xref: /linux/drivers/iommu/amd/init.c (revision dd8a3c6cd531dca5917111a94fa3074077f6ba5a)
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/syscore_ops.h>
16 #include <linux/interrupt.h>
17 #include <linux/msi.h>
18 #include <linux/irq.h>
19 #include <linux/amd-iommu.h>
20 #include <linux/export.h>
21 #include <linux/kmemleak.h>
22 #include <linux/cc_platform.h>
23 #include <linux/iopoll.h>
24 #include <asm/pci-direct.h>
25 #include <asm/iommu.h>
26 #include <asm/apic.h>
27 #include <asm/gart.h>
28 #include <asm/x86_init.h>
29 #include <asm/io_apic.h>
30 #include <asm/irq_remapping.h>
31 #include <asm/set_memory.h>
32 #include <asm/sev.h>
33 
34 #include <linux/crash_dump.h>
35 
36 #include "amd_iommu.h"
37 #include "../irq_remapping.h"
38 #include "../iommu-pages.h"
39 
40 /*
41  * definitions for the ACPI scanning code
42  */
43 #define IVRS_HEADER_LENGTH 48
44 
45 #define ACPI_IVHD_TYPE_MAX_SUPPORTED	0x40
46 #define ACPI_IVMD_TYPE_ALL              0x20
47 #define ACPI_IVMD_TYPE                  0x21
48 #define ACPI_IVMD_TYPE_RANGE            0x22
49 
50 #define IVHD_DEV_ALL                    0x01
51 #define IVHD_DEV_SELECT                 0x02
52 #define IVHD_DEV_SELECT_RANGE_START     0x03
53 #define IVHD_DEV_RANGE_END              0x04
54 #define IVHD_DEV_ALIAS                  0x42
55 #define IVHD_DEV_ALIAS_RANGE            0x43
56 #define IVHD_DEV_EXT_SELECT             0x46
57 #define IVHD_DEV_EXT_SELECT_RANGE       0x47
58 #define IVHD_DEV_SPECIAL		0x48
59 #define IVHD_DEV_ACPI_HID		0xf0
60 
61 #define UID_NOT_PRESENT                 0
62 #define UID_IS_INTEGER                  1
63 #define UID_IS_CHARACTER                2
64 
65 #define IVHD_SPECIAL_IOAPIC		1
66 #define IVHD_SPECIAL_HPET		2
67 
68 #define IVHD_FLAG_HT_TUN_EN_MASK        0x01
69 #define IVHD_FLAG_PASSPW_EN_MASK        0x02
70 #define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
71 #define IVHD_FLAG_ISOC_EN_MASK          0x08
72 
73 #define IVMD_FLAG_EXCL_RANGE            0x08
74 #define IVMD_FLAG_IW                    0x04
75 #define IVMD_FLAG_IR                    0x02
76 #define IVMD_FLAG_UNITY_MAP             0x01
77 
78 #define ACPI_DEVFLAG_INITPASS           0x01
79 #define ACPI_DEVFLAG_EXTINT             0x02
80 #define ACPI_DEVFLAG_NMI                0x04
81 #define ACPI_DEVFLAG_SYSMGT1            0x10
82 #define ACPI_DEVFLAG_SYSMGT2            0x20
83 #define ACPI_DEVFLAG_LINT0              0x40
84 #define ACPI_DEVFLAG_LINT1              0x80
85 #define ACPI_DEVFLAG_ATSDIS             0x10000000
86 
87 #define IVRS_GET_SBDF_ID(seg, bus, dev, fn)	(((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \
88 						 | ((dev & 0x1f) << 3) | (fn & 0x7))
89 
90 /*
91  * ACPI table definitions
92  *
93  * These data structures are laid over the table to parse the important values
94  * out of it.
95  */
96 
97 /*
98  * structure describing one IOMMU in the ACPI table. Typically followed by one
99  * or more ivhd_entrys.
100  */
101 struct ivhd_header {
102 	u8 type;
103 	u8 flags;
104 	u16 length;
105 	u16 devid;
106 	u16 cap_ptr;
107 	u64 mmio_phys;
108 	u16 pci_seg;
109 	u16 info;
110 	u32 efr_attr;
111 
112 	/* Following only valid on IVHD type 11h and 40h */
113 	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
114 	u64 efr_reg2;
115 } __attribute__((packed));
116 
117 /*
118  * A device entry describing which devices a specific IOMMU translates and
119  * which requestor ids they use.
120  */
121 struct ivhd_entry {
122 	u8 type;
123 	u16 devid;
124 	u8 flags;
125 	struct_group(ext_hid,
126 		u32 ext;
127 		u32 hidh;
128 	);
129 	u64 cid;
130 	u8 uidf;
131 	u8 uidl;
132 	u8 uid;
133 } __attribute__((packed));
134 
135 int amd_iommu_evtlog_size = EVTLOG_SIZE_DEF;
136 int amd_iommu_pprlog_size = PPRLOG_SIZE_DEF;
137 
138 /*
139  * An AMD IOMMU memory definition structure. It defines things like exclusion
140  * ranges for devices and regions that should be unity mapped.
141  */
142 struct ivmd_header {
143 	u8 type;
144 	u8 flags;
145 	u16 length;
146 	u16 devid;
147 	u16 aux;
148 	u16 pci_seg;
149 	u8  resv[6];
150 	u64 range_start;
151 	u64 range_length;
152 } __attribute__((packed));
153 
154 bool amd_iommu_dump;
155 bool amd_iommu_irq_remap __read_mostly;
156 
157 enum protection_domain_mode amd_iommu_pgtable = PD_MODE_V1;
158 /* Virtual address size */
159 u8 amd_iommu_hpt_vasize;
160 /* Guest page table level */
161 int amd_iommu_gpt_level = PAGE_MODE_4_LEVEL;
162 
163 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
164 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
165 
166 static bool amd_iommu_detected;
167 static bool amd_iommu_disabled __initdata;
168 static bool amd_iommu_force_enable __initdata;
169 static bool amd_iommu_irtcachedis;
170 static int amd_iommu_target_ivhd_type;
171 
172 /* Global EFR and EFR2 registers */
173 u64 amd_iommu_efr;
174 u64 amd_iommu_efr2;
175 
176 /* Host (v1) page table is not supported*/
177 bool amd_iommu_hatdis;
178 
179 /* SNP is enabled on the system? */
180 bool amd_iommu_snp_en;
181 EXPORT_SYMBOL(amd_iommu_snp_en);
182 
183 LIST_HEAD(amd_iommu_pci_seg_list);	/* list of all PCI segments */
184 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the system */
185 LIST_HEAD(amd_ivhd_dev_flags_list);	/* list of all IVHD device entry settings */
186 
187 /* Number of IOMMUs present in the system */
188 static int amd_iommus_present;
189 
190 /* IOMMUs have a non-present cache? */
191 bool amd_iommu_np_cache __read_mostly;
192 bool amd_iommu_iotlb_sup __read_mostly = true;
193 
194 static bool amd_iommu_pc_present __read_mostly;
195 bool amdr_ivrs_remap_support __read_mostly;
196 
197 bool amd_iommu_force_isolation __read_mostly;
198 
199 unsigned long amd_iommu_pgsize_bitmap __ro_after_init = AMD_IOMMU_PGSIZES;
200 
201 enum iommu_init_state {
202 	IOMMU_START_STATE,
203 	IOMMU_IVRS_DETECTED,
204 	IOMMU_ACPI_FINISHED,
205 	IOMMU_ENABLED,
206 	IOMMU_PCI_INIT,
207 	IOMMU_INTERRUPTS_EN,
208 	IOMMU_INITIALIZED,
209 	IOMMU_NOT_FOUND,
210 	IOMMU_INIT_ERROR,
211 	IOMMU_CMDLINE_DISABLED,
212 };
213 
214 /* Early ioapic and hpet maps from kernel command line */
215 #define EARLY_MAP_SIZE		4
216 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
217 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
218 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
219 
220 static int __initdata early_ioapic_map_size;
221 static int __initdata early_hpet_map_size;
222 static int __initdata early_acpihid_map_size;
223 
224 static bool __initdata cmdline_maps;
225 
226 static enum iommu_init_state init_state = IOMMU_START_STATE;
227 
228 static int amd_iommu_enable_interrupts(void);
229 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
230 
231 static bool amd_iommu_pre_enabled = true;
232 
233 static u32 amd_iommu_ivinfo __initdata;
234 
235 bool translation_pre_enabled(struct amd_iommu *iommu)
236 {
237 	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
238 }
239 
240 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
241 {
242 	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
243 }
244 
245 static void init_translation_status(struct amd_iommu *iommu)
246 {
247 	u64 ctrl;
248 
249 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
250 	if (ctrl & (1<<CONTROL_IOMMU_EN))
251 		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
252 }
253 
254 int amd_iommu_get_num_iommus(void)
255 {
256 	return amd_iommus_present;
257 }
258 
259 bool amd_iommu_ht_range_ignore(void)
260 {
261 	return check_feature2(FEATURE_HT_RANGE_IGNORE);
262 }
263 
264 /*
265  * Iterate through all the IOMMUs to get common EFR
266  * masks among all IOMMUs and warn if found inconsistency.
267  */
268 static __init void get_global_efr(void)
269 {
270 	struct amd_iommu *iommu;
271 
272 	for_each_iommu(iommu) {
273 		u64 tmp = iommu->features;
274 		u64 tmp2 = iommu->features2;
275 
276 		if (list_is_first(&iommu->list, &amd_iommu_list)) {
277 			amd_iommu_efr = tmp;
278 			amd_iommu_efr2 = tmp2;
279 			continue;
280 		}
281 
282 		if (amd_iommu_efr == tmp &&
283 		    amd_iommu_efr2 == tmp2)
284 			continue;
285 
286 		pr_err(FW_BUG
287 		       "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n",
288 		       tmp, tmp2, amd_iommu_efr, amd_iommu_efr2,
289 		       iommu->index, iommu->pci_seg->id,
290 		       PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid),
291 		       PCI_FUNC(iommu->devid));
292 
293 		amd_iommu_efr &= tmp;
294 		amd_iommu_efr2 &= tmp2;
295 	}
296 
297 	pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2);
298 }
299 
300 /*
301  * For IVHD type 0x11/0x40, EFR is also available via IVHD.
302  * Default to IVHD EFR since it is available sooner
303  * (i.e. before PCI init).
304  */
305 static void __init early_iommu_features_init(struct amd_iommu *iommu,
306 					     struct ivhd_header *h)
307 {
308 	if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) {
309 		iommu->features = h->efr_reg;
310 		iommu->features2 = h->efr_reg2;
311 	}
312 	if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
313 		amdr_ivrs_remap_support = true;
314 }
315 
316 /* Access to l1 and l2 indexed register spaces */
317 
318 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
319 {
320 	u32 val;
321 
322 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
323 	pci_read_config_dword(iommu->dev, 0xfc, &val);
324 	return val;
325 }
326 
327 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
328 {
329 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
330 	pci_write_config_dword(iommu->dev, 0xfc, val);
331 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
332 }
333 
334 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
335 {
336 	u32 val;
337 
338 	pci_write_config_dword(iommu->dev, 0xf0, address);
339 	pci_read_config_dword(iommu->dev, 0xf4, &val);
340 	return val;
341 }
342 
343 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
344 {
345 	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
346 	pci_write_config_dword(iommu->dev, 0xf4, val);
347 }
348 
349 /****************************************************************************
350  *
351  * AMD IOMMU MMIO register space handling functions
352  *
353  * These functions are used to program the IOMMU device registers in
354  * MMIO space required for that driver.
355  *
356  ****************************************************************************/
357 
358 static void iommu_set_cwwb_range(struct amd_iommu *iommu)
359 {
360 	u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
361 	u64 entry = start & PM_ADDR_MASK;
362 
363 	if (!check_feature(FEATURE_SNP))
364 		return;
365 
366 	/* Note:
367 	 * Re-purpose Exclusion base/limit registers for Completion wait
368 	 * write-back base/limit.
369 	 */
370 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
371 		    &entry, sizeof(entry));
372 
373 	/* Note:
374 	 * Default to 4 Kbytes, which can be specified by setting base
375 	 * address equal to the limit address.
376 	 */
377 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
378 		    &entry, sizeof(entry));
379 }
380 
381 /* Programs the physical address of the device table into the IOMMU hardware */
382 static void iommu_set_device_table(struct amd_iommu *iommu)
383 {
384 	u64 entry;
385 	u32 dev_table_size = iommu->pci_seg->dev_table_size;
386 	void *dev_table = (void *)get_dev_table(iommu);
387 
388 	BUG_ON(iommu->mmio_base == NULL);
389 
390 	if (is_kdump_kernel())
391 		return;
392 
393 	entry = iommu_virt_to_phys(dev_table);
394 	entry |= (dev_table_size >> 12) - 1;
395 	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
396 			&entry, sizeof(entry));
397 }
398 
399 static void iommu_feature_set(struct amd_iommu *iommu, u64 val, u64 mask, u8 shift)
400 {
401 	u64 ctrl;
402 
403 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
404 	mask <<= shift;
405 	ctrl &= ~mask;
406 	ctrl |= (val << shift) & mask;
407 	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
408 }
409 
410 /* Generic functions to enable/disable certain features of the IOMMU. */
411 void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
412 {
413 	iommu_feature_set(iommu, 1ULL, 1ULL, bit);
414 }
415 
416 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
417 {
418 	iommu_feature_set(iommu, 0ULL, 1ULL, bit);
419 }
420 
421 /* Function to enable the hardware */
422 static void iommu_enable(struct amd_iommu *iommu)
423 {
424 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
425 }
426 
427 static void iommu_disable(struct amd_iommu *iommu)
428 {
429 	if (!iommu->mmio_base)
430 		return;
431 
432 	/* Disable command buffer */
433 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
434 
435 	/* Disable event logging and event interrupts */
436 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
437 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
438 
439 	/* Disable IOMMU GA_LOG */
440 	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
441 	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
442 
443 	/* Disable IOMMU PPR logging */
444 	iommu_feature_disable(iommu, CONTROL_PPRLOG_EN);
445 	iommu_feature_disable(iommu, CONTROL_PPRINT_EN);
446 
447 	/* Disable IOMMU hardware itself */
448 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
449 
450 	/* Clear IRTE cache disabling bit */
451 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
452 }
453 
454 /*
455  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
456  * the system has one.
457  */
458 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
459 {
460 	if (!request_mem_region(address, end, "amd_iommu")) {
461 		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
462 			address, end);
463 		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
464 		return NULL;
465 	}
466 
467 	return (u8 __iomem *)ioremap(address, end);
468 }
469 
470 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
471 {
472 	if (iommu->mmio_base)
473 		iounmap(iommu->mmio_base);
474 	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
475 }
476 
477 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
478 {
479 	u32 size = 0;
480 
481 	switch (h->type) {
482 	case 0x10:
483 		size = 24;
484 		break;
485 	case 0x11:
486 	case 0x40:
487 		size = 40;
488 		break;
489 	}
490 	return size;
491 }
492 
493 /****************************************************************************
494  *
495  * The functions below belong to the first pass of AMD IOMMU ACPI table
496  * parsing. In this pass we try to find out the highest device id this
497  * code has to handle. Upon this information the size of the shared data
498  * structures is determined later.
499  *
500  ****************************************************************************/
501 
502 /*
503  * This function calculates the length of a given IVHD entry
504  */
505 static inline int ivhd_entry_length(u8 *ivhd)
506 {
507 	u32 type = ((struct ivhd_entry *)ivhd)->type;
508 
509 	if (type < 0x80) {
510 		return 0x04 << (*ivhd >> 6);
511 	} else if (type == IVHD_DEV_ACPI_HID) {
512 		/* For ACPI_HID, offset 21 is uid len */
513 		return *((u8 *)ivhd + 21) + 22;
514 	}
515 	return 0;
516 }
517 
518 /*
519  * After reading the highest device id from the IOMMU PCI capability header
520  * this function looks if there is a higher device id defined in the ACPI table
521  */
522 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
523 {
524 	u8 *p = (void *)h, *end = (void *)h;
525 	struct ivhd_entry *dev;
526 	int last_devid = -EINVAL;
527 
528 	u32 ivhd_size = get_ivhd_header_size(h);
529 
530 	if (!ivhd_size) {
531 		pr_err("Unsupported IVHD type %#x\n", h->type);
532 		return -EINVAL;
533 	}
534 
535 	p += ivhd_size;
536 	end += h->length;
537 
538 	while (p < end) {
539 		dev = (struct ivhd_entry *)p;
540 		switch (dev->type) {
541 		case IVHD_DEV_ALL:
542 			/* Use maximum BDF value for DEV_ALL */
543 			return 0xffff;
544 		case IVHD_DEV_SELECT:
545 		case IVHD_DEV_RANGE_END:
546 		case IVHD_DEV_ALIAS:
547 		case IVHD_DEV_EXT_SELECT:
548 			/* all the above subfield types refer to device ids */
549 			if (dev->devid > last_devid)
550 				last_devid = dev->devid;
551 			break;
552 		default:
553 			break;
554 		}
555 		p += ivhd_entry_length(p);
556 	}
557 
558 	WARN_ON(p != end);
559 
560 	return last_devid;
561 }
562 
563 static int __init check_ivrs_checksum(struct acpi_table_header *table)
564 {
565 	int i;
566 	u8 checksum = 0, *p = (u8 *)table;
567 
568 	for (i = 0; i < table->length; ++i)
569 		checksum += p[i];
570 	if (checksum != 0) {
571 		/* ACPI table corrupt */
572 		pr_err(FW_BUG "IVRS invalid checksum\n");
573 		return -ENODEV;
574 	}
575 
576 	return 0;
577 }
578 
579 /*
580  * Iterate over all IVHD entries in the ACPI table and find the highest device
581  * id which we need to handle. This is the first of three functions which parse
582  * the ACPI table. So we check the checksum here.
583  */
584 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
585 {
586 	u8 *p = (u8 *)table, *end = (u8 *)table;
587 	struct ivhd_header *h;
588 	int last_devid, last_bdf = 0;
589 
590 	p += IVRS_HEADER_LENGTH;
591 
592 	end += table->length;
593 	while (p < end) {
594 		h = (struct ivhd_header *)p;
595 		if (h->pci_seg == pci_seg &&
596 		    h->type == amd_iommu_target_ivhd_type) {
597 			last_devid = find_last_devid_from_ivhd(h);
598 
599 			if (last_devid < 0)
600 				return -EINVAL;
601 			if (last_devid > last_bdf)
602 				last_bdf = last_devid;
603 		}
604 		p += h->length;
605 	}
606 	WARN_ON(p != end);
607 
608 	return last_bdf;
609 }
610 
611 /****************************************************************************
612  *
613  * The following functions belong to the code path which parses the ACPI table
614  * the second time. In this ACPI parsing iteration we allocate IOMMU specific
615  * data structures, initialize the per PCI segment device/alias/rlookup table
616  * and also basically initialize the hardware.
617  *
618  ****************************************************************************/
619 
620 /* Allocate per PCI segment device table */
621 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
622 {
623 	pci_seg->dev_table = iommu_alloc_pages_sz(GFP_KERNEL | GFP_DMA32,
624 						  pci_seg->dev_table_size);
625 	if (!pci_seg->dev_table)
626 		return -ENOMEM;
627 
628 	return 0;
629 }
630 
631 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
632 {
633 	if (is_kdump_kernel())
634 		memunmap((void *)pci_seg->dev_table);
635 	else
636 		iommu_free_pages(pci_seg->dev_table);
637 	pci_seg->dev_table = NULL;
638 }
639 
640 /* Allocate per PCI segment IOMMU rlookup table. */
641 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
642 {
643 	pci_seg->rlookup_table = kvzalloc_objs(*pci_seg->rlookup_table,
644 					       pci_seg->last_bdf + 1);
645 	if (pci_seg->rlookup_table == NULL)
646 		return -ENOMEM;
647 
648 	return 0;
649 }
650 
651 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
652 {
653 	kvfree(pci_seg->rlookup_table);
654 	pci_seg->rlookup_table = NULL;
655 }
656 
657 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
658 {
659 	pci_seg->irq_lookup_table = kvzalloc_objs(*pci_seg->irq_lookup_table,
660 						  pci_seg->last_bdf + 1);
661 	if (pci_seg->irq_lookup_table == NULL)
662 		return -ENOMEM;
663 
664 	return 0;
665 }
666 
667 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
668 {
669 	kvfree(pci_seg->irq_lookup_table);
670 	pci_seg->irq_lookup_table = NULL;
671 }
672 
673 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
674 {
675 	int i;
676 
677 	pci_seg->alias_table = kvmalloc_objs(*pci_seg->alias_table,
678 					     pci_seg->last_bdf + 1);
679 	if (!pci_seg->alias_table)
680 		return -ENOMEM;
681 
682 	/*
683 	 * let all alias entries point to itself
684 	 */
685 	for (i = 0; i <= pci_seg->last_bdf; ++i)
686 		pci_seg->alias_table[i] = i;
687 
688 	return 0;
689 }
690 
691 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
692 {
693 	kvfree(pci_seg->alias_table);
694 	pci_seg->alias_table = NULL;
695 }
696 
697 static inline void *iommu_memremap(unsigned long paddr, size_t size)
698 {
699 	phys_addr_t phys;
700 
701 	if (!paddr)
702 		return NULL;
703 
704 	/*
705 	 * Obtain true physical address in kdump kernel when SME is enabled.
706 	 * Currently, previous kernel with SME enabled and kdump kernel
707 	 * with SME support disabled is not supported.
708 	 */
709 	phys = __sme_clr(paddr);
710 
711 	if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
712 		return (__force void *)ioremap_encrypted(phys, size);
713 	else
714 		return memremap(phys, size, MEMREMAP_WB);
715 }
716 
717 /*
718  * Allocates the command buffer. This buffer is per AMD IOMMU. We can
719  * write commands to that buffer later and the IOMMU will execute them
720  * asynchronously
721  */
722 static int __init alloc_command_buffer(struct amd_iommu *iommu)
723 {
724 	iommu->cmd_buf = iommu_alloc_pages_sz(GFP_KERNEL, CMD_BUFFER_SIZE);
725 
726 	return iommu->cmd_buf ? 0 : -ENOMEM;
727 }
728 
729 /*
730  * Interrupt handler has processed all pending events and adjusted head
731  * and tail pointer. Reset overflow mask and restart logging again.
732  */
733 void amd_iommu_restart_log(struct amd_iommu *iommu, const char *evt_type,
734 			   u8 cntrl_intr, u8 cntrl_log,
735 			   u32 status_run_mask, u32 status_overflow_mask)
736 {
737 	u32 status;
738 
739 	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
740 	if (status & status_run_mask)
741 		return;
742 
743 	pr_info_ratelimited("IOMMU %s log restarting\n", evt_type);
744 
745 	iommu_feature_disable(iommu, cntrl_log);
746 	iommu_feature_disable(iommu, cntrl_intr);
747 
748 	writel(status_overflow_mask, iommu->mmio_base + MMIO_STATUS_OFFSET);
749 
750 	iommu_feature_enable(iommu, cntrl_intr);
751 	iommu_feature_enable(iommu, cntrl_log);
752 }
753 
754 /*
755  * This function restarts event logging in case the IOMMU experienced
756  * an event log buffer overflow.
757  */
758 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
759 {
760 	amd_iommu_restart_log(iommu, "Event", CONTROL_EVT_INT_EN,
761 			      CONTROL_EVT_LOG_EN, MMIO_STATUS_EVT_RUN_MASK,
762 			      MMIO_STATUS_EVT_OVERFLOW_MASK);
763 }
764 
765 /*
766  * This function restarts event logging in case the IOMMU experienced
767  * GA log overflow.
768  */
769 void amd_iommu_restart_ga_log(struct amd_iommu *iommu)
770 {
771 	amd_iommu_restart_log(iommu, "GA", CONTROL_GAINT_EN,
772 			      CONTROL_GALOG_EN, MMIO_STATUS_GALOG_RUN_MASK,
773 			      MMIO_STATUS_GALOG_OVERFLOW_MASK);
774 }
775 
776 /*
777  * This function resets the command buffer if the IOMMU stopped fetching
778  * commands from it.
779  */
780 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
781 {
782 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
783 
784 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
785 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
786 	iommu->cmd_buf_head = 0;
787 	iommu->cmd_buf_tail = 0;
788 
789 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
790 }
791 
792 /*
793  * This function writes the command buffer address to the hardware and
794  * enables it.
795  */
796 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
797 {
798 	u64 entry;
799 
800 	BUG_ON(iommu->cmd_buf == NULL);
801 
802 	if (!is_kdump_kernel()) {
803 		/*
804 		 * Command buffer is re-used for kdump kernel and setting
805 		 * of MMIO register is not required.
806 		 */
807 		entry = iommu_virt_to_phys(iommu->cmd_buf);
808 		entry |= MMIO_CMD_SIZE_512;
809 		memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
810 			    &entry, sizeof(entry));
811 	}
812 
813 	amd_iommu_reset_cmd_buffer(iommu);
814 }
815 
816 /*
817  * This function disables the command buffer
818  */
819 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
820 {
821 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
822 }
823 
824 static void __init free_command_buffer(struct amd_iommu *iommu)
825 {
826 	iommu_free_pages(iommu->cmd_buf);
827 }
828 
829 void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
830 				  size_t size)
831 {
832 	int nid = iommu->dev ? dev_to_node(&iommu->dev->dev) : NUMA_NO_NODE;
833 	void *buf;
834 
835 	size = PAGE_ALIGN(size);
836 	buf = iommu_alloc_pages_node_sz(nid, gfp, size);
837 	if (!buf)
838 		return NULL;
839 	if (check_feature(FEATURE_SNP) &&
840 	    set_memory_4k((unsigned long)buf, size / PAGE_SIZE)) {
841 		iommu_free_pages(buf);
842 		return NULL;
843 	}
844 
845 	return buf;
846 }
847 
848 /* allocates the memory where the IOMMU will log its events to */
849 static int __init alloc_event_buffer(void)
850 {
851 	struct amd_iommu *iommu;
852 
853 	for_each_iommu(iommu) {
854 		iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL,
855 						      amd_iommu_evtlog_size);
856 		if (!iommu->evt_buf)
857 			return -ENOMEM;
858 	}
859 
860 	return 0;
861 }
862 
863 static void iommu_enable_event_buffer(void)
864 {
865 	struct amd_iommu *iommu;
866 	u64 entry;
867 
868 	for_each_iommu(iommu) {
869 		BUG_ON(iommu->evt_buf == NULL);
870 
871 		if (!is_kdump_kernel()) {
872 			/*
873 			 * Event buffer is re-used for kdump kernel and setting
874 			 * of MMIO register is not required.
875 			 */
876 			entry = iommu_virt_to_phys(iommu->evt_buf);
877 			entry |= (amd_iommu_evtlog_size == EVTLOG_SIZE_DEF) ?
878 				EVTLOG_LEN_MASK_DEF : EVTLOG_LEN_MASK_MAX;
879 
880 			memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
881 				    &entry, sizeof(entry));
882 		}
883 
884 		/* set head and tail to zero manually */
885 		writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
886 		writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
887 
888 		iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
889 	}
890 }
891 
892 /*
893  * This function disables the event log buffer
894  */
895 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
896 {
897 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
898 }
899 
900 static void __init free_event_buffer(struct amd_iommu *iommu)
901 {
902 	iommu_free_pages(iommu->evt_buf);
903 }
904 
905 static void free_ga_log(struct amd_iommu *iommu)
906 {
907 #ifdef CONFIG_IRQ_REMAP
908 	iommu_free_pages(iommu->ga_log);
909 	iommu_free_pages(iommu->ga_log_tail);
910 #endif
911 }
912 
913 #ifdef CONFIG_IRQ_REMAP
914 static int iommu_ga_log_enable(struct amd_iommu *iommu)
915 {
916 	u32 status, i;
917 	u64 entry;
918 
919 	if (!iommu->ga_log)
920 		return -EINVAL;
921 
922 	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
923 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
924 		    &entry, sizeof(entry));
925 	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
926 		 (BIT_ULL(52)-1)) & ~7ULL;
927 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
928 		    &entry, sizeof(entry));
929 	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
930 	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
931 
932 
933 	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
934 	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
935 
936 	for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
937 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
938 		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
939 			break;
940 		udelay(10);
941 	}
942 
943 	if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
944 		return -EINVAL;
945 
946 	return 0;
947 }
948 
949 static int iommu_init_ga_log(struct amd_iommu *iommu)
950 {
951 	int nid = iommu->dev ? dev_to_node(&iommu->dev->dev) : NUMA_NO_NODE;
952 
953 	if (WARN_ON_ONCE(!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)))
954 		return -EINVAL;
955 
956 	iommu->ga_log = iommu_alloc_pages_node_sz(nid, GFP_KERNEL, GA_LOG_SIZE);
957 	if (!iommu->ga_log)
958 		goto err_out;
959 
960 	iommu->ga_log_tail = iommu_alloc_pages_node_sz(nid, GFP_KERNEL, 8);
961 	if (!iommu->ga_log_tail)
962 		goto err_out;
963 
964 	return 0;
965 err_out:
966 	free_ga_log(iommu);
967 	return -EINVAL;
968 }
969 #endif /* CONFIG_IRQ_REMAP */
970 
971 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
972 {
973 	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL, 1);
974 	if (!iommu->cmd_sem)
975 		return -ENOMEM;
976 	iommu->cmd_sem_paddr = iommu_virt_to_phys((void *)iommu->cmd_sem);
977 	return 0;
978 }
979 
980 static int __init remap_event_buffer(void)
981 {
982 	struct amd_iommu *iommu;
983 	u64 paddr;
984 
985 	pr_info_once("Re-using event buffer from the previous kernel\n");
986 	for_each_iommu(iommu) {
987 		paddr = readq(iommu->mmio_base + MMIO_EVT_BUF_OFFSET) & PM_ADDR_MASK;
988 		iommu->evt_buf = iommu_memremap(paddr, amd_iommu_evtlog_size);
989 		if (!iommu->evt_buf)
990 			return -ENOMEM;
991 	}
992 
993 	return 0;
994 }
995 
996 static int __init remap_command_buffer(struct amd_iommu *iommu)
997 {
998 	u64 paddr;
999 
1000 	pr_info_once("Re-using command buffer from the previous kernel\n");
1001 	paddr = readq(iommu->mmio_base + MMIO_CMD_BUF_OFFSET) & PM_ADDR_MASK;
1002 	iommu->cmd_buf = iommu_memremap(paddr, CMD_BUFFER_SIZE);
1003 
1004 	return iommu->cmd_buf ? 0 : -ENOMEM;
1005 }
1006 
1007 static int __init remap_or_alloc_cwwb_sem(struct amd_iommu *iommu)
1008 {
1009 	u64 paddr;
1010 
1011 	if (check_feature(FEATURE_SNP)) {
1012 		/*
1013 		 * When SNP is enabled, the exclusion base register is used for the
1014 		 * completion wait buffer (CWB) address. Read and re-use it.
1015 		 */
1016 		pr_info_once("Re-using CWB buffers from the previous kernel\n");
1017 		paddr = readq(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET) & PM_ADDR_MASK;
1018 		iommu->cmd_sem = iommu_memremap(paddr, PAGE_SIZE);
1019 		if (!iommu->cmd_sem)
1020 			return -ENOMEM;
1021 		iommu->cmd_sem_paddr = paddr;
1022 	} else {
1023 		return alloc_cwwb_sem(iommu);
1024 	}
1025 
1026 	return 0;
1027 }
1028 
1029 static int __init alloc_iommu_buffers(struct amd_iommu *iommu)
1030 {
1031 	int ret;
1032 
1033 	/*
1034 	 * Reuse/Remap the previous kernel's allocated completion wait
1035 	 * command and event buffers for kdump boot.
1036 	 */
1037 	if (is_kdump_kernel()) {
1038 		ret = remap_or_alloc_cwwb_sem(iommu);
1039 		if (ret)
1040 			return ret;
1041 
1042 		ret = remap_command_buffer(iommu);
1043 		if (ret)
1044 			return ret;
1045 	} else {
1046 		ret = alloc_cwwb_sem(iommu);
1047 		if (ret)
1048 			return ret;
1049 
1050 		ret = alloc_command_buffer(iommu);
1051 		if (ret)
1052 			return ret;
1053 	}
1054 
1055 	return 0;
1056 }
1057 
1058 static void __init free_cwwb_sem(struct amd_iommu *iommu)
1059 {
1060 	if (iommu->cmd_sem)
1061 		iommu_free_pages((void *)iommu->cmd_sem);
1062 }
1063 static void __init unmap_cwwb_sem(struct amd_iommu *iommu)
1064 {
1065 	if (iommu->cmd_sem) {
1066 		if (check_feature(FEATURE_SNP))
1067 			memunmap((void *)iommu->cmd_sem);
1068 		else
1069 			iommu_free_pages((void *)iommu->cmd_sem);
1070 	}
1071 }
1072 
1073 static void __init unmap_command_buffer(struct amd_iommu *iommu)
1074 {
1075 	memunmap((void *)iommu->cmd_buf);
1076 }
1077 
1078 static void __init unmap_event_buffer(struct amd_iommu *iommu)
1079 {
1080 	memunmap(iommu->evt_buf);
1081 }
1082 
1083 static void __init free_iommu_buffers(struct amd_iommu *iommu)
1084 {
1085 	if (is_kdump_kernel()) {
1086 		unmap_cwwb_sem(iommu);
1087 		unmap_command_buffer(iommu);
1088 		unmap_event_buffer(iommu);
1089 	} else {
1090 		free_cwwb_sem(iommu);
1091 		free_command_buffer(iommu);
1092 		free_event_buffer(iommu);
1093 	}
1094 }
1095 
1096 static void iommu_enable_xt(struct amd_iommu *iommu)
1097 {
1098 #ifdef CONFIG_IRQ_REMAP
1099 	/*
1100 	 * XT mode (32-bit APIC destination ID) requires
1101 	 * GA mode (128-bit IRTE support) as a prerequisite.
1102 	 */
1103 	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
1104 	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1105 		iommu_feature_enable(iommu, CONTROL_XT_EN);
1106 #endif /* CONFIG_IRQ_REMAP */
1107 }
1108 
1109 static void iommu_enable_gt(struct amd_iommu *iommu)
1110 {
1111 	if (!check_feature(FEATURE_GT))
1112 		return;
1113 
1114 	iommu_feature_enable(iommu, CONTROL_GT_EN);
1115 
1116 	/*
1117 	 * This feature needs to be enabled prior to a call
1118 	 * to iommu_snp_enable(). Since this function is called
1119 	 * in early_enable_iommu(), it is safe to enable here.
1120 	 */
1121 	if (check_feature2(FEATURE_GCR3TRPMODE))
1122 		iommu_feature_enable(iommu, CONTROL_GCR3TRPMODE);
1123 }
1124 
1125 /* sets a specific bit in the device table entry. */
1126 static void set_dte_bit(struct dev_table_entry *dte, u8 bit)
1127 {
1128 	int i = (bit >> 6) & 0x03;
1129 	int _bit = bit & 0x3f;
1130 
1131 	dte->data[i] |= (1UL << _bit);
1132 }
1133 
1134 static bool __reuse_device_table(struct amd_iommu *iommu)
1135 {
1136 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1137 	struct dev_table_entry *old_dev_tbl_entry;
1138 	u32 lo, hi, old_devtb_size, devid;
1139 	phys_addr_t old_devtb_phys;
1140 	u16 dom_id;
1141 	bool dte_v;
1142 	u64 entry;
1143 
1144 	/* Each IOMMU use separate device table with the same size */
1145 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1146 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1147 	entry = (((u64) hi) << 32) + lo;
1148 
1149 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1150 	if (old_devtb_size != pci_seg->dev_table_size) {
1151 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1152 			iommu->index);
1153 		return false;
1154 	}
1155 
1156 	/*
1157 	 * When SME is enabled in the first kernel, the entry includes the
1158 	 * memory encryption mask(sme_me_mask), we must remove the memory
1159 	 * encryption mask to obtain the true physical address in kdump kernel.
1160 	 */
1161 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1162 
1163 	if (old_devtb_phys >= 0x100000000ULL) {
1164 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1165 		return false;
1166 	}
1167 
1168 	/*
1169 	 * Re-use the previous kernel's device table for kdump.
1170 	 */
1171 	pci_seg->old_dev_tbl_cpy = iommu_memremap(old_devtb_phys, pci_seg->dev_table_size);
1172 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1173 		pr_err("Failed to remap memory for reusing old device table!\n");
1174 		return false;
1175 	}
1176 
1177 	for (devid = 0; devid <= pci_seg->last_bdf; devid++) {
1178 		old_dev_tbl_entry = &pci_seg->old_dev_tbl_cpy[devid];
1179 		dte_v = FIELD_GET(DTE_FLAG_V, old_dev_tbl_entry->data[0]);
1180 		dom_id = FIELD_GET(DTE_DOMID_MASK, old_dev_tbl_entry->data[1]);
1181 
1182 		if (!dte_v || !dom_id)
1183 			continue;
1184 		/*
1185 		 * ID reservation can fail with -ENOSPC when there
1186 		 * are multiple devices present in the same domain,
1187 		 * hence check only for -ENOMEM.
1188 		 */
1189 		if (amd_iommu_pdom_id_reserve(dom_id, GFP_KERNEL) == -ENOMEM)
1190 			return false;
1191 	}
1192 
1193 	return true;
1194 }
1195 
1196 static bool reuse_device_table(void)
1197 {
1198 	struct amd_iommu *iommu;
1199 	struct amd_iommu_pci_seg *pci_seg;
1200 
1201 	if (!amd_iommu_pre_enabled)
1202 		return false;
1203 
1204 	pr_warn("Translation is already enabled - trying to reuse translation structures\n");
1205 
1206 	/*
1207 	 * All IOMMUs within PCI segment shares common device table.
1208 	 * Hence reuse device table only once per PCI segment.
1209 	 */
1210 	for_each_pci_segment(pci_seg) {
1211 		for_each_iommu(iommu) {
1212 			if (pci_seg->id != iommu->pci_seg->id)
1213 				continue;
1214 			if (!__reuse_device_table(iommu))
1215 				return false;
1216 			break;
1217 		}
1218 	}
1219 
1220 	return true;
1221 }
1222 
1223 struct dev_table_entry *amd_iommu_get_ivhd_dte_flags(u16 segid, u16 devid)
1224 {
1225 	struct ivhd_dte_flags *e;
1226 	unsigned int best_len = UINT_MAX;
1227 	struct dev_table_entry *dte = NULL;
1228 
1229 	for_each_ivhd_dte_flags(e) {
1230 		/*
1231 		 * Need to go through the whole list to find the smallest range,
1232 		 * which contains the devid.
1233 		 */
1234 		if ((e->segid == segid) &&
1235 		    (e->devid_first <= devid) && (devid <= e->devid_last)) {
1236 			unsigned int len = e->devid_last - e->devid_first;
1237 
1238 			if (len < best_len) {
1239 				dte = &(e->dte);
1240 				best_len = len;
1241 			}
1242 		}
1243 	}
1244 	return dte;
1245 }
1246 
1247 static bool search_ivhd_dte_flags(u16 segid, u16 first, u16 last)
1248 {
1249 	struct ivhd_dte_flags *e;
1250 
1251 	for_each_ivhd_dte_flags(e) {
1252 		if ((e->segid == segid) &&
1253 		    (e->devid_first == first) &&
1254 		    (e->devid_last == last))
1255 			return true;
1256 	}
1257 	return false;
1258 }
1259 
1260 /*
1261  * This function takes the device specific flags read from the ACPI
1262  * table and sets up the device table entry with that information
1263  */
1264 static void __init
1265 set_dev_entry_from_acpi_range(struct amd_iommu *iommu, u16 first, u16 last,
1266 			      u32 flags, u32 ext_flags)
1267 {
1268 	int i;
1269 	struct dev_table_entry dte = {};
1270 
1271 	/* Parse IVHD DTE setting flags and store information */
1272 	if (flags) {
1273 		struct ivhd_dte_flags *d;
1274 
1275 		if (search_ivhd_dte_flags(iommu->pci_seg->id, first, last))
1276 			return;
1277 
1278 		d = kzalloc_obj(struct ivhd_dte_flags);
1279 		if (!d)
1280 			return;
1281 
1282 		pr_debug("%s: devid range %#x:%#x\n", __func__, first, last);
1283 
1284 		if (flags & ACPI_DEVFLAG_INITPASS)
1285 			set_dte_bit(&dte, DEV_ENTRY_INIT_PASS);
1286 		if (flags & ACPI_DEVFLAG_EXTINT)
1287 			set_dte_bit(&dte, DEV_ENTRY_EINT_PASS);
1288 		if (flags & ACPI_DEVFLAG_NMI)
1289 			set_dte_bit(&dte, DEV_ENTRY_NMI_PASS);
1290 		if (flags & ACPI_DEVFLAG_SYSMGT1)
1291 			set_dte_bit(&dte, DEV_ENTRY_SYSMGT1);
1292 		if (flags & ACPI_DEVFLAG_SYSMGT2)
1293 			set_dte_bit(&dte, DEV_ENTRY_SYSMGT2);
1294 		if (flags & ACPI_DEVFLAG_LINT0)
1295 			set_dte_bit(&dte, DEV_ENTRY_LINT0_PASS);
1296 		if (flags & ACPI_DEVFLAG_LINT1)
1297 			set_dte_bit(&dte, DEV_ENTRY_LINT1_PASS);
1298 
1299 		/* Apply erratum 63, which needs info in initial_dte */
1300 		if (FIELD_GET(DTE_DATA1_SYSMGT_MASK, dte.data[1]) == 0x1)
1301 			dte.data[0] |= DTE_FLAG_IW;
1302 
1303 		memcpy(&d->dte, &dte, sizeof(dte));
1304 		d->segid = iommu->pci_seg->id;
1305 		d->devid_first = first;
1306 		d->devid_last = last;
1307 		list_add_tail(&d->list, &amd_ivhd_dev_flags_list);
1308 	}
1309 
1310 	for (i = first; i <= last; i++)  {
1311 		if (flags) {
1312 			struct dev_table_entry *dev_table = get_dev_table(iommu);
1313 
1314 			memcpy(&dev_table[i], &dte, sizeof(dte));
1315 		}
1316 		amd_iommu_set_rlookup_table(iommu, i);
1317 	}
1318 }
1319 
1320 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1321 					   u16 devid, u32 flags, u32 ext_flags)
1322 {
1323 	set_dev_entry_from_acpi_range(iommu, devid, devid, flags, ext_flags);
1324 }
1325 
1326 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line)
1327 {
1328 	struct devid_map *entry;
1329 	struct list_head *list;
1330 
1331 	if (type == IVHD_SPECIAL_IOAPIC)
1332 		list = &ioapic_map;
1333 	else if (type == IVHD_SPECIAL_HPET)
1334 		list = &hpet_map;
1335 	else
1336 		return -EINVAL;
1337 
1338 	list_for_each_entry(entry, list, list) {
1339 		if (!(entry->id == id && entry->cmd_line))
1340 			continue;
1341 
1342 		pr_info("Command-line override present for %s id %d - ignoring\n",
1343 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1344 
1345 		*devid = entry->devid;
1346 
1347 		return 0;
1348 	}
1349 
1350 	entry = kzalloc_obj(*entry);
1351 	if (!entry)
1352 		return -ENOMEM;
1353 
1354 	entry->id	= id;
1355 	entry->devid	= *devid;
1356 	entry->cmd_line	= cmd_line;
1357 
1358 	list_add_tail(&entry->list, list);
1359 
1360 	return 0;
1361 }
1362 
1363 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid,
1364 				      bool cmd_line)
1365 {
1366 	struct acpihid_map_entry *entry;
1367 	struct list_head *list = &acpihid_map;
1368 
1369 	list_for_each_entry(entry, list, list) {
1370 		if (strcmp(entry->hid, hid) ||
1371 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1372 		    !entry->cmd_line)
1373 			continue;
1374 
1375 		pr_info("Command-line override for hid:%s uid:%s\n",
1376 			hid, uid);
1377 		*devid = entry->devid;
1378 		return 0;
1379 	}
1380 
1381 	entry = kzalloc_obj(*entry);
1382 	if (!entry)
1383 		return -ENOMEM;
1384 
1385 	memcpy(entry->uid, uid, strlen(uid));
1386 	memcpy(entry->hid, hid, strlen(hid));
1387 	entry->devid = *devid;
1388 	entry->cmd_line	= cmd_line;
1389 	entry->root_devid = (entry->devid & (~0x7));
1390 
1391 	pr_info("%s, add hid:%s, uid:%s, rdevid:%#x\n",
1392 		entry->cmd_line ? "cmd" : "ivrs",
1393 		entry->hid, entry->uid, entry->root_devid);
1394 
1395 	list_add_tail(&entry->list, list);
1396 	return 0;
1397 }
1398 
1399 static int __init add_early_maps(void)
1400 {
1401 	int i, ret;
1402 
1403 	for (i = 0; i < early_ioapic_map_size; ++i) {
1404 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1405 					 early_ioapic_map[i].id,
1406 					 &early_ioapic_map[i].devid,
1407 					 early_ioapic_map[i].cmd_line);
1408 		if (ret)
1409 			return ret;
1410 	}
1411 
1412 	for (i = 0; i < early_hpet_map_size; ++i) {
1413 		ret = add_special_device(IVHD_SPECIAL_HPET,
1414 					 early_hpet_map[i].id,
1415 					 &early_hpet_map[i].devid,
1416 					 early_hpet_map[i].cmd_line);
1417 		if (ret)
1418 			return ret;
1419 	}
1420 
1421 	for (i = 0; i < early_acpihid_map_size; ++i) {
1422 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1423 					  early_acpihid_map[i].uid,
1424 					  &early_acpihid_map[i].devid,
1425 					  early_acpihid_map[i].cmd_line);
1426 		if (ret)
1427 			return ret;
1428 	}
1429 
1430 	return 0;
1431 }
1432 
1433 /*
1434  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1435  * initializes the hardware and our data structures with it.
1436  */
1437 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1438 					struct ivhd_header *h)
1439 {
1440 	u8 *p = (u8 *)h;
1441 	u8 *end = p, flags = 0;
1442 	u16 devid = 0, devid_start = 0, devid_to = 0, seg_id;
1443 	u32 dev_i, ext_flags = 0;
1444 	bool alias = false;
1445 	struct ivhd_entry *e;
1446 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1447 	u32 ivhd_size;
1448 	int ret;
1449 
1450 
1451 	ret = add_early_maps();
1452 	if (ret)
1453 		return ret;
1454 
1455 	amd_iommu_apply_ivrs_quirks();
1456 
1457 	/*
1458 	 * First save the recommended feature enable bits from ACPI
1459 	 */
1460 	iommu->acpi_flags = h->flags;
1461 
1462 	/*
1463 	 * Done. Now parse the device entries
1464 	 */
1465 	ivhd_size = get_ivhd_header_size(h);
1466 	if (!ivhd_size) {
1467 		pr_err("Unsupported IVHD type %#x\n", h->type);
1468 		return -EINVAL;
1469 	}
1470 
1471 	p += ivhd_size;
1472 
1473 	end += h->length;
1474 
1475 
1476 	while (p < end) {
1477 		e = (struct ivhd_entry *)p;
1478 		seg_id = pci_seg->id;
1479 
1480 		switch (e->type) {
1481 		case IVHD_DEV_ALL:
1482 
1483 			DUMP_printk("  DEV_ALL\t\t\tsetting: %#02x\n", e->flags);
1484 			set_dev_entry_from_acpi_range(iommu, 0, pci_seg->last_bdf, e->flags, 0);
1485 			break;
1486 		case IVHD_DEV_SELECT:
1487 
1488 			DUMP_printk("  DEV_SELECT\t\t\tdevid: %04x:%02x:%02x.%x flags: %#02x\n",
1489 				    seg_id, PCI_BUS_NUM(e->devid),
1490 				    PCI_SLOT(e->devid),
1491 				    PCI_FUNC(e->devid),
1492 				    e->flags);
1493 
1494 			devid = e->devid;
1495 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1496 			break;
1497 		case IVHD_DEV_SELECT_RANGE_START:
1498 
1499 			DUMP_printk("  DEV_SELECT_RANGE_START\tdevid: %04x:%02x:%02x.%x flags: %#02x\n",
1500 				    seg_id, PCI_BUS_NUM(e->devid),
1501 				    PCI_SLOT(e->devid),
1502 				    PCI_FUNC(e->devid),
1503 				    e->flags);
1504 
1505 			devid_start = e->devid;
1506 			flags = e->flags;
1507 			ext_flags = 0;
1508 			alias = false;
1509 			break;
1510 		case IVHD_DEV_ALIAS:
1511 
1512 			DUMP_printk("  DEV_ALIAS\t\t\tdevid: %04x:%02x:%02x.%x flags: %#02x devid_to: %02x:%02x.%x\n",
1513 				    seg_id, PCI_BUS_NUM(e->devid),
1514 				    PCI_SLOT(e->devid),
1515 				    PCI_FUNC(e->devid),
1516 				    e->flags,
1517 				    PCI_BUS_NUM(e->ext >> 8),
1518 				    PCI_SLOT(e->ext >> 8),
1519 				    PCI_FUNC(e->ext >> 8));
1520 
1521 			devid = e->devid;
1522 			devid_to = e->ext >> 8;
1523 			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1524 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1525 			pci_seg->alias_table[devid] = devid_to;
1526 			break;
1527 		case IVHD_DEV_ALIAS_RANGE:
1528 
1529 			DUMP_printk("  DEV_ALIAS_RANGE\t\tdevid: %04x:%02x:%02x.%x flags: %#02x devid_to: %04x:%02x:%02x.%x\n",
1530 				    seg_id, PCI_BUS_NUM(e->devid),
1531 				    PCI_SLOT(e->devid),
1532 				    PCI_FUNC(e->devid),
1533 				    e->flags,
1534 				    seg_id, PCI_BUS_NUM(e->ext >> 8),
1535 				    PCI_SLOT(e->ext >> 8),
1536 				    PCI_FUNC(e->ext >> 8));
1537 
1538 			devid_start = e->devid;
1539 			flags = e->flags;
1540 			devid_to = e->ext >> 8;
1541 			ext_flags = 0;
1542 			alias = true;
1543 			break;
1544 		case IVHD_DEV_EXT_SELECT:
1545 
1546 			DUMP_printk("  DEV_EXT_SELECT\t\tdevid: %04x:%02x:%02x.%x flags: %#02x ext: %08x\n",
1547 				    seg_id, PCI_BUS_NUM(e->devid),
1548 				    PCI_SLOT(e->devid),
1549 				    PCI_FUNC(e->devid),
1550 				    e->flags, e->ext);
1551 
1552 			devid = e->devid;
1553 			set_dev_entry_from_acpi(iommu, devid, e->flags,
1554 						e->ext);
1555 			break;
1556 		case IVHD_DEV_EXT_SELECT_RANGE:
1557 
1558 			DUMP_printk("  DEV_EXT_SELECT_RANGE\tdevid: %04x:%02x:%02x.%x flags: %#02x ext: %08x\n",
1559 				    seg_id, PCI_BUS_NUM(e->devid),
1560 				    PCI_SLOT(e->devid),
1561 				    PCI_FUNC(e->devid),
1562 				    e->flags, e->ext);
1563 
1564 			devid_start = e->devid;
1565 			flags = e->flags;
1566 			ext_flags = e->ext;
1567 			alias = false;
1568 			break;
1569 		case IVHD_DEV_RANGE_END:
1570 
1571 			DUMP_printk("  DEV_RANGE_END\t\tdevid: %04x:%02x:%02x.%x\n",
1572 				    seg_id, PCI_BUS_NUM(e->devid),
1573 				    PCI_SLOT(e->devid),
1574 				    PCI_FUNC(e->devid));
1575 
1576 			devid = e->devid;
1577 			if (alias) {
1578 				for (dev_i = devid_start; dev_i <= devid; ++dev_i)
1579 					pci_seg->alias_table[dev_i] = devid_to;
1580 				set_dev_entry_from_acpi(iommu, devid_to, flags, ext_flags);
1581 			}
1582 			set_dev_entry_from_acpi_range(iommu, devid_start, devid, flags, ext_flags);
1583 			break;
1584 		case IVHD_DEV_SPECIAL: {
1585 			u8 handle, type;
1586 			const char *var;
1587 			u32 devid;
1588 			int ret;
1589 
1590 			handle = e->ext & 0xff;
1591 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8));
1592 			type   = (e->ext >> 24) & 0xff;
1593 
1594 			if (type == IVHD_SPECIAL_IOAPIC)
1595 				var = "IOAPIC";
1596 			else if (type == IVHD_SPECIAL_HPET)
1597 				var = "HPET";
1598 			else
1599 				var = "UNKNOWN";
1600 
1601 			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x, flags: %#02x\n",
1602 				    var, (int)handle,
1603 				    seg_id, PCI_BUS_NUM(devid),
1604 				    PCI_SLOT(devid),
1605 				    PCI_FUNC(devid),
1606 				    e->flags);
1607 
1608 			ret = add_special_device(type, handle, &devid, false);
1609 			if (ret)
1610 				return ret;
1611 
1612 			/*
1613 			 * add_special_device might update the devid in case a
1614 			 * command-line override is present. So call
1615 			 * set_dev_entry_from_acpi after add_special_device.
1616 			 */
1617 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1618 
1619 			break;
1620 		}
1621 		case IVHD_DEV_ACPI_HID: {
1622 			u32 devid;
1623 			u8 hid[ACPIHID_HID_LEN];
1624 			u8 uid[ACPIHID_UID_LEN];
1625 			int ret;
1626 
1627 			if (h->type != 0x40) {
1628 				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1629 				       e->type);
1630 				break;
1631 			}
1632 
1633 			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1634 			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1635 			hid[ACPIHID_HID_LEN - 1] = '\0';
1636 
1637 			if (!(*hid)) {
1638 				pr_err(FW_BUG "Invalid HID.\n");
1639 				break;
1640 			}
1641 
1642 			uid[0] = '\0';
1643 			switch (e->uidf) {
1644 			case UID_NOT_PRESENT:
1645 
1646 				if (e->uidl != 0)
1647 					pr_warn(FW_BUG "Invalid UID length.\n");
1648 
1649 				break;
1650 			case UID_IS_INTEGER:
1651 
1652 				sprintf(uid, "%d", e->uid);
1653 
1654 				break;
1655 			case UID_IS_CHARACTER:
1656 
1657 				memcpy(uid, &e->uid, e->uidl);
1658 				uid[e->uidl] = '\0';
1659 
1660 				break;
1661 			default:
1662 				break;
1663 			}
1664 
1665 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid);
1666 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x, flags: %#02x\n",
1667 				    hid, uid, seg_id,
1668 				    PCI_BUS_NUM(devid),
1669 				    PCI_SLOT(devid),
1670 				    PCI_FUNC(devid),
1671 				    e->flags);
1672 
1673 			flags = e->flags;
1674 
1675 			ret = add_acpi_hid_device(hid, uid, &devid, false);
1676 			if (ret)
1677 				return ret;
1678 
1679 			/*
1680 			 * add_special_device might update the devid in case a
1681 			 * command-line override is present. So call
1682 			 * set_dev_entry_from_acpi after add_special_device.
1683 			 */
1684 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1685 
1686 			break;
1687 		}
1688 		default:
1689 			break;
1690 		}
1691 
1692 		p += ivhd_entry_length(p);
1693 	}
1694 
1695 	return 0;
1696 }
1697 
1698 /* Allocate PCI segment data structure */
1699 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1700 					  struct acpi_table_header *ivrs_base)
1701 {
1702 	struct amd_iommu_pci_seg *pci_seg;
1703 	int last_bdf;
1704 
1705 	/*
1706 	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1707 	 * handle in this PCI segment. Upon this information the shared data
1708 	 * structures for the PCI segments in the system will be allocated.
1709 	 */
1710 	last_bdf = find_last_devid_acpi(ivrs_base, id);
1711 	if (last_bdf < 0)
1712 		return NULL;
1713 
1714 	pci_seg = kzalloc_obj(struct amd_iommu_pci_seg);
1715 	if (pci_seg == NULL)
1716 		return NULL;
1717 
1718 	pci_seg->last_bdf = last_bdf;
1719 	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1720 	pci_seg->dev_table_size =
1721 		max(roundup_pow_of_two((last_bdf + 1) * DEV_TABLE_ENTRY_SIZE),
1722 		    SZ_4K);
1723 
1724 	pci_seg->id = id;
1725 	init_llist_head(&pci_seg->dev_data_list);
1726 	INIT_LIST_HEAD(&pci_seg->unity_map);
1727 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1728 
1729 	if (alloc_dev_table(pci_seg))
1730 		goto err_free_pci_seg;
1731 	if (alloc_alias_table(pci_seg))
1732 		goto err_free_dev_table;
1733 	if (alloc_rlookup_table(pci_seg))
1734 		goto err_free_alias_table;
1735 
1736 	return pci_seg;
1737 
1738 err_free_alias_table:
1739 	free_alias_table(pci_seg);
1740 err_free_dev_table:
1741 	free_dev_table(pci_seg);
1742 err_free_pci_seg:
1743 	list_del(&pci_seg->list);
1744 	kfree(pci_seg);
1745 	return NULL;
1746 }
1747 
1748 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1749 					struct acpi_table_header *ivrs_base)
1750 {
1751 	struct amd_iommu_pci_seg *pci_seg;
1752 
1753 	for_each_pci_segment(pci_seg) {
1754 		if (pci_seg->id == id)
1755 			return pci_seg;
1756 	}
1757 
1758 	return alloc_pci_segment(id, ivrs_base);
1759 }
1760 
1761 static void __init free_pci_segments(void)
1762 {
1763 	struct amd_iommu_pci_seg *pci_seg, *next;
1764 
1765 	for_each_pci_segment_safe(pci_seg, next) {
1766 		list_del(&pci_seg->list);
1767 		free_irq_lookup_table(pci_seg);
1768 		free_rlookup_table(pci_seg);
1769 		free_alias_table(pci_seg);
1770 		free_dev_table(pci_seg);
1771 		kfree(pci_seg);
1772 	}
1773 }
1774 
1775 static void __init free_sysfs(struct amd_iommu *iommu)
1776 {
1777 	if (iommu->iommu.dev) {
1778 		iommu_device_unregister(&iommu->iommu);
1779 		iommu_device_sysfs_remove(&iommu->iommu);
1780 	}
1781 }
1782 
1783 static void __init free_iommu_one(struct amd_iommu *iommu)
1784 {
1785 	free_sysfs(iommu);
1786 	free_iommu_buffers(iommu);
1787 	amd_iommu_free_ppr_log(iommu);
1788 	free_ga_log(iommu);
1789 	iommu_unmap_mmio_space(iommu);
1790 	amd_iommu_iopf_uninit(iommu);
1791 }
1792 
1793 static void __init free_iommu_all(void)
1794 {
1795 	struct amd_iommu *iommu, *next;
1796 
1797 	for_each_iommu_safe(iommu, next) {
1798 		list_del(&iommu->list);
1799 		free_iommu_one(iommu);
1800 		kfree(iommu);
1801 	}
1802 }
1803 
1804 /*
1805  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1806  * Workaround:
1807  *     BIOS should disable L2B micellaneous clock gating by setting
1808  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1809  */
1810 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1811 {
1812 	u32 value;
1813 
1814 	if ((boot_cpu_data.x86 != 0x15) ||
1815 	    (boot_cpu_data.x86_model < 0x10) ||
1816 	    (boot_cpu_data.x86_model > 0x1f))
1817 		return;
1818 
1819 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1820 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1821 
1822 	if (value & BIT(2))
1823 		return;
1824 
1825 	/* Select NB indirect register 0x90 and enable writing */
1826 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1827 
1828 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1829 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1830 
1831 	/* Clear the enable writing bit */
1832 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1833 }
1834 
1835 /*
1836  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1837  * Workaround:
1838  *     BIOS should enable ATS write permission check by setting
1839  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1840  */
1841 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1842 {
1843 	u32 value;
1844 
1845 	if ((boot_cpu_data.x86 != 0x15) ||
1846 	    (boot_cpu_data.x86_model < 0x30) ||
1847 	    (boot_cpu_data.x86_model > 0x3f))
1848 		return;
1849 
1850 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1851 	value = iommu_read_l2(iommu, 0x47);
1852 
1853 	if (value & BIT(0))
1854 		return;
1855 
1856 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1857 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1858 
1859 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1860 }
1861 
1862 /*
1863  * This function glues the initialization function for one IOMMU
1864  * together and also allocates the command buffer and programs the
1865  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1866  */
1867 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1868 				 struct acpi_table_header *ivrs_base)
1869 {
1870 	struct amd_iommu_pci_seg *pci_seg;
1871 
1872 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1873 	if (pci_seg == NULL)
1874 		return -ENOMEM;
1875 	iommu->pci_seg = pci_seg;
1876 
1877 	raw_spin_lock_init(&iommu->lock);
1878 	iommu->cmd_sem_val = 0;
1879 
1880 	/* Add IOMMU to internal data structures */
1881 	list_add_tail(&iommu->list, &amd_iommu_list);
1882 	iommu->index = amd_iommus_present++;
1883 
1884 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1885 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1886 		return -ENOSYS;
1887 	}
1888 
1889 	/*
1890 	 * Copy data from ACPI table entry to the iommu struct
1891 	 */
1892 	iommu->devid   = h->devid;
1893 	iommu->cap_ptr = h->cap_ptr;
1894 	iommu->mmio_phys = h->mmio_phys;
1895 
1896 	switch (h->type) {
1897 	case 0x10:
1898 		/* Check if IVHD EFR contains proper max banks/counters */
1899 		if ((h->efr_attr != 0) &&
1900 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1901 		    ((h->efr_attr & (0x3F << 17)) != 0))
1902 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1903 		else
1904 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1905 
1906 		/* GAM requires GA mode. */
1907 		if ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0)
1908 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1909 		break;
1910 	case 0x11:
1911 	case 0x40:
1912 		if (h->efr_reg & (1 << 9))
1913 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1914 		else
1915 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1916 
1917 		/* XT and GAM require GA mode. */
1918 		if ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0) {
1919 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1920 		} else {
1921 			if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1922 				amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1923 		}
1924 
1925 		if (h->efr_attr & BIT(IOMMU_IVHD_ATTR_HATDIS_SHIFT)) {
1926 			pr_warn_once("Host Address Translation is not supported.\n");
1927 			amd_iommu_hatdis = true;
1928 		}
1929 
1930 		early_iommu_features_init(iommu, h);
1931 
1932 		break;
1933 	default:
1934 		return -EINVAL;
1935 	}
1936 
1937 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1938 						iommu->mmio_phys_end);
1939 	if (!iommu->mmio_base)
1940 		return -ENOMEM;
1941 
1942 	return init_iommu_from_acpi(iommu, h);
1943 }
1944 
1945 static int __init init_iommu_one_late(struct amd_iommu *iommu)
1946 {
1947 	int ret;
1948 
1949 	ret = alloc_iommu_buffers(iommu);
1950 	if (ret)
1951 		return ret;
1952 
1953 	iommu->int_enabled = false;
1954 
1955 	init_translation_status(iommu);
1956 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1957 		iommu_disable(iommu);
1958 		clear_translation_pre_enabled(iommu);
1959 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1960 			iommu->index);
1961 	}
1962 	if (amd_iommu_pre_enabled)
1963 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1964 
1965 	if (amd_iommu_irq_remap) {
1966 		ret = amd_iommu_create_irq_domain(iommu);
1967 		if (ret)
1968 			return ret;
1969 	}
1970 
1971 	/*
1972 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1973 	 * table tells us so, but this is a lie!
1974 	 */
1975 	iommu->pci_seg->rlookup_table[iommu->devid] = NULL;
1976 
1977 	return 0;
1978 }
1979 
1980 /**
1981  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1982  * @ivrs: Pointer to the IVRS header
1983  *
1984  * This function search through all IVDB of the maximum supported IVHD
1985  */
1986 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1987 {
1988 	u8 *base = (u8 *)ivrs;
1989 	struct ivhd_header *ivhd = (struct ivhd_header *)
1990 					(base + IVRS_HEADER_LENGTH);
1991 	u8 last_type = ivhd->type;
1992 	u16 devid = ivhd->devid;
1993 
1994 	while (((u8 *)ivhd - base < ivrs->length) &&
1995 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1996 		u8 *p = (u8 *) ivhd;
1997 
1998 		if (ivhd->devid == devid)
1999 			last_type = ivhd->type;
2000 		ivhd = (struct ivhd_header *)(p + ivhd->length);
2001 	}
2002 
2003 	return last_type;
2004 }
2005 
2006 /*
2007  * Iterates over all IOMMU entries in the ACPI table, allocates the
2008  * IOMMU structure and initializes it with init_iommu_one()
2009  */
2010 static int __init init_iommu_all(struct acpi_table_header *table)
2011 {
2012 	u8 *p = (u8 *)table, *end = (u8 *)table;
2013 	struct ivhd_header *h;
2014 	struct amd_iommu *iommu;
2015 	int ret;
2016 
2017 	end += table->length;
2018 	p += IVRS_HEADER_LENGTH;
2019 
2020 	/* Phase 1: Process all IVHD blocks */
2021 	while (p < end) {
2022 		h = (struct ivhd_header *)p;
2023 		if (*p == amd_iommu_target_ivhd_type) {
2024 
2025 			DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x "
2026 				    "flags: %01x info %04x\n",
2027 				    h->pci_seg, PCI_BUS_NUM(h->devid),
2028 				    PCI_SLOT(h->devid), PCI_FUNC(h->devid),
2029 				    h->cap_ptr, h->flags, h->info);
2030 			DUMP_printk("       mmio-addr: %016llx\n",
2031 				    h->mmio_phys);
2032 
2033 			iommu = kzalloc_obj(struct amd_iommu);
2034 			if (iommu == NULL)
2035 				return -ENOMEM;
2036 
2037 			ret = init_iommu_one(iommu, h, table);
2038 			if (ret)
2039 				return ret;
2040 		}
2041 		p += h->length;
2042 
2043 	}
2044 	WARN_ON(p != end);
2045 
2046 	/* Phase 2 : Early feature support check */
2047 	get_global_efr();
2048 
2049 	/* Phase 3 : Enabling IOMMU features */
2050 	for_each_iommu(iommu) {
2051 		ret = init_iommu_one_late(iommu);
2052 		if (ret)
2053 			return ret;
2054 	}
2055 
2056 	return 0;
2057 }
2058 
2059 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
2060 {
2061 	u64 val;
2062 	struct pci_dev *pdev = iommu->dev;
2063 
2064 	if (!check_feature(FEATURE_PC))
2065 		return;
2066 
2067 	amd_iommu_pc_present = true;
2068 
2069 	pci_info(pdev, "IOMMU performance counters supported\n");
2070 
2071 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
2072 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
2073 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
2074 
2075 	return;
2076 }
2077 
2078 static ssize_t amd_iommu_show_cap(struct device *dev,
2079 				  struct device_attribute *attr,
2080 				  char *buf)
2081 {
2082 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
2083 	return sysfs_emit(buf, "%x\n", iommu->cap);
2084 }
2085 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
2086 
2087 static ssize_t amd_iommu_show_features(struct device *dev,
2088 				       struct device_attribute *attr,
2089 				       char *buf)
2090 {
2091 	return sysfs_emit(buf, "%llx:%llx\n", amd_iommu_efr, amd_iommu_efr2);
2092 }
2093 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
2094 
2095 static struct attribute *amd_iommu_attrs[] = {
2096 	&dev_attr_cap.attr,
2097 	&dev_attr_features.attr,
2098 	NULL,
2099 };
2100 
2101 static struct attribute_group amd_iommu_group = {
2102 	.name = "amd-iommu",
2103 	.attrs = amd_iommu_attrs,
2104 };
2105 
2106 static const struct attribute_group *amd_iommu_groups[] = {
2107 	&amd_iommu_group,
2108 	NULL,
2109 };
2110 
2111 /*
2112  * Note: IVHD 0x11 and 0x40 also contains exact copy
2113  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
2114  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
2115  */
2116 static void __init late_iommu_features_init(struct amd_iommu *iommu)
2117 {
2118 	u64 features, features2;
2119 
2120 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
2121 		return;
2122 
2123 	/* read extended feature bits */
2124 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
2125 	features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2);
2126 
2127 	if (!amd_iommu_efr) {
2128 		amd_iommu_efr = features;
2129 		amd_iommu_efr2 = features2;
2130 		return;
2131 	}
2132 
2133 	/*
2134 	 * Sanity check and warn if EFR values from
2135 	 * IVHD and MMIO conflict.
2136 	 */
2137 	if (features != amd_iommu_efr ||
2138 	    features2 != amd_iommu_efr2) {
2139 		pr_warn(FW_WARN
2140 			"EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n",
2141 			features, amd_iommu_efr,
2142 			features2, amd_iommu_efr2);
2143 	}
2144 }
2145 
2146 static int __init iommu_init_pci(struct amd_iommu *iommu)
2147 {
2148 	int cap_ptr = iommu->cap_ptr;
2149 	int ret;
2150 
2151 	iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2152 						 PCI_BUS_NUM(iommu->devid),
2153 						 iommu->devid & 0xff);
2154 	if (!iommu->dev)
2155 		return -ENODEV;
2156 
2157 	/* ACPI _PRT won't have an IRQ for IOMMU */
2158 	iommu->dev->irq_managed = 1;
2159 
2160 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2161 			      &iommu->cap);
2162 
2163 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2164 		amd_iommu_iotlb_sup = false;
2165 
2166 	late_iommu_features_init(iommu);
2167 
2168 	if (check_feature(FEATURE_GT)) {
2169 		int glxval;
2170 		u64 pasmax;
2171 
2172 		pasmax = FIELD_GET(FEATURE_PASMAX, amd_iommu_efr);
2173 		iommu->iommu.max_pasids = (1 << (pasmax + 1)) - 1;
2174 
2175 		BUG_ON(iommu->iommu.max_pasids & ~PASID_MASK);
2176 
2177 		glxval = FIELD_GET(FEATURE_GLX, amd_iommu_efr);
2178 
2179 		if (amd_iommu_max_glx_val == -1)
2180 			amd_iommu_max_glx_val = glxval;
2181 		else
2182 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2183 
2184 		iommu_enable_gt(iommu);
2185 	}
2186 
2187 	if (check_feature(FEATURE_PPR) && amd_iommu_alloc_ppr_log(iommu))
2188 		return -ENOMEM;
2189 
2190 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2191 		pr_info("Using strict mode due to virtualization\n");
2192 		iommu_set_dma_strict();
2193 		amd_iommu_np_cache = true;
2194 	}
2195 
2196 	init_iommu_perf_ctr(iommu);
2197 
2198 	if (is_rd890_iommu(iommu->dev)) {
2199 		int i, j;
2200 
2201 		iommu->root_pdev =
2202 			pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2203 						    iommu->dev->bus->number,
2204 						    PCI_DEVFN(0, 0));
2205 
2206 		/*
2207 		 * Some rd890 systems may not be fully reconfigured by the
2208 		 * BIOS, so it's necessary for us to store this information so
2209 		 * it can be reprogrammed on resume
2210 		 */
2211 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2212 				&iommu->stored_addr_lo);
2213 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2214 				&iommu->stored_addr_hi);
2215 
2216 		/* Low bit locks writes to configuration space */
2217 		iommu->stored_addr_lo &= ~1;
2218 
2219 		for (i = 0; i < 6; i++)
2220 			for (j = 0; j < 0x12; j++)
2221 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2222 
2223 		for (i = 0; i < 0x83; i++)
2224 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2225 	}
2226 
2227 	amd_iommu_erratum_746_workaround(iommu);
2228 	amd_iommu_ats_write_check_workaround(iommu);
2229 
2230 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2231 			       amd_iommu_groups, "ivhd%d", iommu->index);
2232 	if (ret)
2233 		return ret;
2234 
2235 	/*
2236 	 * Allocate per IOMMU IOPF queue here so that in attach device path,
2237 	 * PRI capable device can be added to IOPF queue
2238 	 */
2239 	if (amd_iommu_gt_ppr_supported()) {
2240 		ret = amd_iommu_iopf_init(iommu);
2241 		if (ret)
2242 			return ret;
2243 	}
2244 
2245 	ret = iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2246 	if (ret || amd_iommu_pgtable == PD_MODE_NONE) {
2247 		/*
2248 		 * Remove sysfs if DMA translation is not supported by the
2249 		 * IOMMU. Do not return an error to enable IRQ remapping
2250 		 * in state_next(), DTE[V, TV] must eventually be set to 0.
2251 		 */
2252 		iommu_device_sysfs_remove(&iommu->iommu);
2253 	}
2254 
2255 	return pci_enable_device(iommu->dev);
2256 }
2257 
2258 static void print_iommu_info(void)
2259 {
2260 	int i;
2261 	static const char * const feat_str[] = {
2262 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2263 		"IA", "GA", "HE", "PC"
2264 	};
2265 
2266 	if (amd_iommu_efr) {
2267 		pr_info("Extended features (%#llx, %#llx):", amd_iommu_efr, amd_iommu_efr2);
2268 
2269 		for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2270 			if (check_feature(1ULL << i))
2271 				pr_cont(" %s", feat_str[i]);
2272 		}
2273 
2274 		if (check_feature(FEATURE_GAM_VAPIC))
2275 			pr_cont(" GA_vAPIC");
2276 
2277 		if (check_feature(FEATURE_SNP))
2278 			pr_cont(" SNP");
2279 
2280 		if (check_feature2(FEATURE_SEVSNPIO_SUP))
2281 			pr_cont(" SEV-TIO");
2282 
2283 		pr_cont("\n");
2284 	}
2285 
2286 	if (irq_remapping_enabled) {
2287 		pr_info("Interrupt remapping enabled\n");
2288 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2289 			pr_info("X2APIC enabled\n");
2290 	}
2291 	if (amd_iommu_pgtable == PD_MODE_V2) {
2292 		pr_info("V2 page table enabled (Paging mode : %d level)\n",
2293 			amd_iommu_gpt_level);
2294 	}
2295 }
2296 
2297 static int __init amd_iommu_init_pci(void)
2298 {
2299 	struct amd_iommu *iommu;
2300 	struct amd_iommu_pci_seg *pci_seg;
2301 	int ret;
2302 
2303 	/* Init global identity domain before registering IOMMU */
2304 	amd_iommu_init_identity_domain();
2305 
2306 	for_each_iommu(iommu) {
2307 		ret = iommu_init_pci(iommu);
2308 		if (ret) {
2309 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2310 			       iommu->index, ret);
2311 			goto out;
2312 		}
2313 		/* Need to setup range after PCI init */
2314 		iommu_set_cwwb_range(iommu);
2315 	}
2316 
2317 	/*
2318 	 * Order is important here to make sure any unity map requirements are
2319 	 * fulfilled. The unity mappings are created and written to the device
2320 	 * table during the iommu_init_pci() call.
2321 	 *
2322 	 * After that we call init_device_table_dma() to make sure any
2323 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2324 	 * of all IOMMUs to make sure the changes to the device table are
2325 	 * active.
2326 	 */
2327 	for_each_pci_segment(pci_seg)
2328 		init_device_table_dma(pci_seg);
2329 
2330 	for_each_iommu(iommu)
2331 		amd_iommu_flush_all_caches(iommu);
2332 
2333 	print_iommu_info();
2334 
2335 out:
2336 	return ret;
2337 }
2338 
2339 /****************************************************************************
2340  *
2341  * The following functions initialize the MSI interrupts for all IOMMUs
2342  * in the system. It's a bit challenging because there could be multiple
2343  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2344  * pci_dev.
2345  *
2346  ****************************************************************************/
2347 
2348 static int iommu_setup_msi(struct amd_iommu *iommu)
2349 {
2350 	int r;
2351 
2352 	r = pci_enable_msi(iommu->dev);
2353 	if (r)
2354 		return r;
2355 
2356 	r = request_threaded_irq(iommu->dev->irq, NULL, amd_iommu_int_thread,
2357 				 IRQF_ONESHOT, "AMD-Vi", iommu);
2358 	if (r) {
2359 		pci_disable_msi(iommu->dev);
2360 		return r;
2361 	}
2362 
2363 	return 0;
2364 }
2365 
2366 union intcapxt {
2367 	u64	capxt;
2368 	struct {
2369 		u64	reserved_0		:  2,
2370 			dest_mode_logical	:  1,
2371 			reserved_1		:  5,
2372 			destid_0_23		: 24,
2373 			vector			:  8,
2374 			reserved_2		: 16,
2375 			destid_24_31		:  8;
2376 	};
2377 } __attribute__ ((packed));
2378 
2379 
2380 static struct irq_chip intcapxt_controller;
2381 
2382 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2383 				       struct irq_data *irqd, bool reserve)
2384 {
2385 	return 0;
2386 }
2387 
2388 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2389 					  struct irq_data *irqd)
2390 {
2391 }
2392 
2393 
2394 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2395 				    unsigned int nr_irqs, void *arg)
2396 {
2397 	struct irq_alloc_info *info = arg;
2398 	int i, ret;
2399 
2400 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2401 		return -EINVAL;
2402 
2403 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2404 	if (ret < 0)
2405 		return ret;
2406 
2407 	for (i = virq; i < virq + nr_irqs; i++) {
2408 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2409 
2410 		irqd->chip = &intcapxt_controller;
2411 		irqd->hwirq = info->hwirq;
2412 		irqd->chip_data = info->data;
2413 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2414 	}
2415 
2416 	return ret;
2417 }
2418 
2419 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2420 				    unsigned int nr_irqs)
2421 {
2422 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2423 }
2424 
2425 
2426 static void intcapxt_unmask_irq(struct irq_data *irqd)
2427 {
2428 	struct amd_iommu *iommu = irqd->chip_data;
2429 	struct irq_cfg *cfg = irqd_cfg(irqd);
2430 	union intcapxt xt;
2431 
2432 	xt.capxt = 0ULL;
2433 	xt.dest_mode_logical = apic->dest_mode_logical;
2434 	xt.vector = cfg->vector;
2435 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2436 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2437 
2438 	writeq(xt.capxt, iommu->mmio_base + irqd->hwirq);
2439 }
2440 
2441 static void intcapxt_mask_irq(struct irq_data *irqd)
2442 {
2443 	struct amd_iommu *iommu = irqd->chip_data;
2444 
2445 	writeq(0, iommu->mmio_base + irqd->hwirq);
2446 }
2447 
2448 
2449 static int intcapxt_set_affinity(struct irq_data *irqd,
2450 				 const struct cpumask *mask, bool force)
2451 {
2452 	struct irq_data *parent = irqd->parent_data;
2453 	int ret;
2454 
2455 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2456 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2457 		return ret;
2458 	return 0;
2459 }
2460 
2461 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2462 {
2463 	return on ? -EOPNOTSUPP : 0;
2464 }
2465 
2466 static struct irq_chip intcapxt_controller = {
2467 	.name			= "IOMMU-MSI",
2468 	.irq_unmask		= intcapxt_unmask_irq,
2469 	.irq_mask		= intcapxt_mask_irq,
2470 	.irq_ack		= irq_chip_ack_parent,
2471 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2472 	.irq_set_affinity       = intcapxt_set_affinity,
2473 	.irq_set_wake		= intcapxt_set_wake,
2474 	.flags			= IRQCHIP_MASK_ON_SUSPEND | IRQCHIP_MOVE_DEFERRED,
2475 };
2476 
2477 static const struct irq_domain_ops intcapxt_domain_ops = {
2478 	.alloc			= intcapxt_irqdomain_alloc,
2479 	.free			= intcapxt_irqdomain_free,
2480 	.activate		= intcapxt_irqdomain_activate,
2481 	.deactivate		= intcapxt_irqdomain_deactivate,
2482 };
2483 
2484 
2485 static struct irq_domain *iommu_irqdomain;
2486 
2487 static struct irq_domain *iommu_get_irqdomain(void)
2488 {
2489 	struct fwnode_handle *fn;
2490 
2491 	/* No need for locking here (yet) as the init is single-threaded */
2492 	if (iommu_irqdomain)
2493 		return iommu_irqdomain;
2494 
2495 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2496 	if (!fn)
2497 		return NULL;
2498 
2499 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2500 						      fn, &intcapxt_domain_ops,
2501 						      NULL);
2502 	if (!iommu_irqdomain)
2503 		irq_domain_free_fwnode(fn);
2504 
2505 	return iommu_irqdomain;
2506 }
2507 
2508 static int __iommu_setup_intcapxt(struct amd_iommu *iommu, const char *devname,
2509 				  int hwirq, irq_handler_t thread_fn)
2510 {
2511 	struct irq_domain *domain;
2512 	struct irq_alloc_info info;
2513 	int irq, ret;
2514 	int node = dev_to_node(&iommu->dev->dev);
2515 
2516 	domain = iommu_get_irqdomain();
2517 	if (!domain)
2518 		return -ENXIO;
2519 
2520 	init_irq_alloc_info(&info, NULL);
2521 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2522 	info.data = iommu;
2523 	info.hwirq = hwirq;
2524 
2525 	irq = irq_domain_alloc_irqs(domain, 1, node, &info);
2526 	if (irq < 0) {
2527 		irq_domain_remove(domain);
2528 		return irq;
2529 	}
2530 
2531 	ret = request_threaded_irq(irq, NULL, thread_fn, IRQF_ONESHOT, devname,
2532 				   iommu);
2533 	if (ret) {
2534 		irq_domain_free_irqs(irq, 1);
2535 		irq_domain_remove(domain);
2536 		return ret;
2537 	}
2538 
2539 	return 0;
2540 }
2541 
2542 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2543 {
2544 	int ret;
2545 
2546 	snprintf(iommu->evt_irq_name, sizeof(iommu->evt_irq_name),
2547 		 "AMD-Vi%d-Evt", iommu->index);
2548 	ret = __iommu_setup_intcapxt(iommu, iommu->evt_irq_name,
2549 				     MMIO_INTCAPXT_EVT_OFFSET,
2550 				     amd_iommu_int_thread_evtlog);
2551 	if (ret)
2552 		return ret;
2553 
2554 	snprintf(iommu->ppr_irq_name, sizeof(iommu->ppr_irq_name),
2555 		 "AMD-Vi%d-PPR", iommu->index);
2556 	ret = __iommu_setup_intcapxt(iommu, iommu->ppr_irq_name,
2557 				     MMIO_INTCAPXT_PPR_OFFSET,
2558 				     amd_iommu_int_thread_pprlog);
2559 	if (ret)
2560 		return ret;
2561 
2562 #ifdef CONFIG_IRQ_REMAP
2563 	snprintf(iommu->ga_irq_name, sizeof(iommu->ga_irq_name),
2564 		 "AMD-Vi%d-GA", iommu->index);
2565 	ret = __iommu_setup_intcapxt(iommu, iommu->ga_irq_name,
2566 				     MMIO_INTCAPXT_GALOG_OFFSET,
2567 				     amd_iommu_int_thread_galog);
2568 #endif
2569 
2570 	return ret;
2571 }
2572 
2573 static int iommu_init_irq(struct amd_iommu *iommu)
2574 {
2575 	int ret;
2576 
2577 	if (iommu->int_enabled)
2578 		goto enable_faults;
2579 
2580 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2581 		ret = iommu_setup_intcapxt(iommu);
2582 	else if (iommu->dev->msi_cap)
2583 		ret = iommu_setup_msi(iommu);
2584 	else
2585 		ret = -ENODEV;
2586 
2587 	if (ret)
2588 		return ret;
2589 
2590 	iommu->int_enabled = true;
2591 enable_faults:
2592 
2593 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2594 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2595 
2596 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2597 
2598 	return 0;
2599 }
2600 
2601 /****************************************************************************
2602  *
2603  * The next functions belong to the third pass of parsing the ACPI
2604  * table. In this last pass the memory mapping requirements are
2605  * gathered (like exclusion and unity mapping ranges).
2606  *
2607  ****************************************************************************/
2608 
2609 static void __init free_unity_maps(void)
2610 {
2611 	struct unity_map_entry *entry, *next;
2612 	struct amd_iommu_pci_seg *p, *pci_seg;
2613 
2614 	for_each_pci_segment_safe(pci_seg, p) {
2615 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2616 			list_del(&entry->list);
2617 			kfree(entry);
2618 		}
2619 	}
2620 }
2621 
2622 /* called for unity map ACPI definition */
2623 static int __init init_unity_map_range(struct ivmd_header *m,
2624 				       struct acpi_table_header *ivrs_base)
2625 {
2626 	struct unity_map_entry *e = NULL;
2627 	struct amd_iommu_pci_seg *pci_seg;
2628 	char *s;
2629 
2630 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2631 	if (pci_seg == NULL)
2632 		return -ENOMEM;
2633 
2634 	e = kzalloc_obj(*e);
2635 	if (e == NULL)
2636 		return -ENOMEM;
2637 
2638 	switch (m->type) {
2639 	default:
2640 		kfree(e);
2641 		return 0;
2642 	case ACPI_IVMD_TYPE:
2643 		s = "IVMD_TYPEi\t\t\t";
2644 		e->devid_start = e->devid_end = m->devid;
2645 		break;
2646 	case ACPI_IVMD_TYPE_ALL:
2647 		s = "IVMD_TYPE_ALL\t\t";
2648 		e->devid_start = 0;
2649 		e->devid_end = pci_seg->last_bdf;
2650 		break;
2651 	case ACPI_IVMD_TYPE_RANGE:
2652 		s = "IVMD_TYPE_RANGE\t\t";
2653 		e->devid_start = m->devid;
2654 		e->devid_end = m->aux;
2655 		break;
2656 	}
2657 	e->address_start = PAGE_ALIGN(m->range_start);
2658 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2659 	e->prot = m->flags >> 1;
2660 
2661 	/*
2662 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2663 	 * since some buggy BIOSes might lead to the overwritten exclusion
2664 	 * range (exclusion_start and exclusion_length members). This
2665 	 * happens when there are multiple exclusion ranges (IVMD entries)
2666 	 * defined in ACPI table.
2667 	 */
2668 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2669 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2670 
2671 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2672 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2673 		    " flags: %x\n", s, m->pci_seg,
2674 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2675 		    PCI_FUNC(e->devid_start), m->pci_seg,
2676 		    PCI_BUS_NUM(e->devid_end),
2677 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2678 		    e->address_start, e->address_end, m->flags);
2679 
2680 	list_add_tail(&e->list, &pci_seg->unity_map);
2681 
2682 	return 0;
2683 }
2684 
2685 /* iterates over all memory definitions we find in the ACPI table */
2686 static int __init init_memory_definitions(struct acpi_table_header *table)
2687 {
2688 	u8 *p = (u8 *)table, *end = (u8 *)table;
2689 	struct ivmd_header *m;
2690 
2691 	end += table->length;
2692 	p += IVRS_HEADER_LENGTH;
2693 
2694 	while (p < end) {
2695 		m = (struct ivmd_header *)p;
2696 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2697 			init_unity_map_range(m, table);
2698 
2699 		p += m->length;
2700 	}
2701 
2702 	return 0;
2703 }
2704 
2705 /*
2706  * Init the device table to not allow DMA access for devices
2707  */
2708 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2709 {
2710 	u32 devid;
2711 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2712 
2713 	if (!dev_table || amd_iommu_pgtable == PD_MODE_NONE)
2714 		return;
2715 
2716 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2717 		set_dte_bit(&dev_table[devid], DEV_ENTRY_VALID);
2718 		if (!amd_iommu_snp_en)
2719 			set_dte_bit(&dev_table[devid], DEV_ENTRY_TRANSLATION);
2720 	}
2721 }
2722 
2723 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2724 {
2725 	u32 devid;
2726 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2727 
2728 	if (dev_table == NULL)
2729 		return;
2730 
2731 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2732 		dev_table[devid].data[0] = 0ULL;
2733 		dev_table[devid].data[1] = 0ULL;
2734 	}
2735 }
2736 
2737 static void init_device_table(void)
2738 {
2739 	struct amd_iommu_pci_seg *pci_seg;
2740 	u32 devid;
2741 
2742 	if (!amd_iommu_irq_remap)
2743 		return;
2744 
2745 	for_each_pci_segment(pci_seg) {
2746 		for (devid = 0; devid <= pci_seg->last_bdf; ++devid)
2747 			set_dte_bit(&pci_seg->dev_table[devid], DEV_ENTRY_IRQ_TBL_EN);
2748 	}
2749 }
2750 
2751 static void iommu_init_flags(struct amd_iommu *iommu)
2752 {
2753 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2754 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2755 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2756 
2757 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2758 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2759 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2760 
2761 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2762 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2763 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2764 
2765 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2766 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2767 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2768 
2769 	/*
2770 	 * make IOMMU memory accesses cache coherent
2771 	 */
2772 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2773 
2774 	/* Set IOTLB invalidation timeout to 1s */
2775 	iommu_feature_set(iommu, CTRL_INV_TO_1S, CTRL_INV_TO_MASK, CONTROL_INV_TIMEOUT);
2776 
2777 	/* Enable Enhanced Peripheral Page Request Handling */
2778 	if (check_feature(FEATURE_EPHSUP))
2779 		iommu_feature_enable(iommu, CONTROL_EPH_EN);
2780 }
2781 
2782 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2783 {
2784 	int i, j;
2785 	u32 ioc_feature_control;
2786 	struct pci_dev *pdev = iommu->root_pdev;
2787 
2788 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2789 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2790 		return;
2791 
2792 	/*
2793 	 * First, we need to ensure that the iommu is enabled. This is
2794 	 * controlled by a register in the northbridge
2795 	 */
2796 
2797 	/* Select Northbridge indirect register 0x75 and enable writing */
2798 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2799 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2800 
2801 	/* Enable the iommu */
2802 	if (!(ioc_feature_control & 0x1))
2803 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2804 
2805 	/* Restore the iommu BAR */
2806 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2807 			       iommu->stored_addr_lo);
2808 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2809 			       iommu->stored_addr_hi);
2810 
2811 	/* Restore the l1 indirect regs for each of the 6 l1s */
2812 	for (i = 0; i < 6; i++)
2813 		for (j = 0; j < 0x12; j++)
2814 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2815 
2816 	/* Restore the l2 indirect regs */
2817 	for (i = 0; i < 0x83; i++)
2818 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2819 
2820 	/* Lock PCI setup registers */
2821 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2822 			       iommu->stored_addr_lo | 1);
2823 }
2824 
2825 static void iommu_enable_ga(struct amd_iommu *iommu)
2826 {
2827 #ifdef CONFIG_IRQ_REMAP
2828 	switch (amd_iommu_guest_ir) {
2829 	case AMD_IOMMU_GUEST_IR_VAPIC:
2830 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2831 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2832 		iommu->irte_ops = &irte_128_ops;
2833 		break;
2834 	default:
2835 		iommu->irte_ops = &irte_32_ops;
2836 		break;
2837 	}
2838 #endif
2839 }
2840 
2841 static void iommu_disable_irtcachedis(struct amd_iommu *iommu)
2842 {
2843 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
2844 }
2845 
2846 static void iommu_enable_irtcachedis(struct amd_iommu *iommu)
2847 {
2848 	u64 ctrl;
2849 
2850 	if (!amd_iommu_irtcachedis)
2851 		return;
2852 
2853 	/*
2854 	 * Note:
2855 	 * The support for IRTCacheDis feature is dertermined by
2856 	 * checking if the bit is writable.
2857 	 */
2858 	iommu_feature_enable(iommu, CONTROL_IRTCACHEDIS);
2859 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
2860 	ctrl &= (1ULL << CONTROL_IRTCACHEDIS);
2861 	if (ctrl)
2862 		iommu->irtcachedis_enabled = true;
2863 	pr_info("iommu%d (%#06x) : IRT cache is %s\n",
2864 		iommu->index, iommu->devid,
2865 		iommu->irtcachedis_enabled ? "disabled" : "enabled");
2866 }
2867 
2868 static void iommu_enable_2k_int(struct amd_iommu *iommu)
2869 {
2870 	if (!FEATURE_NUM_INT_REMAP_SUP_2K(amd_iommu_efr2))
2871 		return;
2872 
2873 	iommu_feature_set(iommu,
2874 			  CONTROL_NUM_INT_REMAP_MODE_2K,
2875 			  CONTROL_NUM_INT_REMAP_MODE_MASK,
2876 			  CONTROL_NUM_INT_REMAP_MODE);
2877 }
2878 
2879 static void early_enable_iommu(struct amd_iommu *iommu)
2880 {
2881 	iommu_disable(iommu);
2882 	iommu_init_flags(iommu);
2883 	iommu_set_device_table(iommu);
2884 	iommu_enable_command_buffer(iommu);
2885 	iommu_enable_gt(iommu);
2886 	iommu_enable_ga(iommu);
2887 	iommu_enable_xt(iommu);
2888 	iommu_enable_irtcachedis(iommu);
2889 	iommu_enable_2k_int(iommu);
2890 	iommu_enable(iommu);
2891 	amd_iommu_flush_all_caches(iommu);
2892 }
2893 
2894 /*
2895  * This function finally enables all IOMMUs found in the system after
2896  * they have been initialized.
2897  *
2898  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to reuse
2899  * the old content of device table entries. Not this case or reuse failed,
2900  * just continue as normal kernel does.
2901  */
2902 static void early_enable_iommus(void)
2903 {
2904 	struct amd_iommu *iommu;
2905 	struct amd_iommu_pci_seg *pci_seg;
2906 
2907 	if (!reuse_device_table()) {
2908 		/*
2909 		 * If come here because of failure in reusing device table from old
2910 		 * kernel with all IOMMUs enabled, print error message and try to
2911 		 * free allocated old_dev_tbl_cpy.
2912 		 */
2913 		if (amd_iommu_pre_enabled) {
2914 			pr_err("Failed to reuse DEV table from previous kernel.\n");
2915 			/*
2916 			 * Bail out early if unable to remap/reuse DEV table from
2917 			 * previous kernel if SNP enabled as IOMMU commands will
2918 			 * time out without DEV table and cause kdump boot panic.
2919 			 */
2920 			BUG_ON(check_feature(FEATURE_SNP));
2921 		}
2922 
2923 		for_each_pci_segment(pci_seg) {
2924 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2925 				memunmap((void *)pci_seg->old_dev_tbl_cpy);
2926 				pci_seg->old_dev_tbl_cpy = NULL;
2927 			}
2928 		}
2929 
2930 		for_each_iommu(iommu) {
2931 			clear_translation_pre_enabled(iommu);
2932 			early_enable_iommu(iommu);
2933 		}
2934 	} else {
2935 		pr_info("Reused DEV table from previous kernel.\n");
2936 
2937 		for_each_pci_segment(pci_seg) {
2938 			iommu_free_pages(pci_seg->dev_table);
2939 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2940 		}
2941 
2942 		for_each_iommu(iommu) {
2943 			iommu_disable_command_buffer(iommu);
2944 			iommu_disable_event_buffer(iommu);
2945 			iommu_disable_irtcachedis(iommu);
2946 			iommu_enable_command_buffer(iommu);
2947 			iommu_enable_ga(iommu);
2948 			iommu_enable_xt(iommu);
2949 			iommu_enable_irtcachedis(iommu);
2950 			iommu_enable_2k_int(iommu);
2951 			iommu_set_device_table(iommu);
2952 			amd_iommu_flush_all_caches(iommu);
2953 		}
2954 	}
2955 }
2956 
2957 static void enable_iommus_ppr(void)
2958 {
2959 	struct amd_iommu *iommu;
2960 
2961 	if (!amd_iommu_gt_ppr_supported())
2962 		return;
2963 
2964 	for_each_iommu(iommu)
2965 		amd_iommu_enable_ppr_log(iommu);
2966 }
2967 
2968 static void enable_iommus_vapic(void)
2969 {
2970 #ifdef CONFIG_IRQ_REMAP
2971 	u32 status, i;
2972 	struct amd_iommu *iommu;
2973 
2974 	for_each_iommu(iommu) {
2975 		/*
2976 		 * Disable GALog if already running. It could have been enabled
2977 		 * in the previous boot before kdump.
2978 		 */
2979 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2980 		if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2981 			continue;
2982 
2983 		iommu_feature_disable(iommu, CONTROL_GALOG_EN);
2984 		iommu_feature_disable(iommu, CONTROL_GAINT_EN);
2985 
2986 		/*
2987 		 * Need to set and poll check the GALOGRun bit to zero before
2988 		 * we can set/ modify GA Log registers safely.
2989 		 */
2990 		for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
2991 			status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2992 			if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2993 				break;
2994 			udelay(10);
2995 		}
2996 
2997 		if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
2998 			return;
2999 	}
3000 
3001 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
3002 		return;
3003 
3004 	if (!check_feature(FEATURE_GAM_VAPIC)) {
3005 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3006 		return;
3007 	}
3008 
3009 	if (amd_iommu_snp_en &&
3010 	    !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) {
3011 		pr_warn("Force to disable Virtual APIC due to SNP\n");
3012 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3013 		return;
3014 	}
3015 
3016 	/* Enabling GAM and SNPAVIC support */
3017 	for_each_iommu(iommu) {
3018 		if (iommu_init_ga_log(iommu) ||
3019 		    iommu_ga_log_enable(iommu))
3020 			return;
3021 
3022 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
3023 		if (amd_iommu_snp_en)
3024 			iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN);
3025 	}
3026 
3027 	amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
3028 	pr_info("Virtual APIC enabled\n");
3029 #endif
3030 }
3031 
3032 static void disable_iommus(void)
3033 {
3034 	struct amd_iommu *iommu;
3035 
3036 	for_each_iommu(iommu)
3037 		iommu_disable(iommu);
3038 
3039 #ifdef CONFIG_IRQ_REMAP
3040 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
3041 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
3042 #endif
3043 }
3044 
3045 /*
3046  * Suspend/Resume support
3047  * disable suspend until real resume implemented
3048  */
3049 
3050 static void amd_iommu_resume(void *data)
3051 {
3052 	struct amd_iommu *iommu;
3053 
3054 	for_each_iommu(iommu)
3055 		iommu_apply_resume_quirks(iommu);
3056 
3057 	/* re-load the hardware */
3058 	for_each_iommu(iommu)
3059 		early_enable_iommu(iommu);
3060 
3061 	iommu_enable_event_buffer();
3062 	amd_iommu_enable_interrupts();
3063 }
3064 
3065 static int amd_iommu_suspend(void *data)
3066 {
3067 	/* disable IOMMUs to go out of the way for BIOS */
3068 	disable_iommus();
3069 
3070 	return 0;
3071 }
3072 
3073 static const struct syscore_ops amd_iommu_syscore_ops = {
3074 	.suspend = amd_iommu_suspend,
3075 	.resume = amd_iommu_resume,
3076 };
3077 
3078 static struct syscore amd_iommu_syscore = {
3079 	.ops = &amd_iommu_syscore_ops,
3080 };
3081 
3082 static void __init free_iommu_resources(void)
3083 {
3084 	free_iommu_all();
3085 	free_pci_segments();
3086 }
3087 
3088 /* SB IOAPIC is always on this device in AMD systems */
3089 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
3090 
3091 /* SB IOAPIC for Hygon family 18h model 4h is on the device 0xb */
3092 #define IOAPIC_SB_DEVID_FAM18H_M4H	((0x00 << 8) | PCI_DEVFN(0xb, 0))
3093 
3094 static bool __init check_ioapic_information(void)
3095 {
3096 	const char *fw_bug = FW_BUG;
3097 	bool ret, has_sb_ioapic;
3098 	int idx;
3099 
3100 	has_sb_ioapic = false;
3101 	ret           = false;
3102 
3103 	/*
3104 	 * If we have map overrides on the kernel command line the
3105 	 * messages in this function might not describe firmware bugs
3106 	 * anymore - so be careful
3107 	 */
3108 	if (cmdline_maps)
3109 		fw_bug = "";
3110 
3111 	for (idx = 0; idx < nr_ioapics; idx++) {
3112 		int devid, id = mpc_ioapic_id(idx);
3113 
3114 		devid = get_ioapic_devid(id);
3115 		if (devid < 0) {
3116 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
3117 				fw_bug, id);
3118 			ret = false;
3119 		} else if (devid == IOAPIC_SB_DEVID ||
3120 			   (boot_cpu_data.x86_vendor == X86_VENDOR_HYGON &&
3121 			    boot_cpu_data.x86 == 0x18 &&
3122 			    boot_cpu_data.x86_model >= 0x4 &&
3123 			    boot_cpu_data.x86_model <= 0xf &&
3124 			    devid == IOAPIC_SB_DEVID_FAM18H_M4H)) {
3125 			has_sb_ioapic = true;
3126 			ret           = true;
3127 		}
3128 	}
3129 
3130 	if (!has_sb_ioapic) {
3131 		/*
3132 		 * We expect the SB IOAPIC to be listed in the IVRS
3133 		 * table. The system timer is connected to the SB IOAPIC
3134 		 * and if we don't have it in the list the system will
3135 		 * panic at boot time.  This situation usually happens
3136 		 * when the BIOS is buggy and provides us the wrong
3137 		 * device id for the IOAPIC in the system.
3138 		 */
3139 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
3140 	}
3141 
3142 	if (!ret)
3143 		pr_err("Disabling interrupt remapping\n");
3144 
3145 	return ret;
3146 }
3147 
3148 static void __init free_dma_resources(void)
3149 {
3150 	amd_iommu_pdom_id_destroy();
3151 	free_unity_maps();
3152 }
3153 
3154 static void __init ivinfo_init(void *ivrs)
3155 {
3156 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
3157 }
3158 
3159 /*
3160  * This is the hardware init function for AMD IOMMU in the system.
3161  * This function is called either from amd_iommu_init or from the interrupt
3162  * remapping setup code.
3163  *
3164  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
3165  * four times:
3166  *
3167  *	1 pass) Discover the most comprehensive IVHD type to use.
3168  *
3169  *	2 pass) Find the highest PCI device id the driver has to handle.
3170  *		Upon this information the size of the data structures is
3171  *		determined that needs to be allocated.
3172  *
3173  *	3 pass) Initialize the data structures just allocated with the
3174  *		information in the ACPI table about available AMD IOMMUs
3175  *		in the system. It also maps the PCI devices in the
3176  *		system to specific IOMMUs
3177  *
3178  *	4 pass) After the basic data structures are allocated and
3179  *		initialized we update them with information about memory
3180  *		remapping requirements parsed out of the ACPI table in
3181  *		this last pass.
3182  *
3183  * After everything is set up the IOMMUs are enabled and the necessary
3184  * hotplug and suspend notifiers are registered.
3185  */
3186 static int __init early_amd_iommu_init(void)
3187 {
3188 	struct acpi_table_header *ivrs_base;
3189 	int ret;
3190 	acpi_status status;
3191 	u8 efr_hats, max_vasize;
3192 
3193 	if (!amd_iommu_detected)
3194 		return -ENODEV;
3195 
3196 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3197 	if (status == AE_NOT_FOUND)
3198 		return -ENODEV;
3199 	else if (ACPI_FAILURE(status)) {
3200 		const char *err = acpi_format_exception(status);
3201 		pr_err("IVRS table error: %s\n", err);
3202 		return -EINVAL;
3203 	}
3204 
3205 	if (!boot_cpu_has(X86_FEATURE_CX16)) {
3206 		pr_err("Failed to initialize. The CMPXCHG16B feature is required.\n");
3207 		ret = -EINVAL;
3208 		goto out;
3209 	}
3210 
3211 	/*
3212 	 * Validate checksum here so we don't need to do it when
3213 	 * we actually parse the table
3214 	 */
3215 	ret = check_ivrs_checksum(ivrs_base);
3216 	if (ret)
3217 		goto out;
3218 
3219 	ivinfo_init(ivrs_base);
3220 
3221 	max_vasize = FIELD_GET(IOMMU_IVINFO_VASIZE, amd_iommu_ivinfo);
3222 	if (!max_vasize)
3223 		max_vasize = 64;
3224 
3225 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
3226 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
3227 
3228 	/*
3229 	 * now the data structures are allocated and basically initialized
3230 	 * start the real acpi table scan
3231 	 */
3232 	ret = init_iommu_all(ivrs_base);
3233 	if (ret)
3234 		goto out;
3235 
3236 	/* 5 level guest page table */
3237 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3238 	    FIELD_GET(FEATURE_GATS, amd_iommu_efr) == GUEST_PGTABLE_5_LEVEL)
3239 		amd_iommu_gpt_level = PAGE_MODE_5_LEVEL;
3240 
3241 	efr_hats = FIELD_GET(FEATURE_HATS, amd_iommu_efr);
3242 	if (efr_hats != 0x3) {
3243 		/*
3244 		 * efr[HATS] bits specify the maximum host translation level
3245 		 * supported, with LEVEL 4 being initial max level.
3246 		 */
3247 		amd_iommu_hpt_vasize = min_t(unsigned int, max_vasize,
3248 					 (efr_hats + PAGE_MODE_4_LEVEL - 1) * 9 + 21);
3249 	} else {
3250 		pr_warn_once(FW_BUG "Disable host address translation due to invalid translation level (%#x).\n",
3251 			     efr_hats);
3252 		amd_iommu_hatdis = true;
3253 	}
3254 
3255 	if (amd_iommu_pgtable == PD_MODE_V2) {
3256 		if (!amd_iommu_v2_pgtbl_supported()) {
3257 			pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n");
3258 			amd_iommu_pgtable = PD_MODE_V1;
3259 		}
3260 	}
3261 
3262 	if (amd_iommu_hatdis) {
3263 		/*
3264 		 * Host (v1) page table is not available. Attempt to use
3265 		 * Guest (v2) page table.
3266 		 */
3267 		if (amd_iommu_v2_pgtbl_supported())
3268 			amd_iommu_pgtable = PD_MODE_V2;
3269 		else
3270 			amd_iommu_pgtable = PD_MODE_NONE;
3271 	}
3272 
3273 	/* Disable any previously enabled IOMMUs */
3274 	if (!is_kdump_kernel() || amd_iommu_disabled)
3275 		disable_iommus();
3276 
3277 	if (amd_iommu_irq_remap)
3278 		amd_iommu_irq_remap = check_ioapic_information();
3279 
3280 	if (amd_iommu_irq_remap) {
3281 		struct amd_iommu_pci_seg *pci_seg;
3282 		ret = -ENOMEM;
3283 		for_each_pci_segment(pci_seg) {
3284 			if (alloc_irq_lookup_table(pci_seg))
3285 				goto out;
3286 		}
3287 	}
3288 
3289 	ret = init_memory_definitions(ivrs_base);
3290 	if (ret)
3291 		goto out;
3292 
3293 	/* init the device table */
3294 	init_device_table();
3295 
3296 out:
3297 	/* Don't leak any ACPI memory */
3298 	acpi_put_table(ivrs_base);
3299 
3300 	return ret;
3301 }
3302 
3303 static int amd_iommu_enable_interrupts(void)
3304 {
3305 	struct amd_iommu *iommu;
3306 	int ret = 0;
3307 
3308 	for_each_iommu(iommu) {
3309 		ret = iommu_init_irq(iommu);
3310 		if (ret)
3311 			goto out;
3312 	}
3313 
3314 	/*
3315 	 * Interrupt handler is ready to process interrupts. Enable
3316 	 * PPR and GA log interrupt for all IOMMUs.
3317 	 */
3318 	enable_iommus_vapic();
3319 	enable_iommus_ppr();
3320 
3321 out:
3322 	return ret;
3323 }
3324 
3325 static bool __init detect_ivrs(void)
3326 {
3327 	struct acpi_table_header *ivrs_base;
3328 	acpi_status status;
3329 	int i;
3330 
3331 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3332 	if (status == AE_NOT_FOUND)
3333 		return false;
3334 	else if (ACPI_FAILURE(status)) {
3335 		const char *err = acpi_format_exception(status);
3336 		pr_err("IVRS table error: %s\n", err);
3337 		return false;
3338 	}
3339 
3340 	acpi_put_table(ivrs_base);
3341 
3342 	if (amd_iommu_force_enable)
3343 		goto out;
3344 
3345 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3346 	for (i = 0; i < 32; i++) {
3347 		u32 pci_id;
3348 
3349 		pci_id = read_pci_config(0, i, 0, 0);
3350 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3351 			pr_info("Disable IOMMU on Stoney Ridge\n");
3352 			return false;
3353 		}
3354 	}
3355 
3356 out:
3357 	/* Make sure ACS will be enabled during PCI probe */
3358 	pci_request_acs();
3359 
3360 	return true;
3361 }
3362 
3363 static __init void iommu_snp_enable(void)
3364 {
3365 #ifdef CONFIG_KVM_AMD_SEV
3366 	if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3367 		return;
3368 	/*
3369 	 * The SNP support requires that IOMMU must be enabled, and is
3370 	 * configured with V1 page table (DTE[Mode] = 0 is not supported).
3371 	 */
3372 	if (no_iommu || iommu_default_passthrough()) {
3373 		pr_warn("SNP: IOMMU disabled or configured in passthrough mode, SNP cannot be supported.\n");
3374 		goto disable_snp;
3375 	}
3376 
3377 	if (amd_iommu_pgtable != PD_MODE_V1) {
3378 		pr_warn("SNP: IOMMU is configured with V2 page table mode, SNP cannot be supported.\n");
3379 		goto disable_snp;
3380 	}
3381 
3382 	amd_iommu_snp_en = check_feature(FEATURE_SNP);
3383 	if (!amd_iommu_snp_en) {
3384 		pr_warn("SNP: IOMMU SNP feature not enabled, SNP cannot be supported.\n");
3385 		goto disable_snp;
3386 	}
3387 
3388 	/*
3389 	 * Enable host SNP support once SNP support is checked on IOMMU.
3390 	 */
3391 	if (snp_rmptable_init()) {
3392 		pr_warn("SNP: RMP initialization failed, SNP cannot be supported.\n");
3393 		goto disable_snp;
3394 	}
3395 
3396 	pr_info("IOMMU SNP support enabled.\n");
3397 	return;
3398 
3399 disable_snp:
3400 	cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3401 #endif
3402 }
3403 
3404 static void amd_iommu_apply_erratum_snp(void)
3405 {
3406 #ifdef CONFIG_KVM_AMD_SEV
3407 	if (!amd_iommu_snp_en)
3408 		return;
3409 
3410 	/* Errata fix for Family 0x19 */
3411 	if (boot_cpu_data.x86 != 0x19)
3412 		return;
3413 
3414 	/* Set event log buffer size to max */
3415 	amd_iommu_evtlog_size = EVTLOG_SIZE_MAX;
3416 	pr_info("Applying erratum: Increase Event log size to 0x%x\n",
3417 		amd_iommu_evtlog_size);
3418 
3419 	/*
3420 	 * Set PPR log buffer size to max.
3421 	 * (Family 0x19, model < 0x10 doesn't support PPR when SNP is enabled).
3422 	 */
3423 	if (boot_cpu_data.x86_model >= 0x10) {
3424 		amd_iommu_pprlog_size = PPRLOG_SIZE_MAX;
3425 		pr_info("Applying erratum: Increase PPR log size to 0x%x\n",
3426 			amd_iommu_pprlog_size);
3427 	}
3428 #endif
3429 }
3430 
3431 /****************************************************************************
3432  *
3433  * AMD IOMMU Initialization State Machine
3434  *
3435  ****************************************************************************/
3436 
3437 static int __init state_next(void)
3438 {
3439 	int ret = 0;
3440 
3441 	switch (init_state) {
3442 	case IOMMU_START_STATE:
3443 		if (!detect_ivrs()) {
3444 			init_state	= IOMMU_NOT_FOUND;
3445 			ret		= -ENODEV;
3446 		} else {
3447 			init_state	= IOMMU_IVRS_DETECTED;
3448 		}
3449 		break;
3450 	case IOMMU_IVRS_DETECTED:
3451 		if (amd_iommu_disabled) {
3452 			init_state = IOMMU_CMDLINE_DISABLED;
3453 			ret = -EINVAL;
3454 		} else {
3455 			ret = early_amd_iommu_init();
3456 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3457 		}
3458 		break;
3459 	case IOMMU_ACPI_FINISHED:
3460 		early_enable_iommus();
3461 		x86_platform.iommu_shutdown = disable_iommus;
3462 		init_state = IOMMU_ENABLED;
3463 		break;
3464 	case IOMMU_ENABLED:
3465 		register_syscore(&amd_iommu_syscore);
3466 		iommu_snp_enable();
3467 
3468 		amd_iommu_apply_erratum_snp();
3469 
3470 		/* Allocate/enable event log buffer */
3471 		if (is_kdump_kernel())
3472 			ret = remap_event_buffer();
3473 		else
3474 			ret = alloc_event_buffer();
3475 
3476 		if (ret) {
3477 			init_state = IOMMU_INIT_ERROR;
3478 			break;
3479 		}
3480 		iommu_enable_event_buffer();
3481 
3482 		ret = amd_iommu_init_pci();
3483 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3484 		break;
3485 	case IOMMU_PCI_INIT:
3486 		ret = amd_iommu_enable_interrupts();
3487 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3488 		break;
3489 	case IOMMU_INTERRUPTS_EN:
3490 		init_state = IOMMU_INITIALIZED;
3491 		break;
3492 	case IOMMU_INITIALIZED:
3493 		/* Nothing to do */
3494 		break;
3495 	case IOMMU_NOT_FOUND:
3496 	case IOMMU_INIT_ERROR:
3497 	case IOMMU_CMDLINE_DISABLED:
3498 		/* Error states => do nothing */
3499 		ret = -EINVAL;
3500 		break;
3501 	default:
3502 		/* Unknown state */
3503 		BUG();
3504 	}
3505 
3506 	if (ret) {
3507 		free_dma_resources();
3508 		if (!irq_remapping_enabled) {
3509 			disable_iommus();
3510 			free_iommu_resources();
3511 		} else {
3512 			struct amd_iommu *iommu;
3513 			struct amd_iommu_pci_seg *pci_seg;
3514 
3515 			for_each_pci_segment(pci_seg)
3516 				uninit_device_table_dma(pci_seg);
3517 
3518 			for_each_iommu(iommu)
3519 				amd_iommu_flush_all_caches(iommu);
3520 		}
3521 	}
3522 	return ret;
3523 }
3524 
3525 static int __init iommu_go_to_state(enum iommu_init_state state)
3526 {
3527 	int ret = -EINVAL;
3528 
3529 	while (init_state != state) {
3530 		if (init_state == IOMMU_NOT_FOUND         ||
3531 		    init_state == IOMMU_INIT_ERROR        ||
3532 		    init_state == IOMMU_CMDLINE_DISABLED)
3533 			break;
3534 		ret = state_next();
3535 	}
3536 
3537 	/*
3538 	 * SNP platform initilazation requires IOMMUs to be fully configured.
3539 	 * If the SNP support on IOMMUs has NOT been checked, simply mark SNP
3540 	 * as unsupported. If the SNP support on IOMMUs has been checked and
3541 	 * host SNP support enabled but RMP enforcement has not been enabled
3542 	 * in IOMMUs, then the system is in a half-baked state, but can limp
3543 	 * along as all memory should be Hypervisor-Owned in the RMP. WARN,
3544 	 * but leave SNP as "supported" to avoid confusing the kernel.
3545 	 */
3546 	if (ret && cc_platform_has(CC_ATTR_HOST_SEV_SNP) &&
3547 	    !WARN_ON_ONCE(amd_iommu_snp_en))
3548 		cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3549 
3550 	return ret;
3551 }
3552 
3553 #ifdef CONFIG_IRQ_REMAP
3554 int __init amd_iommu_prepare(void)
3555 {
3556 	int ret;
3557 
3558 	amd_iommu_irq_remap = true;
3559 
3560 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3561 	if (ret) {
3562 		amd_iommu_irq_remap = false;
3563 		return ret;
3564 	}
3565 
3566 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3567 }
3568 
3569 int __init amd_iommu_enable(void)
3570 {
3571 	int ret;
3572 
3573 	ret = iommu_go_to_state(IOMMU_ENABLED);
3574 	if (ret)
3575 		return ret;
3576 
3577 	irq_remapping_enabled = 1;
3578 	return amd_iommu_xt_mode;
3579 }
3580 
3581 void amd_iommu_disable(void)
3582 {
3583 	amd_iommu_suspend(NULL);
3584 }
3585 
3586 int amd_iommu_reenable(int mode)
3587 {
3588 	amd_iommu_resume(NULL);
3589 
3590 	return 0;
3591 }
3592 
3593 int amd_iommu_enable_faulting(unsigned int cpu)
3594 {
3595 	/* We enable MSI later when PCI is initialized */
3596 	return 0;
3597 }
3598 #endif
3599 
3600 /*
3601  * This is the core init function for AMD IOMMU hardware in the system.
3602  * This function is called from the generic x86 DMA layer initialization
3603  * code.
3604  */
3605 static int __init amd_iommu_init(void)
3606 {
3607 	int ret;
3608 
3609 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3610 #ifdef CONFIG_GART_IOMMU
3611 	if (ret && list_empty(&amd_iommu_list)) {
3612 		/*
3613 		 * We failed to initialize the AMD IOMMU - try fallback
3614 		 * to GART if possible.
3615 		 */
3616 		gart_iommu_init();
3617 	}
3618 #endif
3619 
3620 	if (!ret)
3621 		amd_iommu_debugfs_setup();
3622 
3623 	return ret;
3624 }
3625 
3626 static bool amd_iommu_sme_check(void)
3627 {
3628 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3629 	    (boot_cpu_data.x86 != 0x17))
3630 		return true;
3631 
3632 	/* For Fam17h, a specific level of support is required */
3633 	if (boot_cpu_data.microcode >= 0x08001205)
3634 		return true;
3635 
3636 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3637 	    (boot_cpu_data.microcode <= 0x080011ff))
3638 		return true;
3639 
3640 	pr_notice("IOMMU not currently supported when SME is active\n");
3641 
3642 	return false;
3643 }
3644 
3645 /****************************************************************************
3646  *
3647  * Early detect code. This code runs at IOMMU detection time in the DMA
3648  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3649  * IOMMUs
3650  *
3651  ****************************************************************************/
3652 void __init amd_iommu_detect(void)
3653 {
3654 	int ret;
3655 
3656 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3657 		goto disable_snp;
3658 
3659 	if (!amd_iommu_sme_check())
3660 		goto disable_snp;
3661 
3662 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3663 	if (ret)
3664 		goto disable_snp;
3665 
3666 	amd_iommu_detected = true;
3667 	iommu_detected = 1;
3668 	x86_init.iommu.iommu_init = amd_iommu_init;
3669 	return;
3670 
3671 disable_snp:
3672 	if (cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3673 		cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3674 }
3675 
3676 /****************************************************************************
3677  *
3678  * Parsing functions for the AMD IOMMU specific kernel command line
3679  * options.
3680  *
3681  ****************************************************************************/
3682 
3683 static int __init parse_amd_iommu_dump(char *str)
3684 {
3685 	amd_iommu_dump = true;
3686 
3687 	return 1;
3688 }
3689 
3690 static int __init parse_amd_iommu_intr(char *str)
3691 {
3692 	for (; *str; ++str) {
3693 		if (strncmp(str, "legacy", 6) == 0) {
3694 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3695 			break;
3696 		}
3697 		if (strncmp(str, "vapic", 5) == 0) {
3698 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3699 			break;
3700 		}
3701 	}
3702 	return 1;
3703 }
3704 
3705 static int __init parse_amd_iommu_options(char *str)
3706 {
3707 	if (!str)
3708 		return -EINVAL;
3709 
3710 	while (*str) {
3711 		if (strncmp(str, "fullflush", 9) == 0) {
3712 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3713 			iommu_set_dma_strict();
3714 		} else if (strncmp(str, "force_enable", 12) == 0) {
3715 			amd_iommu_force_enable = true;
3716 		} else if (strncmp(str, "off", 3) == 0) {
3717 			amd_iommu_disabled = true;
3718 		} else if (strncmp(str, "force_isolation", 15) == 0) {
3719 			amd_iommu_force_isolation = true;
3720 		} else if (strncmp(str, "pgtbl_v1", 8) == 0) {
3721 			amd_iommu_pgtable = PD_MODE_V1;
3722 		} else if (strncmp(str, "pgtbl_v2", 8) == 0) {
3723 			amd_iommu_pgtable = PD_MODE_V2;
3724 		} else if (strncmp(str, "irtcachedis", 11) == 0) {
3725 			amd_iommu_irtcachedis = true;
3726 		} else if (strncmp(str, "nohugepages", 11) == 0) {
3727 			pr_info("Restricting V1 page-sizes to 4KiB");
3728 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_4K;
3729 		} else if (strncmp(str, "v2_pgsizes_only", 15) == 0) {
3730 			pr_info("Restricting V1 page-sizes to 4KiB/2MiB/1GiB");
3731 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_V2;
3732 		} else {
3733 			pr_notice("Unknown option - '%s'\n", str);
3734 		}
3735 
3736 		str += strcspn(str, ",");
3737 		while (*str == ',')
3738 			str++;
3739 	}
3740 
3741 	return 1;
3742 }
3743 
3744 static int __init parse_ivrs_ioapic(char *str)
3745 {
3746 	u32 seg = 0, bus, dev, fn;
3747 	int id, i;
3748 	u32 devid;
3749 
3750 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3751 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3752 		goto found;
3753 
3754 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3755 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3756 		pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n",
3757 			str, id, seg, bus, dev, fn);
3758 		goto found;
3759 	}
3760 
3761 	pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3762 	return 1;
3763 
3764 found:
3765 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3766 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3767 			str);
3768 		return 1;
3769 	}
3770 
3771 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3772 
3773 	cmdline_maps			= true;
3774 	i				= early_ioapic_map_size++;
3775 	early_ioapic_map[i].id		= id;
3776 	early_ioapic_map[i].devid	= devid;
3777 	early_ioapic_map[i].cmd_line	= true;
3778 
3779 	return 1;
3780 }
3781 
3782 static int __init parse_ivrs_hpet(char *str)
3783 {
3784 	u32 seg = 0, bus, dev, fn;
3785 	int id, i;
3786 	u32 devid;
3787 
3788 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3789 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3790 		goto found;
3791 
3792 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3793 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3794 		pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n",
3795 			str, id, seg, bus, dev, fn);
3796 		goto found;
3797 	}
3798 
3799 	pr_err("Invalid command line: ivrs_hpet%s\n", str);
3800 	return 1;
3801 
3802 found:
3803 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3804 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3805 			str);
3806 		return 1;
3807 	}
3808 
3809 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3810 
3811 	cmdline_maps			= true;
3812 	i				= early_hpet_map_size++;
3813 	early_hpet_map[i].id		= id;
3814 	early_hpet_map[i].devid		= devid;
3815 	early_hpet_map[i].cmd_line	= true;
3816 
3817 	return 1;
3818 }
3819 
3820 #define ACPIID_LEN (ACPIHID_UID_LEN + ACPIHID_HID_LEN)
3821 
3822 static int __init parse_ivrs_acpihid(char *str)
3823 {
3824 	u32 seg = 0, bus, dev, fn;
3825 	char *hid, *uid, *p, *addr;
3826 	char acpiid[ACPIID_LEN + 1] = { }; /* size with NULL terminator */
3827 	int i;
3828 
3829 	addr = strchr(str, '@');
3830 	if (!addr) {
3831 		addr = strchr(str, '=');
3832 		if (!addr)
3833 			goto not_found;
3834 
3835 		++addr;
3836 
3837 		if (strlen(addr) > ACPIID_LEN)
3838 			goto not_found;
3839 
3840 		if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 ||
3841 		    sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) {
3842 			pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n",
3843 				str, acpiid, seg, bus, dev, fn);
3844 			goto found;
3845 		}
3846 		goto not_found;
3847 	}
3848 
3849 	/* We have the '@', make it the terminator to get just the acpiid */
3850 	*addr++ = 0;
3851 
3852 	if (strlen(str) > ACPIID_LEN)
3853 		goto not_found;
3854 
3855 	if (sscanf(str, "=%s", acpiid) != 1)
3856 		goto not_found;
3857 
3858 	if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 ||
3859 	    sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4)
3860 		goto found;
3861 
3862 not_found:
3863 	pr_err("Invalid command line: ivrs_acpihid%s\n", str);
3864 	return 1;
3865 
3866 found:
3867 	p = acpiid;
3868 	hid = strsep(&p, ":");
3869 	uid = p;
3870 
3871 	if (!hid || !(*hid) || !uid) {
3872 		pr_err("Invalid command line: hid or uid\n");
3873 		return 1;
3874 	}
3875 
3876 	/*
3877 	 * Ignore leading zeroes after ':', so e.g., AMDI0095:00
3878 	 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match
3879 	 */
3880 	while (*uid == '0' && *(uid + 1))
3881 		uid++;
3882 
3883 	if (strlen(hid) >= ACPIHID_HID_LEN) {
3884 		pr_err("Invalid command line: hid is too long\n");
3885 		return 1;
3886 	} else if (strlen(uid) >= ACPIHID_UID_LEN) {
3887 		pr_err("Invalid command line: uid is too long\n");
3888 		return 1;
3889 	}
3890 
3891 	i = early_acpihid_map_size++;
3892 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3893 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3894 	early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3895 	early_acpihid_map[i].cmd_line	= true;
3896 
3897 	return 1;
3898 }
3899 
3900 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3901 __setup("amd_iommu=",		parse_amd_iommu_options);
3902 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3903 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3904 __setup("ivrs_hpet",		parse_ivrs_hpet);
3905 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3906 
3907 bool amd_iommu_pasid_supported(void)
3908 {
3909 	/* CPU page table size should match IOMMU guest page table size */
3910 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3911 	    amd_iommu_gpt_level != PAGE_MODE_5_LEVEL)
3912 		return false;
3913 
3914 	/*
3915 	 * Since DTE[Mode]=0 is prohibited on SNP-enabled system
3916 	 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without
3917 	 * setting up IOMMUv1 page table.
3918 	 */
3919 	return amd_iommu_gt_ppr_supported() && !amd_iommu_snp_en;
3920 }
3921 
3922 struct amd_iommu *get_amd_iommu(unsigned int idx)
3923 {
3924 	unsigned int i = 0;
3925 	struct amd_iommu *iommu;
3926 
3927 	for_each_iommu(iommu)
3928 		if (i++ == idx)
3929 			return iommu;
3930 	return NULL;
3931 }
3932 
3933 /****************************************************************************
3934  *
3935  * IOMMU EFR Performance Counter support functionality. This code allows
3936  * access to the IOMMU PC functionality.
3937  *
3938  ****************************************************************************/
3939 
3940 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3941 {
3942 	struct amd_iommu *iommu = get_amd_iommu(idx);
3943 
3944 	if (iommu)
3945 		return iommu->max_banks;
3946 
3947 	return 0;
3948 }
3949 
3950 bool amd_iommu_pc_supported(void)
3951 {
3952 	return amd_iommu_pc_present;
3953 }
3954 
3955 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3956 {
3957 	struct amd_iommu *iommu = get_amd_iommu(idx);
3958 
3959 	if (iommu)
3960 		return iommu->max_counters;
3961 
3962 	return 0;
3963 }
3964 
3965 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3966 				u8 fxn, u64 *value, bool is_write)
3967 {
3968 	u32 offset;
3969 	u32 max_offset_lim;
3970 
3971 	/* Make sure the IOMMU PC resource is available */
3972 	if (!amd_iommu_pc_present)
3973 		return -ENODEV;
3974 
3975 	/* Check for valid iommu and pc register indexing */
3976 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3977 		return -ENODEV;
3978 
3979 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3980 
3981 	/* Limit the offset to the hw defined mmio region aperture */
3982 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3983 				(iommu->max_counters << 8) | 0x28);
3984 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3985 	    (offset > max_offset_lim))
3986 		return -EINVAL;
3987 
3988 	if (is_write) {
3989 		u64 val = *value & GENMASK_ULL(47, 0);
3990 
3991 		writel((u32)val, iommu->mmio_base + offset);
3992 		writel((val >> 32), iommu->mmio_base + offset + 4);
3993 	} else {
3994 		*value = readl(iommu->mmio_base + offset + 4);
3995 		*value <<= 32;
3996 		*value |= readl(iommu->mmio_base + offset);
3997 		*value &= GENMASK_ULL(47, 0);
3998 	}
3999 
4000 	return 0;
4001 }
4002 
4003 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
4004 {
4005 	if (!iommu)
4006 		return -EINVAL;
4007 
4008 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
4009 }
4010 
4011 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
4012 {
4013 	if (!iommu)
4014 		return -EINVAL;
4015 
4016 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
4017 }
4018 
4019 #ifdef CONFIG_KVM_AMD_SEV
4020 static int iommu_page_make_shared(void *page)
4021 {
4022 	unsigned long paddr, pfn;
4023 
4024 	paddr = iommu_virt_to_phys(page);
4025 	/* Cbit maybe set in the paddr */
4026 	pfn = __sme_clr(paddr) >> PAGE_SHIFT;
4027 
4028 	if (!(pfn % PTRS_PER_PMD)) {
4029 		int ret, level;
4030 		bool assigned;
4031 
4032 		ret = snp_lookup_rmpentry(pfn, &assigned, &level);
4033 		if (ret) {
4034 			pr_warn("IOMMU PFN %lx RMP lookup failed, ret %d\n", pfn, ret);
4035 			return ret;
4036 		}
4037 
4038 		if (!assigned) {
4039 			pr_warn("IOMMU PFN %lx not assigned in RMP table\n", pfn);
4040 			return -EINVAL;
4041 		}
4042 
4043 		if (level > PG_LEVEL_4K) {
4044 			ret = psmash(pfn);
4045 			if (!ret)
4046 				goto done;
4047 
4048 			pr_warn("PSMASH failed for IOMMU PFN %lx huge RMP entry, ret: %d, level: %d\n",
4049 				pfn, ret, level);
4050 			return ret;
4051 		}
4052 	}
4053 
4054 done:
4055 	return rmp_make_shared(pfn, PG_LEVEL_4K);
4056 }
4057 
4058 static int iommu_make_shared(void *va, size_t size)
4059 {
4060 	void *page;
4061 	int ret;
4062 
4063 	if (!va)
4064 		return 0;
4065 
4066 	for (page = va; page < (va + size); page += PAGE_SIZE) {
4067 		ret = iommu_page_make_shared(page);
4068 		if (ret)
4069 			return ret;
4070 	}
4071 
4072 	return 0;
4073 }
4074 
4075 int amd_iommu_snp_disable(void)
4076 {
4077 	struct amd_iommu *iommu;
4078 	int ret;
4079 
4080 	if (!amd_iommu_snp_en)
4081 		return 0;
4082 
4083 	for_each_iommu(iommu) {
4084 		ret = iommu_make_shared(iommu->evt_buf, amd_iommu_evtlog_size);
4085 		if (ret)
4086 			return ret;
4087 
4088 		ret = iommu_make_shared(iommu->ppr_log, amd_iommu_pprlog_size);
4089 		if (ret)
4090 			return ret;
4091 
4092 		ret = iommu_make_shared((void *)iommu->cmd_sem, PAGE_SIZE);
4093 		if (ret)
4094 			return ret;
4095 	}
4096 
4097 	return 0;
4098 }
4099 EXPORT_SYMBOL_GPL(amd_iommu_snp_disable);
4100 
4101 bool amd_iommu_sev_tio_supported(void)
4102 {
4103 	return check_feature2(FEATURE_SEVSNPIO_SUP);
4104 }
4105 EXPORT_SYMBOL_GPL(amd_iommu_sev_tio_supported);
4106 #endif
4107