2897  *	1 pass) Discover the most comprehensive IVHD type to use.
2898  *
2899  *	2 pass) Find the highest PCI device id the driver has to handle.
2900  *		Upon this information the size of the data structures is
2901  *		determined that needs to be allocated.
2902  *
2903  *	3 pass) Initialize the data structures just allocated with the
2904  *		information in the ACPI table about available AMD IOMMUs
2905  *		in the system. It also maps the PCI devices in the
2906  *		system to specific IOMMUs
2907  *
2908  *	4 pass) After the basic data structures are allocated and
2909  *		initialized we update them with information about memory
2910  *		remapping requirements parsed out of the ACPI table in
2911  *		this last pass.
2912  *
2913  * After everything is set up the IOMMUs are enabled and the necessary
2914  * hotplug and suspend notifiers are registered.
2915  */
2916 static int __init early_amd_iommu_init(void)
2917 {
2918 	struct acpi_table_header *ivrs_base;
2919 	int remap_cache_sz, ret;
2920 	acpi_status status;
2921 
2922 	if (!amd_iommu_detected)
2923 		return -ENODEV;
2924 
2925 	status = acpi_get_table("IVRS", 0, &ivrs_base);
2926 	if (status == AE_NOT_FOUND)
2927 		return -ENODEV;
2928 	else if (ACPI_FAILURE(status)) {
2929 		const char *err = acpi_format_exception(status);
2930 		pr_err("IVRS table error: %s\n", err);
2931 		return -EINVAL;
2932 	}
2933 
2934 	/*
2935 	 * Validate checksum here so we don't need to do it when
2936 	 * we actually parse the table
2937 	 */
2938 	ret = check_ivrs_checksum(ivrs_base);
2939 	if (ret)
2940 		goto out;
2941 
2942 	ivinfo_init(ivrs_base);
2943 
2944 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2945 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2946 
2947 	dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE);
2948 	alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2949 
2950 	/* Device table - directly used by all IOMMUs */
2951 	ret = -ENOMEM;
2952 	amd_iommu_dev_table = (void *)__get_free_pages(
2953 				      GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2954 				      get_order(dev_table_size));
2955 	if (amd_iommu_dev_table == NULL)
2956 		goto out;
2957 
2958 	/*
2959 	 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2960 	 * IOMMU see for that device
2961 	 */
2962 	amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2963 			get_order(alias_table_size));
2964 	if (amd_iommu_alias_table == NULL)
2965 		goto out;
2966 
2967 	amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2968 					    GFP_KERNEL | __GFP_ZERO,
2969 					    get_order(MAX_DOMAIN_ID/8));
2970 	if (amd_iommu_pd_alloc_bitmap == NULL)
2971 		goto out;
2972 
2973 	/*
2974 	 * never allocate domain 0 because its used as the non-allocated and
2975 	 * error value placeholder
2976 	 */
2977 	__set_bit(0, amd_iommu_pd_alloc_bitmap);
2978 
2979 	/*
2980 	 * now the data structures are allocated and basically initialized
2981 	 * start the real acpi table scan
2982 	 */
2983 	ret = init_iommu_all(ivrs_base);
2984 	if (ret)
2985 		goto out;
2986 
2987 	/* Disable any previously enabled IOMMUs */
2988 	if (!is_kdump_kernel() || amd_iommu_disabled)
2989 		disable_iommus();
2990 
2991 	if (amd_iommu_irq_remap)
2992 		amd_iommu_irq_remap = check_ioapic_information();
2993 
2994 	if (amd_iommu_irq_remap) {
2995 		struct amd_iommu_pci_seg *pci_seg;
2996 		/*
2997 		 * Interrupt remapping enabled, create kmem_cache for the
2998 		 * remapping tables.
2999 		 */
3000 		ret = -ENOMEM;
3001 		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
3002 			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
3003 		else
3004 			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
3005 		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
3006 							remap_cache_sz,
3007 							DTE_INTTAB_ALIGNMENT,
3008 							0, NULL);
3009 		if (!amd_iommu_irq_cache)
3010 			goto out;
3011 
3012 		for_each_pci_segment(pci_seg) {
3013 			if (alloc_irq_lookup_table(pci_seg))
3014 				goto out;
3015 		}
3016 	}
3017 
3018 	ret = init_memory_definitions(ivrs_base);
3019 	if (ret)
3020 		goto out;
3021 
3022 	/* init the device table */
3023 	init_device_table();
3024 
3025 out:
3026 	/* Don't leak any ACPI memory */
3027 	acpi_put_table(ivrs_base);
3028 
3029 	return ret;
3030 }
3031 
3032 static int amd_iommu_enable_interrupts(void)
3033 {
3034 	struct amd_iommu *iommu;
3035 	int ret = 0;
3036 
3037 	for_each_iommu(iommu) {
3038 		ret = iommu_init_irq(iommu);
3039 		if (ret)
3040 			goto out;
3041 	}
3042 
3043 out:
3044 	return ret;
3045 }
3046 
3047 static bool __init detect_ivrs(void)
3048 {
3049 	struct acpi_table_header *ivrs_base;
3050 	acpi_status status;
3051 	int i;
3052 
3053 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3054 	if (status == AE_NOT_FOUND)
3055 		return false;
3056 	else if (ACPI_FAILURE(status)) {
3057 		const char *err = acpi_format_exception(status);
3058 		pr_err("IVRS table error: %s\n", err);
3059 		return false;
3060 	}
3061 
3062 	acpi_put_table(ivrs_base);
3063 
3064 	if (amd_iommu_force_enable)
3065 		goto out;
3066 
3067 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3068 	for (i = 0; i < 32; i++) {
3069 		u32 pci_id;
3070 
3071 		pci_id = read_pci_config(0, i, 0, 0);
3072 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3073 			pr_info("Disable IOMMU on Stoney Ridge\n");
3074 			return false;
3075 		}
3076 	}
3077 
3078 out:
3079 	/* Make sure ACS will be enabled during PCI probe */
3080 	pci_request_acs();
3081 
3082 	return true;
3083 }
3084 
3085 /****************************************************************************
3086  *
3087  * AMD IOMMU Initialization State Machine
3088  *
3089  ****************************************************************************/
3090 
3091 static int __init state_next(void)
3092 {
3093 	int ret = 0;
3094 
3095 	switch (init_state) {
3096 	case IOMMU_START_STATE:
3097 		if (!detect_ivrs()) {
3098 			init_state	= IOMMU_NOT_FOUND;
3099 			ret		= -ENODEV;
3100 		} else {
3101 			init_state	= IOMMU_IVRS_DETECTED;
3102 		}
3103 		break;
3104 	case IOMMU_IVRS_DETECTED:
3105 		if (amd_iommu_disabled) {
3106 			init_state = IOMMU_CMDLINE_DISABLED;
3107 			ret = -EINVAL;
3108 		} else {
3109 			ret = early_amd_iommu_init();
3110 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3111 		}
3112 		break;
3113 	case IOMMU_ACPI_FINISHED:
3114 		early_enable_iommus();
3115 		x86_platform.iommu_shutdown = disable_iommus;
3116 		init_state = IOMMU_ENABLED;
3117 		break;
3118 	case IOMMU_ENABLED:
3119 		register_syscore_ops(&amd_iommu_syscore_ops);
3120 		ret = amd_iommu_init_pci();
3121 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3122 		enable_iommus_v2();
3123 		break;
3124 	case IOMMU_PCI_INIT:
3125 		ret = amd_iommu_enable_interrupts();
3126 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3127 		break;
3128 	case IOMMU_INTERRUPTS_EN:
3129 		init_state = IOMMU_INITIALIZED;
3130 		break;
3131 	case IOMMU_INITIALIZED:
3132 		/* Nothing to do */
3133 		break;
3134 	case IOMMU_NOT_FOUND:
3135 	case IOMMU_INIT_ERROR:
3136 	case IOMMU_CMDLINE_DISABLED:
3137 		/* Error states => do nothing */
3138 		ret = -EINVAL;
3139 		break;
3140 	default:
3141 		/* Unknown state */
3142 		BUG();
3143 	}
3144 
3145 	if (ret) {
3146 		free_dma_resources();
3147 		if (!irq_remapping_enabled) {
3148 			disable_iommus();
3149 			free_iommu_resources();
3150 		} else {
3151 			struct amd_iommu *iommu;
3152 			struct amd_iommu_pci_seg *pci_seg;
3153 
3154 			for_each_pci_segment(pci_seg)
3155 				uninit_device_table_dma(pci_seg);
3156 
3157 			for_each_iommu(iommu)
3158 				iommu_flush_all_caches(iommu);
3159 		}
3160 	}
3161 	return ret;
3162 }
3163 
3164 static int __init iommu_go_to_state(enum iommu_init_state state)
3165 {
3166 	int ret = -EINVAL;
3167 
3168 	while (init_state != state) {
3169 		if (init_state == IOMMU_NOT_FOUND         ||
3170 		    init_state == IOMMU_INIT_ERROR        ||
3171 		    init_state == IOMMU_CMDLINE_DISABLED)
3172 			break;
3173 		ret = state_next();
3174 	}
3175 
3176 	return ret;
3177 }
3178 
3179 #ifdef CONFIG_IRQ_REMAP
3180 int __init amd_iommu_prepare(void)
3181 {
3182 	int ret;
3183 
3184 	amd_iommu_irq_remap = true;
3185 
3186 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3187 	if (ret) {
3188 		amd_iommu_irq_remap = false;
3189 		return ret;
3190 	}
3191 
3192 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3193 }
3194 
3195 int __init amd_iommu_enable(void)
3196 {
3197 	int ret;
3198 
3199 	ret = iommu_go_to_state(IOMMU_ENABLED);
3200 	if (ret)
3201 		return ret;
3202 
3203 	irq_remapping_enabled = 1;
3204 	return amd_iommu_xt_mode;
3205 }
3206 
3207 void amd_iommu_disable(void)
3208 {
3209 	amd_iommu_suspend();
3210 }
3211 
3212 int amd_iommu_reenable(int mode)
3213 {
3214 	amd_iommu_resume();
3215 
3216 	return 0;
3217 }
3218 
3219 int __init amd_iommu_enable_faulting(void)
3220 {
3221 	/* We enable MSI later when PCI is initialized */
3222 	return 0;
3223 }
3224 #endif
3225 
3226 /*
3227  * This is the core init function for AMD IOMMU hardware in the system.
3228  * This function is called from the generic x86 DMA layer initialization
3229  * code.
3230  */
3231 static int __init amd_iommu_init(void)
3232 {
3233 	struct amd_iommu *iommu;
3234 	int ret;
3235 
3236 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3237 #ifdef CONFIG_GART_IOMMU
3238 	if (ret && list_empty(&amd_iommu_list)) {
3239 		/*
3240 		 * We failed to initialize the AMD IOMMU - try fallback
3241 		 * to GART if possible.
3242 		 */
3243 		gart_iommu_init();
3244 	}
3245 #endif
3246 
3247 	for_each_iommu(iommu)
3248 		amd_iommu_debugfs_setup(iommu);
3249 
3250 	return ret;
3251 }
3252 
3253 static bool amd_iommu_sme_check(void)
3254 {
3255 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3256 	    (boot_cpu_data.x86 != 0x17))
3257 		return true;
3258 
3259 	/* For Fam17h, a specific level of support is required */
3260 	if (boot_cpu_data.microcode >= 0x08001205)
3261 		return true;
3262 
3263 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3264 	    (boot_cpu_data.microcode <= 0x080011ff))
3265 		return true;
3266 
3267 	pr_notice("IOMMU not currently supported when SME is active\n");
3268 
3269 	return false;
3270 }
3271 
3272 /****************************************************************************
3273  *
3274  * Early detect code. This code runs at IOMMU detection time in the DMA
3275  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3276  * IOMMUs
3277  *
3278  ****************************************************************************/
3279 int __init amd_iommu_detect(void)
3280 {
3281 	int ret;
3282 
3283 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3284 		return -ENODEV;
3285 
3286 	if (!amd_iommu_sme_check())
3287 		return -ENODEV;
3288 
3289 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3290 	if (ret)
3291 		return ret;
3292 
3293 	amd_iommu_detected = true;
3294 	iommu_detected = 1;
3295 	x86_init.iommu.iommu_init = amd_iommu_init;
3296 
3297 	return 1;
3298 }
3299 
3300 /****************************************************************************
3301  *
3302  * Parsing functions for the AMD IOMMU specific kernel command line
3303  * options.
3304  *
3305  ****************************************************************************/
3306 
3307 static int __init parse_amd_iommu_dump(char *str)
3308 {
3309 	amd_iommu_dump = true;
3310 
3311 	return 1;
3312 }
3313 
3314 static int __init parse_amd_iommu_intr(char *str)
3315 {
3316 	for (; *str; ++str) {
3317 		if (strncmp(str, "legacy", 6) == 0) {
3318 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3319 			break;
3320 		}
3321 		if (strncmp(str, "vapic", 5) == 0) {
3322 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3323 			break;
3324 		}
3325 	}
3326 	return 1;
3327 }
3328 
3329 static int __init parse_amd_iommu_options(char *str)
3330 {
3331 	for (; *str; ++str) {
3332 		if (strncmp(str, "fullflush", 9) == 0) {
3333 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3334 			iommu_set_dma_strict();
3335 		}
3336 		if (strncmp(str, "force_enable", 12) == 0)
3337 			amd_iommu_force_enable = true;
3338 		if (strncmp(str, "off", 3) == 0)
3339 			amd_iommu_disabled = true;
3340 		if (strncmp(str, "force_isolation", 15) == 0)
3341 			amd_iommu_force_isolation = true;
3342 	}
3343 
3344 	return 1;
3345 }
3346 
3347 static int __init parse_ivrs_ioapic(char *str)
3348 {
3349 	unsigned int bus, dev, fn;
3350 	int ret, id, i;
3351 	u16 devid;
3352 
3353 	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
3354 
3355 	if (ret != 4) {
3356 		pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3357 		return 1;
3358 	}
3359 
3360 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3361 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3362 			str);
3363 		return 1;
3364 	}
3365 
3366 	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3367 
3368 	cmdline_maps			= true;
3369 	i				= early_ioapic_map_size++;
3370 	early_ioapic_map[i].id		= id;
3371 	early_ioapic_map[i].devid	= devid;
3372 	early_ioapic_map[i].cmd_line	= true;
3373 
3374 	return 1;
3375 }
3376 
3377 static int __init parse_ivrs_hpet(char *str)
3378 {
3379 	unsigned int bus, dev, fn;
3380 	int ret, id, i;
3381 	u16 devid;
3382 
3383 	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
3384 
3385 	if (ret != 4) {
3386 		pr_err("Invalid command line: ivrs_hpet%s\n", str);
3387 		return 1;
3388 	}
3389 
3390 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3391 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3392 			str);
3393 		return 1;
3394 	}
3395 
3396 	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3397 
3398 	cmdline_maps			= true;
3399 	i				= early_hpet_map_size++;
3400 	early_hpet_map[i].id		= id;
3401 	early_hpet_map[i].devid		= devid;
3402 	early_hpet_map[i].cmd_line	= true;
3403 
3404 	return 1;
3405 }
3406 
3407 static int __init parse_ivrs_acpihid(char *str)
3408 {
3409 	u32 bus, dev, fn;
3410 	char *hid, *uid, *p;
3411 	char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3412 	int ret, i;
3413 
3414 	ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
3415 	if (ret != 4) {
3416 		pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
3417 		return 1;
3418 	}
3419 
3420 	p = acpiid;
3421 	hid = strsep(&p, ":");
3422 	uid = p;
3423 
3424 	if (!hid || !(*hid) || !uid) {
3425 		pr_err("Invalid command line: hid or uid\n");
3426 		return 1;
3427 	}
3428 
3429 	i = early_acpihid_map_size++;
3430 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3431 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3432 	early_acpihid_map[i].devid =
3433 		((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3434 	early_acpihid_map[i].cmd_line	= true;
3435 
3436 	return 1;
3437 }
3438 
3439 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3440 __setup("amd_iommu=",		parse_amd_iommu_options);
3441 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3442 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3443 __setup("ivrs_hpet",		parse_ivrs_hpet);
3444 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3445 
3446 bool amd_iommu_v2_supported(void)
3447 {
3448 	return amd_iommu_v2_present;
3449 }
3450 EXPORT_SYMBOL(amd_iommu_v2_supported);
3451 
3452 struct amd_iommu *get_amd_iommu(unsigned int idx)
3453 {
3454 	unsigned int i = 0;
3455 	struct amd_iommu *iommu;
3456 
3457 	for_each_iommu(iommu)
3458 		if (i++ == idx)
3459 			return iommu;
3460 	return NULL;
3461 }
3462 
3463 /****************************************************************************
3464  *
3465  * IOMMU EFR Performance Counter support functionality. This code allows
3466  * access to the IOMMU PC functionality.
3467  *
3468  ****************************************************************************/
3469 
3470 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3471 {
3472 	struct amd_iommu *iommu = get_amd_iommu(idx);
3473 
3474 	if (iommu)
3475 		return iommu->max_banks;
3476 
3477 	return 0;
3478 }
3479 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3480 
3481 bool amd_iommu_pc_supported(void)
3482 {
3483 	return amd_iommu_pc_present;
3484 }
3485 EXPORT_SYMBOL(amd_iommu_pc_supported);
3486 
3487 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3488 {
3489 	struct amd_iommu *iommu = get_amd_iommu(idx);
3490 
3491 	if (iommu)
3492 		return iommu->max_counters;
3493 
3494 	return 0;
3495 }
3496 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3497 
3498 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3499 				u8 fxn, u64 *value, bool is_write)
3500 {
3501 	u32 offset;
3502 	u32 max_offset_lim;
3503 
3504 	/* Make sure the IOMMU PC resource is available */
3505 	if (!amd_iommu_pc_present)
3506 		return -ENODEV;
3507 
3508 	/* Check for valid iommu and pc register indexing */
3509 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3510 		return -ENODEV;
3511 
3512 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3513 
3514 	/* Limit the offset to the hw defined mmio region aperture */
3515 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3516 				(iommu->max_counters << 8) | 0x28);
3517 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3518 	    (offset > max_offset_lim))
3519 		return -EINVAL;
3520 
3521 	if (is_write) {
3522 		u64 val = *value & GENMASK_ULL(47, 0);
3523 
3524 		writel((u32)val, iommu->mmio_base + offset);
3525 		writel((val >> 32), iommu->mmio_base + offset + 4);
3526 	} else {
3527 		*value = readl(iommu->mmio_base + offset + 4);
3528 		*value <<= 32;
3529 		*value |= readl(iommu->mmio_base + offset);
3530 		*value &= GENMASK_ULL(47, 0);
3531 	}
3532 
3533 	return 0;
3534 }
3535 
3536 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3537 {
3538 	if (!iommu)
3539 		return -EINVAL;
3540 
3541 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3542 }
3543 
3544 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3545 {
3546 	if (!iommu)
3547 		return -EINVAL;
3548 
3549 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3550 }
3551