xref: /linux/drivers/iommu/amd/init.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
4  * Author: Joerg Roedel <jroedel@suse.de>
5  *         Leo Duran <leo.duran@amd.com>
6  */
7 
8 #define pr_fmt(fmt)     "AMD-Vi: " fmt
9 #define dev_fmt(fmt)    pr_fmt(fmt)
10 
11 #include <linux/pci.h>
12 #include <linux/acpi.h>
13 #include <linux/list.h>
14 #include <linux/bitmap.h>
15 #include <linux/slab.h>
16 #include <linux/syscore_ops.h>
17 #include <linux/interrupt.h>
18 #include <linux/msi.h>
19 #include <linux/irq.h>
20 #include <linux/amd-iommu.h>
21 #include <linux/export.h>
22 #include <linux/kmemleak.h>
23 #include <linux/cc_platform.h>
24 #include <linux/iopoll.h>
25 #include <asm/pci-direct.h>
26 #include <asm/iommu.h>
27 #include <asm/apic.h>
28 #include <asm/gart.h>
29 #include <asm/x86_init.h>
30 #include <asm/io_apic.h>
31 #include <asm/irq_remapping.h>
32 #include <asm/set_memory.h>
33 #include <asm/sev.h>
34 
35 #include <linux/crash_dump.h>
36 
37 #include "amd_iommu.h"
38 #include "../irq_remapping.h"
39 #include "../iommu-pages.h"
40 
41 /*
42  * definitions for the ACPI scanning code
43  */
44 #define IVRS_HEADER_LENGTH 48
45 
46 #define ACPI_IVHD_TYPE_MAX_SUPPORTED	0x40
47 #define ACPI_IVMD_TYPE_ALL              0x20
48 #define ACPI_IVMD_TYPE                  0x21
49 #define ACPI_IVMD_TYPE_RANGE            0x22
50 
51 #define IVHD_DEV_ALL                    0x01
52 #define IVHD_DEV_SELECT                 0x02
53 #define IVHD_DEV_SELECT_RANGE_START     0x03
54 #define IVHD_DEV_RANGE_END              0x04
55 #define IVHD_DEV_ALIAS                  0x42
56 #define IVHD_DEV_ALIAS_RANGE            0x43
57 #define IVHD_DEV_EXT_SELECT             0x46
58 #define IVHD_DEV_EXT_SELECT_RANGE       0x47
59 #define IVHD_DEV_SPECIAL		0x48
60 #define IVHD_DEV_ACPI_HID		0xf0
61 
62 #define UID_NOT_PRESENT                 0
63 #define UID_IS_INTEGER                  1
64 #define UID_IS_CHARACTER                2
65 
66 #define IVHD_SPECIAL_IOAPIC		1
67 #define IVHD_SPECIAL_HPET		2
68 
69 #define IVHD_FLAG_HT_TUN_EN_MASK        0x01
70 #define IVHD_FLAG_PASSPW_EN_MASK        0x02
71 #define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
72 #define IVHD_FLAG_ISOC_EN_MASK          0x08
73 
74 #define IVMD_FLAG_EXCL_RANGE            0x08
75 #define IVMD_FLAG_IW                    0x04
76 #define IVMD_FLAG_IR                    0x02
77 #define IVMD_FLAG_UNITY_MAP             0x01
78 
79 #define ACPI_DEVFLAG_INITPASS           0x01
80 #define ACPI_DEVFLAG_EXTINT             0x02
81 #define ACPI_DEVFLAG_NMI                0x04
82 #define ACPI_DEVFLAG_SYSMGT1            0x10
83 #define ACPI_DEVFLAG_SYSMGT2            0x20
84 #define ACPI_DEVFLAG_LINT0              0x40
85 #define ACPI_DEVFLAG_LINT1              0x80
86 #define ACPI_DEVFLAG_ATSDIS             0x10000000
87 
88 #define IVRS_GET_SBDF_ID(seg, bus, dev, fn)	(((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \
89 						 | ((dev & 0x1f) << 3) | (fn & 0x7))
90 
91 /*
92  * ACPI table definitions
93  *
94  * These data structures are laid over the table to parse the important values
95  * out of it.
96  */
97 
98 /*
99  * structure describing one IOMMU in the ACPI table. Typically followed by one
100  * or more ivhd_entrys.
101  */
102 struct ivhd_header {
103 	u8 type;
104 	u8 flags;
105 	u16 length;
106 	u16 devid;
107 	u16 cap_ptr;
108 	u64 mmio_phys;
109 	u16 pci_seg;
110 	u16 info;
111 	u32 efr_attr;
112 
113 	/* Following only valid on IVHD type 11h and 40h */
114 	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
115 	u64 efr_reg2;
116 } __attribute__((packed));
117 
118 /*
119  * A device entry describing which devices a specific IOMMU translates and
120  * which requestor ids they use.
121  */
122 struct ivhd_entry {
123 	u8 type;
124 	u16 devid;
125 	u8 flags;
126 	struct_group(ext_hid,
127 		u32 ext;
128 		u32 hidh;
129 	);
130 	u64 cid;
131 	u8 uidf;
132 	u8 uidl;
133 	u8 uid;
134 } __attribute__((packed));
135 
136 /*
137  * An AMD IOMMU memory definition structure. It defines things like exclusion
138  * ranges for devices and regions that should be unity mapped.
139  */
140 struct ivmd_header {
141 	u8 type;
142 	u8 flags;
143 	u16 length;
144 	u16 devid;
145 	u16 aux;
146 	u16 pci_seg;
147 	u8  resv[6];
148 	u64 range_start;
149 	u64 range_length;
150 } __attribute__((packed));
151 
152 bool amd_iommu_dump;
153 bool amd_iommu_irq_remap __read_mostly;
154 
155 enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1;
156 /* Guest page table level */
157 int amd_iommu_gpt_level = PAGE_MODE_4_LEVEL;
158 
159 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
160 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
161 
162 static bool amd_iommu_detected;
163 static bool amd_iommu_disabled __initdata;
164 static bool amd_iommu_force_enable __initdata;
165 static bool amd_iommu_irtcachedis;
166 static int amd_iommu_target_ivhd_type;
167 
168 /* Global EFR and EFR2 registers */
169 u64 amd_iommu_efr;
170 u64 amd_iommu_efr2;
171 
172 /* SNP is enabled on the system? */
173 bool amd_iommu_snp_en;
174 EXPORT_SYMBOL(amd_iommu_snp_en);
175 
176 LIST_HEAD(amd_iommu_pci_seg_list);	/* list of all PCI segments */
177 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
178 					   system */
179 
180 /* Number of IOMMUs present in the system */
181 static int amd_iommus_present;
182 
183 /* IOMMUs have a non-present cache? */
184 bool amd_iommu_np_cache __read_mostly;
185 bool amd_iommu_iotlb_sup __read_mostly = true;
186 
187 static bool amd_iommu_pc_present __read_mostly;
188 bool amdr_ivrs_remap_support __read_mostly;
189 
190 bool amd_iommu_force_isolation __read_mostly;
191 
192 unsigned long amd_iommu_pgsize_bitmap __ro_after_init = AMD_IOMMU_PGSIZES;
193 
194 enum iommu_init_state {
195 	IOMMU_START_STATE,
196 	IOMMU_IVRS_DETECTED,
197 	IOMMU_ACPI_FINISHED,
198 	IOMMU_ENABLED,
199 	IOMMU_PCI_INIT,
200 	IOMMU_INTERRUPTS_EN,
201 	IOMMU_INITIALIZED,
202 	IOMMU_NOT_FOUND,
203 	IOMMU_INIT_ERROR,
204 	IOMMU_CMDLINE_DISABLED,
205 };
206 
207 /* Early ioapic and hpet maps from kernel command line */
208 #define EARLY_MAP_SIZE		4
209 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
210 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
211 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
212 
213 static int __initdata early_ioapic_map_size;
214 static int __initdata early_hpet_map_size;
215 static int __initdata early_acpihid_map_size;
216 
217 static bool __initdata cmdline_maps;
218 
219 static enum iommu_init_state init_state = IOMMU_START_STATE;
220 
221 static int amd_iommu_enable_interrupts(void);
222 static int __init iommu_go_to_state(enum iommu_init_state state);
223 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
224 
225 static bool amd_iommu_pre_enabled = true;
226 
227 static u32 amd_iommu_ivinfo __initdata;
228 
229 bool translation_pre_enabled(struct amd_iommu *iommu)
230 {
231 	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
232 }
233 
234 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
235 {
236 	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
237 }
238 
239 static void init_translation_status(struct amd_iommu *iommu)
240 {
241 	u64 ctrl;
242 
243 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
244 	if (ctrl & (1<<CONTROL_IOMMU_EN))
245 		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
246 }
247 
248 static inline unsigned long tbl_size(int entry_size, int last_bdf)
249 {
250 	unsigned shift = PAGE_SHIFT +
251 			 get_order((last_bdf + 1) * entry_size);
252 
253 	return 1UL << shift;
254 }
255 
256 int amd_iommu_get_num_iommus(void)
257 {
258 	return amd_iommus_present;
259 }
260 
261 /*
262  * Iterate through all the IOMMUs to get common EFR
263  * masks among all IOMMUs and warn if found inconsistency.
264  */
265 static __init void get_global_efr(void)
266 {
267 	struct amd_iommu *iommu;
268 
269 	for_each_iommu(iommu) {
270 		u64 tmp = iommu->features;
271 		u64 tmp2 = iommu->features2;
272 
273 		if (list_is_first(&iommu->list, &amd_iommu_list)) {
274 			amd_iommu_efr = tmp;
275 			amd_iommu_efr2 = tmp2;
276 			continue;
277 		}
278 
279 		if (amd_iommu_efr == tmp &&
280 		    amd_iommu_efr2 == tmp2)
281 			continue;
282 
283 		pr_err(FW_BUG
284 		       "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n",
285 		       tmp, tmp2, amd_iommu_efr, amd_iommu_efr2,
286 		       iommu->index, iommu->pci_seg->id,
287 		       PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid),
288 		       PCI_FUNC(iommu->devid));
289 
290 		amd_iommu_efr &= tmp;
291 		amd_iommu_efr2 &= tmp2;
292 	}
293 
294 	pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2);
295 }
296 
297 /*
298  * For IVHD type 0x11/0x40, EFR is also available via IVHD.
299  * Default to IVHD EFR since it is available sooner
300  * (i.e. before PCI init).
301  */
302 static void __init early_iommu_features_init(struct amd_iommu *iommu,
303 					     struct ivhd_header *h)
304 {
305 	if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) {
306 		iommu->features = h->efr_reg;
307 		iommu->features2 = h->efr_reg2;
308 	}
309 	if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
310 		amdr_ivrs_remap_support = true;
311 }
312 
313 /* Access to l1 and l2 indexed register spaces */
314 
315 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
316 {
317 	u32 val;
318 
319 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
320 	pci_read_config_dword(iommu->dev, 0xfc, &val);
321 	return val;
322 }
323 
324 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
325 {
326 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
327 	pci_write_config_dword(iommu->dev, 0xfc, val);
328 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
329 }
330 
331 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
332 {
333 	u32 val;
334 
335 	pci_write_config_dword(iommu->dev, 0xf0, address);
336 	pci_read_config_dword(iommu->dev, 0xf4, &val);
337 	return val;
338 }
339 
340 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
341 {
342 	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
343 	pci_write_config_dword(iommu->dev, 0xf4, val);
344 }
345 
346 /****************************************************************************
347  *
348  * AMD IOMMU MMIO register space handling functions
349  *
350  * These functions are used to program the IOMMU device registers in
351  * MMIO space required for that driver.
352  *
353  ****************************************************************************/
354 
355 /*
356  * This function set the exclusion range in the IOMMU. DMA accesses to the
357  * exclusion range are passed through untranslated
358  */
359 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
360 {
361 	u64 start = iommu->exclusion_start & PAGE_MASK;
362 	u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
363 	u64 entry;
364 
365 	if (!iommu->exclusion_start)
366 		return;
367 
368 	entry = start | MMIO_EXCL_ENABLE_MASK;
369 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
370 			&entry, sizeof(entry));
371 
372 	entry = limit;
373 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
374 			&entry, sizeof(entry));
375 }
376 
377 static void iommu_set_cwwb_range(struct amd_iommu *iommu)
378 {
379 	u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
380 	u64 entry = start & PM_ADDR_MASK;
381 
382 	if (!check_feature(FEATURE_SNP))
383 		return;
384 
385 	/* Note:
386 	 * Re-purpose Exclusion base/limit registers for Completion wait
387 	 * write-back base/limit.
388 	 */
389 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
390 		    &entry, sizeof(entry));
391 
392 	/* Note:
393 	 * Default to 4 Kbytes, which can be specified by setting base
394 	 * address equal to the limit address.
395 	 */
396 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
397 		    &entry, sizeof(entry));
398 }
399 
400 /* Programs the physical address of the device table into the IOMMU hardware */
401 static void iommu_set_device_table(struct amd_iommu *iommu)
402 {
403 	u64 entry;
404 	u32 dev_table_size = iommu->pci_seg->dev_table_size;
405 	void *dev_table = (void *)get_dev_table(iommu);
406 
407 	BUG_ON(iommu->mmio_base == NULL);
408 
409 	entry = iommu_virt_to_phys(dev_table);
410 	entry |= (dev_table_size >> 12) - 1;
411 	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
412 			&entry, sizeof(entry));
413 }
414 
415 /* Generic functions to enable/disable certain features of the IOMMU. */
416 void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
417 {
418 	u64 ctrl;
419 
420 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
421 	ctrl |= (1ULL << bit);
422 	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
423 }
424 
425 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
426 {
427 	u64 ctrl;
428 
429 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
430 	ctrl &= ~(1ULL << bit);
431 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
432 }
433 
434 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
435 {
436 	u64 ctrl;
437 
438 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
439 	ctrl &= ~CTRL_INV_TO_MASK;
440 	ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
441 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
442 }
443 
444 /* Function to enable the hardware */
445 static void iommu_enable(struct amd_iommu *iommu)
446 {
447 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
448 }
449 
450 static void iommu_disable(struct amd_iommu *iommu)
451 {
452 	if (!iommu->mmio_base)
453 		return;
454 
455 	/* Disable command buffer */
456 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
457 
458 	/* Disable event logging and event interrupts */
459 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
460 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
461 
462 	/* Disable IOMMU GA_LOG */
463 	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
464 	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
465 
466 	/* Disable IOMMU PPR logging */
467 	iommu_feature_disable(iommu, CONTROL_PPRLOG_EN);
468 	iommu_feature_disable(iommu, CONTROL_PPRINT_EN);
469 
470 	/* Disable IOMMU hardware itself */
471 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
472 
473 	/* Clear IRTE cache disabling bit */
474 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
475 }
476 
477 /*
478  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
479  * the system has one.
480  */
481 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
482 {
483 	if (!request_mem_region(address, end, "amd_iommu")) {
484 		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
485 			address, end);
486 		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
487 		return NULL;
488 	}
489 
490 	return (u8 __iomem *)ioremap(address, end);
491 }
492 
493 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
494 {
495 	if (iommu->mmio_base)
496 		iounmap(iommu->mmio_base);
497 	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
498 }
499 
500 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
501 {
502 	u32 size = 0;
503 
504 	switch (h->type) {
505 	case 0x10:
506 		size = 24;
507 		break;
508 	case 0x11:
509 	case 0x40:
510 		size = 40;
511 		break;
512 	}
513 	return size;
514 }
515 
516 /****************************************************************************
517  *
518  * The functions below belong to the first pass of AMD IOMMU ACPI table
519  * parsing. In this pass we try to find out the highest device id this
520  * code has to handle. Upon this information the size of the shared data
521  * structures is determined later.
522  *
523  ****************************************************************************/
524 
525 /*
526  * This function calculates the length of a given IVHD entry
527  */
528 static inline int ivhd_entry_length(u8 *ivhd)
529 {
530 	u32 type = ((struct ivhd_entry *)ivhd)->type;
531 
532 	if (type < 0x80) {
533 		return 0x04 << (*ivhd >> 6);
534 	} else if (type == IVHD_DEV_ACPI_HID) {
535 		/* For ACPI_HID, offset 21 is uid len */
536 		return *((u8 *)ivhd + 21) + 22;
537 	}
538 	return 0;
539 }
540 
541 /*
542  * After reading the highest device id from the IOMMU PCI capability header
543  * this function looks if there is a higher device id defined in the ACPI table
544  */
545 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
546 {
547 	u8 *p = (void *)h, *end = (void *)h;
548 	struct ivhd_entry *dev;
549 	int last_devid = -EINVAL;
550 
551 	u32 ivhd_size = get_ivhd_header_size(h);
552 
553 	if (!ivhd_size) {
554 		pr_err("Unsupported IVHD type %#x\n", h->type);
555 		return -EINVAL;
556 	}
557 
558 	p += ivhd_size;
559 	end += h->length;
560 
561 	while (p < end) {
562 		dev = (struct ivhd_entry *)p;
563 		switch (dev->type) {
564 		case IVHD_DEV_ALL:
565 			/* Use maximum BDF value for DEV_ALL */
566 			return 0xffff;
567 		case IVHD_DEV_SELECT:
568 		case IVHD_DEV_RANGE_END:
569 		case IVHD_DEV_ALIAS:
570 		case IVHD_DEV_EXT_SELECT:
571 			/* all the above subfield types refer to device ids */
572 			if (dev->devid > last_devid)
573 				last_devid = dev->devid;
574 			break;
575 		default:
576 			break;
577 		}
578 		p += ivhd_entry_length(p);
579 	}
580 
581 	WARN_ON(p != end);
582 
583 	return last_devid;
584 }
585 
586 static int __init check_ivrs_checksum(struct acpi_table_header *table)
587 {
588 	int i;
589 	u8 checksum = 0, *p = (u8 *)table;
590 
591 	for (i = 0; i < table->length; ++i)
592 		checksum += p[i];
593 	if (checksum != 0) {
594 		/* ACPI table corrupt */
595 		pr_err(FW_BUG "IVRS invalid checksum\n");
596 		return -ENODEV;
597 	}
598 
599 	return 0;
600 }
601 
602 /*
603  * Iterate over all IVHD entries in the ACPI table and find the highest device
604  * id which we need to handle. This is the first of three functions which parse
605  * the ACPI table. So we check the checksum here.
606  */
607 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
608 {
609 	u8 *p = (u8 *)table, *end = (u8 *)table;
610 	struct ivhd_header *h;
611 	int last_devid, last_bdf = 0;
612 
613 	p += IVRS_HEADER_LENGTH;
614 
615 	end += table->length;
616 	while (p < end) {
617 		h = (struct ivhd_header *)p;
618 		if (h->pci_seg == pci_seg &&
619 		    h->type == amd_iommu_target_ivhd_type) {
620 			last_devid = find_last_devid_from_ivhd(h);
621 
622 			if (last_devid < 0)
623 				return -EINVAL;
624 			if (last_devid > last_bdf)
625 				last_bdf = last_devid;
626 		}
627 		p += h->length;
628 	}
629 	WARN_ON(p != end);
630 
631 	return last_bdf;
632 }
633 
634 /****************************************************************************
635  *
636  * The following functions belong to the code path which parses the ACPI table
637  * the second time. In this ACPI parsing iteration we allocate IOMMU specific
638  * data structures, initialize the per PCI segment device/alias/rlookup table
639  * and also basically initialize the hardware.
640  *
641  ****************************************************************************/
642 
643 /* Allocate per PCI segment device table */
644 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
645 {
646 	pci_seg->dev_table = iommu_alloc_pages(GFP_KERNEL | GFP_DMA32,
647 					       get_order(pci_seg->dev_table_size));
648 	if (!pci_seg->dev_table)
649 		return -ENOMEM;
650 
651 	return 0;
652 }
653 
654 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
655 {
656 	iommu_free_pages(pci_seg->dev_table,
657 			 get_order(pci_seg->dev_table_size));
658 	pci_seg->dev_table = NULL;
659 }
660 
661 /* Allocate per PCI segment IOMMU rlookup table. */
662 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
663 {
664 	pci_seg->rlookup_table = iommu_alloc_pages(GFP_KERNEL,
665 						   get_order(pci_seg->rlookup_table_size));
666 	if (pci_seg->rlookup_table == NULL)
667 		return -ENOMEM;
668 
669 	return 0;
670 }
671 
672 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
673 {
674 	iommu_free_pages(pci_seg->rlookup_table,
675 			 get_order(pci_seg->rlookup_table_size));
676 	pci_seg->rlookup_table = NULL;
677 }
678 
679 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
680 {
681 	pci_seg->irq_lookup_table = iommu_alloc_pages(GFP_KERNEL,
682 						      get_order(pci_seg->rlookup_table_size));
683 	kmemleak_alloc(pci_seg->irq_lookup_table,
684 		       pci_seg->rlookup_table_size, 1, GFP_KERNEL);
685 	if (pci_seg->irq_lookup_table == NULL)
686 		return -ENOMEM;
687 
688 	return 0;
689 }
690 
691 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
692 {
693 	kmemleak_free(pci_seg->irq_lookup_table);
694 	iommu_free_pages(pci_seg->irq_lookup_table,
695 			 get_order(pci_seg->rlookup_table_size));
696 	pci_seg->irq_lookup_table = NULL;
697 }
698 
699 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
700 {
701 	int i;
702 
703 	pci_seg->alias_table = iommu_alloc_pages(GFP_KERNEL,
704 						 get_order(pci_seg->alias_table_size));
705 	if (!pci_seg->alias_table)
706 		return -ENOMEM;
707 
708 	/*
709 	 * let all alias entries point to itself
710 	 */
711 	for (i = 0; i <= pci_seg->last_bdf; ++i)
712 		pci_seg->alias_table[i] = i;
713 
714 	return 0;
715 }
716 
717 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
718 {
719 	iommu_free_pages(pci_seg->alias_table,
720 			 get_order(pci_seg->alias_table_size));
721 	pci_seg->alias_table = NULL;
722 }
723 
724 /*
725  * Allocates the command buffer. This buffer is per AMD IOMMU. We can
726  * write commands to that buffer later and the IOMMU will execute them
727  * asynchronously
728  */
729 static int __init alloc_command_buffer(struct amd_iommu *iommu)
730 {
731 	iommu->cmd_buf = iommu_alloc_pages(GFP_KERNEL,
732 					   get_order(CMD_BUFFER_SIZE));
733 
734 	return iommu->cmd_buf ? 0 : -ENOMEM;
735 }
736 
737 /*
738  * Interrupt handler has processed all pending events and adjusted head
739  * and tail pointer. Reset overflow mask and restart logging again.
740  */
741 void amd_iommu_restart_log(struct amd_iommu *iommu, const char *evt_type,
742 			   u8 cntrl_intr, u8 cntrl_log,
743 			   u32 status_run_mask, u32 status_overflow_mask)
744 {
745 	u32 status;
746 
747 	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
748 	if (status & status_run_mask)
749 		return;
750 
751 	pr_info_ratelimited("IOMMU %s log restarting\n", evt_type);
752 
753 	iommu_feature_disable(iommu, cntrl_log);
754 	iommu_feature_disable(iommu, cntrl_intr);
755 
756 	writel(status_overflow_mask, iommu->mmio_base + MMIO_STATUS_OFFSET);
757 
758 	iommu_feature_enable(iommu, cntrl_intr);
759 	iommu_feature_enable(iommu, cntrl_log);
760 }
761 
762 /*
763  * This function restarts event logging in case the IOMMU experienced
764  * an event log buffer overflow.
765  */
766 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
767 {
768 	amd_iommu_restart_log(iommu, "Event", CONTROL_EVT_INT_EN,
769 			      CONTROL_EVT_LOG_EN, MMIO_STATUS_EVT_RUN_MASK,
770 			      MMIO_STATUS_EVT_OVERFLOW_MASK);
771 }
772 
773 /*
774  * This function restarts event logging in case the IOMMU experienced
775  * GA log overflow.
776  */
777 void amd_iommu_restart_ga_log(struct amd_iommu *iommu)
778 {
779 	amd_iommu_restart_log(iommu, "GA", CONTROL_GAINT_EN,
780 			      CONTROL_GALOG_EN, MMIO_STATUS_GALOG_RUN_MASK,
781 			      MMIO_STATUS_GALOG_OVERFLOW_MASK);
782 }
783 
784 /*
785  * This function resets the command buffer if the IOMMU stopped fetching
786  * commands from it.
787  */
788 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
789 {
790 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
791 
792 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
793 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
794 	iommu->cmd_buf_head = 0;
795 	iommu->cmd_buf_tail = 0;
796 
797 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
798 }
799 
800 /*
801  * This function writes the command buffer address to the hardware and
802  * enables it.
803  */
804 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
805 {
806 	u64 entry;
807 
808 	BUG_ON(iommu->cmd_buf == NULL);
809 
810 	entry = iommu_virt_to_phys(iommu->cmd_buf);
811 	entry |= MMIO_CMD_SIZE_512;
812 
813 	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
814 		    &entry, sizeof(entry));
815 
816 	amd_iommu_reset_cmd_buffer(iommu);
817 }
818 
819 /*
820  * This function disables the command buffer
821  */
822 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
823 {
824 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
825 }
826 
827 static void __init free_command_buffer(struct amd_iommu *iommu)
828 {
829 	iommu_free_pages(iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
830 }
831 
832 void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
833 				  size_t size)
834 {
835 	int order = get_order(size);
836 	void *buf = iommu_alloc_pages(gfp, order);
837 
838 	if (buf &&
839 	    check_feature(FEATURE_SNP) &&
840 	    set_memory_4k((unsigned long)buf, (1 << order))) {
841 		iommu_free_pages(buf, order);
842 		buf = 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(struct amd_iommu *iommu)
850 {
851 	iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL,
852 					      EVT_BUFFER_SIZE);
853 
854 	return iommu->evt_buf ? 0 : -ENOMEM;
855 }
856 
857 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
858 {
859 	u64 entry;
860 
861 	BUG_ON(iommu->evt_buf == NULL);
862 
863 	entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
864 
865 	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
866 		    &entry, sizeof(entry));
867 
868 	/* set head and tail to zero manually */
869 	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
870 	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
871 
872 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
873 }
874 
875 /*
876  * This function disables the event log buffer
877  */
878 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
879 {
880 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
881 }
882 
883 static void __init free_event_buffer(struct amd_iommu *iommu)
884 {
885 	iommu_free_pages(iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
886 }
887 
888 static void free_ga_log(struct amd_iommu *iommu)
889 {
890 #ifdef CONFIG_IRQ_REMAP
891 	iommu_free_pages(iommu->ga_log, get_order(GA_LOG_SIZE));
892 	iommu_free_pages(iommu->ga_log_tail, get_order(8));
893 #endif
894 }
895 
896 #ifdef CONFIG_IRQ_REMAP
897 static int iommu_ga_log_enable(struct amd_iommu *iommu)
898 {
899 	u32 status, i;
900 	u64 entry;
901 
902 	if (!iommu->ga_log)
903 		return -EINVAL;
904 
905 	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
906 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
907 		    &entry, sizeof(entry));
908 	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
909 		 (BIT_ULL(52)-1)) & ~7ULL;
910 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
911 		    &entry, sizeof(entry));
912 	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
913 	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
914 
915 
916 	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
917 	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
918 
919 	for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
920 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
921 		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
922 			break;
923 		udelay(10);
924 	}
925 
926 	if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
927 		return -EINVAL;
928 
929 	return 0;
930 }
931 
932 static int iommu_init_ga_log(struct amd_iommu *iommu)
933 {
934 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
935 		return 0;
936 
937 	iommu->ga_log = iommu_alloc_pages(GFP_KERNEL, get_order(GA_LOG_SIZE));
938 	if (!iommu->ga_log)
939 		goto err_out;
940 
941 	iommu->ga_log_tail = iommu_alloc_pages(GFP_KERNEL, get_order(8));
942 	if (!iommu->ga_log_tail)
943 		goto err_out;
944 
945 	return 0;
946 err_out:
947 	free_ga_log(iommu);
948 	return -EINVAL;
949 }
950 #endif /* CONFIG_IRQ_REMAP */
951 
952 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
953 {
954 	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL, 1);
955 
956 	return iommu->cmd_sem ? 0 : -ENOMEM;
957 }
958 
959 static void __init free_cwwb_sem(struct amd_iommu *iommu)
960 {
961 	if (iommu->cmd_sem)
962 		iommu_free_page((void *)iommu->cmd_sem);
963 }
964 
965 static void iommu_enable_xt(struct amd_iommu *iommu)
966 {
967 #ifdef CONFIG_IRQ_REMAP
968 	/*
969 	 * XT mode (32-bit APIC destination ID) requires
970 	 * GA mode (128-bit IRTE support) as a prerequisite.
971 	 */
972 	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
973 	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
974 		iommu_feature_enable(iommu, CONTROL_XT_EN);
975 #endif /* CONFIG_IRQ_REMAP */
976 }
977 
978 static void iommu_enable_gt(struct amd_iommu *iommu)
979 {
980 	if (!check_feature(FEATURE_GT))
981 		return;
982 
983 	iommu_feature_enable(iommu, CONTROL_GT_EN);
984 }
985 
986 /* sets a specific bit in the device table entry. */
987 static void __set_dev_entry_bit(struct dev_table_entry *dev_table,
988 				u16 devid, u8 bit)
989 {
990 	int i = (bit >> 6) & 0x03;
991 	int _bit = bit & 0x3f;
992 
993 	dev_table[devid].data[i] |= (1UL << _bit);
994 }
995 
996 static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
997 {
998 	struct dev_table_entry *dev_table = get_dev_table(iommu);
999 
1000 	return __set_dev_entry_bit(dev_table, devid, bit);
1001 }
1002 
1003 static int __get_dev_entry_bit(struct dev_table_entry *dev_table,
1004 			       u16 devid, u8 bit)
1005 {
1006 	int i = (bit >> 6) & 0x03;
1007 	int _bit = bit & 0x3f;
1008 
1009 	return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1010 }
1011 
1012 static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1013 {
1014 	struct dev_table_entry *dev_table = get_dev_table(iommu);
1015 
1016 	return __get_dev_entry_bit(dev_table, devid, bit);
1017 }
1018 
1019 static bool __copy_device_table(struct amd_iommu *iommu)
1020 {
1021 	u64 int_ctl, int_tab_len, entry = 0;
1022 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1023 	struct dev_table_entry *old_devtb = NULL;
1024 	u32 lo, hi, devid, old_devtb_size;
1025 	phys_addr_t old_devtb_phys;
1026 	u16 dom_id, dte_v, irq_v;
1027 	u64 tmp;
1028 
1029 	/* Each IOMMU use separate device table with the same size */
1030 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1031 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1032 	entry = (((u64) hi) << 32) + lo;
1033 
1034 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1035 	if (old_devtb_size != pci_seg->dev_table_size) {
1036 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1037 			iommu->index);
1038 		return false;
1039 	}
1040 
1041 	/*
1042 	 * When SME is enabled in the first kernel, the entry includes the
1043 	 * memory encryption mask(sme_me_mask), we must remove the memory
1044 	 * encryption mask to obtain the true physical address in kdump kernel.
1045 	 */
1046 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1047 
1048 	if (old_devtb_phys >= 0x100000000ULL) {
1049 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1050 		return false;
1051 	}
1052 	old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1053 		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
1054 							pci_seg->dev_table_size)
1055 		    : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB);
1056 
1057 	if (!old_devtb)
1058 		return false;
1059 
1060 	pci_seg->old_dev_tbl_cpy = iommu_alloc_pages(GFP_KERNEL | GFP_DMA32,
1061 						     get_order(pci_seg->dev_table_size));
1062 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1063 		pr_err("Failed to allocate memory for copying old device table!\n");
1064 		memunmap(old_devtb);
1065 		return false;
1066 	}
1067 
1068 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
1069 		pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1070 		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1071 		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1072 
1073 		if (dte_v && dom_id) {
1074 			pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1075 			pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1076 			/* Reserve the Domain IDs used by previous kernel */
1077 			if (ida_alloc_range(&pdom_ids, dom_id, dom_id, GFP_ATOMIC) != dom_id) {
1078 				pr_err("Failed to reserve domain ID 0x%x\n", dom_id);
1079 				memunmap(old_devtb);
1080 				return false;
1081 			}
1082 			/* If gcr3 table existed, mask it out */
1083 			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1084 				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1085 				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1086 				pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1087 				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1088 				tmp |= DTE_FLAG_GV;
1089 				pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1090 			}
1091 		}
1092 
1093 		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1094 		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1095 		int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1096 		if (irq_v && (int_ctl || int_tab_len)) {
1097 			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1098 			    (int_tab_len != DTE_INTTABLEN)) {
1099 				pr_err("Wrong old irq remapping flag: %#x\n", devid);
1100 				memunmap(old_devtb);
1101 				return false;
1102 			}
1103 
1104 			pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1105 		}
1106 	}
1107 	memunmap(old_devtb);
1108 
1109 	return true;
1110 }
1111 
1112 static bool copy_device_table(void)
1113 {
1114 	struct amd_iommu *iommu;
1115 	struct amd_iommu_pci_seg *pci_seg;
1116 
1117 	if (!amd_iommu_pre_enabled)
1118 		return false;
1119 
1120 	pr_warn("Translation is already enabled - trying to copy translation structures\n");
1121 
1122 	/*
1123 	 * All IOMMUs within PCI segment shares common device table.
1124 	 * Hence copy device table only once per PCI segment.
1125 	 */
1126 	for_each_pci_segment(pci_seg) {
1127 		for_each_iommu(iommu) {
1128 			if (pci_seg->id != iommu->pci_seg->id)
1129 				continue;
1130 			if (!__copy_device_table(iommu))
1131 				return false;
1132 			break;
1133 		}
1134 	}
1135 
1136 	return true;
1137 }
1138 
1139 void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid)
1140 {
1141 	int sysmgt;
1142 
1143 	sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) |
1144 		 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1);
1145 
1146 	if (sysmgt == 0x01)
1147 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW);
1148 }
1149 
1150 /*
1151  * This function takes the device specific flags read from the ACPI
1152  * table and sets up the device table entry with that information
1153  */
1154 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1155 					   u16 devid, u32 flags, u32 ext_flags)
1156 {
1157 	if (flags & ACPI_DEVFLAG_INITPASS)
1158 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS);
1159 	if (flags & ACPI_DEVFLAG_EXTINT)
1160 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS);
1161 	if (flags & ACPI_DEVFLAG_NMI)
1162 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS);
1163 	if (flags & ACPI_DEVFLAG_SYSMGT1)
1164 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1);
1165 	if (flags & ACPI_DEVFLAG_SYSMGT2)
1166 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2);
1167 	if (flags & ACPI_DEVFLAG_LINT0)
1168 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS);
1169 	if (flags & ACPI_DEVFLAG_LINT1)
1170 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS);
1171 
1172 	amd_iommu_apply_erratum_63(iommu, devid);
1173 
1174 	amd_iommu_set_rlookup_table(iommu, devid);
1175 }
1176 
1177 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line)
1178 {
1179 	struct devid_map *entry;
1180 	struct list_head *list;
1181 
1182 	if (type == IVHD_SPECIAL_IOAPIC)
1183 		list = &ioapic_map;
1184 	else if (type == IVHD_SPECIAL_HPET)
1185 		list = &hpet_map;
1186 	else
1187 		return -EINVAL;
1188 
1189 	list_for_each_entry(entry, list, list) {
1190 		if (!(entry->id == id && entry->cmd_line))
1191 			continue;
1192 
1193 		pr_info("Command-line override present for %s id %d - ignoring\n",
1194 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1195 
1196 		*devid = entry->devid;
1197 
1198 		return 0;
1199 	}
1200 
1201 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1202 	if (!entry)
1203 		return -ENOMEM;
1204 
1205 	entry->id	= id;
1206 	entry->devid	= *devid;
1207 	entry->cmd_line	= cmd_line;
1208 
1209 	list_add_tail(&entry->list, list);
1210 
1211 	return 0;
1212 }
1213 
1214 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid,
1215 				      bool cmd_line)
1216 {
1217 	struct acpihid_map_entry *entry;
1218 	struct list_head *list = &acpihid_map;
1219 
1220 	list_for_each_entry(entry, list, list) {
1221 		if (strcmp(entry->hid, hid) ||
1222 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1223 		    !entry->cmd_line)
1224 			continue;
1225 
1226 		pr_info("Command-line override for hid:%s uid:%s\n",
1227 			hid, uid);
1228 		*devid = entry->devid;
1229 		return 0;
1230 	}
1231 
1232 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1233 	if (!entry)
1234 		return -ENOMEM;
1235 
1236 	memcpy(entry->uid, uid, strlen(uid));
1237 	memcpy(entry->hid, hid, strlen(hid));
1238 	entry->devid = *devid;
1239 	entry->cmd_line	= cmd_line;
1240 	entry->root_devid = (entry->devid & (~0x7));
1241 
1242 	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1243 		entry->cmd_line ? "cmd" : "ivrs",
1244 		entry->hid, entry->uid, entry->root_devid);
1245 
1246 	list_add_tail(&entry->list, list);
1247 	return 0;
1248 }
1249 
1250 static int __init add_early_maps(void)
1251 {
1252 	int i, ret;
1253 
1254 	for (i = 0; i < early_ioapic_map_size; ++i) {
1255 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1256 					 early_ioapic_map[i].id,
1257 					 &early_ioapic_map[i].devid,
1258 					 early_ioapic_map[i].cmd_line);
1259 		if (ret)
1260 			return ret;
1261 	}
1262 
1263 	for (i = 0; i < early_hpet_map_size; ++i) {
1264 		ret = add_special_device(IVHD_SPECIAL_HPET,
1265 					 early_hpet_map[i].id,
1266 					 &early_hpet_map[i].devid,
1267 					 early_hpet_map[i].cmd_line);
1268 		if (ret)
1269 			return ret;
1270 	}
1271 
1272 	for (i = 0; i < early_acpihid_map_size; ++i) {
1273 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1274 					  early_acpihid_map[i].uid,
1275 					  &early_acpihid_map[i].devid,
1276 					  early_acpihid_map[i].cmd_line);
1277 		if (ret)
1278 			return ret;
1279 	}
1280 
1281 	return 0;
1282 }
1283 
1284 /*
1285  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1286  * initializes the hardware and our data structures with it.
1287  */
1288 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1289 					struct ivhd_header *h)
1290 {
1291 	u8 *p = (u8 *)h;
1292 	u8 *end = p, flags = 0;
1293 	u16 devid = 0, devid_start = 0, devid_to = 0, seg_id;
1294 	u32 dev_i, ext_flags = 0;
1295 	bool alias = false;
1296 	struct ivhd_entry *e;
1297 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1298 	u32 ivhd_size;
1299 	int ret;
1300 
1301 
1302 	ret = add_early_maps();
1303 	if (ret)
1304 		return ret;
1305 
1306 	amd_iommu_apply_ivrs_quirks();
1307 
1308 	/*
1309 	 * First save the recommended feature enable bits from ACPI
1310 	 */
1311 	iommu->acpi_flags = h->flags;
1312 
1313 	/*
1314 	 * Done. Now parse the device entries
1315 	 */
1316 	ivhd_size = get_ivhd_header_size(h);
1317 	if (!ivhd_size) {
1318 		pr_err("Unsupported IVHD type %#x\n", h->type);
1319 		return -EINVAL;
1320 	}
1321 
1322 	p += ivhd_size;
1323 
1324 	end += h->length;
1325 
1326 
1327 	while (p < end) {
1328 		e = (struct ivhd_entry *)p;
1329 		seg_id = pci_seg->id;
1330 
1331 		switch (e->type) {
1332 		case IVHD_DEV_ALL:
1333 
1334 			DUMP_printk("  DEV_ALL\t\t\tflags: %02x\n", e->flags);
1335 
1336 			for (dev_i = 0; dev_i <= pci_seg->last_bdf; ++dev_i)
1337 				set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1338 			break;
1339 		case IVHD_DEV_SELECT:
1340 
1341 			DUMP_printk("  DEV_SELECT\t\t\t devid: %04x:%02x:%02x.%x "
1342 				    "flags: %02x\n",
1343 				    seg_id, PCI_BUS_NUM(e->devid),
1344 				    PCI_SLOT(e->devid),
1345 				    PCI_FUNC(e->devid),
1346 				    e->flags);
1347 
1348 			devid = e->devid;
1349 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1350 			break;
1351 		case IVHD_DEV_SELECT_RANGE_START:
1352 
1353 			DUMP_printk("  DEV_SELECT_RANGE_START\t "
1354 				    "devid: %04x:%02x:%02x.%x flags: %02x\n",
1355 				    seg_id, PCI_BUS_NUM(e->devid),
1356 				    PCI_SLOT(e->devid),
1357 				    PCI_FUNC(e->devid),
1358 				    e->flags);
1359 
1360 			devid_start = e->devid;
1361 			flags = e->flags;
1362 			ext_flags = 0;
1363 			alias = false;
1364 			break;
1365 		case IVHD_DEV_ALIAS:
1366 
1367 			DUMP_printk("  DEV_ALIAS\t\t\t devid: %04x:%02x:%02x.%x "
1368 				    "flags: %02x devid_to: %02x:%02x.%x\n",
1369 				    seg_id, PCI_BUS_NUM(e->devid),
1370 				    PCI_SLOT(e->devid),
1371 				    PCI_FUNC(e->devid),
1372 				    e->flags,
1373 				    PCI_BUS_NUM(e->ext >> 8),
1374 				    PCI_SLOT(e->ext >> 8),
1375 				    PCI_FUNC(e->ext >> 8));
1376 
1377 			devid = e->devid;
1378 			devid_to = e->ext >> 8;
1379 			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1380 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1381 			pci_seg->alias_table[devid] = devid_to;
1382 			break;
1383 		case IVHD_DEV_ALIAS_RANGE:
1384 
1385 			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
1386 				    "devid: %04x:%02x:%02x.%x flags: %02x "
1387 				    "devid_to: %04x:%02x:%02x.%x\n",
1388 				    seg_id, PCI_BUS_NUM(e->devid),
1389 				    PCI_SLOT(e->devid),
1390 				    PCI_FUNC(e->devid),
1391 				    e->flags,
1392 				    seg_id, PCI_BUS_NUM(e->ext >> 8),
1393 				    PCI_SLOT(e->ext >> 8),
1394 				    PCI_FUNC(e->ext >> 8));
1395 
1396 			devid_start = e->devid;
1397 			flags = e->flags;
1398 			devid_to = e->ext >> 8;
1399 			ext_flags = 0;
1400 			alias = true;
1401 			break;
1402 		case IVHD_DEV_EXT_SELECT:
1403 
1404 			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %04x:%02x:%02x.%x "
1405 				    "flags: %02x ext: %08x\n",
1406 				    seg_id, PCI_BUS_NUM(e->devid),
1407 				    PCI_SLOT(e->devid),
1408 				    PCI_FUNC(e->devid),
1409 				    e->flags, e->ext);
1410 
1411 			devid = e->devid;
1412 			set_dev_entry_from_acpi(iommu, devid, e->flags,
1413 						e->ext);
1414 			break;
1415 		case IVHD_DEV_EXT_SELECT_RANGE:
1416 
1417 			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
1418 				    "%04x:%02x:%02x.%x flags: %02x ext: %08x\n",
1419 				    seg_id, PCI_BUS_NUM(e->devid),
1420 				    PCI_SLOT(e->devid),
1421 				    PCI_FUNC(e->devid),
1422 				    e->flags, e->ext);
1423 
1424 			devid_start = e->devid;
1425 			flags = e->flags;
1426 			ext_flags = e->ext;
1427 			alias = false;
1428 			break;
1429 		case IVHD_DEV_RANGE_END:
1430 
1431 			DUMP_printk("  DEV_RANGE_END\t\t devid: %04x:%02x:%02x.%x\n",
1432 				    seg_id, PCI_BUS_NUM(e->devid),
1433 				    PCI_SLOT(e->devid),
1434 				    PCI_FUNC(e->devid));
1435 
1436 			devid = e->devid;
1437 			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1438 				if (alias) {
1439 					pci_seg->alias_table[dev_i] = devid_to;
1440 					set_dev_entry_from_acpi(iommu,
1441 						devid_to, flags, ext_flags);
1442 				}
1443 				set_dev_entry_from_acpi(iommu, dev_i,
1444 							flags, ext_flags);
1445 			}
1446 			break;
1447 		case IVHD_DEV_SPECIAL: {
1448 			u8 handle, type;
1449 			const char *var;
1450 			u32 devid;
1451 			int ret;
1452 
1453 			handle = e->ext & 0xff;
1454 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8));
1455 			type   = (e->ext >> 24) & 0xff;
1456 
1457 			if (type == IVHD_SPECIAL_IOAPIC)
1458 				var = "IOAPIC";
1459 			else if (type == IVHD_SPECIAL_HPET)
1460 				var = "HPET";
1461 			else
1462 				var = "UNKNOWN";
1463 
1464 			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x\n",
1465 				    var, (int)handle,
1466 				    seg_id, PCI_BUS_NUM(devid),
1467 				    PCI_SLOT(devid),
1468 				    PCI_FUNC(devid));
1469 
1470 			ret = add_special_device(type, handle, &devid, false);
1471 			if (ret)
1472 				return ret;
1473 
1474 			/*
1475 			 * add_special_device might update the devid in case a
1476 			 * command-line override is present. So call
1477 			 * set_dev_entry_from_acpi after add_special_device.
1478 			 */
1479 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1480 
1481 			break;
1482 		}
1483 		case IVHD_DEV_ACPI_HID: {
1484 			u32 devid;
1485 			u8 hid[ACPIHID_HID_LEN];
1486 			u8 uid[ACPIHID_UID_LEN];
1487 			int ret;
1488 
1489 			if (h->type != 0x40) {
1490 				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1491 				       e->type);
1492 				break;
1493 			}
1494 
1495 			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1496 			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1497 			hid[ACPIHID_HID_LEN - 1] = '\0';
1498 
1499 			if (!(*hid)) {
1500 				pr_err(FW_BUG "Invalid HID.\n");
1501 				break;
1502 			}
1503 
1504 			uid[0] = '\0';
1505 			switch (e->uidf) {
1506 			case UID_NOT_PRESENT:
1507 
1508 				if (e->uidl != 0)
1509 					pr_warn(FW_BUG "Invalid UID length.\n");
1510 
1511 				break;
1512 			case UID_IS_INTEGER:
1513 
1514 				sprintf(uid, "%d", e->uid);
1515 
1516 				break;
1517 			case UID_IS_CHARACTER:
1518 
1519 				memcpy(uid, &e->uid, e->uidl);
1520 				uid[e->uidl] = '\0';
1521 
1522 				break;
1523 			default:
1524 				break;
1525 			}
1526 
1527 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid);
1528 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x\n",
1529 				    hid, uid, seg_id,
1530 				    PCI_BUS_NUM(devid),
1531 				    PCI_SLOT(devid),
1532 				    PCI_FUNC(devid));
1533 
1534 			flags = e->flags;
1535 
1536 			ret = add_acpi_hid_device(hid, uid, &devid, false);
1537 			if (ret)
1538 				return ret;
1539 
1540 			/*
1541 			 * add_special_device might update the devid in case a
1542 			 * command-line override is present. So call
1543 			 * set_dev_entry_from_acpi after add_special_device.
1544 			 */
1545 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1546 
1547 			break;
1548 		}
1549 		default:
1550 			break;
1551 		}
1552 
1553 		p += ivhd_entry_length(p);
1554 	}
1555 
1556 	return 0;
1557 }
1558 
1559 /* Allocate PCI segment data structure */
1560 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1561 					  struct acpi_table_header *ivrs_base)
1562 {
1563 	struct amd_iommu_pci_seg *pci_seg;
1564 	int last_bdf;
1565 
1566 	/*
1567 	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1568 	 * handle in this PCI segment. Upon this information the shared data
1569 	 * structures for the PCI segments in the system will be allocated.
1570 	 */
1571 	last_bdf = find_last_devid_acpi(ivrs_base, id);
1572 	if (last_bdf < 0)
1573 		return NULL;
1574 
1575 	pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1576 	if (pci_seg == NULL)
1577 		return NULL;
1578 
1579 	pci_seg->last_bdf = last_bdf;
1580 	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1581 	pci_seg->dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE, last_bdf);
1582 	pci_seg->alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE, last_bdf);
1583 	pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE, last_bdf);
1584 
1585 	pci_seg->id = id;
1586 	init_llist_head(&pci_seg->dev_data_list);
1587 	INIT_LIST_HEAD(&pci_seg->unity_map);
1588 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1589 
1590 	if (alloc_dev_table(pci_seg))
1591 		return NULL;
1592 	if (alloc_alias_table(pci_seg))
1593 		return NULL;
1594 	if (alloc_rlookup_table(pci_seg))
1595 		return NULL;
1596 
1597 	return pci_seg;
1598 }
1599 
1600 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1601 					struct acpi_table_header *ivrs_base)
1602 {
1603 	struct amd_iommu_pci_seg *pci_seg;
1604 
1605 	for_each_pci_segment(pci_seg) {
1606 		if (pci_seg->id == id)
1607 			return pci_seg;
1608 	}
1609 
1610 	return alloc_pci_segment(id, ivrs_base);
1611 }
1612 
1613 static void __init free_pci_segments(void)
1614 {
1615 	struct amd_iommu_pci_seg *pci_seg, *next;
1616 
1617 	for_each_pci_segment_safe(pci_seg, next) {
1618 		list_del(&pci_seg->list);
1619 		free_irq_lookup_table(pci_seg);
1620 		free_rlookup_table(pci_seg);
1621 		free_alias_table(pci_seg);
1622 		free_dev_table(pci_seg);
1623 		kfree(pci_seg);
1624 	}
1625 }
1626 
1627 static void __init free_sysfs(struct amd_iommu *iommu)
1628 {
1629 	if (iommu->iommu.dev) {
1630 		iommu_device_unregister(&iommu->iommu);
1631 		iommu_device_sysfs_remove(&iommu->iommu);
1632 	}
1633 }
1634 
1635 static void __init free_iommu_one(struct amd_iommu *iommu)
1636 {
1637 	free_sysfs(iommu);
1638 	free_cwwb_sem(iommu);
1639 	free_command_buffer(iommu);
1640 	free_event_buffer(iommu);
1641 	amd_iommu_free_ppr_log(iommu);
1642 	free_ga_log(iommu);
1643 	iommu_unmap_mmio_space(iommu);
1644 	amd_iommu_iopf_uninit(iommu);
1645 }
1646 
1647 static void __init free_iommu_all(void)
1648 {
1649 	struct amd_iommu *iommu, *next;
1650 
1651 	for_each_iommu_safe(iommu, next) {
1652 		list_del(&iommu->list);
1653 		free_iommu_one(iommu);
1654 		kfree(iommu);
1655 	}
1656 }
1657 
1658 /*
1659  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1660  * Workaround:
1661  *     BIOS should disable L2B micellaneous clock gating by setting
1662  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1663  */
1664 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1665 {
1666 	u32 value;
1667 
1668 	if ((boot_cpu_data.x86 != 0x15) ||
1669 	    (boot_cpu_data.x86_model < 0x10) ||
1670 	    (boot_cpu_data.x86_model > 0x1f))
1671 		return;
1672 
1673 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1674 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1675 
1676 	if (value & BIT(2))
1677 		return;
1678 
1679 	/* Select NB indirect register 0x90 and enable writing */
1680 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1681 
1682 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1683 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1684 
1685 	/* Clear the enable writing bit */
1686 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1687 }
1688 
1689 /*
1690  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1691  * Workaround:
1692  *     BIOS should enable ATS write permission check by setting
1693  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1694  */
1695 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1696 {
1697 	u32 value;
1698 
1699 	if ((boot_cpu_data.x86 != 0x15) ||
1700 	    (boot_cpu_data.x86_model < 0x30) ||
1701 	    (boot_cpu_data.x86_model > 0x3f))
1702 		return;
1703 
1704 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1705 	value = iommu_read_l2(iommu, 0x47);
1706 
1707 	if (value & BIT(0))
1708 		return;
1709 
1710 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1711 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1712 
1713 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1714 }
1715 
1716 /*
1717  * This function glues the initialization function for one IOMMU
1718  * together and also allocates the command buffer and programs the
1719  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1720  */
1721 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1722 				 struct acpi_table_header *ivrs_base)
1723 {
1724 	struct amd_iommu_pci_seg *pci_seg;
1725 
1726 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1727 	if (pci_seg == NULL)
1728 		return -ENOMEM;
1729 	iommu->pci_seg = pci_seg;
1730 
1731 	raw_spin_lock_init(&iommu->lock);
1732 	atomic64_set(&iommu->cmd_sem_val, 0);
1733 
1734 	/* Add IOMMU to internal data structures */
1735 	list_add_tail(&iommu->list, &amd_iommu_list);
1736 	iommu->index = amd_iommus_present++;
1737 
1738 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1739 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1740 		return -ENOSYS;
1741 	}
1742 
1743 	/*
1744 	 * Copy data from ACPI table entry to the iommu struct
1745 	 */
1746 	iommu->devid   = h->devid;
1747 	iommu->cap_ptr = h->cap_ptr;
1748 	iommu->mmio_phys = h->mmio_phys;
1749 
1750 	switch (h->type) {
1751 	case 0x10:
1752 		/* Check if IVHD EFR contains proper max banks/counters */
1753 		if ((h->efr_attr != 0) &&
1754 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1755 		    ((h->efr_attr & (0x3F << 17)) != 0))
1756 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1757 		else
1758 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1759 
1760 		/*
1761 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1762 		 * GAM also requires GA mode. Therefore, we need to
1763 		 * check cmpxchg16b support before enabling it.
1764 		 */
1765 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1766 		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1767 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1768 		break;
1769 	case 0x11:
1770 	case 0x40:
1771 		if (h->efr_reg & (1 << 9))
1772 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1773 		else
1774 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1775 
1776 		/*
1777 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1778 		 * XT, GAM also requires GA mode. Therefore, we need to
1779 		 * check cmpxchg16b support before enabling them.
1780 		 */
1781 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1782 		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1783 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1784 			break;
1785 		}
1786 
1787 		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1788 			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1789 
1790 		early_iommu_features_init(iommu, h);
1791 
1792 		break;
1793 	default:
1794 		return -EINVAL;
1795 	}
1796 
1797 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1798 						iommu->mmio_phys_end);
1799 	if (!iommu->mmio_base)
1800 		return -ENOMEM;
1801 
1802 	return init_iommu_from_acpi(iommu, h);
1803 }
1804 
1805 static int __init init_iommu_one_late(struct amd_iommu *iommu)
1806 {
1807 	int ret;
1808 
1809 	if (alloc_cwwb_sem(iommu))
1810 		return -ENOMEM;
1811 
1812 	if (alloc_command_buffer(iommu))
1813 		return -ENOMEM;
1814 
1815 	if (alloc_event_buffer(iommu))
1816 		return -ENOMEM;
1817 
1818 	iommu->int_enabled = false;
1819 
1820 	init_translation_status(iommu);
1821 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1822 		iommu_disable(iommu);
1823 		clear_translation_pre_enabled(iommu);
1824 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1825 			iommu->index);
1826 	}
1827 	if (amd_iommu_pre_enabled)
1828 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1829 
1830 	if (amd_iommu_irq_remap) {
1831 		ret = amd_iommu_create_irq_domain(iommu);
1832 		if (ret)
1833 			return ret;
1834 	}
1835 
1836 	/*
1837 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1838 	 * table tells us so, but this is a lie!
1839 	 */
1840 	iommu->pci_seg->rlookup_table[iommu->devid] = NULL;
1841 
1842 	return 0;
1843 }
1844 
1845 /**
1846  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1847  * @ivrs: Pointer to the IVRS header
1848  *
1849  * This function search through all IVDB of the maximum supported IVHD
1850  */
1851 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1852 {
1853 	u8 *base = (u8 *)ivrs;
1854 	struct ivhd_header *ivhd = (struct ivhd_header *)
1855 					(base + IVRS_HEADER_LENGTH);
1856 	u8 last_type = ivhd->type;
1857 	u16 devid = ivhd->devid;
1858 
1859 	while (((u8 *)ivhd - base < ivrs->length) &&
1860 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1861 		u8 *p = (u8 *) ivhd;
1862 
1863 		if (ivhd->devid == devid)
1864 			last_type = ivhd->type;
1865 		ivhd = (struct ivhd_header *)(p + ivhd->length);
1866 	}
1867 
1868 	return last_type;
1869 }
1870 
1871 /*
1872  * Iterates over all IOMMU entries in the ACPI table, allocates the
1873  * IOMMU structure and initializes it with init_iommu_one()
1874  */
1875 static int __init init_iommu_all(struct acpi_table_header *table)
1876 {
1877 	u8 *p = (u8 *)table, *end = (u8 *)table;
1878 	struct ivhd_header *h;
1879 	struct amd_iommu *iommu;
1880 	int ret;
1881 
1882 	end += table->length;
1883 	p += IVRS_HEADER_LENGTH;
1884 
1885 	/* Phase 1: Process all IVHD blocks */
1886 	while (p < end) {
1887 		h = (struct ivhd_header *)p;
1888 		if (*p == amd_iommu_target_ivhd_type) {
1889 
1890 			DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x "
1891 				    "flags: %01x info %04x\n",
1892 				    h->pci_seg, PCI_BUS_NUM(h->devid),
1893 				    PCI_SLOT(h->devid), PCI_FUNC(h->devid),
1894 				    h->cap_ptr, h->flags, h->info);
1895 			DUMP_printk("       mmio-addr: %016llx\n",
1896 				    h->mmio_phys);
1897 
1898 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1899 			if (iommu == NULL)
1900 				return -ENOMEM;
1901 
1902 			ret = init_iommu_one(iommu, h, table);
1903 			if (ret)
1904 				return ret;
1905 		}
1906 		p += h->length;
1907 
1908 	}
1909 	WARN_ON(p != end);
1910 
1911 	/* Phase 2 : Early feature support check */
1912 	get_global_efr();
1913 
1914 	/* Phase 3 : Enabling IOMMU features */
1915 	for_each_iommu(iommu) {
1916 		ret = init_iommu_one_late(iommu);
1917 		if (ret)
1918 			return ret;
1919 	}
1920 
1921 	return 0;
1922 }
1923 
1924 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1925 {
1926 	u64 val;
1927 	struct pci_dev *pdev = iommu->dev;
1928 
1929 	if (!check_feature(FEATURE_PC))
1930 		return;
1931 
1932 	amd_iommu_pc_present = true;
1933 
1934 	pci_info(pdev, "IOMMU performance counters supported\n");
1935 
1936 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1937 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1938 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1939 
1940 	return;
1941 }
1942 
1943 static ssize_t amd_iommu_show_cap(struct device *dev,
1944 				  struct device_attribute *attr,
1945 				  char *buf)
1946 {
1947 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1948 	return sysfs_emit(buf, "%x\n", iommu->cap);
1949 }
1950 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1951 
1952 static ssize_t amd_iommu_show_features(struct device *dev,
1953 				       struct device_attribute *attr,
1954 				       char *buf)
1955 {
1956 	return sysfs_emit(buf, "%llx:%llx\n", amd_iommu_efr, amd_iommu_efr2);
1957 }
1958 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1959 
1960 static struct attribute *amd_iommu_attrs[] = {
1961 	&dev_attr_cap.attr,
1962 	&dev_attr_features.attr,
1963 	NULL,
1964 };
1965 
1966 static struct attribute_group amd_iommu_group = {
1967 	.name = "amd-iommu",
1968 	.attrs = amd_iommu_attrs,
1969 };
1970 
1971 static const struct attribute_group *amd_iommu_groups[] = {
1972 	&amd_iommu_group,
1973 	NULL,
1974 };
1975 
1976 /*
1977  * Note: IVHD 0x11 and 0x40 also contains exact copy
1978  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1979  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1980  */
1981 static void __init late_iommu_features_init(struct amd_iommu *iommu)
1982 {
1983 	u64 features, features2;
1984 
1985 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1986 		return;
1987 
1988 	/* read extended feature bits */
1989 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1990 	features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2);
1991 
1992 	if (!amd_iommu_efr) {
1993 		amd_iommu_efr = features;
1994 		amd_iommu_efr2 = features2;
1995 		return;
1996 	}
1997 
1998 	/*
1999 	 * Sanity check and warn if EFR values from
2000 	 * IVHD and MMIO conflict.
2001 	 */
2002 	if (features != amd_iommu_efr ||
2003 	    features2 != amd_iommu_efr2) {
2004 		pr_warn(FW_WARN
2005 			"EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n",
2006 			features, amd_iommu_efr,
2007 			features2, amd_iommu_efr2);
2008 	}
2009 }
2010 
2011 static int __init iommu_init_pci(struct amd_iommu *iommu)
2012 {
2013 	int cap_ptr = iommu->cap_ptr;
2014 	int ret;
2015 
2016 	iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2017 						 PCI_BUS_NUM(iommu->devid),
2018 						 iommu->devid & 0xff);
2019 	if (!iommu->dev)
2020 		return -ENODEV;
2021 
2022 	/* Prevent binding other PCI device drivers to IOMMU devices */
2023 	iommu->dev->match_driver = false;
2024 
2025 	/* ACPI _PRT won't have an IRQ for IOMMU */
2026 	iommu->dev->irq_managed = 1;
2027 
2028 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2029 			      &iommu->cap);
2030 
2031 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2032 		amd_iommu_iotlb_sup = false;
2033 
2034 	late_iommu_features_init(iommu);
2035 
2036 	if (check_feature(FEATURE_GT)) {
2037 		int glxval;
2038 		u64 pasmax;
2039 
2040 		pasmax = FIELD_GET(FEATURE_PASMAX, amd_iommu_efr);
2041 		iommu->iommu.max_pasids = (1 << (pasmax + 1)) - 1;
2042 
2043 		BUG_ON(iommu->iommu.max_pasids & ~PASID_MASK);
2044 
2045 		glxval = FIELD_GET(FEATURE_GLX, amd_iommu_efr);
2046 
2047 		if (amd_iommu_max_glx_val == -1)
2048 			amd_iommu_max_glx_val = glxval;
2049 		else
2050 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2051 
2052 		iommu_enable_gt(iommu);
2053 	}
2054 
2055 	if (check_feature(FEATURE_PPR) && amd_iommu_alloc_ppr_log(iommu))
2056 		return -ENOMEM;
2057 
2058 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2059 		pr_info("Using strict mode due to virtualization\n");
2060 		iommu_set_dma_strict();
2061 		amd_iommu_np_cache = true;
2062 	}
2063 
2064 	init_iommu_perf_ctr(iommu);
2065 
2066 	if (is_rd890_iommu(iommu->dev)) {
2067 		int i, j;
2068 
2069 		iommu->root_pdev =
2070 			pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2071 						    iommu->dev->bus->number,
2072 						    PCI_DEVFN(0, 0));
2073 
2074 		/*
2075 		 * Some rd890 systems may not be fully reconfigured by the
2076 		 * BIOS, so it's necessary for us to store this information so
2077 		 * it can be reprogrammed on resume
2078 		 */
2079 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2080 				&iommu->stored_addr_lo);
2081 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2082 				&iommu->stored_addr_hi);
2083 
2084 		/* Low bit locks writes to configuration space */
2085 		iommu->stored_addr_lo &= ~1;
2086 
2087 		for (i = 0; i < 6; i++)
2088 			for (j = 0; j < 0x12; j++)
2089 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2090 
2091 		for (i = 0; i < 0x83; i++)
2092 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2093 	}
2094 
2095 	amd_iommu_erratum_746_workaround(iommu);
2096 	amd_iommu_ats_write_check_workaround(iommu);
2097 
2098 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2099 			       amd_iommu_groups, "ivhd%d", iommu->index);
2100 	if (ret)
2101 		return ret;
2102 
2103 	/*
2104 	 * Allocate per IOMMU IOPF queue here so that in attach device path,
2105 	 * PRI capable device can be added to IOPF queue
2106 	 */
2107 	if (amd_iommu_gt_ppr_supported()) {
2108 		ret = amd_iommu_iopf_init(iommu);
2109 		if (ret)
2110 			return ret;
2111 	}
2112 
2113 	iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2114 
2115 	return pci_enable_device(iommu->dev);
2116 }
2117 
2118 static void print_iommu_info(void)
2119 {
2120 	int i;
2121 	static const char * const feat_str[] = {
2122 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2123 		"IA", "GA", "HE", "PC"
2124 	};
2125 
2126 	if (amd_iommu_efr) {
2127 		pr_info("Extended features (%#llx, %#llx):", amd_iommu_efr, amd_iommu_efr2);
2128 
2129 		for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2130 			if (check_feature(1ULL << i))
2131 				pr_cont(" %s", feat_str[i]);
2132 		}
2133 
2134 		if (check_feature(FEATURE_GAM_VAPIC))
2135 			pr_cont(" GA_vAPIC");
2136 
2137 		if (check_feature(FEATURE_SNP))
2138 			pr_cont(" SNP");
2139 
2140 		pr_cont("\n");
2141 	}
2142 
2143 	if (irq_remapping_enabled) {
2144 		pr_info("Interrupt remapping enabled\n");
2145 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2146 			pr_info("X2APIC enabled\n");
2147 	}
2148 	if (amd_iommu_pgtable == AMD_IOMMU_V2) {
2149 		pr_info("V2 page table enabled (Paging mode : %d level)\n",
2150 			amd_iommu_gpt_level);
2151 	}
2152 }
2153 
2154 static int __init amd_iommu_init_pci(void)
2155 {
2156 	struct amd_iommu *iommu;
2157 	struct amd_iommu_pci_seg *pci_seg;
2158 	int ret;
2159 
2160 	/* Init global identity domain before registering IOMMU */
2161 	amd_iommu_init_identity_domain();
2162 
2163 	for_each_iommu(iommu) {
2164 		ret = iommu_init_pci(iommu);
2165 		if (ret) {
2166 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2167 			       iommu->index, ret);
2168 			goto out;
2169 		}
2170 		/* Need to setup range after PCI init */
2171 		iommu_set_cwwb_range(iommu);
2172 	}
2173 
2174 	/*
2175 	 * Order is important here to make sure any unity map requirements are
2176 	 * fulfilled. The unity mappings are created and written to the device
2177 	 * table during the iommu_init_pci() call.
2178 	 *
2179 	 * After that we call init_device_table_dma() to make sure any
2180 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2181 	 * of all IOMMUs to make sure the changes to the device table are
2182 	 * active.
2183 	 */
2184 	for_each_pci_segment(pci_seg)
2185 		init_device_table_dma(pci_seg);
2186 
2187 	for_each_iommu(iommu)
2188 		amd_iommu_flush_all_caches(iommu);
2189 
2190 	print_iommu_info();
2191 
2192 out:
2193 	return ret;
2194 }
2195 
2196 /****************************************************************************
2197  *
2198  * The following functions initialize the MSI interrupts for all IOMMUs
2199  * in the system. It's a bit challenging because there could be multiple
2200  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2201  * pci_dev.
2202  *
2203  ****************************************************************************/
2204 
2205 static int iommu_setup_msi(struct amd_iommu *iommu)
2206 {
2207 	int r;
2208 
2209 	r = pci_enable_msi(iommu->dev);
2210 	if (r)
2211 		return r;
2212 
2213 	r = request_threaded_irq(iommu->dev->irq,
2214 				 amd_iommu_int_handler,
2215 				 amd_iommu_int_thread,
2216 				 0, "AMD-Vi",
2217 				 iommu);
2218 
2219 	if (r) {
2220 		pci_disable_msi(iommu->dev);
2221 		return r;
2222 	}
2223 
2224 	return 0;
2225 }
2226 
2227 union intcapxt {
2228 	u64	capxt;
2229 	struct {
2230 		u64	reserved_0		:  2,
2231 			dest_mode_logical	:  1,
2232 			reserved_1		:  5,
2233 			destid_0_23		: 24,
2234 			vector			:  8,
2235 			reserved_2		: 16,
2236 			destid_24_31		:  8;
2237 	};
2238 } __attribute__ ((packed));
2239 
2240 
2241 static struct irq_chip intcapxt_controller;
2242 
2243 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2244 				       struct irq_data *irqd, bool reserve)
2245 {
2246 	return 0;
2247 }
2248 
2249 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2250 					  struct irq_data *irqd)
2251 {
2252 }
2253 
2254 
2255 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2256 				    unsigned int nr_irqs, void *arg)
2257 {
2258 	struct irq_alloc_info *info = arg;
2259 	int i, ret;
2260 
2261 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2262 		return -EINVAL;
2263 
2264 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2265 	if (ret < 0)
2266 		return ret;
2267 
2268 	for (i = virq; i < virq + nr_irqs; i++) {
2269 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2270 
2271 		irqd->chip = &intcapxt_controller;
2272 		irqd->hwirq = info->hwirq;
2273 		irqd->chip_data = info->data;
2274 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2275 	}
2276 
2277 	return ret;
2278 }
2279 
2280 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2281 				    unsigned int nr_irqs)
2282 {
2283 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2284 }
2285 
2286 
2287 static void intcapxt_unmask_irq(struct irq_data *irqd)
2288 {
2289 	struct amd_iommu *iommu = irqd->chip_data;
2290 	struct irq_cfg *cfg = irqd_cfg(irqd);
2291 	union intcapxt xt;
2292 
2293 	xt.capxt = 0ULL;
2294 	xt.dest_mode_logical = apic->dest_mode_logical;
2295 	xt.vector = cfg->vector;
2296 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2297 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2298 
2299 	writeq(xt.capxt, iommu->mmio_base + irqd->hwirq);
2300 }
2301 
2302 static void intcapxt_mask_irq(struct irq_data *irqd)
2303 {
2304 	struct amd_iommu *iommu = irqd->chip_data;
2305 
2306 	writeq(0, iommu->mmio_base + irqd->hwirq);
2307 }
2308 
2309 
2310 static int intcapxt_set_affinity(struct irq_data *irqd,
2311 				 const struct cpumask *mask, bool force)
2312 {
2313 	struct irq_data *parent = irqd->parent_data;
2314 	int ret;
2315 
2316 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2317 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2318 		return ret;
2319 	return 0;
2320 }
2321 
2322 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2323 {
2324 	return on ? -EOPNOTSUPP : 0;
2325 }
2326 
2327 static struct irq_chip intcapxt_controller = {
2328 	.name			= "IOMMU-MSI",
2329 	.irq_unmask		= intcapxt_unmask_irq,
2330 	.irq_mask		= intcapxt_mask_irq,
2331 	.irq_ack		= irq_chip_ack_parent,
2332 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2333 	.irq_set_affinity       = intcapxt_set_affinity,
2334 	.irq_set_wake		= intcapxt_set_wake,
2335 	.flags			= IRQCHIP_MASK_ON_SUSPEND,
2336 };
2337 
2338 static const struct irq_domain_ops intcapxt_domain_ops = {
2339 	.alloc			= intcapxt_irqdomain_alloc,
2340 	.free			= intcapxt_irqdomain_free,
2341 	.activate		= intcapxt_irqdomain_activate,
2342 	.deactivate		= intcapxt_irqdomain_deactivate,
2343 };
2344 
2345 
2346 static struct irq_domain *iommu_irqdomain;
2347 
2348 static struct irq_domain *iommu_get_irqdomain(void)
2349 {
2350 	struct fwnode_handle *fn;
2351 
2352 	/* No need for locking here (yet) as the init is single-threaded */
2353 	if (iommu_irqdomain)
2354 		return iommu_irqdomain;
2355 
2356 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2357 	if (!fn)
2358 		return NULL;
2359 
2360 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2361 						      fn, &intcapxt_domain_ops,
2362 						      NULL);
2363 	if (!iommu_irqdomain)
2364 		irq_domain_free_fwnode(fn);
2365 
2366 	return iommu_irqdomain;
2367 }
2368 
2369 static int __iommu_setup_intcapxt(struct amd_iommu *iommu, const char *devname,
2370 				  int hwirq, irq_handler_t thread_fn)
2371 {
2372 	struct irq_domain *domain;
2373 	struct irq_alloc_info info;
2374 	int irq, ret;
2375 	int node = dev_to_node(&iommu->dev->dev);
2376 
2377 	domain = iommu_get_irqdomain();
2378 	if (!domain)
2379 		return -ENXIO;
2380 
2381 	init_irq_alloc_info(&info, NULL);
2382 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2383 	info.data = iommu;
2384 	info.hwirq = hwirq;
2385 
2386 	irq = irq_domain_alloc_irqs(domain, 1, node, &info);
2387 	if (irq < 0) {
2388 		irq_domain_remove(domain);
2389 		return irq;
2390 	}
2391 
2392 	ret = request_threaded_irq(irq, amd_iommu_int_handler,
2393 				   thread_fn, 0, devname, iommu);
2394 	if (ret) {
2395 		irq_domain_free_irqs(irq, 1);
2396 		irq_domain_remove(domain);
2397 		return ret;
2398 	}
2399 
2400 	return 0;
2401 }
2402 
2403 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2404 {
2405 	int ret;
2406 
2407 	snprintf(iommu->evt_irq_name, sizeof(iommu->evt_irq_name),
2408 		 "AMD-Vi%d-Evt", iommu->index);
2409 	ret = __iommu_setup_intcapxt(iommu, iommu->evt_irq_name,
2410 				     MMIO_INTCAPXT_EVT_OFFSET,
2411 				     amd_iommu_int_thread_evtlog);
2412 	if (ret)
2413 		return ret;
2414 
2415 	snprintf(iommu->ppr_irq_name, sizeof(iommu->ppr_irq_name),
2416 		 "AMD-Vi%d-PPR", iommu->index);
2417 	ret = __iommu_setup_intcapxt(iommu, iommu->ppr_irq_name,
2418 				     MMIO_INTCAPXT_PPR_OFFSET,
2419 				     amd_iommu_int_thread_pprlog);
2420 	if (ret)
2421 		return ret;
2422 
2423 #ifdef CONFIG_IRQ_REMAP
2424 	snprintf(iommu->ga_irq_name, sizeof(iommu->ga_irq_name),
2425 		 "AMD-Vi%d-GA", iommu->index);
2426 	ret = __iommu_setup_intcapxt(iommu, iommu->ga_irq_name,
2427 				     MMIO_INTCAPXT_GALOG_OFFSET,
2428 				     amd_iommu_int_thread_galog);
2429 #endif
2430 
2431 	return ret;
2432 }
2433 
2434 static int iommu_init_irq(struct amd_iommu *iommu)
2435 {
2436 	int ret;
2437 
2438 	if (iommu->int_enabled)
2439 		goto enable_faults;
2440 
2441 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2442 		ret = iommu_setup_intcapxt(iommu);
2443 	else if (iommu->dev->msi_cap)
2444 		ret = iommu_setup_msi(iommu);
2445 	else
2446 		ret = -ENODEV;
2447 
2448 	if (ret)
2449 		return ret;
2450 
2451 	iommu->int_enabled = true;
2452 enable_faults:
2453 
2454 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2455 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2456 
2457 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2458 
2459 	return 0;
2460 }
2461 
2462 /****************************************************************************
2463  *
2464  * The next functions belong to the third pass of parsing the ACPI
2465  * table. In this last pass the memory mapping requirements are
2466  * gathered (like exclusion and unity mapping ranges).
2467  *
2468  ****************************************************************************/
2469 
2470 static void __init free_unity_maps(void)
2471 {
2472 	struct unity_map_entry *entry, *next;
2473 	struct amd_iommu_pci_seg *p, *pci_seg;
2474 
2475 	for_each_pci_segment_safe(pci_seg, p) {
2476 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2477 			list_del(&entry->list);
2478 			kfree(entry);
2479 		}
2480 	}
2481 }
2482 
2483 /* called for unity map ACPI definition */
2484 static int __init init_unity_map_range(struct ivmd_header *m,
2485 				       struct acpi_table_header *ivrs_base)
2486 {
2487 	struct unity_map_entry *e = NULL;
2488 	struct amd_iommu_pci_seg *pci_seg;
2489 	char *s;
2490 
2491 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2492 	if (pci_seg == NULL)
2493 		return -ENOMEM;
2494 
2495 	e = kzalloc(sizeof(*e), GFP_KERNEL);
2496 	if (e == NULL)
2497 		return -ENOMEM;
2498 
2499 	switch (m->type) {
2500 	default:
2501 		kfree(e);
2502 		return 0;
2503 	case ACPI_IVMD_TYPE:
2504 		s = "IVMD_TYPEi\t\t\t";
2505 		e->devid_start = e->devid_end = m->devid;
2506 		break;
2507 	case ACPI_IVMD_TYPE_ALL:
2508 		s = "IVMD_TYPE_ALL\t\t";
2509 		e->devid_start = 0;
2510 		e->devid_end = pci_seg->last_bdf;
2511 		break;
2512 	case ACPI_IVMD_TYPE_RANGE:
2513 		s = "IVMD_TYPE_RANGE\t\t";
2514 		e->devid_start = m->devid;
2515 		e->devid_end = m->aux;
2516 		break;
2517 	}
2518 	e->address_start = PAGE_ALIGN(m->range_start);
2519 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2520 	e->prot = m->flags >> 1;
2521 
2522 	/*
2523 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2524 	 * since some buggy BIOSes might lead to the overwritten exclusion
2525 	 * range (exclusion_start and exclusion_length members). This
2526 	 * happens when there are multiple exclusion ranges (IVMD entries)
2527 	 * defined in ACPI table.
2528 	 */
2529 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2530 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2531 
2532 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2533 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2534 		    " flags: %x\n", s, m->pci_seg,
2535 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2536 		    PCI_FUNC(e->devid_start), m->pci_seg,
2537 		    PCI_BUS_NUM(e->devid_end),
2538 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2539 		    e->address_start, e->address_end, m->flags);
2540 
2541 	list_add_tail(&e->list, &pci_seg->unity_map);
2542 
2543 	return 0;
2544 }
2545 
2546 /* iterates over all memory definitions we find in the ACPI table */
2547 static int __init init_memory_definitions(struct acpi_table_header *table)
2548 {
2549 	u8 *p = (u8 *)table, *end = (u8 *)table;
2550 	struct ivmd_header *m;
2551 
2552 	end += table->length;
2553 	p += IVRS_HEADER_LENGTH;
2554 
2555 	while (p < end) {
2556 		m = (struct ivmd_header *)p;
2557 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2558 			init_unity_map_range(m, table);
2559 
2560 		p += m->length;
2561 	}
2562 
2563 	return 0;
2564 }
2565 
2566 /*
2567  * Init the device table to not allow DMA access for devices
2568  */
2569 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2570 {
2571 	u32 devid;
2572 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2573 
2574 	if (dev_table == NULL)
2575 		return;
2576 
2577 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2578 		__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID);
2579 		if (!amd_iommu_snp_en)
2580 			__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION);
2581 	}
2582 }
2583 
2584 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2585 {
2586 	u32 devid;
2587 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2588 
2589 	if (dev_table == NULL)
2590 		return;
2591 
2592 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2593 		dev_table[devid].data[0] = 0ULL;
2594 		dev_table[devid].data[1] = 0ULL;
2595 	}
2596 }
2597 
2598 static void init_device_table(void)
2599 {
2600 	struct amd_iommu_pci_seg *pci_seg;
2601 	u32 devid;
2602 
2603 	if (!amd_iommu_irq_remap)
2604 		return;
2605 
2606 	for_each_pci_segment(pci_seg) {
2607 		for (devid = 0; devid <= pci_seg->last_bdf; ++devid)
2608 			__set_dev_entry_bit(pci_seg->dev_table,
2609 					    devid, DEV_ENTRY_IRQ_TBL_EN);
2610 	}
2611 }
2612 
2613 static void iommu_init_flags(struct amd_iommu *iommu)
2614 {
2615 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2616 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2617 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2618 
2619 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2620 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2621 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2622 
2623 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2624 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2625 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2626 
2627 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2628 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2629 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2630 
2631 	/*
2632 	 * make IOMMU memory accesses cache coherent
2633 	 */
2634 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2635 
2636 	/* Set IOTLB invalidation timeout to 1s */
2637 	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2638 }
2639 
2640 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2641 {
2642 	int i, j;
2643 	u32 ioc_feature_control;
2644 	struct pci_dev *pdev = iommu->root_pdev;
2645 
2646 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2647 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2648 		return;
2649 
2650 	/*
2651 	 * First, we need to ensure that the iommu is enabled. This is
2652 	 * controlled by a register in the northbridge
2653 	 */
2654 
2655 	/* Select Northbridge indirect register 0x75 and enable writing */
2656 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2657 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2658 
2659 	/* Enable the iommu */
2660 	if (!(ioc_feature_control & 0x1))
2661 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2662 
2663 	/* Restore the iommu BAR */
2664 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2665 			       iommu->stored_addr_lo);
2666 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2667 			       iommu->stored_addr_hi);
2668 
2669 	/* Restore the l1 indirect regs for each of the 6 l1s */
2670 	for (i = 0; i < 6; i++)
2671 		for (j = 0; j < 0x12; j++)
2672 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2673 
2674 	/* Restore the l2 indirect regs */
2675 	for (i = 0; i < 0x83; i++)
2676 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2677 
2678 	/* Lock PCI setup registers */
2679 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2680 			       iommu->stored_addr_lo | 1);
2681 }
2682 
2683 static void iommu_enable_ga(struct amd_iommu *iommu)
2684 {
2685 #ifdef CONFIG_IRQ_REMAP
2686 	switch (amd_iommu_guest_ir) {
2687 	case AMD_IOMMU_GUEST_IR_VAPIC:
2688 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2689 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2690 		iommu->irte_ops = &irte_128_ops;
2691 		break;
2692 	default:
2693 		iommu->irte_ops = &irte_32_ops;
2694 		break;
2695 	}
2696 #endif
2697 }
2698 
2699 static void iommu_disable_irtcachedis(struct amd_iommu *iommu)
2700 {
2701 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
2702 }
2703 
2704 static void iommu_enable_irtcachedis(struct amd_iommu *iommu)
2705 {
2706 	u64 ctrl;
2707 
2708 	if (!amd_iommu_irtcachedis)
2709 		return;
2710 
2711 	/*
2712 	 * Note:
2713 	 * The support for IRTCacheDis feature is dertermined by
2714 	 * checking if the bit is writable.
2715 	 */
2716 	iommu_feature_enable(iommu, CONTROL_IRTCACHEDIS);
2717 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
2718 	ctrl &= (1ULL << CONTROL_IRTCACHEDIS);
2719 	if (ctrl)
2720 		iommu->irtcachedis_enabled = true;
2721 	pr_info("iommu%d (%#06x) : IRT cache is %s\n",
2722 		iommu->index, iommu->devid,
2723 		iommu->irtcachedis_enabled ? "disabled" : "enabled");
2724 }
2725 
2726 static void early_enable_iommu(struct amd_iommu *iommu)
2727 {
2728 	iommu_disable(iommu);
2729 	iommu_init_flags(iommu);
2730 	iommu_set_device_table(iommu);
2731 	iommu_enable_command_buffer(iommu);
2732 	iommu_enable_event_buffer(iommu);
2733 	iommu_set_exclusion_range(iommu);
2734 	iommu_enable_gt(iommu);
2735 	iommu_enable_ga(iommu);
2736 	iommu_enable_xt(iommu);
2737 	iommu_enable_irtcachedis(iommu);
2738 	iommu_enable(iommu);
2739 	amd_iommu_flush_all_caches(iommu);
2740 }
2741 
2742 /*
2743  * This function finally enables all IOMMUs found in the system after
2744  * they have been initialized.
2745  *
2746  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2747  * the old content of device table entries. Not this case or copy failed,
2748  * just continue as normal kernel does.
2749  */
2750 static void early_enable_iommus(void)
2751 {
2752 	struct amd_iommu *iommu;
2753 	struct amd_iommu_pci_seg *pci_seg;
2754 
2755 	if (!copy_device_table()) {
2756 		/*
2757 		 * If come here because of failure in copying device table from old
2758 		 * kernel with all IOMMUs enabled, print error message and try to
2759 		 * free allocated old_dev_tbl_cpy.
2760 		 */
2761 		if (amd_iommu_pre_enabled)
2762 			pr_err("Failed to copy DEV table from previous kernel.\n");
2763 
2764 		for_each_pci_segment(pci_seg) {
2765 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2766 				iommu_free_pages(pci_seg->old_dev_tbl_cpy,
2767 						 get_order(pci_seg->dev_table_size));
2768 				pci_seg->old_dev_tbl_cpy = NULL;
2769 			}
2770 		}
2771 
2772 		for_each_iommu(iommu) {
2773 			clear_translation_pre_enabled(iommu);
2774 			early_enable_iommu(iommu);
2775 		}
2776 	} else {
2777 		pr_info("Copied DEV table from previous kernel.\n");
2778 
2779 		for_each_pci_segment(pci_seg) {
2780 			iommu_free_pages(pci_seg->dev_table,
2781 					 get_order(pci_seg->dev_table_size));
2782 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2783 		}
2784 
2785 		for_each_iommu(iommu) {
2786 			iommu_disable_command_buffer(iommu);
2787 			iommu_disable_event_buffer(iommu);
2788 			iommu_disable_irtcachedis(iommu);
2789 			iommu_enable_command_buffer(iommu);
2790 			iommu_enable_event_buffer(iommu);
2791 			iommu_enable_ga(iommu);
2792 			iommu_enable_xt(iommu);
2793 			iommu_enable_irtcachedis(iommu);
2794 			iommu_set_device_table(iommu);
2795 			amd_iommu_flush_all_caches(iommu);
2796 		}
2797 	}
2798 }
2799 
2800 static void enable_iommus_ppr(void)
2801 {
2802 	struct amd_iommu *iommu;
2803 
2804 	if (!amd_iommu_gt_ppr_supported())
2805 		return;
2806 
2807 	for_each_iommu(iommu)
2808 		amd_iommu_enable_ppr_log(iommu);
2809 }
2810 
2811 static void enable_iommus_vapic(void)
2812 {
2813 #ifdef CONFIG_IRQ_REMAP
2814 	u32 status, i;
2815 	struct amd_iommu *iommu;
2816 
2817 	for_each_iommu(iommu) {
2818 		/*
2819 		 * Disable GALog if already running. It could have been enabled
2820 		 * in the previous boot before kdump.
2821 		 */
2822 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2823 		if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2824 			continue;
2825 
2826 		iommu_feature_disable(iommu, CONTROL_GALOG_EN);
2827 		iommu_feature_disable(iommu, CONTROL_GAINT_EN);
2828 
2829 		/*
2830 		 * Need to set and poll check the GALOGRun bit to zero before
2831 		 * we can set/ modify GA Log registers safely.
2832 		 */
2833 		for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
2834 			status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2835 			if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2836 				break;
2837 			udelay(10);
2838 		}
2839 
2840 		if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
2841 			return;
2842 	}
2843 
2844 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2845 	    !check_feature(FEATURE_GAM_VAPIC)) {
2846 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2847 		return;
2848 	}
2849 
2850 	if (amd_iommu_snp_en &&
2851 	    !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) {
2852 		pr_warn("Force to disable Virtual APIC due to SNP\n");
2853 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2854 		return;
2855 	}
2856 
2857 	/* Enabling GAM and SNPAVIC support */
2858 	for_each_iommu(iommu) {
2859 		if (iommu_init_ga_log(iommu) ||
2860 		    iommu_ga_log_enable(iommu))
2861 			return;
2862 
2863 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2864 		if (amd_iommu_snp_en)
2865 			iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN);
2866 	}
2867 
2868 	amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2869 	pr_info("Virtual APIC enabled\n");
2870 #endif
2871 }
2872 
2873 static void disable_iommus(void)
2874 {
2875 	struct amd_iommu *iommu;
2876 
2877 	for_each_iommu(iommu)
2878 		iommu_disable(iommu);
2879 
2880 #ifdef CONFIG_IRQ_REMAP
2881 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2882 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2883 #endif
2884 }
2885 
2886 /*
2887  * Suspend/Resume support
2888  * disable suspend until real resume implemented
2889  */
2890 
2891 static void amd_iommu_resume(void)
2892 {
2893 	struct amd_iommu *iommu;
2894 
2895 	for_each_iommu(iommu)
2896 		iommu_apply_resume_quirks(iommu);
2897 
2898 	/* re-load the hardware */
2899 	for_each_iommu(iommu)
2900 		early_enable_iommu(iommu);
2901 
2902 	amd_iommu_enable_interrupts();
2903 }
2904 
2905 static int amd_iommu_suspend(void)
2906 {
2907 	/* disable IOMMUs to go out of the way for BIOS */
2908 	disable_iommus();
2909 
2910 	return 0;
2911 }
2912 
2913 static struct syscore_ops amd_iommu_syscore_ops = {
2914 	.suspend = amd_iommu_suspend,
2915 	.resume = amd_iommu_resume,
2916 };
2917 
2918 static void __init free_iommu_resources(void)
2919 {
2920 	kmem_cache_destroy(amd_iommu_irq_cache);
2921 	amd_iommu_irq_cache = NULL;
2922 
2923 	free_iommu_all();
2924 	free_pci_segments();
2925 }
2926 
2927 /* SB IOAPIC is always on this device in AMD systems */
2928 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2929 
2930 static bool __init check_ioapic_information(void)
2931 {
2932 	const char *fw_bug = FW_BUG;
2933 	bool ret, has_sb_ioapic;
2934 	int idx;
2935 
2936 	has_sb_ioapic = false;
2937 	ret           = false;
2938 
2939 	/*
2940 	 * If we have map overrides on the kernel command line the
2941 	 * messages in this function might not describe firmware bugs
2942 	 * anymore - so be careful
2943 	 */
2944 	if (cmdline_maps)
2945 		fw_bug = "";
2946 
2947 	for (idx = 0; idx < nr_ioapics; idx++) {
2948 		int devid, id = mpc_ioapic_id(idx);
2949 
2950 		devid = get_ioapic_devid(id);
2951 		if (devid < 0) {
2952 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2953 				fw_bug, id);
2954 			ret = false;
2955 		} else if (devid == IOAPIC_SB_DEVID) {
2956 			has_sb_ioapic = true;
2957 			ret           = true;
2958 		}
2959 	}
2960 
2961 	if (!has_sb_ioapic) {
2962 		/*
2963 		 * We expect the SB IOAPIC to be listed in the IVRS
2964 		 * table. The system timer is connected to the SB IOAPIC
2965 		 * and if we don't have it in the list the system will
2966 		 * panic at boot time.  This situation usually happens
2967 		 * when the BIOS is buggy and provides us the wrong
2968 		 * device id for the IOAPIC in the system.
2969 		 */
2970 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2971 	}
2972 
2973 	if (!ret)
2974 		pr_err("Disabling interrupt remapping\n");
2975 
2976 	return ret;
2977 }
2978 
2979 static void __init free_dma_resources(void)
2980 {
2981 	ida_destroy(&pdom_ids);
2982 
2983 	free_unity_maps();
2984 }
2985 
2986 static void __init ivinfo_init(void *ivrs)
2987 {
2988 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
2989 }
2990 
2991 /*
2992  * This is the hardware init function for AMD IOMMU in the system.
2993  * This function is called either from amd_iommu_init or from the interrupt
2994  * remapping setup code.
2995  *
2996  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2997  * four times:
2998  *
2999  *	1 pass) Discover the most comprehensive IVHD type to use.
3000  *
3001  *	2 pass) Find the highest PCI device id the driver has to handle.
3002  *		Upon this information the size of the data structures is
3003  *		determined that needs to be allocated.
3004  *
3005  *	3 pass) Initialize the data structures just allocated with the
3006  *		information in the ACPI table about available AMD IOMMUs
3007  *		in the system. It also maps the PCI devices in the
3008  *		system to specific IOMMUs
3009  *
3010  *	4 pass) After the basic data structures are allocated and
3011  *		initialized we update them with information about memory
3012  *		remapping requirements parsed out of the ACPI table in
3013  *		this last pass.
3014  *
3015  * After everything is set up the IOMMUs are enabled and the necessary
3016  * hotplug and suspend notifiers are registered.
3017  */
3018 static int __init early_amd_iommu_init(void)
3019 {
3020 	struct acpi_table_header *ivrs_base;
3021 	int remap_cache_sz, ret;
3022 	acpi_status status;
3023 
3024 	if (!amd_iommu_detected)
3025 		return -ENODEV;
3026 
3027 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3028 	if (status == AE_NOT_FOUND)
3029 		return -ENODEV;
3030 	else if (ACPI_FAILURE(status)) {
3031 		const char *err = acpi_format_exception(status);
3032 		pr_err("IVRS table error: %s\n", err);
3033 		return -EINVAL;
3034 	}
3035 
3036 	/*
3037 	 * Validate checksum here so we don't need to do it when
3038 	 * we actually parse the table
3039 	 */
3040 	ret = check_ivrs_checksum(ivrs_base);
3041 	if (ret)
3042 		goto out;
3043 
3044 	ivinfo_init(ivrs_base);
3045 
3046 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
3047 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
3048 
3049 	/*
3050 	 * now the data structures are allocated and basically initialized
3051 	 * start the real acpi table scan
3052 	 */
3053 	ret = init_iommu_all(ivrs_base);
3054 	if (ret)
3055 		goto out;
3056 
3057 	/* 5 level guest page table */
3058 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3059 	    FIELD_GET(FEATURE_GATS, amd_iommu_efr) == GUEST_PGTABLE_5_LEVEL)
3060 		amd_iommu_gpt_level = PAGE_MODE_5_LEVEL;
3061 
3062 	if (amd_iommu_pgtable == AMD_IOMMU_V2) {
3063 		if (!amd_iommu_v2_pgtbl_supported()) {
3064 			pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n");
3065 			amd_iommu_pgtable = AMD_IOMMU_V1;
3066 		}
3067 	}
3068 
3069 	/* Disable any previously enabled IOMMUs */
3070 	if (!is_kdump_kernel() || amd_iommu_disabled)
3071 		disable_iommus();
3072 
3073 	if (amd_iommu_irq_remap)
3074 		amd_iommu_irq_remap = check_ioapic_information();
3075 
3076 	if (amd_iommu_irq_remap) {
3077 		struct amd_iommu_pci_seg *pci_seg;
3078 		/*
3079 		 * Interrupt remapping enabled, create kmem_cache for the
3080 		 * remapping tables.
3081 		 */
3082 		ret = -ENOMEM;
3083 		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
3084 			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
3085 		else
3086 			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
3087 		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
3088 							remap_cache_sz,
3089 							DTE_INTTAB_ALIGNMENT,
3090 							0, NULL);
3091 		if (!amd_iommu_irq_cache)
3092 			goto out;
3093 
3094 		for_each_pci_segment(pci_seg) {
3095 			if (alloc_irq_lookup_table(pci_seg))
3096 				goto out;
3097 		}
3098 	}
3099 
3100 	ret = init_memory_definitions(ivrs_base);
3101 	if (ret)
3102 		goto out;
3103 
3104 	/* init the device table */
3105 	init_device_table();
3106 
3107 out:
3108 	/* Don't leak any ACPI memory */
3109 	acpi_put_table(ivrs_base);
3110 
3111 	return ret;
3112 }
3113 
3114 static int amd_iommu_enable_interrupts(void)
3115 {
3116 	struct amd_iommu *iommu;
3117 	int ret = 0;
3118 
3119 	for_each_iommu(iommu) {
3120 		ret = iommu_init_irq(iommu);
3121 		if (ret)
3122 			goto out;
3123 	}
3124 
3125 	/*
3126 	 * Interrupt handler is ready to process interrupts. Enable
3127 	 * PPR and GA log interrupt for all IOMMUs.
3128 	 */
3129 	enable_iommus_vapic();
3130 	enable_iommus_ppr();
3131 
3132 out:
3133 	return ret;
3134 }
3135 
3136 static bool __init detect_ivrs(void)
3137 {
3138 	struct acpi_table_header *ivrs_base;
3139 	acpi_status status;
3140 	int i;
3141 
3142 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3143 	if (status == AE_NOT_FOUND)
3144 		return false;
3145 	else if (ACPI_FAILURE(status)) {
3146 		const char *err = acpi_format_exception(status);
3147 		pr_err("IVRS table error: %s\n", err);
3148 		return false;
3149 	}
3150 
3151 	acpi_put_table(ivrs_base);
3152 
3153 	if (amd_iommu_force_enable)
3154 		goto out;
3155 
3156 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3157 	for (i = 0; i < 32; i++) {
3158 		u32 pci_id;
3159 
3160 		pci_id = read_pci_config(0, i, 0, 0);
3161 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3162 			pr_info("Disable IOMMU on Stoney Ridge\n");
3163 			return false;
3164 		}
3165 	}
3166 
3167 out:
3168 	/* Make sure ACS will be enabled during PCI probe */
3169 	pci_request_acs();
3170 
3171 	return true;
3172 }
3173 
3174 static void iommu_snp_enable(void)
3175 {
3176 #ifdef CONFIG_KVM_AMD_SEV
3177 	if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3178 		return;
3179 	/*
3180 	 * The SNP support requires that IOMMU must be enabled, and is
3181 	 * configured with V1 page table (DTE[Mode] = 0 is not supported).
3182 	 */
3183 	if (no_iommu || iommu_default_passthrough()) {
3184 		pr_warn("SNP: IOMMU disabled or configured in passthrough mode, SNP cannot be supported.\n");
3185 		goto disable_snp;
3186 	}
3187 
3188 	if (amd_iommu_pgtable != AMD_IOMMU_V1) {
3189 		pr_warn("SNP: IOMMU is configured with V2 page table mode, SNP cannot be supported.\n");
3190 		goto disable_snp;
3191 	}
3192 
3193 	amd_iommu_snp_en = check_feature(FEATURE_SNP);
3194 	if (!amd_iommu_snp_en) {
3195 		pr_warn("SNP: IOMMU SNP feature not enabled, SNP cannot be supported.\n");
3196 		goto disable_snp;
3197 	}
3198 
3199 	pr_info("IOMMU SNP support enabled.\n");
3200 	return;
3201 
3202 disable_snp:
3203 	cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3204 #endif
3205 }
3206 
3207 /****************************************************************************
3208  *
3209  * AMD IOMMU Initialization State Machine
3210  *
3211  ****************************************************************************/
3212 
3213 static int __init state_next(void)
3214 {
3215 	int ret = 0;
3216 
3217 	switch (init_state) {
3218 	case IOMMU_START_STATE:
3219 		if (!detect_ivrs()) {
3220 			init_state	= IOMMU_NOT_FOUND;
3221 			ret		= -ENODEV;
3222 		} else {
3223 			init_state	= IOMMU_IVRS_DETECTED;
3224 		}
3225 		break;
3226 	case IOMMU_IVRS_DETECTED:
3227 		if (amd_iommu_disabled) {
3228 			init_state = IOMMU_CMDLINE_DISABLED;
3229 			ret = -EINVAL;
3230 		} else {
3231 			ret = early_amd_iommu_init();
3232 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3233 		}
3234 		break;
3235 	case IOMMU_ACPI_FINISHED:
3236 		early_enable_iommus();
3237 		x86_platform.iommu_shutdown = disable_iommus;
3238 		init_state = IOMMU_ENABLED;
3239 		break;
3240 	case IOMMU_ENABLED:
3241 		register_syscore_ops(&amd_iommu_syscore_ops);
3242 		iommu_snp_enable();
3243 		ret = amd_iommu_init_pci();
3244 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3245 		break;
3246 	case IOMMU_PCI_INIT:
3247 		ret = amd_iommu_enable_interrupts();
3248 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3249 		break;
3250 	case IOMMU_INTERRUPTS_EN:
3251 		init_state = IOMMU_INITIALIZED;
3252 		break;
3253 	case IOMMU_INITIALIZED:
3254 		/* Nothing to do */
3255 		break;
3256 	case IOMMU_NOT_FOUND:
3257 	case IOMMU_INIT_ERROR:
3258 	case IOMMU_CMDLINE_DISABLED:
3259 		/* Error states => do nothing */
3260 		ret = -EINVAL;
3261 		break;
3262 	default:
3263 		/* Unknown state */
3264 		BUG();
3265 	}
3266 
3267 	if (ret) {
3268 		free_dma_resources();
3269 		if (!irq_remapping_enabled) {
3270 			disable_iommus();
3271 			free_iommu_resources();
3272 		} else {
3273 			struct amd_iommu *iommu;
3274 			struct amd_iommu_pci_seg *pci_seg;
3275 
3276 			for_each_pci_segment(pci_seg)
3277 				uninit_device_table_dma(pci_seg);
3278 
3279 			for_each_iommu(iommu)
3280 				amd_iommu_flush_all_caches(iommu);
3281 		}
3282 	}
3283 	return ret;
3284 }
3285 
3286 static int __init iommu_go_to_state(enum iommu_init_state state)
3287 {
3288 	int ret = -EINVAL;
3289 
3290 	while (init_state != state) {
3291 		if (init_state == IOMMU_NOT_FOUND         ||
3292 		    init_state == IOMMU_INIT_ERROR        ||
3293 		    init_state == IOMMU_CMDLINE_DISABLED)
3294 			break;
3295 		ret = state_next();
3296 	}
3297 
3298 	return ret;
3299 }
3300 
3301 #ifdef CONFIG_IRQ_REMAP
3302 int __init amd_iommu_prepare(void)
3303 {
3304 	int ret;
3305 
3306 	amd_iommu_irq_remap = true;
3307 
3308 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3309 	if (ret) {
3310 		amd_iommu_irq_remap = false;
3311 		return ret;
3312 	}
3313 
3314 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3315 }
3316 
3317 int __init amd_iommu_enable(void)
3318 {
3319 	int ret;
3320 
3321 	ret = iommu_go_to_state(IOMMU_ENABLED);
3322 	if (ret)
3323 		return ret;
3324 
3325 	irq_remapping_enabled = 1;
3326 	return amd_iommu_xt_mode;
3327 }
3328 
3329 void amd_iommu_disable(void)
3330 {
3331 	amd_iommu_suspend();
3332 }
3333 
3334 int amd_iommu_reenable(int mode)
3335 {
3336 	amd_iommu_resume();
3337 
3338 	return 0;
3339 }
3340 
3341 int amd_iommu_enable_faulting(unsigned int cpu)
3342 {
3343 	/* We enable MSI later when PCI is initialized */
3344 	return 0;
3345 }
3346 #endif
3347 
3348 /*
3349  * This is the core init function for AMD IOMMU hardware in the system.
3350  * This function is called from the generic x86 DMA layer initialization
3351  * code.
3352  */
3353 static int __init amd_iommu_init(void)
3354 {
3355 	struct amd_iommu *iommu;
3356 	int ret;
3357 
3358 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3359 #ifdef CONFIG_GART_IOMMU
3360 	if (ret && list_empty(&amd_iommu_list)) {
3361 		/*
3362 		 * We failed to initialize the AMD IOMMU - try fallback
3363 		 * to GART if possible.
3364 		 */
3365 		gart_iommu_init();
3366 	}
3367 #endif
3368 
3369 	for_each_iommu(iommu)
3370 		amd_iommu_debugfs_setup(iommu);
3371 
3372 	return ret;
3373 }
3374 
3375 static bool amd_iommu_sme_check(void)
3376 {
3377 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3378 	    (boot_cpu_data.x86 != 0x17))
3379 		return true;
3380 
3381 	/* For Fam17h, a specific level of support is required */
3382 	if (boot_cpu_data.microcode >= 0x08001205)
3383 		return true;
3384 
3385 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3386 	    (boot_cpu_data.microcode <= 0x080011ff))
3387 		return true;
3388 
3389 	pr_notice("IOMMU not currently supported when SME is active\n");
3390 
3391 	return false;
3392 }
3393 
3394 /****************************************************************************
3395  *
3396  * Early detect code. This code runs at IOMMU detection time in the DMA
3397  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3398  * IOMMUs
3399  *
3400  ****************************************************************************/
3401 int __init amd_iommu_detect(void)
3402 {
3403 	int ret;
3404 
3405 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3406 		return -ENODEV;
3407 
3408 	if (!amd_iommu_sme_check())
3409 		return -ENODEV;
3410 
3411 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3412 	if (ret)
3413 		return ret;
3414 
3415 	amd_iommu_detected = true;
3416 	iommu_detected = 1;
3417 	x86_init.iommu.iommu_init = amd_iommu_init;
3418 
3419 	return 1;
3420 }
3421 
3422 /****************************************************************************
3423  *
3424  * Parsing functions for the AMD IOMMU specific kernel command line
3425  * options.
3426  *
3427  ****************************************************************************/
3428 
3429 static int __init parse_amd_iommu_dump(char *str)
3430 {
3431 	amd_iommu_dump = true;
3432 
3433 	return 1;
3434 }
3435 
3436 static int __init parse_amd_iommu_intr(char *str)
3437 {
3438 	for (; *str; ++str) {
3439 		if (strncmp(str, "legacy", 6) == 0) {
3440 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3441 			break;
3442 		}
3443 		if (strncmp(str, "vapic", 5) == 0) {
3444 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3445 			break;
3446 		}
3447 	}
3448 	return 1;
3449 }
3450 
3451 static int __init parse_amd_iommu_options(char *str)
3452 {
3453 	if (!str)
3454 		return -EINVAL;
3455 
3456 	while (*str) {
3457 		if (strncmp(str, "fullflush", 9) == 0) {
3458 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3459 			iommu_set_dma_strict();
3460 		} else if (strncmp(str, "force_enable", 12) == 0) {
3461 			amd_iommu_force_enable = true;
3462 		} else if (strncmp(str, "off", 3) == 0) {
3463 			amd_iommu_disabled = true;
3464 		} else if (strncmp(str, "force_isolation", 15) == 0) {
3465 			amd_iommu_force_isolation = true;
3466 		} else if (strncmp(str, "pgtbl_v1", 8) == 0) {
3467 			amd_iommu_pgtable = AMD_IOMMU_V1;
3468 		} else if (strncmp(str, "pgtbl_v2", 8) == 0) {
3469 			amd_iommu_pgtable = AMD_IOMMU_V2;
3470 		} else if (strncmp(str, "irtcachedis", 11) == 0) {
3471 			amd_iommu_irtcachedis = true;
3472 		} else if (strncmp(str, "nohugepages", 11) == 0) {
3473 			pr_info("Restricting V1 page-sizes to 4KiB");
3474 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_4K;
3475 		} else if (strncmp(str, "v2_pgsizes_only", 15) == 0) {
3476 			pr_info("Restricting V1 page-sizes to 4KiB/2MiB/1GiB");
3477 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_V2;
3478 		} else {
3479 			pr_notice("Unknown option - '%s'\n", str);
3480 		}
3481 
3482 		str += strcspn(str, ",");
3483 		while (*str == ',')
3484 			str++;
3485 	}
3486 
3487 	return 1;
3488 }
3489 
3490 static int __init parse_ivrs_ioapic(char *str)
3491 {
3492 	u32 seg = 0, bus, dev, fn;
3493 	int id, i;
3494 	u32 devid;
3495 
3496 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3497 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3498 		goto found;
3499 
3500 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3501 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3502 		pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n",
3503 			str, id, seg, bus, dev, fn);
3504 		goto found;
3505 	}
3506 
3507 	pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3508 	return 1;
3509 
3510 found:
3511 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3512 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3513 			str);
3514 		return 1;
3515 	}
3516 
3517 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3518 
3519 	cmdline_maps			= true;
3520 	i				= early_ioapic_map_size++;
3521 	early_ioapic_map[i].id		= id;
3522 	early_ioapic_map[i].devid	= devid;
3523 	early_ioapic_map[i].cmd_line	= true;
3524 
3525 	return 1;
3526 }
3527 
3528 static int __init parse_ivrs_hpet(char *str)
3529 {
3530 	u32 seg = 0, bus, dev, fn;
3531 	int id, i;
3532 	u32 devid;
3533 
3534 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3535 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3536 		goto found;
3537 
3538 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3539 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3540 		pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n",
3541 			str, id, seg, bus, dev, fn);
3542 		goto found;
3543 	}
3544 
3545 	pr_err("Invalid command line: ivrs_hpet%s\n", str);
3546 	return 1;
3547 
3548 found:
3549 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3550 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3551 			str);
3552 		return 1;
3553 	}
3554 
3555 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3556 
3557 	cmdline_maps			= true;
3558 	i				= early_hpet_map_size++;
3559 	early_hpet_map[i].id		= id;
3560 	early_hpet_map[i].devid		= devid;
3561 	early_hpet_map[i].cmd_line	= true;
3562 
3563 	return 1;
3564 }
3565 
3566 #define ACPIID_LEN (ACPIHID_UID_LEN + ACPIHID_HID_LEN)
3567 
3568 static int __init parse_ivrs_acpihid(char *str)
3569 {
3570 	u32 seg = 0, bus, dev, fn;
3571 	char *hid, *uid, *p, *addr;
3572 	char acpiid[ACPIID_LEN] = {0};
3573 	int i;
3574 
3575 	addr = strchr(str, '@');
3576 	if (!addr) {
3577 		addr = strchr(str, '=');
3578 		if (!addr)
3579 			goto not_found;
3580 
3581 		++addr;
3582 
3583 		if (strlen(addr) > ACPIID_LEN)
3584 			goto not_found;
3585 
3586 		if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 ||
3587 		    sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) {
3588 			pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n",
3589 				str, acpiid, seg, bus, dev, fn);
3590 			goto found;
3591 		}
3592 		goto not_found;
3593 	}
3594 
3595 	/* We have the '@', make it the terminator to get just the acpiid */
3596 	*addr++ = 0;
3597 
3598 	if (strlen(str) > ACPIID_LEN + 1)
3599 		goto not_found;
3600 
3601 	if (sscanf(str, "=%s", acpiid) != 1)
3602 		goto not_found;
3603 
3604 	if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 ||
3605 	    sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4)
3606 		goto found;
3607 
3608 not_found:
3609 	pr_err("Invalid command line: ivrs_acpihid%s\n", str);
3610 	return 1;
3611 
3612 found:
3613 	p = acpiid;
3614 	hid = strsep(&p, ":");
3615 	uid = p;
3616 
3617 	if (!hid || !(*hid) || !uid) {
3618 		pr_err("Invalid command line: hid or uid\n");
3619 		return 1;
3620 	}
3621 
3622 	/*
3623 	 * Ignore leading zeroes after ':', so e.g., AMDI0095:00
3624 	 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match
3625 	 */
3626 	while (*uid == '0' && *(uid + 1))
3627 		uid++;
3628 
3629 	i = early_acpihid_map_size++;
3630 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3631 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3632 	early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3633 	early_acpihid_map[i].cmd_line	= true;
3634 
3635 	return 1;
3636 }
3637 
3638 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3639 __setup("amd_iommu=",		parse_amd_iommu_options);
3640 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3641 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3642 __setup("ivrs_hpet",		parse_ivrs_hpet);
3643 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3644 
3645 bool amd_iommu_pasid_supported(void)
3646 {
3647 	/* CPU page table size should match IOMMU guest page table size */
3648 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3649 	    amd_iommu_gpt_level != PAGE_MODE_5_LEVEL)
3650 		return false;
3651 
3652 	/*
3653 	 * Since DTE[Mode]=0 is prohibited on SNP-enabled system
3654 	 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without
3655 	 * setting up IOMMUv1 page table.
3656 	 */
3657 	return amd_iommu_gt_ppr_supported() && !amd_iommu_snp_en;
3658 }
3659 
3660 struct amd_iommu *get_amd_iommu(unsigned int idx)
3661 {
3662 	unsigned int i = 0;
3663 	struct amd_iommu *iommu;
3664 
3665 	for_each_iommu(iommu)
3666 		if (i++ == idx)
3667 			return iommu;
3668 	return NULL;
3669 }
3670 
3671 /****************************************************************************
3672  *
3673  * IOMMU EFR Performance Counter support functionality. This code allows
3674  * access to the IOMMU PC functionality.
3675  *
3676  ****************************************************************************/
3677 
3678 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3679 {
3680 	struct amd_iommu *iommu = get_amd_iommu(idx);
3681 
3682 	if (iommu)
3683 		return iommu->max_banks;
3684 
3685 	return 0;
3686 }
3687 
3688 bool amd_iommu_pc_supported(void)
3689 {
3690 	return amd_iommu_pc_present;
3691 }
3692 
3693 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3694 {
3695 	struct amd_iommu *iommu = get_amd_iommu(idx);
3696 
3697 	if (iommu)
3698 		return iommu->max_counters;
3699 
3700 	return 0;
3701 }
3702 
3703 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3704 				u8 fxn, u64 *value, bool is_write)
3705 {
3706 	u32 offset;
3707 	u32 max_offset_lim;
3708 
3709 	/* Make sure the IOMMU PC resource is available */
3710 	if (!amd_iommu_pc_present)
3711 		return -ENODEV;
3712 
3713 	/* Check for valid iommu and pc register indexing */
3714 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3715 		return -ENODEV;
3716 
3717 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3718 
3719 	/* Limit the offset to the hw defined mmio region aperture */
3720 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3721 				(iommu->max_counters << 8) | 0x28);
3722 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3723 	    (offset > max_offset_lim))
3724 		return -EINVAL;
3725 
3726 	if (is_write) {
3727 		u64 val = *value & GENMASK_ULL(47, 0);
3728 
3729 		writel((u32)val, iommu->mmio_base + offset);
3730 		writel((val >> 32), iommu->mmio_base + offset + 4);
3731 	} else {
3732 		*value = readl(iommu->mmio_base + offset + 4);
3733 		*value <<= 32;
3734 		*value |= readl(iommu->mmio_base + offset);
3735 		*value &= GENMASK_ULL(47, 0);
3736 	}
3737 
3738 	return 0;
3739 }
3740 
3741 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3742 {
3743 	if (!iommu)
3744 		return -EINVAL;
3745 
3746 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3747 }
3748 
3749 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3750 {
3751 	if (!iommu)
3752 		return -EINVAL;
3753 
3754 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3755 }
3756 
3757 #ifdef CONFIG_KVM_AMD_SEV
3758 static int iommu_page_make_shared(void *page)
3759 {
3760 	unsigned long paddr, pfn;
3761 
3762 	paddr = iommu_virt_to_phys(page);
3763 	/* Cbit maybe set in the paddr */
3764 	pfn = __sme_clr(paddr) >> PAGE_SHIFT;
3765 
3766 	if (!(pfn % PTRS_PER_PMD)) {
3767 		int ret, level;
3768 		bool assigned;
3769 
3770 		ret = snp_lookup_rmpentry(pfn, &assigned, &level);
3771 		if (ret) {
3772 			pr_warn("IOMMU PFN %lx RMP lookup failed, ret %d\n", pfn, ret);
3773 			return ret;
3774 		}
3775 
3776 		if (!assigned) {
3777 			pr_warn("IOMMU PFN %lx not assigned in RMP table\n", pfn);
3778 			return -EINVAL;
3779 		}
3780 
3781 		if (level > PG_LEVEL_4K) {
3782 			ret = psmash(pfn);
3783 			if (!ret)
3784 				goto done;
3785 
3786 			pr_warn("PSMASH failed for IOMMU PFN %lx huge RMP entry, ret: %d, level: %d\n",
3787 				pfn, ret, level);
3788 			return ret;
3789 		}
3790 	}
3791 
3792 done:
3793 	return rmp_make_shared(pfn, PG_LEVEL_4K);
3794 }
3795 
3796 static int iommu_make_shared(void *va, size_t size)
3797 {
3798 	void *page;
3799 	int ret;
3800 
3801 	if (!va)
3802 		return 0;
3803 
3804 	for (page = va; page < (va + size); page += PAGE_SIZE) {
3805 		ret = iommu_page_make_shared(page);
3806 		if (ret)
3807 			return ret;
3808 	}
3809 
3810 	return 0;
3811 }
3812 
3813 int amd_iommu_snp_disable(void)
3814 {
3815 	struct amd_iommu *iommu;
3816 	int ret;
3817 
3818 	if (!amd_iommu_snp_en)
3819 		return 0;
3820 
3821 	for_each_iommu(iommu) {
3822 		ret = iommu_make_shared(iommu->evt_buf, EVT_BUFFER_SIZE);
3823 		if (ret)
3824 			return ret;
3825 
3826 		ret = iommu_make_shared(iommu->ppr_log, PPR_LOG_SIZE);
3827 		if (ret)
3828 			return ret;
3829 
3830 		ret = iommu_make_shared((void *)iommu->cmd_sem, PAGE_SIZE);
3831 		if (ret)
3832 			return ret;
3833 	}
3834 
3835 	return 0;
3836 }
3837 EXPORT_SYMBOL_GPL(amd_iommu_snp_disable);
3838 #endif
3839