xref: /openbmc/linux/drivers/iommu/amd/init.c (revision 1ab5a15334529d3980a85abb2e06498c8e5ac8cc)
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(u16 devid, u8 bit)
993 {
994 	int i = (bit >> 6) & 0x03;
995 	int _bit = bit & 0x3f;
996 
997 	amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
998 }
999 
1000 static int get_dev_entry_bit(u16 devid, u8 bit)
1001 {
1002 	int i = (bit >> 6) & 0x03;
1003 	int _bit = bit & 0x3f;
1004 
1005 	return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1006 }
1007 
1008 
1009 static bool __copy_device_table(struct amd_iommu *iommu)
1010 {
1011 	u64 int_ctl, int_tab_len, entry = 0;
1012 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1013 	struct dev_table_entry *old_devtb = NULL;
1014 	u32 lo, hi, devid, old_devtb_size;
1015 	phys_addr_t old_devtb_phys;
1016 	u16 dom_id, dte_v, irq_v;
1017 	gfp_t gfp_flag;
1018 	u64 tmp;
1019 
1020 	/* Each IOMMU use separate device table with the same size */
1021 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1022 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1023 	entry = (((u64) hi) << 32) + lo;
1024 
1025 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1026 	if (old_devtb_size != pci_seg->dev_table_size) {
1027 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1028 			iommu->index);
1029 		return false;
1030 	}
1031 
1032 	/*
1033 	 * When SME is enabled in the first kernel, the entry includes the
1034 	 * memory encryption mask(sme_me_mask), we must remove the memory
1035 	 * encryption mask to obtain the true physical address in kdump kernel.
1036 	 */
1037 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1038 
1039 	if (old_devtb_phys >= 0x100000000ULL) {
1040 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1041 		return false;
1042 	}
1043 	old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1044 		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
1045 							pci_seg->dev_table_size)
1046 		    : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB);
1047 
1048 	if (!old_devtb)
1049 		return false;
1050 
1051 	gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
1052 	pci_seg->old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
1053 						    get_order(pci_seg->dev_table_size));
1054 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1055 		pr_err("Failed to allocate memory for copying old device table!\n");
1056 		memunmap(old_devtb);
1057 		return false;
1058 	}
1059 
1060 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1061 		pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1062 		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1063 		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1064 
1065 		if (dte_v && dom_id) {
1066 			pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1067 			pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1068 			__set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
1069 			/* If gcr3 table existed, mask it out */
1070 			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1071 				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1072 				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1073 				pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1074 				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1075 				tmp |= DTE_FLAG_GV;
1076 				pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1077 			}
1078 		}
1079 
1080 		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1081 		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1082 		int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1083 		if (irq_v && (int_ctl || int_tab_len)) {
1084 			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1085 			    (int_tab_len != DTE_INTTABLEN)) {
1086 				pr_err("Wrong old irq remapping flag: %#x\n", devid);
1087 				memunmap(old_devtb);
1088 				return false;
1089 			}
1090 
1091 			pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1092 		}
1093 	}
1094 	memunmap(old_devtb);
1095 
1096 	return true;
1097 }
1098 
1099 static bool copy_device_table(void)
1100 {
1101 	struct amd_iommu *iommu;
1102 	struct amd_iommu_pci_seg *pci_seg;
1103 
1104 	if (!amd_iommu_pre_enabled)
1105 		return false;
1106 
1107 	pr_warn("Translation is already enabled - trying to copy translation structures\n");
1108 
1109 	/*
1110 	 * All IOMMUs within PCI segment shares common device table.
1111 	 * Hence copy device table only once per PCI segment.
1112 	 */
1113 	for_each_pci_segment(pci_seg) {
1114 		for_each_iommu(iommu) {
1115 			if (pci_seg->id != iommu->pci_seg->id)
1116 				continue;
1117 			if (!__copy_device_table(iommu))
1118 				return false;
1119 			break;
1120 		}
1121 	}
1122 
1123 	return true;
1124 }
1125 
1126 void amd_iommu_apply_erratum_63(u16 devid)
1127 {
1128 	int sysmgt;
1129 
1130 	sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
1131 		 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
1132 
1133 	if (sysmgt == 0x01)
1134 		set_dev_entry_bit(devid, DEV_ENTRY_IW);
1135 }
1136 
1137 /* Writes the specific IOMMU for a device into the rlookup table */
1138 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
1139 {
1140 	iommu->pci_seg->rlookup_table[devid] = iommu;
1141 }
1142 
1143 /*
1144  * This function takes the device specific flags read from the ACPI
1145  * table and sets up the device table entry with that information
1146  */
1147 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1148 					   u16 devid, u32 flags, u32 ext_flags)
1149 {
1150 	if (flags & ACPI_DEVFLAG_INITPASS)
1151 		set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
1152 	if (flags & ACPI_DEVFLAG_EXTINT)
1153 		set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
1154 	if (flags & ACPI_DEVFLAG_NMI)
1155 		set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
1156 	if (flags & ACPI_DEVFLAG_SYSMGT1)
1157 		set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
1158 	if (flags & ACPI_DEVFLAG_SYSMGT2)
1159 		set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
1160 	if (flags & ACPI_DEVFLAG_LINT0)
1161 		set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
1162 	if (flags & ACPI_DEVFLAG_LINT1)
1163 		set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
1164 
1165 	amd_iommu_apply_erratum_63(devid);
1166 
1167 	set_iommu_for_device(iommu, devid);
1168 }
1169 
1170 int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1171 {
1172 	struct devid_map *entry;
1173 	struct list_head *list;
1174 
1175 	if (type == IVHD_SPECIAL_IOAPIC)
1176 		list = &ioapic_map;
1177 	else if (type == IVHD_SPECIAL_HPET)
1178 		list = &hpet_map;
1179 	else
1180 		return -EINVAL;
1181 
1182 	list_for_each_entry(entry, list, list) {
1183 		if (!(entry->id == id && entry->cmd_line))
1184 			continue;
1185 
1186 		pr_info("Command-line override present for %s id %d - ignoring\n",
1187 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1188 
1189 		*devid = entry->devid;
1190 
1191 		return 0;
1192 	}
1193 
1194 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1195 	if (!entry)
1196 		return -ENOMEM;
1197 
1198 	entry->id	= id;
1199 	entry->devid	= *devid;
1200 	entry->cmd_line	= cmd_line;
1201 
1202 	list_add_tail(&entry->list, list);
1203 
1204 	return 0;
1205 }
1206 
1207 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1208 				      bool cmd_line)
1209 {
1210 	struct acpihid_map_entry *entry;
1211 	struct list_head *list = &acpihid_map;
1212 
1213 	list_for_each_entry(entry, list, list) {
1214 		if (strcmp(entry->hid, hid) ||
1215 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1216 		    !entry->cmd_line)
1217 			continue;
1218 
1219 		pr_info("Command-line override for hid:%s uid:%s\n",
1220 			hid, uid);
1221 		*devid = entry->devid;
1222 		return 0;
1223 	}
1224 
1225 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1226 	if (!entry)
1227 		return -ENOMEM;
1228 
1229 	memcpy(entry->uid, uid, strlen(uid));
1230 	memcpy(entry->hid, hid, strlen(hid));
1231 	entry->devid = *devid;
1232 	entry->cmd_line	= cmd_line;
1233 	entry->root_devid = (entry->devid & (~0x7));
1234 
1235 	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1236 		entry->cmd_line ? "cmd" : "ivrs",
1237 		entry->hid, entry->uid, entry->root_devid);
1238 
1239 	list_add_tail(&entry->list, list);
1240 	return 0;
1241 }
1242 
1243 static int __init add_early_maps(void)
1244 {
1245 	int i, ret;
1246 
1247 	for (i = 0; i < early_ioapic_map_size; ++i) {
1248 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1249 					 early_ioapic_map[i].id,
1250 					 &early_ioapic_map[i].devid,
1251 					 early_ioapic_map[i].cmd_line);
1252 		if (ret)
1253 			return ret;
1254 	}
1255 
1256 	for (i = 0; i < early_hpet_map_size; ++i) {
1257 		ret = add_special_device(IVHD_SPECIAL_HPET,
1258 					 early_hpet_map[i].id,
1259 					 &early_hpet_map[i].devid,
1260 					 early_hpet_map[i].cmd_line);
1261 		if (ret)
1262 			return ret;
1263 	}
1264 
1265 	for (i = 0; i < early_acpihid_map_size; ++i) {
1266 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1267 					  early_acpihid_map[i].uid,
1268 					  &early_acpihid_map[i].devid,
1269 					  early_acpihid_map[i].cmd_line);
1270 		if (ret)
1271 			return ret;
1272 	}
1273 
1274 	return 0;
1275 }
1276 
1277 /*
1278  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1279  * initializes the hardware and our data structures with it.
1280  */
1281 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1282 					struct ivhd_header *h)
1283 {
1284 	u8 *p = (u8 *)h;
1285 	u8 *end = p, flags = 0;
1286 	u16 devid = 0, devid_start = 0, devid_to = 0;
1287 	u32 dev_i, ext_flags = 0;
1288 	bool alias = false;
1289 	struct ivhd_entry *e;
1290 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1291 	u32 ivhd_size;
1292 	int ret;
1293 
1294 
1295 	ret = add_early_maps();
1296 	if (ret)
1297 		return ret;
1298 
1299 	amd_iommu_apply_ivrs_quirks();
1300 
1301 	/*
1302 	 * First save the recommended feature enable bits from ACPI
1303 	 */
1304 	iommu->acpi_flags = h->flags;
1305 
1306 	/*
1307 	 * Done. Now parse the device entries
1308 	 */
1309 	ivhd_size = get_ivhd_header_size(h);
1310 	if (!ivhd_size) {
1311 		pr_err("Unsupported IVHD type %#x\n", h->type);
1312 		return -EINVAL;
1313 	}
1314 
1315 	p += ivhd_size;
1316 
1317 	end += h->length;
1318 
1319 
1320 	while (p < end) {
1321 		e = (struct ivhd_entry *)p;
1322 		switch (e->type) {
1323 		case IVHD_DEV_ALL:
1324 
1325 			DUMP_printk("  DEV_ALL\t\t\tflags: %02x\n", e->flags);
1326 
1327 			for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1328 				set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1329 			break;
1330 		case IVHD_DEV_SELECT:
1331 
1332 			DUMP_printk("  DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1333 				    "flags: %02x\n",
1334 				    PCI_BUS_NUM(e->devid),
1335 				    PCI_SLOT(e->devid),
1336 				    PCI_FUNC(e->devid),
1337 				    e->flags);
1338 
1339 			devid = e->devid;
1340 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1341 			break;
1342 		case IVHD_DEV_SELECT_RANGE_START:
1343 
1344 			DUMP_printk("  DEV_SELECT_RANGE_START\t "
1345 				    "devid: %02x:%02x.%x flags: %02x\n",
1346 				    PCI_BUS_NUM(e->devid),
1347 				    PCI_SLOT(e->devid),
1348 				    PCI_FUNC(e->devid),
1349 				    e->flags);
1350 
1351 			devid_start = e->devid;
1352 			flags = e->flags;
1353 			ext_flags = 0;
1354 			alias = false;
1355 			break;
1356 		case IVHD_DEV_ALIAS:
1357 
1358 			DUMP_printk("  DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1359 				    "flags: %02x devid_to: %02x:%02x.%x\n",
1360 				    PCI_BUS_NUM(e->devid),
1361 				    PCI_SLOT(e->devid),
1362 				    PCI_FUNC(e->devid),
1363 				    e->flags,
1364 				    PCI_BUS_NUM(e->ext >> 8),
1365 				    PCI_SLOT(e->ext >> 8),
1366 				    PCI_FUNC(e->ext >> 8));
1367 
1368 			devid = e->devid;
1369 			devid_to = e->ext >> 8;
1370 			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1371 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1372 			pci_seg->alias_table[devid] = devid_to;
1373 			break;
1374 		case IVHD_DEV_ALIAS_RANGE:
1375 
1376 			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
1377 				    "devid: %02x:%02x.%x flags: %02x "
1378 				    "devid_to: %02x:%02x.%x\n",
1379 				    PCI_BUS_NUM(e->devid),
1380 				    PCI_SLOT(e->devid),
1381 				    PCI_FUNC(e->devid),
1382 				    e->flags,
1383 				    PCI_BUS_NUM(e->ext >> 8),
1384 				    PCI_SLOT(e->ext >> 8),
1385 				    PCI_FUNC(e->ext >> 8));
1386 
1387 			devid_start = e->devid;
1388 			flags = e->flags;
1389 			devid_to = e->ext >> 8;
1390 			ext_flags = 0;
1391 			alias = true;
1392 			break;
1393 		case IVHD_DEV_EXT_SELECT:
1394 
1395 			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1396 				    "flags: %02x ext: %08x\n",
1397 				    PCI_BUS_NUM(e->devid),
1398 				    PCI_SLOT(e->devid),
1399 				    PCI_FUNC(e->devid),
1400 				    e->flags, e->ext);
1401 
1402 			devid = e->devid;
1403 			set_dev_entry_from_acpi(iommu, devid, e->flags,
1404 						e->ext);
1405 			break;
1406 		case IVHD_DEV_EXT_SELECT_RANGE:
1407 
1408 			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
1409 				    "%02x:%02x.%x flags: %02x ext: %08x\n",
1410 				    PCI_BUS_NUM(e->devid),
1411 				    PCI_SLOT(e->devid),
1412 				    PCI_FUNC(e->devid),
1413 				    e->flags, e->ext);
1414 
1415 			devid_start = e->devid;
1416 			flags = e->flags;
1417 			ext_flags = e->ext;
1418 			alias = false;
1419 			break;
1420 		case IVHD_DEV_RANGE_END:
1421 
1422 			DUMP_printk("  DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1423 				    PCI_BUS_NUM(e->devid),
1424 				    PCI_SLOT(e->devid),
1425 				    PCI_FUNC(e->devid));
1426 
1427 			devid = e->devid;
1428 			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1429 				if (alias) {
1430 					pci_seg->alias_table[dev_i] = devid_to;
1431 					set_dev_entry_from_acpi(iommu,
1432 						devid_to, flags, ext_flags);
1433 				}
1434 				set_dev_entry_from_acpi(iommu, dev_i,
1435 							flags, ext_flags);
1436 			}
1437 			break;
1438 		case IVHD_DEV_SPECIAL: {
1439 			u8 handle, type;
1440 			const char *var;
1441 			u16 devid;
1442 			int ret;
1443 
1444 			handle = e->ext & 0xff;
1445 			devid  = (e->ext >>  8) & 0xffff;
1446 			type   = (e->ext >> 24) & 0xff;
1447 
1448 			if (type == IVHD_SPECIAL_IOAPIC)
1449 				var = "IOAPIC";
1450 			else if (type == IVHD_SPECIAL_HPET)
1451 				var = "HPET";
1452 			else
1453 				var = "UNKNOWN";
1454 
1455 			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1456 				    var, (int)handle,
1457 				    PCI_BUS_NUM(devid),
1458 				    PCI_SLOT(devid),
1459 				    PCI_FUNC(devid));
1460 
1461 			ret = add_special_device(type, handle, &devid, false);
1462 			if (ret)
1463 				return ret;
1464 
1465 			/*
1466 			 * add_special_device might update the devid in case a
1467 			 * command-line override is present. So call
1468 			 * set_dev_entry_from_acpi after add_special_device.
1469 			 */
1470 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1471 
1472 			break;
1473 		}
1474 		case IVHD_DEV_ACPI_HID: {
1475 			u16 devid;
1476 			u8 hid[ACPIHID_HID_LEN];
1477 			u8 uid[ACPIHID_UID_LEN];
1478 			int ret;
1479 
1480 			if (h->type != 0x40) {
1481 				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1482 				       e->type);
1483 				break;
1484 			}
1485 
1486 			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1487 			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1488 			hid[ACPIHID_HID_LEN - 1] = '\0';
1489 
1490 			if (!(*hid)) {
1491 				pr_err(FW_BUG "Invalid HID.\n");
1492 				break;
1493 			}
1494 
1495 			uid[0] = '\0';
1496 			switch (e->uidf) {
1497 			case UID_NOT_PRESENT:
1498 
1499 				if (e->uidl != 0)
1500 					pr_warn(FW_BUG "Invalid UID length.\n");
1501 
1502 				break;
1503 			case UID_IS_INTEGER:
1504 
1505 				sprintf(uid, "%d", e->uid);
1506 
1507 				break;
1508 			case UID_IS_CHARACTER:
1509 
1510 				memcpy(uid, &e->uid, e->uidl);
1511 				uid[e->uidl] = '\0';
1512 
1513 				break;
1514 			default:
1515 				break;
1516 			}
1517 
1518 			devid = e->devid;
1519 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1520 				    hid, uid,
1521 				    PCI_BUS_NUM(devid),
1522 				    PCI_SLOT(devid),
1523 				    PCI_FUNC(devid));
1524 
1525 			flags = e->flags;
1526 
1527 			ret = add_acpi_hid_device(hid, uid, &devid, false);
1528 			if (ret)
1529 				return ret;
1530 
1531 			/*
1532 			 * add_special_device might update the devid in case a
1533 			 * command-line override is present. So call
1534 			 * set_dev_entry_from_acpi after add_special_device.
1535 			 */
1536 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1537 
1538 			break;
1539 		}
1540 		default:
1541 			break;
1542 		}
1543 
1544 		p += ivhd_entry_length(p);
1545 	}
1546 
1547 	return 0;
1548 }
1549 
1550 /* Allocate PCI segment data structure */
1551 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1552 					  struct acpi_table_header *ivrs_base)
1553 {
1554 	struct amd_iommu_pci_seg *pci_seg;
1555 	int last_bdf;
1556 
1557 	/*
1558 	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1559 	 * handle in this PCI segment. Upon this information the shared data
1560 	 * structures for the PCI segments in the system will be allocated.
1561 	 */
1562 	last_bdf = find_last_devid_acpi(ivrs_base, id);
1563 	if (last_bdf < 0)
1564 		return NULL;
1565 
1566 	pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1567 	if (pci_seg == NULL)
1568 		return NULL;
1569 
1570 	pci_seg->last_bdf = last_bdf;
1571 	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1572 	pci_seg->dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE);
1573 	pci_seg->alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1574 	pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1575 
1576 	pci_seg->id = id;
1577 	init_llist_head(&pci_seg->dev_data_list);
1578 	INIT_LIST_HEAD(&pci_seg->unity_map);
1579 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1580 
1581 	if (alloc_dev_table(pci_seg))
1582 		return NULL;
1583 	if (alloc_alias_table(pci_seg))
1584 		return NULL;
1585 	if (alloc_rlookup_table(pci_seg))
1586 		return NULL;
1587 
1588 	return pci_seg;
1589 }
1590 
1591 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1592 					struct acpi_table_header *ivrs_base)
1593 {
1594 	struct amd_iommu_pci_seg *pci_seg;
1595 
1596 	for_each_pci_segment(pci_seg) {
1597 		if (pci_seg->id == id)
1598 			return pci_seg;
1599 	}
1600 
1601 	return alloc_pci_segment(id, ivrs_base);
1602 }
1603 
1604 static void __init free_pci_segments(void)
1605 {
1606 	struct amd_iommu_pci_seg *pci_seg, *next;
1607 
1608 	for_each_pci_segment_safe(pci_seg, next) {
1609 		list_del(&pci_seg->list);
1610 		free_irq_lookup_table(pci_seg);
1611 		free_rlookup_table(pci_seg);
1612 		free_alias_table(pci_seg);
1613 		free_dev_table(pci_seg);
1614 		kfree(pci_seg);
1615 	}
1616 }
1617 
1618 static void __init free_iommu_one(struct amd_iommu *iommu)
1619 {
1620 	free_cwwb_sem(iommu);
1621 	free_command_buffer(iommu);
1622 	free_event_buffer(iommu);
1623 	free_ppr_log(iommu);
1624 	free_ga_log(iommu);
1625 	iommu_unmap_mmio_space(iommu);
1626 }
1627 
1628 static void __init free_iommu_all(void)
1629 {
1630 	struct amd_iommu *iommu, *next;
1631 
1632 	for_each_iommu_safe(iommu, next) {
1633 		list_del(&iommu->list);
1634 		free_iommu_one(iommu);
1635 		kfree(iommu);
1636 	}
1637 }
1638 
1639 /*
1640  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1641  * Workaround:
1642  *     BIOS should disable L2B micellaneous clock gating by setting
1643  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1644  */
1645 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1646 {
1647 	u32 value;
1648 
1649 	if ((boot_cpu_data.x86 != 0x15) ||
1650 	    (boot_cpu_data.x86_model < 0x10) ||
1651 	    (boot_cpu_data.x86_model > 0x1f))
1652 		return;
1653 
1654 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1655 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1656 
1657 	if (value & BIT(2))
1658 		return;
1659 
1660 	/* Select NB indirect register 0x90 and enable writing */
1661 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1662 
1663 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1664 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1665 
1666 	/* Clear the enable writing bit */
1667 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1668 }
1669 
1670 /*
1671  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1672  * Workaround:
1673  *     BIOS should enable ATS write permission check by setting
1674  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1675  */
1676 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1677 {
1678 	u32 value;
1679 
1680 	if ((boot_cpu_data.x86 != 0x15) ||
1681 	    (boot_cpu_data.x86_model < 0x30) ||
1682 	    (boot_cpu_data.x86_model > 0x3f))
1683 		return;
1684 
1685 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1686 	value = iommu_read_l2(iommu, 0x47);
1687 
1688 	if (value & BIT(0))
1689 		return;
1690 
1691 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1692 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1693 
1694 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1695 }
1696 
1697 /*
1698  * This function glues the initialization function for one IOMMU
1699  * together and also allocates the command buffer and programs the
1700  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1701  */
1702 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1703 				 struct acpi_table_header *ivrs_base)
1704 {
1705 	struct amd_iommu_pci_seg *pci_seg;
1706 	int ret;
1707 
1708 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1709 	if (pci_seg == NULL)
1710 		return -ENOMEM;
1711 	iommu->pci_seg = pci_seg;
1712 
1713 	raw_spin_lock_init(&iommu->lock);
1714 	iommu->cmd_sem_val = 0;
1715 
1716 	/* Add IOMMU to internal data structures */
1717 	list_add_tail(&iommu->list, &amd_iommu_list);
1718 	iommu->index = amd_iommus_present++;
1719 
1720 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1721 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1722 		return -ENOSYS;
1723 	}
1724 
1725 	/* Index is fine - add IOMMU to the array */
1726 	amd_iommus[iommu->index] = iommu;
1727 
1728 	/*
1729 	 * Copy data from ACPI table entry to the iommu struct
1730 	 */
1731 	iommu->devid   = h->devid;
1732 	iommu->cap_ptr = h->cap_ptr;
1733 	iommu->mmio_phys = h->mmio_phys;
1734 
1735 	switch (h->type) {
1736 	case 0x10:
1737 		/* Check if IVHD EFR contains proper max banks/counters */
1738 		if ((h->efr_attr != 0) &&
1739 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1740 		    ((h->efr_attr & (0x3F << 17)) != 0))
1741 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1742 		else
1743 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1744 
1745 		/*
1746 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1747 		 * GAM also requires GA mode. Therefore, we need to
1748 		 * check cmpxchg16b support before enabling it.
1749 		 */
1750 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1751 		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1752 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1753 		break;
1754 	case 0x11:
1755 	case 0x40:
1756 		if (h->efr_reg & (1 << 9))
1757 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1758 		else
1759 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1760 
1761 		/*
1762 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1763 		 * XT, GAM also requires GA mode. Therefore, we need to
1764 		 * check cmpxchg16b support before enabling them.
1765 		 */
1766 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1767 		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1768 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1769 			break;
1770 		}
1771 
1772 		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1773 			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1774 
1775 		early_iommu_features_init(iommu, h);
1776 
1777 		break;
1778 	default:
1779 		return -EINVAL;
1780 	}
1781 
1782 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1783 						iommu->mmio_phys_end);
1784 	if (!iommu->mmio_base)
1785 		return -ENOMEM;
1786 
1787 	if (alloc_cwwb_sem(iommu))
1788 		return -ENOMEM;
1789 
1790 	if (alloc_command_buffer(iommu))
1791 		return -ENOMEM;
1792 
1793 	if (alloc_event_buffer(iommu))
1794 		return -ENOMEM;
1795 
1796 	iommu->int_enabled = false;
1797 
1798 	init_translation_status(iommu);
1799 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1800 		iommu_disable(iommu);
1801 		clear_translation_pre_enabled(iommu);
1802 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1803 			iommu->index);
1804 	}
1805 	if (amd_iommu_pre_enabled)
1806 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1807 
1808 	ret = init_iommu_from_acpi(iommu, h);
1809 	if (ret)
1810 		return ret;
1811 
1812 	if (amd_iommu_irq_remap) {
1813 		ret = amd_iommu_create_irq_domain(iommu);
1814 		if (ret)
1815 			return ret;
1816 	}
1817 
1818 	/*
1819 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1820 	 * table tells us so, but this is a lie!
1821 	 */
1822 	pci_seg->rlookup_table[iommu->devid] = NULL;
1823 
1824 	return 0;
1825 }
1826 
1827 /**
1828  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1829  * @ivrs: Pointer to the IVRS header
1830  *
1831  * This function search through all IVDB of the maximum supported IVHD
1832  */
1833 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1834 {
1835 	u8 *base = (u8 *)ivrs;
1836 	struct ivhd_header *ivhd = (struct ivhd_header *)
1837 					(base + IVRS_HEADER_LENGTH);
1838 	u8 last_type = ivhd->type;
1839 	u16 devid = ivhd->devid;
1840 
1841 	while (((u8 *)ivhd - base < ivrs->length) &&
1842 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1843 		u8 *p = (u8 *) ivhd;
1844 
1845 		if (ivhd->devid == devid)
1846 			last_type = ivhd->type;
1847 		ivhd = (struct ivhd_header *)(p + ivhd->length);
1848 	}
1849 
1850 	return last_type;
1851 }
1852 
1853 /*
1854  * Iterates over all IOMMU entries in the ACPI table, allocates the
1855  * IOMMU structure and initializes it with init_iommu_one()
1856  */
1857 static int __init init_iommu_all(struct acpi_table_header *table)
1858 {
1859 	u8 *p = (u8 *)table, *end = (u8 *)table;
1860 	struct ivhd_header *h;
1861 	struct amd_iommu *iommu;
1862 	int ret;
1863 
1864 	end += table->length;
1865 	p += IVRS_HEADER_LENGTH;
1866 
1867 	while (p < end) {
1868 		h = (struct ivhd_header *)p;
1869 		if (*p == amd_iommu_target_ivhd_type) {
1870 
1871 			DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1872 				    "seg: %d flags: %01x info %04x\n",
1873 				    PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1874 				    PCI_FUNC(h->devid), h->cap_ptr,
1875 				    h->pci_seg, h->flags, h->info);
1876 			DUMP_printk("       mmio-addr: %016llx\n",
1877 				    h->mmio_phys);
1878 
1879 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1880 			if (iommu == NULL)
1881 				return -ENOMEM;
1882 
1883 			ret = init_iommu_one(iommu, h, table);
1884 			if (ret)
1885 				return ret;
1886 		}
1887 		p += h->length;
1888 
1889 	}
1890 	WARN_ON(p != end);
1891 
1892 	return 0;
1893 }
1894 
1895 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1896 {
1897 	u64 val;
1898 	struct pci_dev *pdev = iommu->dev;
1899 
1900 	if (!iommu_feature(iommu, FEATURE_PC))
1901 		return;
1902 
1903 	amd_iommu_pc_present = true;
1904 
1905 	pci_info(pdev, "IOMMU performance counters supported\n");
1906 
1907 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1908 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1909 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1910 
1911 	return;
1912 }
1913 
1914 static ssize_t amd_iommu_show_cap(struct device *dev,
1915 				  struct device_attribute *attr,
1916 				  char *buf)
1917 {
1918 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1919 	return sprintf(buf, "%x\n", iommu->cap);
1920 }
1921 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1922 
1923 static ssize_t amd_iommu_show_features(struct device *dev,
1924 				       struct device_attribute *attr,
1925 				       char *buf)
1926 {
1927 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1928 	return sprintf(buf, "%llx\n", iommu->features);
1929 }
1930 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1931 
1932 static struct attribute *amd_iommu_attrs[] = {
1933 	&dev_attr_cap.attr,
1934 	&dev_attr_features.attr,
1935 	NULL,
1936 };
1937 
1938 static struct attribute_group amd_iommu_group = {
1939 	.name = "amd-iommu",
1940 	.attrs = amd_iommu_attrs,
1941 };
1942 
1943 static const struct attribute_group *amd_iommu_groups[] = {
1944 	&amd_iommu_group,
1945 	NULL,
1946 };
1947 
1948 /*
1949  * Note: IVHD 0x11 and 0x40 also contains exact copy
1950  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1951  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1952  */
1953 static void __init late_iommu_features_init(struct amd_iommu *iommu)
1954 {
1955 	u64 features;
1956 
1957 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1958 		return;
1959 
1960 	/* read extended feature bits */
1961 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1962 
1963 	if (!iommu->features) {
1964 		iommu->features = features;
1965 		return;
1966 	}
1967 
1968 	/*
1969 	 * Sanity check and warn if EFR values from
1970 	 * IVHD and MMIO conflict.
1971 	 */
1972 	if (features != iommu->features)
1973 		pr_warn(FW_WARN "EFR mismatch. Use IVHD EFR (%#llx : %#llx).\n",
1974 			features, iommu->features);
1975 }
1976 
1977 static int __init iommu_init_pci(struct amd_iommu *iommu)
1978 {
1979 	int cap_ptr = iommu->cap_ptr;
1980 	int ret;
1981 
1982 	iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1983 						 iommu->devid & 0xff);
1984 	if (!iommu->dev)
1985 		return -ENODEV;
1986 
1987 	/* Prevent binding other PCI device drivers to IOMMU devices */
1988 	iommu->dev->match_driver = false;
1989 
1990 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1991 			      &iommu->cap);
1992 
1993 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1994 		amd_iommu_iotlb_sup = false;
1995 
1996 	late_iommu_features_init(iommu);
1997 
1998 	if (iommu_feature(iommu, FEATURE_GT)) {
1999 		int glxval;
2000 		u32 max_pasid;
2001 		u64 pasmax;
2002 
2003 		pasmax = iommu->features & FEATURE_PASID_MASK;
2004 		pasmax >>= FEATURE_PASID_SHIFT;
2005 		max_pasid  = (1 << (pasmax + 1)) - 1;
2006 
2007 		amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
2008 
2009 		BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
2010 
2011 		glxval   = iommu->features & FEATURE_GLXVAL_MASK;
2012 		glxval >>= FEATURE_GLXVAL_SHIFT;
2013 
2014 		if (amd_iommu_max_glx_val == -1)
2015 			amd_iommu_max_glx_val = glxval;
2016 		else
2017 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2018 	}
2019 
2020 	if (iommu_feature(iommu, FEATURE_GT) &&
2021 	    iommu_feature(iommu, FEATURE_PPR)) {
2022 		iommu->is_iommu_v2   = true;
2023 		amd_iommu_v2_present = true;
2024 	}
2025 
2026 	if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
2027 		return -ENOMEM;
2028 
2029 	ret = iommu_init_ga_log(iommu);
2030 	if (ret)
2031 		return ret;
2032 
2033 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2034 		pr_info("Using strict mode due to virtualization\n");
2035 		iommu_set_dma_strict();
2036 		amd_iommu_np_cache = true;
2037 	}
2038 
2039 	init_iommu_perf_ctr(iommu);
2040 
2041 	if (is_rd890_iommu(iommu->dev)) {
2042 		int i, j;
2043 
2044 		iommu->root_pdev =
2045 			pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
2046 						    PCI_DEVFN(0, 0));
2047 
2048 		/*
2049 		 * Some rd890 systems may not be fully reconfigured by the
2050 		 * BIOS, so it's necessary for us to store this information so
2051 		 * it can be reprogrammed on resume
2052 		 */
2053 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2054 				&iommu->stored_addr_lo);
2055 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2056 				&iommu->stored_addr_hi);
2057 
2058 		/* Low bit locks writes to configuration space */
2059 		iommu->stored_addr_lo &= ~1;
2060 
2061 		for (i = 0; i < 6; i++)
2062 			for (j = 0; j < 0x12; j++)
2063 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2064 
2065 		for (i = 0; i < 0x83; i++)
2066 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2067 	}
2068 
2069 	amd_iommu_erratum_746_workaround(iommu);
2070 	amd_iommu_ats_write_check_workaround(iommu);
2071 
2072 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2073 			       amd_iommu_groups, "ivhd%d", iommu->index);
2074 	if (ret)
2075 		return ret;
2076 
2077 	iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2078 
2079 	return pci_enable_device(iommu->dev);
2080 }
2081 
2082 static void print_iommu_info(void)
2083 {
2084 	static const char * const feat_str[] = {
2085 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2086 		"IA", "GA", "HE", "PC"
2087 	};
2088 	struct amd_iommu *iommu;
2089 
2090 	for_each_iommu(iommu) {
2091 		struct pci_dev *pdev = iommu->dev;
2092 		int i;
2093 
2094 		pci_info(pdev, "Found IOMMU cap 0x%x\n", iommu->cap_ptr);
2095 
2096 		if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
2097 			pr_info("Extended features (%#llx):", iommu->features);
2098 
2099 			for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2100 				if (iommu_feature(iommu, (1ULL << i)))
2101 					pr_cont(" %s", feat_str[i]);
2102 			}
2103 
2104 			if (iommu->features & FEATURE_GAM_VAPIC)
2105 				pr_cont(" GA_vAPIC");
2106 
2107 			pr_cont("\n");
2108 		}
2109 	}
2110 	if (irq_remapping_enabled) {
2111 		pr_info("Interrupt remapping enabled\n");
2112 		if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2113 			pr_info("Virtual APIC enabled\n");
2114 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2115 			pr_info("X2APIC enabled\n");
2116 	}
2117 }
2118 
2119 static int __init amd_iommu_init_pci(void)
2120 {
2121 	struct amd_iommu *iommu;
2122 	struct amd_iommu_pci_seg *pci_seg;
2123 	int ret;
2124 
2125 	for_each_iommu(iommu) {
2126 		ret = iommu_init_pci(iommu);
2127 		if (ret) {
2128 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2129 			       iommu->index, ret);
2130 			goto out;
2131 		}
2132 		/* Need to setup range after PCI init */
2133 		iommu_set_cwwb_range(iommu);
2134 	}
2135 
2136 	/*
2137 	 * Order is important here to make sure any unity map requirements are
2138 	 * fulfilled. The unity mappings are created and written to the device
2139 	 * table during the amd_iommu_init_api() call.
2140 	 *
2141 	 * After that we call init_device_table_dma() to make sure any
2142 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2143 	 * of all IOMMUs to make sure the changes to the device table are
2144 	 * active.
2145 	 */
2146 	ret = amd_iommu_init_api();
2147 	if (ret) {
2148 		pr_err("IOMMU: Failed to initialize IOMMU-API interface (error=%d)!\n",
2149 		       ret);
2150 		goto out;
2151 	}
2152 
2153 	for_each_pci_segment(pci_seg)
2154 		init_device_table_dma(pci_seg);
2155 
2156 	for_each_iommu(iommu)
2157 		iommu_flush_all_caches(iommu);
2158 
2159 	print_iommu_info();
2160 
2161 out:
2162 	return ret;
2163 }
2164 
2165 /****************************************************************************
2166  *
2167  * The following functions initialize the MSI interrupts for all IOMMUs
2168  * in the system. It's a bit challenging because there could be multiple
2169  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2170  * pci_dev.
2171  *
2172  ****************************************************************************/
2173 
2174 static int iommu_setup_msi(struct amd_iommu *iommu)
2175 {
2176 	int r;
2177 
2178 	r = pci_enable_msi(iommu->dev);
2179 	if (r)
2180 		return r;
2181 
2182 	r = request_threaded_irq(iommu->dev->irq,
2183 				 amd_iommu_int_handler,
2184 				 amd_iommu_int_thread,
2185 				 0, "AMD-Vi",
2186 				 iommu);
2187 
2188 	if (r) {
2189 		pci_disable_msi(iommu->dev);
2190 		return r;
2191 	}
2192 
2193 	return 0;
2194 }
2195 
2196 union intcapxt {
2197 	u64	capxt;
2198 	struct {
2199 		u64	reserved_0		:  2,
2200 			dest_mode_logical	:  1,
2201 			reserved_1		:  5,
2202 			destid_0_23		: 24,
2203 			vector			:  8,
2204 			reserved_2		: 16,
2205 			destid_24_31		:  8;
2206 	};
2207 } __attribute__ ((packed));
2208 
2209 
2210 static struct irq_chip intcapxt_controller;
2211 
2212 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2213 				       struct irq_data *irqd, bool reserve)
2214 {
2215 	return 0;
2216 }
2217 
2218 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2219 					  struct irq_data *irqd)
2220 {
2221 }
2222 
2223 
2224 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2225 				    unsigned int nr_irqs, void *arg)
2226 {
2227 	struct irq_alloc_info *info = arg;
2228 	int i, ret;
2229 
2230 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2231 		return -EINVAL;
2232 
2233 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2234 	if (ret < 0)
2235 		return ret;
2236 
2237 	for (i = virq; i < virq + nr_irqs; i++) {
2238 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2239 
2240 		irqd->chip = &intcapxt_controller;
2241 		irqd->chip_data = info->data;
2242 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2243 	}
2244 
2245 	return ret;
2246 }
2247 
2248 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2249 				    unsigned int nr_irqs)
2250 {
2251 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2252 }
2253 
2254 
2255 static void intcapxt_unmask_irq(struct irq_data *irqd)
2256 {
2257 	struct amd_iommu *iommu = irqd->chip_data;
2258 	struct irq_cfg *cfg = irqd_cfg(irqd);
2259 	union intcapxt xt;
2260 
2261 	xt.capxt = 0ULL;
2262 	xt.dest_mode_logical = apic->dest_mode_logical;
2263 	xt.vector = cfg->vector;
2264 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2265 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2266 
2267 	/**
2268 	 * Current IOMMU implementation uses the same IRQ for all
2269 	 * 3 IOMMU interrupts.
2270 	 */
2271 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2272 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2273 	writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2274 }
2275 
2276 static void intcapxt_mask_irq(struct irq_data *irqd)
2277 {
2278 	struct amd_iommu *iommu = irqd->chip_data;
2279 
2280 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET);
2281 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET);
2282 	writeq(0, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET);
2283 }
2284 
2285 
2286 static int intcapxt_set_affinity(struct irq_data *irqd,
2287 				 const struct cpumask *mask, bool force)
2288 {
2289 	struct irq_data *parent = irqd->parent_data;
2290 	int ret;
2291 
2292 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2293 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2294 		return ret;
2295 	return 0;
2296 }
2297 
2298 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2299 {
2300 	return on ? -EOPNOTSUPP : 0;
2301 }
2302 
2303 static struct irq_chip intcapxt_controller = {
2304 	.name			= "IOMMU-MSI",
2305 	.irq_unmask		= intcapxt_unmask_irq,
2306 	.irq_mask		= intcapxt_mask_irq,
2307 	.irq_ack		= irq_chip_ack_parent,
2308 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2309 	.irq_set_affinity       = intcapxt_set_affinity,
2310 	.irq_set_wake		= intcapxt_set_wake,
2311 	.flags			= IRQCHIP_MASK_ON_SUSPEND,
2312 };
2313 
2314 static const struct irq_domain_ops intcapxt_domain_ops = {
2315 	.alloc			= intcapxt_irqdomain_alloc,
2316 	.free			= intcapxt_irqdomain_free,
2317 	.activate		= intcapxt_irqdomain_activate,
2318 	.deactivate		= intcapxt_irqdomain_deactivate,
2319 };
2320 
2321 
2322 static struct irq_domain *iommu_irqdomain;
2323 
2324 static struct irq_domain *iommu_get_irqdomain(void)
2325 {
2326 	struct fwnode_handle *fn;
2327 
2328 	/* No need for locking here (yet) as the init is single-threaded */
2329 	if (iommu_irqdomain)
2330 		return iommu_irqdomain;
2331 
2332 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2333 	if (!fn)
2334 		return NULL;
2335 
2336 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2337 						      fn, &intcapxt_domain_ops,
2338 						      NULL);
2339 	if (!iommu_irqdomain)
2340 		irq_domain_free_fwnode(fn);
2341 
2342 	return iommu_irqdomain;
2343 }
2344 
2345 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2346 {
2347 	struct irq_domain *domain;
2348 	struct irq_alloc_info info;
2349 	int irq, ret;
2350 
2351 	domain = iommu_get_irqdomain();
2352 	if (!domain)
2353 		return -ENXIO;
2354 
2355 	init_irq_alloc_info(&info, NULL);
2356 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2357 	info.data = iommu;
2358 
2359 	irq = irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
2360 	if (irq < 0) {
2361 		irq_domain_remove(domain);
2362 		return irq;
2363 	}
2364 
2365 	ret = request_threaded_irq(irq, amd_iommu_int_handler,
2366 				   amd_iommu_int_thread, 0, "AMD-Vi", iommu);
2367 	if (ret) {
2368 		irq_domain_free_irqs(irq, 1);
2369 		irq_domain_remove(domain);
2370 		return ret;
2371 	}
2372 
2373 	return 0;
2374 }
2375 
2376 static int iommu_init_irq(struct amd_iommu *iommu)
2377 {
2378 	int ret;
2379 
2380 	if (iommu->int_enabled)
2381 		goto enable_faults;
2382 
2383 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2384 		ret = iommu_setup_intcapxt(iommu);
2385 	else if (iommu->dev->msi_cap)
2386 		ret = iommu_setup_msi(iommu);
2387 	else
2388 		ret = -ENODEV;
2389 
2390 	if (ret)
2391 		return ret;
2392 
2393 	iommu->int_enabled = true;
2394 enable_faults:
2395 
2396 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2397 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2398 
2399 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2400 
2401 	if (iommu->ppr_log != NULL)
2402 		iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
2403 
2404 	iommu_ga_log_enable(iommu);
2405 
2406 	return 0;
2407 }
2408 
2409 /****************************************************************************
2410  *
2411  * The next functions belong to the third pass of parsing the ACPI
2412  * table. In this last pass the memory mapping requirements are
2413  * gathered (like exclusion and unity mapping ranges).
2414  *
2415  ****************************************************************************/
2416 
2417 static void __init free_unity_maps(void)
2418 {
2419 	struct unity_map_entry *entry, *next;
2420 	struct amd_iommu_pci_seg *p, *pci_seg;
2421 
2422 	for_each_pci_segment_safe(pci_seg, p) {
2423 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2424 			list_del(&entry->list);
2425 			kfree(entry);
2426 		}
2427 	}
2428 }
2429 
2430 /* called for unity map ACPI definition */
2431 static int __init init_unity_map_range(struct ivmd_header *m,
2432 				       struct acpi_table_header *ivrs_base)
2433 {
2434 	struct unity_map_entry *e = NULL;
2435 	struct amd_iommu_pci_seg *pci_seg;
2436 	char *s;
2437 
2438 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2439 	if (pci_seg == NULL)
2440 		return -ENOMEM;
2441 
2442 	e = kzalloc(sizeof(*e), GFP_KERNEL);
2443 	if (e == NULL)
2444 		return -ENOMEM;
2445 
2446 	switch (m->type) {
2447 	default:
2448 		kfree(e);
2449 		return 0;
2450 	case ACPI_IVMD_TYPE:
2451 		s = "IVMD_TYPEi\t\t\t";
2452 		e->devid_start = e->devid_end = m->devid;
2453 		break;
2454 	case ACPI_IVMD_TYPE_ALL:
2455 		s = "IVMD_TYPE_ALL\t\t";
2456 		e->devid_start = 0;
2457 		e->devid_end = amd_iommu_last_bdf;
2458 		break;
2459 	case ACPI_IVMD_TYPE_RANGE:
2460 		s = "IVMD_TYPE_RANGE\t\t";
2461 		e->devid_start = m->devid;
2462 		e->devid_end = m->aux;
2463 		break;
2464 	}
2465 	e->address_start = PAGE_ALIGN(m->range_start);
2466 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2467 	e->prot = m->flags >> 1;
2468 
2469 	/*
2470 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2471 	 * since some buggy BIOSes might lead to the overwritten exclusion
2472 	 * range (exclusion_start and exclusion_length members). This
2473 	 * happens when there are multiple exclusion ranges (IVMD entries)
2474 	 * defined in ACPI table.
2475 	 */
2476 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2477 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2478 
2479 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2480 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2481 		    " flags: %x\n", s, m->pci_seg,
2482 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2483 		    PCI_FUNC(e->devid_start), m->pci_seg,
2484 		    PCI_BUS_NUM(e->devid_end),
2485 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2486 		    e->address_start, e->address_end, m->flags);
2487 
2488 	list_add_tail(&e->list, &pci_seg->unity_map);
2489 
2490 	return 0;
2491 }
2492 
2493 /* iterates over all memory definitions we find in the ACPI table */
2494 static int __init init_memory_definitions(struct acpi_table_header *table)
2495 {
2496 	u8 *p = (u8 *)table, *end = (u8 *)table;
2497 	struct ivmd_header *m;
2498 
2499 	end += table->length;
2500 	p += IVRS_HEADER_LENGTH;
2501 
2502 	while (p < end) {
2503 		m = (struct ivmd_header *)p;
2504 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2505 			init_unity_map_range(m, table);
2506 
2507 		p += m->length;
2508 	}
2509 
2510 	return 0;
2511 }
2512 
2513 /*
2514  * Init the device table to not allow DMA access for devices
2515  */
2516 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2517 {
2518 	u32 devid;
2519 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2520 
2521 	if (dev_table == NULL)
2522 		return;
2523 
2524 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2525 		set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2526 		set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2527 	}
2528 }
2529 
2530 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2531 {
2532 	u32 devid;
2533 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2534 
2535 	if (dev_table == NULL)
2536 		return;
2537 
2538 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2539 		dev_table[devid].data[0] = 0ULL;
2540 		dev_table[devid].data[1] = 0ULL;
2541 	}
2542 }
2543 
2544 static void init_device_table(void)
2545 {
2546 	u32 devid;
2547 
2548 	if (!amd_iommu_irq_remap)
2549 		return;
2550 
2551 	for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2552 		set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2553 }
2554 
2555 static void iommu_init_flags(struct amd_iommu *iommu)
2556 {
2557 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2558 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2559 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2560 
2561 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2562 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2563 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2564 
2565 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2566 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2567 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2568 
2569 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2570 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2571 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2572 
2573 	/*
2574 	 * make IOMMU memory accesses cache coherent
2575 	 */
2576 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2577 
2578 	/* Set IOTLB invalidation timeout to 1s */
2579 	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2580 }
2581 
2582 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2583 {
2584 	int i, j;
2585 	u32 ioc_feature_control;
2586 	struct pci_dev *pdev = iommu->root_pdev;
2587 
2588 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2589 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2590 		return;
2591 
2592 	/*
2593 	 * First, we need to ensure that the iommu is enabled. This is
2594 	 * controlled by a register in the northbridge
2595 	 */
2596 
2597 	/* Select Northbridge indirect register 0x75 and enable writing */
2598 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2599 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2600 
2601 	/* Enable the iommu */
2602 	if (!(ioc_feature_control & 0x1))
2603 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2604 
2605 	/* Restore the iommu BAR */
2606 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2607 			       iommu->stored_addr_lo);
2608 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2609 			       iommu->stored_addr_hi);
2610 
2611 	/* Restore the l1 indirect regs for each of the 6 l1s */
2612 	for (i = 0; i < 6; i++)
2613 		for (j = 0; j < 0x12; j++)
2614 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2615 
2616 	/* Restore the l2 indirect regs */
2617 	for (i = 0; i < 0x83; i++)
2618 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2619 
2620 	/* Lock PCI setup registers */
2621 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2622 			       iommu->stored_addr_lo | 1);
2623 }
2624 
2625 static void iommu_enable_ga(struct amd_iommu *iommu)
2626 {
2627 #ifdef CONFIG_IRQ_REMAP
2628 	switch (amd_iommu_guest_ir) {
2629 	case AMD_IOMMU_GUEST_IR_VAPIC:
2630 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2631 		fallthrough;
2632 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2633 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2634 		iommu->irte_ops = &irte_128_ops;
2635 		break;
2636 	default:
2637 		iommu->irte_ops = &irte_32_ops;
2638 		break;
2639 	}
2640 #endif
2641 }
2642 
2643 static void early_enable_iommu(struct amd_iommu *iommu)
2644 {
2645 	iommu_disable(iommu);
2646 	iommu_init_flags(iommu);
2647 	iommu_set_device_table(iommu);
2648 	iommu_enable_command_buffer(iommu);
2649 	iommu_enable_event_buffer(iommu);
2650 	iommu_set_exclusion_range(iommu);
2651 	iommu_enable_ga(iommu);
2652 	iommu_enable_xt(iommu);
2653 	iommu_enable(iommu);
2654 	iommu_flush_all_caches(iommu);
2655 }
2656 
2657 /*
2658  * This function finally enables all IOMMUs found in the system after
2659  * they have been initialized.
2660  *
2661  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2662  * the old content of device table entries. Not this case or copy failed,
2663  * just continue as normal kernel does.
2664  */
2665 static void early_enable_iommus(void)
2666 {
2667 	struct amd_iommu *iommu;
2668 	struct amd_iommu_pci_seg *pci_seg;
2669 
2670 	if (!copy_device_table()) {
2671 		/*
2672 		 * If come here because of failure in copying device table from old
2673 		 * kernel with all IOMMUs enabled, print error message and try to
2674 		 * free allocated old_dev_tbl_cpy.
2675 		 */
2676 		if (amd_iommu_pre_enabled)
2677 			pr_err("Failed to copy DEV table from previous kernel.\n");
2678 
2679 		for_each_pci_segment(pci_seg) {
2680 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2681 				free_pages((unsigned long)pci_seg->old_dev_tbl_cpy,
2682 						get_order(pci_seg->dev_table_size));
2683 				pci_seg->old_dev_tbl_cpy = NULL;
2684 			}
2685 		}
2686 
2687 		for_each_iommu(iommu) {
2688 			clear_translation_pre_enabled(iommu);
2689 			early_enable_iommu(iommu);
2690 		}
2691 	} else {
2692 		pr_info("Copied DEV table from previous kernel.\n");
2693 
2694 		for_each_pci_segment(pci_seg) {
2695 			free_pages((unsigned long)pci_seg->dev_table,
2696 				   get_order(pci_seg->dev_table_size));
2697 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2698 		}
2699 
2700 		for_each_iommu(iommu) {
2701 			iommu_disable_command_buffer(iommu);
2702 			iommu_disable_event_buffer(iommu);
2703 			iommu_enable_command_buffer(iommu);
2704 			iommu_enable_event_buffer(iommu);
2705 			iommu_enable_ga(iommu);
2706 			iommu_enable_xt(iommu);
2707 			iommu_set_device_table(iommu);
2708 			iommu_flush_all_caches(iommu);
2709 		}
2710 	}
2711 
2712 #ifdef CONFIG_IRQ_REMAP
2713 	/*
2714 	 * Note: We have already checked GASup from IVRS table.
2715 	 *       Now, we need to make sure that GAMSup is set.
2716 	 */
2717 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2718 	    !check_feature_on_all_iommus(FEATURE_GAM_VAPIC))
2719 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2720 
2721 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2722 		amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2723 #endif
2724 }
2725 
2726 static void enable_iommus_v2(void)
2727 {
2728 	struct amd_iommu *iommu;
2729 
2730 	for_each_iommu(iommu) {
2731 		iommu_enable_ppr_log(iommu);
2732 		iommu_enable_gt(iommu);
2733 	}
2734 }
2735 
2736 static void enable_iommus(void)
2737 {
2738 	early_enable_iommus();
2739 
2740 	enable_iommus_v2();
2741 }
2742 
2743 static void disable_iommus(void)
2744 {
2745 	struct amd_iommu *iommu;
2746 
2747 	for_each_iommu(iommu)
2748 		iommu_disable(iommu);
2749 
2750 #ifdef CONFIG_IRQ_REMAP
2751 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2752 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2753 #endif
2754 }
2755 
2756 /*
2757  * Suspend/Resume support
2758  * disable suspend until real resume implemented
2759  */
2760 
2761 static void amd_iommu_resume(void)
2762 {
2763 	struct amd_iommu *iommu;
2764 
2765 	for_each_iommu(iommu)
2766 		iommu_apply_resume_quirks(iommu);
2767 
2768 	/* re-load the hardware */
2769 	enable_iommus();
2770 
2771 	amd_iommu_enable_interrupts();
2772 }
2773 
2774 static int amd_iommu_suspend(void)
2775 {
2776 	/* disable IOMMUs to go out of the way for BIOS */
2777 	disable_iommus();
2778 
2779 	return 0;
2780 }
2781 
2782 static struct syscore_ops amd_iommu_syscore_ops = {
2783 	.suspend = amd_iommu_suspend,
2784 	.resume = amd_iommu_resume,
2785 };
2786 
2787 static void __init free_iommu_resources(void)
2788 {
2789 	kmem_cache_destroy(amd_iommu_irq_cache);
2790 	amd_iommu_irq_cache = NULL;
2791 
2792 	free_pages((unsigned long)amd_iommu_alias_table,
2793 		   get_order(alias_table_size));
2794 	amd_iommu_alias_table = NULL;
2795 
2796 	free_pages((unsigned long)amd_iommu_dev_table,
2797 		   get_order(dev_table_size));
2798 	amd_iommu_dev_table = NULL;
2799 
2800 	free_iommu_all();
2801 	free_pci_segments();
2802 }
2803 
2804 /* SB IOAPIC is always on this device in AMD systems */
2805 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2806 
2807 static bool __init check_ioapic_information(void)
2808 {
2809 	const char *fw_bug = FW_BUG;
2810 	bool ret, has_sb_ioapic;
2811 	int idx;
2812 
2813 	has_sb_ioapic = false;
2814 	ret           = false;
2815 
2816 	/*
2817 	 * If we have map overrides on the kernel command line the
2818 	 * messages in this function might not describe firmware bugs
2819 	 * anymore - so be careful
2820 	 */
2821 	if (cmdline_maps)
2822 		fw_bug = "";
2823 
2824 	for (idx = 0; idx < nr_ioapics; idx++) {
2825 		int devid, id = mpc_ioapic_id(idx);
2826 
2827 		devid = get_ioapic_devid(id);
2828 		if (devid < 0) {
2829 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2830 				fw_bug, id);
2831 			ret = false;
2832 		} else if (devid == IOAPIC_SB_DEVID) {
2833 			has_sb_ioapic = true;
2834 			ret           = true;
2835 		}
2836 	}
2837 
2838 	if (!has_sb_ioapic) {
2839 		/*
2840 		 * We expect the SB IOAPIC to be listed in the IVRS
2841 		 * table. The system timer is connected to the SB IOAPIC
2842 		 * and if we don't have it in the list the system will
2843 		 * panic at boot time.  This situation usually happens
2844 		 * when the BIOS is buggy and provides us the wrong
2845 		 * device id for the IOAPIC in the system.
2846 		 */
2847 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2848 	}
2849 
2850 	if (!ret)
2851 		pr_err("Disabling interrupt remapping\n");
2852 
2853 	return ret;
2854 }
2855 
2856 static void __init free_dma_resources(void)
2857 {
2858 	free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2859 		   get_order(MAX_DOMAIN_ID/8));
2860 	amd_iommu_pd_alloc_bitmap = NULL;
2861 
2862 	free_unity_maps();
2863 }
2864 
2865 static void __init ivinfo_init(void *ivrs)
2866 {
2867 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
2868 }
2869 
2870 /*
2871  * This is the hardware init function for AMD IOMMU in the system.
2872  * This function is called either from amd_iommu_init or from the interrupt
2873  * remapping setup code.
2874  *
2875  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2876  * four times:
2877  *
2878  *	1 pass) Discover the most comprehensive IVHD type to use.
2879  *
2880  *	2 pass) Find the highest PCI device id the driver has to handle.
2881  *		Upon this information the size of the data structures is
2882  *		determined that needs to be allocated.
2883  *
2884  *	3 pass) Initialize the data structures just allocated with the
2885  *		information in the ACPI table about available AMD IOMMUs
2886  *		in the system. It also maps the PCI devices in the
2887  *		system to specific IOMMUs
2888  *
2889  *	4 pass) After the basic data structures are allocated and
2890  *		initialized we update them with information about memory
2891  *		remapping requirements parsed out of the ACPI table in
2892  *		this last pass.
2893  *
2894  * After everything is set up the IOMMUs are enabled and the necessary
2895  * hotplug and suspend notifiers are registered.
2896  */
2897 static int __init early_amd_iommu_init(void)
2898 {
2899 	struct acpi_table_header *ivrs_base;
2900 	int remap_cache_sz, ret;
2901 	acpi_status status;
2902 
2903 	if (!amd_iommu_detected)
2904 		return -ENODEV;
2905 
2906 	status = acpi_get_table("IVRS", 0, &ivrs_base);
2907 	if (status == AE_NOT_FOUND)
2908 		return -ENODEV;
2909 	else if (ACPI_FAILURE(status)) {
2910 		const char *err = acpi_format_exception(status);
2911 		pr_err("IVRS table error: %s\n", err);
2912 		return -EINVAL;
2913 	}
2914 
2915 	/*
2916 	 * Validate checksum here so we don't need to do it when
2917 	 * we actually parse the table
2918 	 */
2919 	ret = check_ivrs_checksum(ivrs_base);
2920 	if (ret)
2921 		goto out;
2922 
2923 	ivinfo_init(ivrs_base);
2924 
2925 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2926 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2927 
2928 	dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE);
2929 	alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2930 
2931 	/* Device table - directly used by all IOMMUs */
2932 	ret = -ENOMEM;
2933 	amd_iommu_dev_table = (void *)__get_free_pages(
2934 				      GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2935 				      get_order(dev_table_size));
2936 	if (amd_iommu_dev_table == NULL)
2937 		goto out;
2938 
2939 	/*
2940 	 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2941 	 * IOMMU see for that device
2942 	 */
2943 	amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2944 			get_order(alias_table_size));
2945 	if (amd_iommu_alias_table == NULL)
2946 		goto out;
2947 
2948 	amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2949 					    GFP_KERNEL | __GFP_ZERO,
2950 					    get_order(MAX_DOMAIN_ID/8));
2951 	if (amd_iommu_pd_alloc_bitmap == NULL)
2952 		goto out;
2953 
2954 	/*
2955 	 * never allocate domain 0 because its used as the non-allocated and
2956 	 * error value placeholder
2957 	 */
2958 	__set_bit(0, amd_iommu_pd_alloc_bitmap);
2959 
2960 	/*
2961 	 * now the data structures are allocated and basically initialized
2962 	 * start the real acpi table scan
2963 	 */
2964 	ret = init_iommu_all(ivrs_base);
2965 	if (ret)
2966 		goto out;
2967 
2968 	/* Disable any previously enabled IOMMUs */
2969 	if (!is_kdump_kernel() || amd_iommu_disabled)
2970 		disable_iommus();
2971 
2972 	if (amd_iommu_irq_remap)
2973 		amd_iommu_irq_remap = check_ioapic_information();
2974 
2975 	if (amd_iommu_irq_remap) {
2976 		struct amd_iommu_pci_seg *pci_seg;
2977 		/*
2978 		 * Interrupt remapping enabled, create kmem_cache for the
2979 		 * remapping tables.
2980 		 */
2981 		ret = -ENOMEM;
2982 		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2983 			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2984 		else
2985 			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2986 		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2987 							remap_cache_sz,
2988 							DTE_INTTAB_ALIGNMENT,
2989 							0, NULL);
2990 		if (!amd_iommu_irq_cache)
2991 			goto out;
2992 
2993 		for_each_pci_segment(pci_seg) {
2994 			if (alloc_irq_lookup_table(pci_seg))
2995 				goto out;
2996 		}
2997 	}
2998 
2999 	ret = init_memory_definitions(ivrs_base);
3000 	if (ret)
3001 		goto out;
3002 
3003 	/* init the device table */
3004 	init_device_table();
3005 
3006 out:
3007 	/* Don't leak any ACPI memory */
3008 	acpi_put_table(ivrs_base);
3009 
3010 	return ret;
3011 }
3012 
3013 static int amd_iommu_enable_interrupts(void)
3014 {
3015 	struct amd_iommu *iommu;
3016 	int ret = 0;
3017 
3018 	for_each_iommu(iommu) {
3019 		ret = iommu_init_irq(iommu);
3020 		if (ret)
3021 			goto out;
3022 	}
3023 
3024 out:
3025 	return ret;
3026 }
3027 
3028 static bool __init detect_ivrs(void)
3029 {
3030 	struct acpi_table_header *ivrs_base;
3031 	acpi_status status;
3032 	int i;
3033 
3034 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3035 	if (status == AE_NOT_FOUND)
3036 		return false;
3037 	else if (ACPI_FAILURE(status)) {
3038 		const char *err = acpi_format_exception(status);
3039 		pr_err("IVRS table error: %s\n", err);
3040 		return false;
3041 	}
3042 
3043 	acpi_put_table(ivrs_base);
3044 
3045 	if (amd_iommu_force_enable)
3046 		goto out;
3047 
3048 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3049 	for (i = 0; i < 32; i++) {
3050 		u32 pci_id;
3051 
3052 		pci_id = read_pci_config(0, i, 0, 0);
3053 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3054 			pr_info("Disable IOMMU on Stoney Ridge\n");
3055 			return false;
3056 		}
3057 	}
3058 
3059 out:
3060 	/* Make sure ACS will be enabled during PCI probe */
3061 	pci_request_acs();
3062 
3063 	return true;
3064 }
3065 
3066 /****************************************************************************
3067  *
3068  * AMD IOMMU Initialization State Machine
3069  *
3070  ****************************************************************************/
3071 
3072 static int __init state_next(void)
3073 {
3074 	int ret = 0;
3075 
3076 	switch (init_state) {
3077 	case IOMMU_START_STATE:
3078 		if (!detect_ivrs()) {
3079 			init_state	= IOMMU_NOT_FOUND;
3080 			ret		= -ENODEV;
3081 		} else {
3082 			init_state	= IOMMU_IVRS_DETECTED;
3083 		}
3084 		break;
3085 	case IOMMU_IVRS_DETECTED:
3086 		if (amd_iommu_disabled) {
3087 			init_state = IOMMU_CMDLINE_DISABLED;
3088 			ret = -EINVAL;
3089 		} else {
3090 			ret = early_amd_iommu_init();
3091 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3092 		}
3093 		break;
3094 	case IOMMU_ACPI_FINISHED:
3095 		early_enable_iommus();
3096 		x86_platform.iommu_shutdown = disable_iommus;
3097 		init_state = IOMMU_ENABLED;
3098 		break;
3099 	case IOMMU_ENABLED:
3100 		register_syscore_ops(&amd_iommu_syscore_ops);
3101 		ret = amd_iommu_init_pci();
3102 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3103 		enable_iommus_v2();
3104 		break;
3105 	case IOMMU_PCI_INIT:
3106 		ret = amd_iommu_enable_interrupts();
3107 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3108 		break;
3109 	case IOMMU_INTERRUPTS_EN:
3110 		init_state = IOMMU_INITIALIZED;
3111 		break;
3112 	case IOMMU_INITIALIZED:
3113 		/* Nothing to do */
3114 		break;
3115 	case IOMMU_NOT_FOUND:
3116 	case IOMMU_INIT_ERROR:
3117 	case IOMMU_CMDLINE_DISABLED:
3118 		/* Error states => do nothing */
3119 		ret = -EINVAL;
3120 		break;
3121 	default:
3122 		/* Unknown state */
3123 		BUG();
3124 	}
3125 
3126 	if (ret) {
3127 		free_dma_resources();
3128 		if (!irq_remapping_enabled) {
3129 			disable_iommus();
3130 			free_iommu_resources();
3131 		} else {
3132 			struct amd_iommu *iommu;
3133 			struct amd_iommu_pci_seg *pci_seg;
3134 
3135 			for_each_pci_segment(pci_seg)
3136 				uninit_device_table_dma(pci_seg);
3137 
3138 			for_each_iommu(iommu)
3139 				iommu_flush_all_caches(iommu);
3140 		}
3141 	}
3142 	return ret;
3143 }
3144 
3145 static int __init iommu_go_to_state(enum iommu_init_state state)
3146 {
3147 	int ret = -EINVAL;
3148 
3149 	while (init_state != state) {
3150 		if (init_state == IOMMU_NOT_FOUND         ||
3151 		    init_state == IOMMU_INIT_ERROR        ||
3152 		    init_state == IOMMU_CMDLINE_DISABLED)
3153 			break;
3154 		ret = state_next();
3155 	}
3156 
3157 	return ret;
3158 }
3159 
3160 #ifdef CONFIG_IRQ_REMAP
3161 int __init amd_iommu_prepare(void)
3162 {
3163 	int ret;
3164 
3165 	amd_iommu_irq_remap = true;
3166 
3167 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3168 	if (ret) {
3169 		amd_iommu_irq_remap = false;
3170 		return ret;
3171 	}
3172 
3173 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3174 }
3175 
3176 int __init amd_iommu_enable(void)
3177 {
3178 	int ret;
3179 
3180 	ret = iommu_go_to_state(IOMMU_ENABLED);
3181 	if (ret)
3182 		return ret;
3183 
3184 	irq_remapping_enabled = 1;
3185 	return amd_iommu_xt_mode;
3186 }
3187 
3188 void amd_iommu_disable(void)
3189 {
3190 	amd_iommu_suspend();
3191 }
3192 
3193 int amd_iommu_reenable(int mode)
3194 {
3195 	amd_iommu_resume();
3196 
3197 	return 0;
3198 }
3199 
3200 int __init amd_iommu_enable_faulting(void)
3201 {
3202 	/* We enable MSI later when PCI is initialized */
3203 	return 0;
3204 }
3205 #endif
3206 
3207 /*
3208  * This is the core init function for AMD IOMMU hardware in the system.
3209  * This function is called from the generic x86 DMA layer initialization
3210  * code.
3211  */
3212 static int __init amd_iommu_init(void)
3213 {
3214 	struct amd_iommu *iommu;
3215 	int ret;
3216 
3217 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3218 #ifdef CONFIG_GART_IOMMU
3219 	if (ret && list_empty(&amd_iommu_list)) {
3220 		/*
3221 		 * We failed to initialize the AMD IOMMU - try fallback
3222 		 * to GART if possible.
3223 		 */
3224 		gart_iommu_init();
3225 	}
3226 #endif
3227 
3228 	for_each_iommu(iommu)
3229 		amd_iommu_debugfs_setup(iommu);
3230 
3231 	return ret;
3232 }
3233 
3234 static bool amd_iommu_sme_check(void)
3235 {
3236 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3237 	    (boot_cpu_data.x86 != 0x17))
3238 		return true;
3239 
3240 	/* For Fam17h, a specific level of support is required */
3241 	if (boot_cpu_data.microcode >= 0x08001205)
3242 		return true;
3243 
3244 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3245 	    (boot_cpu_data.microcode <= 0x080011ff))
3246 		return true;
3247 
3248 	pr_notice("IOMMU not currently supported when SME is active\n");
3249 
3250 	return false;
3251 }
3252 
3253 /****************************************************************************
3254  *
3255  * Early detect code. This code runs at IOMMU detection time in the DMA
3256  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3257  * IOMMUs
3258  *
3259  ****************************************************************************/
3260 int __init amd_iommu_detect(void)
3261 {
3262 	int ret;
3263 
3264 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3265 		return -ENODEV;
3266 
3267 	if (!amd_iommu_sme_check())
3268 		return -ENODEV;
3269 
3270 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3271 	if (ret)
3272 		return ret;
3273 
3274 	amd_iommu_detected = true;
3275 	iommu_detected = 1;
3276 	x86_init.iommu.iommu_init = amd_iommu_init;
3277 
3278 	return 1;
3279 }
3280 
3281 /****************************************************************************
3282  *
3283  * Parsing functions for the AMD IOMMU specific kernel command line
3284  * options.
3285  *
3286  ****************************************************************************/
3287 
3288 static int __init parse_amd_iommu_dump(char *str)
3289 {
3290 	amd_iommu_dump = true;
3291 
3292 	return 1;
3293 }
3294 
3295 static int __init parse_amd_iommu_intr(char *str)
3296 {
3297 	for (; *str; ++str) {
3298 		if (strncmp(str, "legacy", 6) == 0) {
3299 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3300 			break;
3301 		}
3302 		if (strncmp(str, "vapic", 5) == 0) {
3303 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3304 			break;
3305 		}
3306 	}
3307 	return 1;
3308 }
3309 
3310 static int __init parse_amd_iommu_options(char *str)
3311 {
3312 	for (; *str; ++str) {
3313 		if (strncmp(str, "fullflush", 9) == 0) {
3314 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3315 			iommu_set_dma_strict();
3316 		}
3317 		if (strncmp(str, "force_enable", 12) == 0)
3318 			amd_iommu_force_enable = true;
3319 		if (strncmp(str, "off", 3) == 0)
3320 			amd_iommu_disabled = true;
3321 		if (strncmp(str, "force_isolation", 15) == 0)
3322 			amd_iommu_force_isolation = true;
3323 	}
3324 
3325 	return 1;
3326 }
3327 
3328 static int __init parse_ivrs_ioapic(char *str)
3329 {
3330 	unsigned int bus, dev, fn;
3331 	int ret, id, i;
3332 	u16 devid;
3333 
3334 	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
3335 
3336 	if (ret != 4) {
3337 		pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3338 		return 1;
3339 	}
3340 
3341 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3342 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3343 			str);
3344 		return 1;
3345 	}
3346 
3347 	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3348 
3349 	cmdline_maps			= true;
3350 	i				= early_ioapic_map_size++;
3351 	early_ioapic_map[i].id		= id;
3352 	early_ioapic_map[i].devid	= devid;
3353 	early_ioapic_map[i].cmd_line	= true;
3354 
3355 	return 1;
3356 }
3357 
3358 static int __init parse_ivrs_hpet(char *str)
3359 {
3360 	unsigned int bus, dev, fn;
3361 	int ret, id, i;
3362 	u16 devid;
3363 
3364 	ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
3365 
3366 	if (ret != 4) {
3367 		pr_err("Invalid command line: ivrs_hpet%s\n", str);
3368 		return 1;
3369 	}
3370 
3371 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3372 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3373 			str);
3374 		return 1;
3375 	}
3376 
3377 	devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3378 
3379 	cmdline_maps			= true;
3380 	i				= early_hpet_map_size++;
3381 	early_hpet_map[i].id		= id;
3382 	early_hpet_map[i].devid		= devid;
3383 	early_hpet_map[i].cmd_line	= true;
3384 
3385 	return 1;
3386 }
3387 
3388 static int __init parse_ivrs_acpihid(char *str)
3389 {
3390 	u32 bus, dev, fn;
3391 	char *hid, *uid, *p;
3392 	char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
3393 	int ret, i;
3394 
3395 	ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
3396 	if (ret != 4) {
3397 		pr_err("Invalid command line: ivrs_acpihid(%s)\n", str);
3398 		return 1;
3399 	}
3400 
3401 	p = acpiid;
3402 	hid = strsep(&p, ":");
3403 	uid = p;
3404 
3405 	if (!hid || !(*hid) || !uid) {
3406 		pr_err("Invalid command line: hid or uid\n");
3407 		return 1;
3408 	}
3409 
3410 	i = early_acpihid_map_size++;
3411 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3412 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3413 	early_acpihid_map[i].devid =
3414 		((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
3415 	early_acpihid_map[i].cmd_line	= true;
3416 
3417 	return 1;
3418 }
3419 
3420 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3421 __setup("amd_iommu=",		parse_amd_iommu_options);
3422 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3423 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3424 __setup("ivrs_hpet",		parse_ivrs_hpet);
3425 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3426 
3427 bool amd_iommu_v2_supported(void)
3428 {
3429 	return amd_iommu_v2_present;
3430 }
3431 EXPORT_SYMBOL(amd_iommu_v2_supported);
3432 
3433 struct amd_iommu *get_amd_iommu(unsigned int idx)
3434 {
3435 	unsigned int i = 0;
3436 	struct amd_iommu *iommu;
3437 
3438 	for_each_iommu(iommu)
3439 		if (i++ == idx)
3440 			return iommu;
3441 	return NULL;
3442 }
3443 
3444 /****************************************************************************
3445  *
3446  * IOMMU EFR Performance Counter support functionality. This code allows
3447  * access to the IOMMU PC functionality.
3448  *
3449  ****************************************************************************/
3450 
3451 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3452 {
3453 	struct amd_iommu *iommu = get_amd_iommu(idx);
3454 
3455 	if (iommu)
3456 		return iommu->max_banks;
3457 
3458 	return 0;
3459 }
3460 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3461 
3462 bool amd_iommu_pc_supported(void)
3463 {
3464 	return amd_iommu_pc_present;
3465 }
3466 EXPORT_SYMBOL(amd_iommu_pc_supported);
3467 
3468 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3469 {
3470 	struct amd_iommu *iommu = get_amd_iommu(idx);
3471 
3472 	if (iommu)
3473 		return iommu->max_counters;
3474 
3475 	return 0;
3476 }
3477 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3478 
3479 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3480 				u8 fxn, u64 *value, bool is_write)
3481 {
3482 	u32 offset;
3483 	u32 max_offset_lim;
3484 
3485 	/* Make sure the IOMMU PC resource is available */
3486 	if (!amd_iommu_pc_present)
3487 		return -ENODEV;
3488 
3489 	/* Check for valid iommu and pc register indexing */
3490 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3491 		return -ENODEV;
3492 
3493 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3494 
3495 	/* Limit the offset to the hw defined mmio region aperture */
3496 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3497 				(iommu->max_counters << 8) | 0x28);
3498 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3499 	    (offset > max_offset_lim))
3500 		return -EINVAL;
3501 
3502 	if (is_write) {
3503 		u64 val = *value & GENMASK_ULL(47, 0);
3504 
3505 		writel((u32)val, iommu->mmio_base + offset);
3506 		writel((val >> 32), iommu->mmio_base + offset + 4);
3507 	} else {
3508 		*value = readl(iommu->mmio_base + offset + 4);
3509 		*value <<= 32;
3510 		*value |= readl(iommu->mmio_base + offset);
3511 		*value &= GENMASK_ULL(47, 0);
3512 	}
3513 
3514 	return 0;
3515 }
3516 
3517 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3518 {
3519 	if (!iommu)
3520 		return -EINVAL;
3521 
3522 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3523 }
3524 
3525 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3526 {
3527 	if (!iommu)
3528 		return -EINVAL;
3529 
3530 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3531 }
3532