xref: /linux/drivers/iommu/amd/init.c (revision 1e0ea4dff0f46a3575b6882941dc7331c232d72c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
4  * Author: Joerg Roedel <jroedel@suse.de>
5  *         Leo Duran <leo.duran@amd.com>
6  */
7 
8 #define pr_fmt(fmt)     "AMD-Vi: " fmt
9 #define dev_fmt(fmt)    pr_fmt(fmt)
10 
11 #include <linux/pci.h>
12 #include <linux/acpi.h>
13 #include <linux/list.h>
14 #include <linux/bitmap.h>
15 #include <linux/syscore_ops.h>
16 #include <linux/interrupt.h>
17 #include <linux/msi.h>
18 #include <linux/irq.h>
19 #include <linux/amd-iommu.h>
20 #include <linux/export.h>
21 #include <linux/kmemleak.h>
22 #include <linux/cc_platform.h>
23 #include <linux/iopoll.h>
24 #include <asm/pci-direct.h>
25 #include <asm/iommu.h>
26 #include <asm/apic.h>
27 #include <asm/gart.h>
28 #include <asm/x86_init.h>
29 #include <asm/io_apic.h>
30 #include <asm/irq_remapping.h>
31 #include <asm/set_memory.h>
32 #include <asm/sev.h>
33 
34 #include <linux/crash_dump.h>
35 
36 #include "amd_iommu.h"
37 #include "../irq_remapping.h"
38 #include "../iommu-pages.h"
39 
40 /*
41  * definitions for the ACPI scanning code
42  */
43 #define IVRS_HEADER_LENGTH 48
44 
45 #define ACPI_IVHD_TYPE_MAX_SUPPORTED	0x40
46 #define ACPI_IVMD_TYPE_ALL              0x20
47 #define ACPI_IVMD_TYPE                  0x21
48 #define ACPI_IVMD_TYPE_RANGE            0x22
49 
50 #define IVHD_DEV_ALL                    0x01
51 #define IVHD_DEV_SELECT                 0x02
52 #define IVHD_DEV_SELECT_RANGE_START     0x03
53 #define IVHD_DEV_RANGE_END              0x04
54 #define IVHD_DEV_ALIAS                  0x42
55 #define IVHD_DEV_ALIAS_RANGE            0x43
56 #define IVHD_DEV_EXT_SELECT             0x46
57 #define IVHD_DEV_EXT_SELECT_RANGE       0x47
58 #define IVHD_DEV_SPECIAL		0x48
59 #define IVHD_DEV_ACPI_HID		0xf0
60 
61 #define UID_NOT_PRESENT                 0
62 #define UID_IS_INTEGER                  1
63 #define UID_IS_CHARACTER                2
64 
65 #define IVHD_SPECIAL_IOAPIC		1
66 #define IVHD_SPECIAL_HPET		2
67 
68 #define IVHD_FLAG_HT_TUN_EN_MASK        0x01
69 #define IVHD_FLAG_PASSPW_EN_MASK        0x02
70 #define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
71 #define IVHD_FLAG_ISOC_EN_MASK          0x08
72 
73 #define IVMD_FLAG_EXCL_RANGE            0x08
74 #define IVMD_FLAG_IW                    0x04
75 #define IVMD_FLAG_IR                    0x02
76 #define IVMD_FLAG_UNITY_MAP             0x01
77 
78 #define ACPI_DEVFLAG_INITPASS           0x01
79 #define ACPI_DEVFLAG_EXTINT             0x02
80 #define ACPI_DEVFLAG_NMI                0x04
81 #define ACPI_DEVFLAG_SYSMGT1            0x10
82 #define ACPI_DEVFLAG_SYSMGT2            0x20
83 #define ACPI_DEVFLAG_LINT0              0x40
84 #define ACPI_DEVFLAG_LINT1              0x80
85 #define ACPI_DEVFLAG_ATSDIS             0x10000000
86 
87 #define IVRS_GET_SBDF_ID(seg, bus, dev, fn)	(((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \
88 						 | ((dev & 0x1f) << 3) | (fn & 0x7))
89 
90 /*
91  * ACPI table definitions
92  *
93  * These data structures are laid over the table to parse the important values
94  * out of it.
95  */
96 
97 /*
98  * structure describing one IOMMU in the ACPI table. Typically followed by one
99  * or more ivhd_entrys.
100  */
101 struct ivhd_header {
102 	u8 type;
103 	u8 flags;
104 	u16 length;
105 	u16 devid;
106 	u16 cap_ptr;
107 	u64 mmio_phys;
108 	u16 pci_seg;
109 	u16 info;
110 	u32 efr_attr;
111 
112 	/* Following only valid on IVHD type 11h and 40h */
113 	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
114 	u64 efr_reg2;
115 } __attribute__((packed));
116 
117 /*
118  * A device entry describing which devices a specific IOMMU translates and
119  * which requestor ids they use.
120  */
121 struct ivhd_entry {
122 	u8 type;
123 	u16 devid;
124 	u8 flags;
125 	struct_group(ext_hid,
126 		u32 ext;
127 		u32 hidh;
128 	);
129 	u64 cid;
130 	u8 uidf;
131 	u8 uidl;
132 	u8 uid;
133 } __attribute__((packed));
134 
135 /*
136  * An AMD IOMMU memory definition structure. It defines things like exclusion
137  * ranges for devices and regions that should be unity mapped.
138  */
139 struct ivmd_header {
140 	u8 type;
141 	u8 flags;
142 	u16 length;
143 	u16 devid;
144 	u16 aux;
145 	u16 pci_seg;
146 	u8  resv[6];
147 	u64 range_start;
148 	u64 range_length;
149 } __attribute__((packed));
150 
151 bool amd_iommu_dump;
152 bool amd_iommu_irq_remap __read_mostly;
153 
154 enum protection_domain_mode amd_iommu_pgtable = PD_MODE_V1;
155 /* Host page table level */
156 u8 amd_iommu_hpt_level;
157 /* Guest page table level */
158 int amd_iommu_gpt_level = PAGE_MODE_4_LEVEL;
159 
160 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
161 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
162 
163 static bool amd_iommu_detected;
164 static bool amd_iommu_disabled __initdata;
165 static bool amd_iommu_force_enable __initdata;
166 static bool amd_iommu_irtcachedis;
167 static int amd_iommu_target_ivhd_type;
168 
169 /* Global EFR and EFR2 registers */
170 u64 amd_iommu_efr;
171 u64 amd_iommu_efr2;
172 
173 /* Host (v1) page table is not supported*/
174 bool amd_iommu_hatdis;
175 
176 /* SNP is enabled on the system? */
177 bool amd_iommu_snp_en;
178 EXPORT_SYMBOL(amd_iommu_snp_en);
179 
180 LIST_HEAD(amd_iommu_pci_seg_list);	/* list of all PCI segments */
181 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the system */
182 LIST_HEAD(amd_ivhd_dev_flags_list);	/* list of all IVHD device entry settings */
183 
184 /* Number of IOMMUs present in the system */
185 static int amd_iommus_present;
186 
187 /* IOMMUs have a non-present cache? */
188 bool amd_iommu_np_cache __read_mostly;
189 bool amd_iommu_iotlb_sup __read_mostly = true;
190 
191 static bool amd_iommu_pc_present __read_mostly;
192 bool amdr_ivrs_remap_support __read_mostly;
193 
194 bool amd_iommu_force_isolation __read_mostly;
195 
196 unsigned long amd_iommu_pgsize_bitmap __ro_after_init = AMD_IOMMU_PGSIZES;
197 
198 enum iommu_init_state {
199 	IOMMU_START_STATE,
200 	IOMMU_IVRS_DETECTED,
201 	IOMMU_ACPI_FINISHED,
202 	IOMMU_ENABLED,
203 	IOMMU_PCI_INIT,
204 	IOMMU_INTERRUPTS_EN,
205 	IOMMU_INITIALIZED,
206 	IOMMU_NOT_FOUND,
207 	IOMMU_INIT_ERROR,
208 	IOMMU_CMDLINE_DISABLED,
209 };
210 
211 /* Early ioapic and hpet maps from kernel command line */
212 #define EARLY_MAP_SIZE		4
213 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
214 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
215 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
216 
217 static int __initdata early_ioapic_map_size;
218 static int __initdata early_hpet_map_size;
219 static int __initdata early_acpihid_map_size;
220 
221 static bool __initdata cmdline_maps;
222 
223 static enum iommu_init_state init_state = IOMMU_START_STATE;
224 
225 static int amd_iommu_enable_interrupts(void);
226 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
227 
228 static bool amd_iommu_pre_enabled = true;
229 
230 static u32 amd_iommu_ivinfo __initdata;
231 
232 bool translation_pre_enabled(struct amd_iommu *iommu)
233 {
234 	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
235 }
236 
237 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
238 {
239 	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
240 }
241 
242 static void init_translation_status(struct amd_iommu *iommu)
243 {
244 	u64 ctrl;
245 
246 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
247 	if (ctrl & (1<<CONTROL_IOMMU_EN))
248 		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
249 }
250 
251 int amd_iommu_get_num_iommus(void)
252 {
253 	return amd_iommus_present;
254 }
255 
256 bool amd_iommu_ht_range_ignore(void)
257 {
258 	return check_feature2(FEATURE_HT_RANGE_IGNORE);
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 	if (is_kdump_kernel())
410 		return;
411 
412 	entry = iommu_virt_to_phys(dev_table);
413 	entry |= (dev_table_size >> 12) - 1;
414 	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
415 			&entry, sizeof(entry));
416 }
417 
418 static void iommu_feature_set(struct amd_iommu *iommu, u64 val, u64 mask, u8 shift)
419 {
420 	u64 ctrl;
421 
422 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
423 	mask <<= shift;
424 	ctrl &= ~mask;
425 	ctrl |= (val << shift) & mask;
426 	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
427 }
428 
429 /* Generic functions to enable/disable certain features of the IOMMU. */
430 void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
431 {
432 	iommu_feature_set(iommu, 1ULL, 1ULL, bit);
433 }
434 
435 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
436 {
437 	iommu_feature_set(iommu, 0ULL, 1ULL, bit);
438 }
439 
440 /* Function to enable the hardware */
441 static void iommu_enable(struct amd_iommu *iommu)
442 {
443 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
444 }
445 
446 static void iommu_disable(struct amd_iommu *iommu)
447 {
448 	if (!iommu->mmio_base)
449 		return;
450 
451 	/* Disable command buffer */
452 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
453 
454 	/* Disable event logging and event interrupts */
455 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
456 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
457 
458 	/* Disable IOMMU GA_LOG */
459 	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
460 	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
461 
462 	/* Disable IOMMU PPR logging */
463 	iommu_feature_disable(iommu, CONTROL_PPRLOG_EN);
464 	iommu_feature_disable(iommu, CONTROL_PPRINT_EN);
465 
466 	/* Disable IOMMU hardware itself */
467 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
468 
469 	/* Clear IRTE cache disabling bit */
470 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
471 }
472 
473 /*
474  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
475  * the system has one.
476  */
477 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
478 {
479 	if (!request_mem_region(address, end, "amd_iommu")) {
480 		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
481 			address, end);
482 		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
483 		return NULL;
484 	}
485 
486 	return (u8 __iomem *)ioremap(address, end);
487 }
488 
489 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
490 {
491 	if (iommu->mmio_base)
492 		iounmap(iommu->mmio_base);
493 	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
494 }
495 
496 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
497 {
498 	u32 size = 0;
499 
500 	switch (h->type) {
501 	case 0x10:
502 		size = 24;
503 		break;
504 	case 0x11:
505 	case 0x40:
506 		size = 40;
507 		break;
508 	}
509 	return size;
510 }
511 
512 /****************************************************************************
513  *
514  * The functions below belong to the first pass of AMD IOMMU ACPI table
515  * parsing. In this pass we try to find out the highest device id this
516  * code has to handle. Upon this information the size of the shared data
517  * structures is determined later.
518  *
519  ****************************************************************************/
520 
521 /*
522  * This function calculates the length of a given IVHD entry
523  */
524 static inline int ivhd_entry_length(u8 *ivhd)
525 {
526 	u32 type = ((struct ivhd_entry *)ivhd)->type;
527 
528 	if (type < 0x80) {
529 		return 0x04 << (*ivhd >> 6);
530 	} else if (type == IVHD_DEV_ACPI_HID) {
531 		/* For ACPI_HID, offset 21 is uid len */
532 		return *((u8 *)ivhd + 21) + 22;
533 	}
534 	return 0;
535 }
536 
537 /*
538  * After reading the highest device id from the IOMMU PCI capability header
539  * this function looks if there is a higher device id defined in the ACPI table
540  */
541 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
542 {
543 	u8 *p = (void *)h, *end = (void *)h;
544 	struct ivhd_entry *dev;
545 	int last_devid = -EINVAL;
546 
547 	u32 ivhd_size = get_ivhd_header_size(h);
548 
549 	if (!ivhd_size) {
550 		pr_err("Unsupported IVHD type %#x\n", h->type);
551 		return -EINVAL;
552 	}
553 
554 	p += ivhd_size;
555 	end += h->length;
556 
557 	while (p < end) {
558 		dev = (struct ivhd_entry *)p;
559 		switch (dev->type) {
560 		case IVHD_DEV_ALL:
561 			/* Use maximum BDF value for DEV_ALL */
562 			return 0xffff;
563 		case IVHD_DEV_SELECT:
564 		case IVHD_DEV_RANGE_END:
565 		case IVHD_DEV_ALIAS:
566 		case IVHD_DEV_EXT_SELECT:
567 			/* all the above subfield types refer to device ids */
568 			if (dev->devid > last_devid)
569 				last_devid = dev->devid;
570 			break;
571 		default:
572 			break;
573 		}
574 		p += ivhd_entry_length(p);
575 	}
576 
577 	WARN_ON(p != end);
578 
579 	return last_devid;
580 }
581 
582 static int __init check_ivrs_checksum(struct acpi_table_header *table)
583 {
584 	int i;
585 	u8 checksum = 0, *p = (u8 *)table;
586 
587 	for (i = 0; i < table->length; ++i)
588 		checksum += p[i];
589 	if (checksum != 0) {
590 		/* ACPI table corrupt */
591 		pr_err(FW_BUG "IVRS invalid checksum\n");
592 		return -ENODEV;
593 	}
594 
595 	return 0;
596 }
597 
598 /*
599  * Iterate over all IVHD entries in the ACPI table and find the highest device
600  * id which we need to handle. This is the first of three functions which parse
601  * the ACPI table. So we check the checksum here.
602  */
603 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
604 {
605 	u8 *p = (u8 *)table, *end = (u8 *)table;
606 	struct ivhd_header *h;
607 	int last_devid, last_bdf = 0;
608 
609 	p += IVRS_HEADER_LENGTH;
610 
611 	end += table->length;
612 	while (p < end) {
613 		h = (struct ivhd_header *)p;
614 		if (h->pci_seg == pci_seg &&
615 		    h->type == amd_iommu_target_ivhd_type) {
616 			last_devid = find_last_devid_from_ivhd(h);
617 
618 			if (last_devid < 0)
619 				return -EINVAL;
620 			if (last_devid > last_bdf)
621 				last_bdf = last_devid;
622 		}
623 		p += h->length;
624 	}
625 	WARN_ON(p != end);
626 
627 	return last_bdf;
628 }
629 
630 /****************************************************************************
631  *
632  * The following functions belong to the code path which parses the ACPI table
633  * the second time. In this ACPI parsing iteration we allocate IOMMU specific
634  * data structures, initialize the per PCI segment device/alias/rlookup table
635  * and also basically initialize the hardware.
636  *
637  ****************************************************************************/
638 
639 /* Allocate per PCI segment device table */
640 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
641 {
642 	pci_seg->dev_table = iommu_alloc_pages_sz(GFP_KERNEL | GFP_DMA32,
643 						  pci_seg->dev_table_size);
644 	if (!pci_seg->dev_table)
645 		return -ENOMEM;
646 
647 	return 0;
648 }
649 
650 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
651 {
652 	if (is_kdump_kernel())
653 		memunmap((void *)pci_seg->dev_table);
654 	else
655 		iommu_free_pages(pci_seg->dev_table);
656 	pci_seg->dev_table = NULL;
657 }
658 
659 /* Allocate per PCI segment IOMMU rlookup table. */
660 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
661 {
662 	pci_seg->rlookup_table = kvcalloc(pci_seg->last_bdf + 1,
663 					  sizeof(*pci_seg->rlookup_table),
664 					  GFP_KERNEL);
665 	if (pci_seg->rlookup_table == NULL)
666 		return -ENOMEM;
667 
668 	return 0;
669 }
670 
671 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
672 {
673 	kvfree(pci_seg->rlookup_table);
674 	pci_seg->rlookup_table = NULL;
675 }
676 
677 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
678 {
679 	pci_seg->irq_lookup_table = kvcalloc(pci_seg->last_bdf + 1,
680 					     sizeof(*pci_seg->irq_lookup_table),
681 					     GFP_KERNEL);
682 	if (pci_seg->irq_lookup_table == NULL)
683 		return -ENOMEM;
684 
685 	return 0;
686 }
687 
688 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
689 {
690 	kvfree(pci_seg->irq_lookup_table);
691 	pci_seg->irq_lookup_table = NULL;
692 }
693 
694 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
695 {
696 	int i;
697 
698 	pci_seg->alias_table = kvmalloc_array(pci_seg->last_bdf + 1,
699 					      sizeof(*pci_seg->alias_table),
700 					      GFP_KERNEL);
701 	if (!pci_seg->alias_table)
702 		return -ENOMEM;
703 
704 	/*
705 	 * let all alias entries point to itself
706 	 */
707 	for (i = 0; i <= pci_seg->last_bdf; ++i)
708 		pci_seg->alias_table[i] = i;
709 
710 	return 0;
711 }
712 
713 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
714 {
715 	kvfree(pci_seg->alias_table);
716 	pci_seg->alias_table = NULL;
717 }
718 
719 static inline void *iommu_memremap(unsigned long paddr, size_t size)
720 {
721 	phys_addr_t phys;
722 
723 	if (!paddr)
724 		return NULL;
725 
726 	/*
727 	 * Obtain true physical address in kdump kernel when SME is enabled.
728 	 * Currently, previous kernel with SME enabled and kdump kernel
729 	 * with SME support disabled is not supported.
730 	 */
731 	phys = __sme_clr(paddr);
732 
733 	if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
734 		return (__force void *)ioremap_encrypted(phys, size);
735 	else
736 		return memremap(phys, size, MEMREMAP_WB);
737 }
738 
739 /*
740  * Allocates the command buffer. This buffer is per AMD IOMMU. We can
741  * write commands to that buffer later and the IOMMU will execute them
742  * asynchronously
743  */
744 static int __init alloc_command_buffer(struct amd_iommu *iommu)
745 {
746 	iommu->cmd_buf = iommu_alloc_pages_sz(GFP_KERNEL, CMD_BUFFER_SIZE);
747 
748 	return iommu->cmd_buf ? 0 : -ENOMEM;
749 }
750 
751 /*
752  * Interrupt handler has processed all pending events and adjusted head
753  * and tail pointer. Reset overflow mask and restart logging again.
754  */
755 void amd_iommu_restart_log(struct amd_iommu *iommu, const char *evt_type,
756 			   u8 cntrl_intr, u8 cntrl_log,
757 			   u32 status_run_mask, u32 status_overflow_mask)
758 {
759 	u32 status;
760 
761 	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
762 	if (status & status_run_mask)
763 		return;
764 
765 	pr_info_ratelimited("IOMMU %s log restarting\n", evt_type);
766 
767 	iommu_feature_disable(iommu, cntrl_log);
768 	iommu_feature_disable(iommu, cntrl_intr);
769 
770 	writel(status_overflow_mask, iommu->mmio_base + MMIO_STATUS_OFFSET);
771 
772 	iommu_feature_enable(iommu, cntrl_intr);
773 	iommu_feature_enable(iommu, cntrl_log);
774 }
775 
776 /*
777  * This function restarts event logging in case the IOMMU experienced
778  * an event log buffer overflow.
779  */
780 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
781 {
782 	amd_iommu_restart_log(iommu, "Event", CONTROL_EVT_INT_EN,
783 			      CONTROL_EVT_LOG_EN, MMIO_STATUS_EVT_RUN_MASK,
784 			      MMIO_STATUS_EVT_OVERFLOW_MASK);
785 }
786 
787 /*
788  * This function restarts event logging in case the IOMMU experienced
789  * GA log overflow.
790  */
791 void amd_iommu_restart_ga_log(struct amd_iommu *iommu)
792 {
793 	amd_iommu_restart_log(iommu, "GA", CONTROL_GAINT_EN,
794 			      CONTROL_GALOG_EN, MMIO_STATUS_GALOG_RUN_MASK,
795 			      MMIO_STATUS_GALOG_OVERFLOW_MASK);
796 }
797 
798 /*
799  * This function resets the command buffer if the IOMMU stopped fetching
800  * commands from it.
801  */
802 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
803 {
804 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
805 
806 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
807 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
808 	iommu->cmd_buf_head = 0;
809 	iommu->cmd_buf_tail = 0;
810 
811 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
812 }
813 
814 /*
815  * This function writes the command buffer address to the hardware and
816  * enables it.
817  */
818 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
819 {
820 	u64 entry;
821 
822 	BUG_ON(iommu->cmd_buf == NULL);
823 
824 	if (!is_kdump_kernel()) {
825 		/*
826 		 * Command buffer is re-used for kdump kernel and setting
827 		 * of MMIO register is not required.
828 		 */
829 		entry = iommu_virt_to_phys(iommu->cmd_buf);
830 		entry |= MMIO_CMD_SIZE_512;
831 		memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
832 			    &entry, sizeof(entry));
833 	}
834 
835 	amd_iommu_reset_cmd_buffer(iommu);
836 }
837 
838 /*
839  * This function disables the command buffer
840  */
841 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
842 {
843 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
844 }
845 
846 static void __init free_command_buffer(struct amd_iommu *iommu)
847 {
848 	iommu_free_pages(iommu->cmd_buf);
849 }
850 
851 void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
852 				  size_t size)
853 {
854 	void *buf;
855 
856 	size = PAGE_ALIGN(size);
857 	buf = iommu_alloc_pages_sz(gfp, size);
858 	if (!buf)
859 		return NULL;
860 	if (check_feature(FEATURE_SNP) &&
861 	    set_memory_4k((unsigned long)buf, size / PAGE_SIZE)) {
862 		iommu_free_pages(buf);
863 		return NULL;
864 	}
865 
866 	return buf;
867 }
868 
869 /* allocates the memory where the IOMMU will log its events to */
870 static int __init alloc_event_buffer(struct amd_iommu *iommu)
871 {
872 	iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL,
873 					      EVT_BUFFER_SIZE);
874 
875 	return iommu->evt_buf ? 0 : -ENOMEM;
876 }
877 
878 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
879 {
880 	u64 entry;
881 
882 	BUG_ON(iommu->evt_buf == NULL);
883 
884 	if (!is_kdump_kernel()) {
885 		/*
886 		 * Event buffer is re-used for kdump kernel and setting
887 		 * of MMIO register is not required.
888 		 */
889 		entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
890 		memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
891 			    &entry, sizeof(entry));
892 	}
893 
894 	/* set head and tail to zero manually */
895 	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
896 	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
897 
898 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
899 }
900 
901 /*
902  * This function disables the event log buffer
903  */
904 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
905 {
906 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
907 }
908 
909 static void __init free_event_buffer(struct amd_iommu *iommu)
910 {
911 	iommu_free_pages(iommu->evt_buf);
912 }
913 
914 static void free_ga_log(struct amd_iommu *iommu)
915 {
916 #ifdef CONFIG_IRQ_REMAP
917 	iommu_free_pages(iommu->ga_log);
918 	iommu_free_pages(iommu->ga_log_tail);
919 #endif
920 }
921 
922 #ifdef CONFIG_IRQ_REMAP
923 static int iommu_ga_log_enable(struct amd_iommu *iommu)
924 {
925 	u32 status, i;
926 	u64 entry;
927 
928 	if (!iommu->ga_log)
929 		return -EINVAL;
930 
931 	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
932 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
933 		    &entry, sizeof(entry));
934 	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
935 		 (BIT_ULL(52)-1)) & ~7ULL;
936 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
937 		    &entry, sizeof(entry));
938 	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
939 	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
940 
941 
942 	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
943 	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
944 
945 	for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
946 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
947 		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
948 			break;
949 		udelay(10);
950 	}
951 
952 	if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
953 		return -EINVAL;
954 
955 	return 0;
956 }
957 
958 static int iommu_init_ga_log(struct amd_iommu *iommu)
959 {
960 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
961 		return 0;
962 
963 	iommu->ga_log = iommu_alloc_pages_sz(GFP_KERNEL, GA_LOG_SIZE);
964 	if (!iommu->ga_log)
965 		goto err_out;
966 
967 	iommu->ga_log_tail = iommu_alloc_pages_sz(GFP_KERNEL, 8);
968 	if (!iommu->ga_log_tail)
969 		goto err_out;
970 
971 	return 0;
972 err_out:
973 	free_ga_log(iommu);
974 	return -EINVAL;
975 }
976 #endif /* CONFIG_IRQ_REMAP */
977 
978 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
979 {
980 	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL, 1);
981 	if (!iommu->cmd_sem)
982 		return -ENOMEM;
983 	iommu->cmd_sem_paddr = iommu_virt_to_phys((void *)iommu->cmd_sem);
984 	return 0;
985 }
986 
987 static int __init remap_event_buffer(struct amd_iommu *iommu)
988 {
989 	u64 paddr;
990 
991 	pr_info_once("Re-using event buffer from the previous kernel\n");
992 	paddr = readq(iommu->mmio_base + MMIO_EVT_BUF_OFFSET) & PM_ADDR_MASK;
993 	iommu->evt_buf = iommu_memremap(paddr, EVT_BUFFER_SIZE);
994 
995 	return iommu->evt_buf ? 0 : -ENOMEM;
996 }
997 
998 static int __init remap_command_buffer(struct amd_iommu *iommu)
999 {
1000 	u64 paddr;
1001 
1002 	pr_info_once("Re-using command buffer from the previous kernel\n");
1003 	paddr = readq(iommu->mmio_base + MMIO_CMD_BUF_OFFSET) & PM_ADDR_MASK;
1004 	iommu->cmd_buf = iommu_memremap(paddr, CMD_BUFFER_SIZE);
1005 
1006 	return iommu->cmd_buf ? 0 : -ENOMEM;
1007 }
1008 
1009 static int __init remap_or_alloc_cwwb_sem(struct amd_iommu *iommu)
1010 {
1011 	u64 paddr;
1012 
1013 	if (check_feature(FEATURE_SNP)) {
1014 		/*
1015 		 * When SNP is enabled, the exclusion base register is used for the
1016 		 * completion wait buffer (CWB) address. Read and re-use it.
1017 		 */
1018 		pr_info_once("Re-using CWB buffers from the previous kernel\n");
1019 		paddr = readq(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET) & PM_ADDR_MASK;
1020 		iommu->cmd_sem = iommu_memremap(paddr, PAGE_SIZE);
1021 		if (!iommu->cmd_sem)
1022 			return -ENOMEM;
1023 		iommu->cmd_sem_paddr = paddr;
1024 	} else {
1025 		return alloc_cwwb_sem(iommu);
1026 	}
1027 
1028 	return 0;
1029 }
1030 
1031 static int __init alloc_iommu_buffers(struct amd_iommu *iommu)
1032 {
1033 	int ret;
1034 
1035 	/*
1036 	 * Reuse/Remap the previous kernel's allocated completion wait
1037 	 * command and event buffers for kdump boot.
1038 	 */
1039 	if (is_kdump_kernel()) {
1040 		ret = remap_or_alloc_cwwb_sem(iommu);
1041 		if (ret)
1042 			return ret;
1043 
1044 		ret = remap_command_buffer(iommu);
1045 		if (ret)
1046 			return ret;
1047 
1048 		ret = remap_event_buffer(iommu);
1049 		if (ret)
1050 			return ret;
1051 	} else {
1052 		ret = alloc_cwwb_sem(iommu);
1053 		if (ret)
1054 			return ret;
1055 
1056 		ret = alloc_command_buffer(iommu);
1057 		if (ret)
1058 			return ret;
1059 
1060 		ret = alloc_event_buffer(iommu);
1061 		if (ret)
1062 			return ret;
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 static void __init free_cwwb_sem(struct amd_iommu *iommu)
1069 {
1070 	if (iommu->cmd_sem)
1071 		iommu_free_pages((void *)iommu->cmd_sem);
1072 }
1073 static void __init unmap_cwwb_sem(struct amd_iommu *iommu)
1074 {
1075 	if (iommu->cmd_sem) {
1076 		if (check_feature(FEATURE_SNP))
1077 			memunmap((void *)iommu->cmd_sem);
1078 		else
1079 			iommu_free_pages((void *)iommu->cmd_sem);
1080 	}
1081 }
1082 
1083 static void __init unmap_command_buffer(struct amd_iommu *iommu)
1084 {
1085 	memunmap((void *)iommu->cmd_buf);
1086 }
1087 
1088 static void __init unmap_event_buffer(struct amd_iommu *iommu)
1089 {
1090 	memunmap(iommu->evt_buf);
1091 }
1092 
1093 static void __init free_iommu_buffers(struct amd_iommu *iommu)
1094 {
1095 	if (is_kdump_kernel()) {
1096 		unmap_cwwb_sem(iommu);
1097 		unmap_command_buffer(iommu);
1098 		unmap_event_buffer(iommu);
1099 	} else {
1100 		free_cwwb_sem(iommu);
1101 		free_command_buffer(iommu);
1102 		free_event_buffer(iommu);
1103 	}
1104 }
1105 
1106 static void iommu_enable_xt(struct amd_iommu *iommu)
1107 {
1108 #ifdef CONFIG_IRQ_REMAP
1109 	/*
1110 	 * XT mode (32-bit APIC destination ID) requires
1111 	 * GA mode (128-bit IRTE support) as a prerequisite.
1112 	 */
1113 	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
1114 	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1115 		iommu_feature_enable(iommu, CONTROL_XT_EN);
1116 #endif /* CONFIG_IRQ_REMAP */
1117 }
1118 
1119 static void iommu_enable_gt(struct amd_iommu *iommu)
1120 {
1121 	if (!check_feature(FEATURE_GT))
1122 		return;
1123 
1124 	iommu_feature_enable(iommu, CONTROL_GT_EN);
1125 
1126 	/*
1127 	 * This feature needs to be enabled prior to a call
1128 	 * to iommu_snp_enable(). Since this function is called
1129 	 * in early_enable_iommu(), it is safe to enable here.
1130 	 */
1131 	if (check_feature2(FEATURE_GCR3TRPMODE))
1132 		iommu_feature_enable(iommu, CONTROL_GCR3TRPMODE);
1133 }
1134 
1135 /* sets a specific bit in the device table entry. */
1136 static void set_dte_bit(struct dev_table_entry *dte, u8 bit)
1137 {
1138 	int i = (bit >> 6) & 0x03;
1139 	int _bit = bit & 0x3f;
1140 
1141 	dte->data[i] |= (1UL << _bit);
1142 }
1143 
1144 static bool __reuse_device_table(struct amd_iommu *iommu)
1145 {
1146 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1147 	struct dev_table_entry *old_dev_tbl_entry;
1148 	u32 lo, hi, old_devtb_size, devid;
1149 	phys_addr_t old_devtb_phys;
1150 	u16 dom_id;
1151 	bool dte_v;
1152 	u64 entry;
1153 
1154 	/* Each IOMMU use separate device table with the same size */
1155 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1156 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1157 	entry = (((u64) hi) << 32) + lo;
1158 
1159 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1160 	if (old_devtb_size != pci_seg->dev_table_size) {
1161 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1162 			iommu->index);
1163 		return false;
1164 	}
1165 
1166 	/*
1167 	 * When SME is enabled in the first kernel, the entry includes the
1168 	 * memory encryption mask(sme_me_mask), we must remove the memory
1169 	 * encryption mask to obtain the true physical address in kdump kernel.
1170 	 */
1171 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1172 
1173 	if (old_devtb_phys >= 0x100000000ULL) {
1174 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1175 		return false;
1176 	}
1177 
1178 	/*
1179 	 * Re-use the previous kernel's device table for kdump.
1180 	 */
1181 	pci_seg->old_dev_tbl_cpy = iommu_memremap(old_devtb_phys, pci_seg->dev_table_size);
1182 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1183 		pr_err("Failed to remap memory for reusing old device table!\n");
1184 		return false;
1185 	}
1186 
1187 	for (devid = 0; devid <= pci_seg->last_bdf; devid++) {
1188 		old_dev_tbl_entry = &pci_seg->old_dev_tbl_cpy[devid];
1189 		dte_v = FIELD_GET(DTE_FLAG_V, old_dev_tbl_entry->data[0]);
1190 		dom_id = FIELD_GET(DTE_DOMID_MASK, old_dev_tbl_entry->data[1]);
1191 
1192 		if (!dte_v || !dom_id)
1193 			continue;
1194 		/*
1195 		 * ID reservation can fail with -ENOSPC when there
1196 		 * are multiple devices present in the same domain,
1197 		 * hence check only for -ENOMEM.
1198 		 */
1199 		if (amd_iommu_pdom_id_reserve(dom_id, GFP_KERNEL) == -ENOMEM)
1200 			return false;
1201 	}
1202 
1203 	return true;
1204 }
1205 
1206 static bool reuse_device_table(void)
1207 {
1208 	struct amd_iommu *iommu;
1209 	struct amd_iommu_pci_seg *pci_seg;
1210 
1211 	if (!amd_iommu_pre_enabled)
1212 		return false;
1213 
1214 	pr_warn("Translation is already enabled - trying to reuse translation structures\n");
1215 
1216 	/*
1217 	 * All IOMMUs within PCI segment shares common device table.
1218 	 * Hence reuse device table only once per PCI segment.
1219 	 */
1220 	for_each_pci_segment(pci_seg) {
1221 		for_each_iommu(iommu) {
1222 			if (pci_seg->id != iommu->pci_seg->id)
1223 				continue;
1224 			if (!__reuse_device_table(iommu))
1225 				return false;
1226 			break;
1227 		}
1228 	}
1229 
1230 	return true;
1231 }
1232 
1233 struct dev_table_entry *amd_iommu_get_ivhd_dte_flags(u16 segid, u16 devid)
1234 {
1235 	struct ivhd_dte_flags *e;
1236 	unsigned int best_len = UINT_MAX;
1237 	struct dev_table_entry *dte = NULL;
1238 
1239 	for_each_ivhd_dte_flags(e) {
1240 		/*
1241 		 * Need to go through the whole list to find the smallest range,
1242 		 * which contains the devid.
1243 		 */
1244 		if ((e->segid == segid) &&
1245 		    (e->devid_first <= devid) && (devid <= e->devid_last)) {
1246 			unsigned int len = e->devid_last - e->devid_first;
1247 
1248 			if (len < best_len) {
1249 				dte = &(e->dte);
1250 				best_len = len;
1251 			}
1252 		}
1253 	}
1254 	return dte;
1255 }
1256 
1257 static bool search_ivhd_dte_flags(u16 segid, u16 first, u16 last)
1258 {
1259 	struct ivhd_dte_flags *e;
1260 
1261 	for_each_ivhd_dte_flags(e) {
1262 		if ((e->segid == segid) &&
1263 		    (e->devid_first == first) &&
1264 		    (e->devid_last == last))
1265 			return true;
1266 	}
1267 	return false;
1268 }
1269 
1270 /*
1271  * This function takes the device specific flags read from the ACPI
1272  * table and sets up the device table entry with that information
1273  */
1274 static void __init
1275 set_dev_entry_from_acpi_range(struct amd_iommu *iommu, u16 first, u16 last,
1276 			      u32 flags, u32 ext_flags)
1277 {
1278 	int i;
1279 	struct dev_table_entry dte = {};
1280 
1281 	/* Parse IVHD DTE setting flags and store information */
1282 	if (flags) {
1283 		struct ivhd_dte_flags *d;
1284 
1285 		if (search_ivhd_dte_flags(iommu->pci_seg->id, first, last))
1286 			return;
1287 
1288 		d = kzalloc(sizeof(struct ivhd_dte_flags), GFP_KERNEL);
1289 		if (!d)
1290 			return;
1291 
1292 		pr_debug("%s: devid range %#x:%#x\n", __func__, first, last);
1293 
1294 		if (flags & ACPI_DEVFLAG_INITPASS)
1295 			set_dte_bit(&dte, DEV_ENTRY_INIT_PASS);
1296 		if (flags & ACPI_DEVFLAG_EXTINT)
1297 			set_dte_bit(&dte, DEV_ENTRY_EINT_PASS);
1298 		if (flags & ACPI_DEVFLAG_NMI)
1299 			set_dte_bit(&dte, DEV_ENTRY_NMI_PASS);
1300 		if (flags & ACPI_DEVFLAG_SYSMGT1)
1301 			set_dte_bit(&dte, DEV_ENTRY_SYSMGT1);
1302 		if (flags & ACPI_DEVFLAG_SYSMGT2)
1303 			set_dte_bit(&dte, DEV_ENTRY_SYSMGT2);
1304 		if (flags & ACPI_DEVFLAG_LINT0)
1305 			set_dte_bit(&dte, DEV_ENTRY_LINT0_PASS);
1306 		if (flags & ACPI_DEVFLAG_LINT1)
1307 			set_dte_bit(&dte, DEV_ENTRY_LINT1_PASS);
1308 
1309 		/* Apply erratum 63, which needs info in initial_dte */
1310 		if (FIELD_GET(DTE_DATA1_SYSMGT_MASK, dte.data[1]) == 0x1)
1311 			dte.data[0] |= DTE_FLAG_IW;
1312 
1313 		memcpy(&d->dte, &dte, sizeof(dte));
1314 		d->segid = iommu->pci_seg->id;
1315 		d->devid_first = first;
1316 		d->devid_last = last;
1317 		list_add_tail(&d->list, &amd_ivhd_dev_flags_list);
1318 	}
1319 
1320 	for (i = first; i <= last; i++)  {
1321 		if (flags) {
1322 			struct dev_table_entry *dev_table = get_dev_table(iommu);
1323 
1324 			memcpy(&dev_table[i], &dte, sizeof(dte));
1325 		}
1326 		amd_iommu_set_rlookup_table(iommu, i);
1327 	}
1328 }
1329 
1330 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1331 					   u16 devid, u32 flags, u32 ext_flags)
1332 {
1333 	set_dev_entry_from_acpi_range(iommu, devid, devid, flags, ext_flags);
1334 }
1335 
1336 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line)
1337 {
1338 	struct devid_map *entry;
1339 	struct list_head *list;
1340 
1341 	if (type == IVHD_SPECIAL_IOAPIC)
1342 		list = &ioapic_map;
1343 	else if (type == IVHD_SPECIAL_HPET)
1344 		list = &hpet_map;
1345 	else
1346 		return -EINVAL;
1347 
1348 	list_for_each_entry(entry, list, list) {
1349 		if (!(entry->id == id && entry->cmd_line))
1350 			continue;
1351 
1352 		pr_info("Command-line override present for %s id %d - ignoring\n",
1353 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1354 
1355 		*devid = entry->devid;
1356 
1357 		return 0;
1358 	}
1359 
1360 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1361 	if (!entry)
1362 		return -ENOMEM;
1363 
1364 	entry->id	= id;
1365 	entry->devid	= *devid;
1366 	entry->cmd_line	= cmd_line;
1367 
1368 	list_add_tail(&entry->list, list);
1369 
1370 	return 0;
1371 }
1372 
1373 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid,
1374 				      bool cmd_line)
1375 {
1376 	struct acpihid_map_entry *entry;
1377 	struct list_head *list = &acpihid_map;
1378 
1379 	list_for_each_entry(entry, list, list) {
1380 		if (strcmp(entry->hid, hid) ||
1381 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1382 		    !entry->cmd_line)
1383 			continue;
1384 
1385 		pr_info("Command-line override for hid:%s uid:%s\n",
1386 			hid, uid);
1387 		*devid = entry->devid;
1388 		return 0;
1389 	}
1390 
1391 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1392 	if (!entry)
1393 		return -ENOMEM;
1394 
1395 	memcpy(entry->uid, uid, strlen(uid));
1396 	memcpy(entry->hid, hid, strlen(hid));
1397 	entry->devid = *devid;
1398 	entry->cmd_line	= cmd_line;
1399 	entry->root_devid = (entry->devid & (~0x7));
1400 
1401 	pr_info("%s, add hid:%s, uid:%s, rdevid:%#x\n",
1402 		entry->cmd_line ? "cmd" : "ivrs",
1403 		entry->hid, entry->uid, entry->root_devid);
1404 
1405 	list_add_tail(&entry->list, list);
1406 	return 0;
1407 }
1408 
1409 static int __init add_early_maps(void)
1410 {
1411 	int i, ret;
1412 
1413 	for (i = 0; i < early_ioapic_map_size; ++i) {
1414 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1415 					 early_ioapic_map[i].id,
1416 					 &early_ioapic_map[i].devid,
1417 					 early_ioapic_map[i].cmd_line);
1418 		if (ret)
1419 			return ret;
1420 	}
1421 
1422 	for (i = 0; i < early_hpet_map_size; ++i) {
1423 		ret = add_special_device(IVHD_SPECIAL_HPET,
1424 					 early_hpet_map[i].id,
1425 					 &early_hpet_map[i].devid,
1426 					 early_hpet_map[i].cmd_line);
1427 		if (ret)
1428 			return ret;
1429 	}
1430 
1431 	for (i = 0; i < early_acpihid_map_size; ++i) {
1432 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1433 					  early_acpihid_map[i].uid,
1434 					  &early_acpihid_map[i].devid,
1435 					  early_acpihid_map[i].cmd_line);
1436 		if (ret)
1437 			return ret;
1438 	}
1439 
1440 	return 0;
1441 }
1442 
1443 /*
1444  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1445  * initializes the hardware and our data structures with it.
1446  */
1447 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1448 					struct ivhd_header *h)
1449 {
1450 	u8 *p = (u8 *)h;
1451 	u8 *end = p, flags = 0;
1452 	u16 devid = 0, devid_start = 0, devid_to = 0, seg_id;
1453 	u32 dev_i, ext_flags = 0;
1454 	bool alias = false;
1455 	struct ivhd_entry *e;
1456 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1457 	u32 ivhd_size;
1458 	int ret;
1459 
1460 
1461 	ret = add_early_maps();
1462 	if (ret)
1463 		return ret;
1464 
1465 	amd_iommu_apply_ivrs_quirks();
1466 
1467 	/*
1468 	 * First save the recommended feature enable bits from ACPI
1469 	 */
1470 	iommu->acpi_flags = h->flags;
1471 
1472 	/*
1473 	 * Done. Now parse the device entries
1474 	 */
1475 	ivhd_size = get_ivhd_header_size(h);
1476 	if (!ivhd_size) {
1477 		pr_err("Unsupported IVHD type %#x\n", h->type);
1478 		return -EINVAL;
1479 	}
1480 
1481 	p += ivhd_size;
1482 
1483 	end += h->length;
1484 
1485 
1486 	while (p < end) {
1487 		e = (struct ivhd_entry *)p;
1488 		seg_id = pci_seg->id;
1489 
1490 		switch (e->type) {
1491 		case IVHD_DEV_ALL:
1492 
1493 			DUMP_printk("  DEV_ALL\t\t\tsetting: %#02x\n", e->flags);
1494 			set_dev_entry_from_acpi_range(iommu, 0, pci_seg->last_bdf, e->flags, 0);
1495 			break;
1496 		case IVHD_DEV_SELECT:
1497 
1498 			DUMP_printk("  DEV_SELECT\t\t\tdevid: %04x:%02x:%02x.%x flags: %#02x\n",
1499 				    seg_id, PCI_BUS_NUM(e->devid),
1500 				    PCI_SLOT(e->devid),
1501 				    PCI_FUNC(e->devid),
1502 				    e->flags);
1503 
1504 			devid = e->devid;
1505 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1506 			break;
1507 		case IVHD_DEV_SELECT_RANGE_START:
1508 
1509 			DUMP_printk("  DEV_SELECT_RANGE_START\tdevid: %04x:%02x:%02x.%x flags: %#02x\n",
1510 				    seg_id, PCI_BUS_NUM(e->devid),
1511 				    PCI_SLOT(e->devid),
1512 				    PCI_FUNC(e->devid),
1513 				    e->flags);
1514 
1515 			devid_start = e->devid;
1516 			flags = e->flags;
1517 			ext_flags = 0;
1518 			alias = false;
1519 			break;
1520 		case IVHD_DEV_ALIAS:
1521 
1522 			DUMP_printk("  DEV_ALIAS\t\t\tdevid: %04x:%02x:%02x.%x flags: %#02x devid_to: %02x:%02x.%x\n",
1523 				    seg_id, PCI_BUS_NUM(e->devid),
1524 				    PCI_SLOT(e->devid),
1525 				    PCI_FUNC(e->devid),
1526 				    e->flags,
1527 				    PCI_BUS_NUM(e->ext >> 8),
1528 				    PCI_SLOT(e->ext >> 8),
1529 				    PCI_FUNC(e->ext >> 8));
1530 
1531 			devid = e->devid;
1532 			devid_to = e->ext >> 8;
1533 			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1534 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1535 			pci_seg->alias_table[devid] = devid_to;
1536 			break;
1537 		case IVHD_DEV_ALIAS_RANGE:
1538 
1539 			DUMP_printk("  DEV_ALIAS_RANGE\t\tdevid: %04x:%02x:%02x.%x flags: %#02x devid_to: %04x:%02x:%02x.%x\n",
1540 				    seg_id, PCI_BUS_NUM(e->devid),
1541 				    PCI_SLOT(e->devid),
1542 				    PCI_FUNC(e->devid),
1543 				    e->flags,
1544 				    seg_id, PCI_BUS_NUM(e->ext >> 8),
1545 				    PCI_SLOT(e->ext >> 8),
1546 				    PCI_FUNC(e->ext >> 8));
1547 
1548 			devid_start = e->devid;
1549 			flags = e->flags;
1550 			devid_to = e->ext >> 8;
1551 			ext_flags = 0;
1552 			alias = true;
1553 			break;
1554 		case IVHD_DEV_EXT_SELECT:
1555 
1556 			DUMP_printk("  DEV_EXT_SELECT\t\tdevid: %04x:%02x:%02x.%x flags: %#02x ext: %08x\n",
1557 				    seg_id, PCI_BUS_NUM(e->devid),
1558 				    PCI_SLOT(e->devid),
1559 				    PCI_FUNC(e->devid),
1560 				    e->flags, e->ext);
1561 
1562 			devid = e->devid;
1563 			set_dev_entry_from_acpi(iommu, devid, e->flags,
1564 						e->ext);
1565 			break;
1566 		case IVHD_DEV_EXT_SELECT_RANGE:
1567 
1568 			DUMP_printk("  DEV_EXT_SELECT_RANGE\tdevid: %04x:%02x:%02x.%x flags: %#02x ext: %08x\n",
1569 				    seg_id, PCI_BUS_NUM(e->devid),
1570 				    PCI_SLOT(e->devid),
1571 				    PCI_FUNC(e->devid),
1572 				    e->flags, e->ext);
1573 
1574 			devid_start = e->devid;
1575 			flags = e->flags;
1576 			ext_flags = e->ext;
1577 			alias = false;
1578 			break;
1579 		case IVHD_DEV_RANGE_END:
1580 
1581 			DUMP_printk("  DEV_RANGE_END\t\tdevid: %04x:%02x:%02x.%x\n",
1582 				    seg_id, PCI_BUS_NUM(e->devid),
1583 				    PCI_SLOT(e->devid),
1584 				    PCI_FUNC(e->devid));
1585 
1586 			devid = e->devid;
1587 			if (alias) {
1588 				for (dev_i = devid_start; dev_i <= devid; ++dev_i)
1589 					pci_seg->alias_table[dev_i] = devid_to;
1590 				set_dev_entry_from_acpi(iommu, devid_to, flags, ext_flags);
1591 			}
1592 			set_dev_entry_from_acpi_range(iommu, devid_start, devid, flags, ext_flags);
1593 			break;
1594 		case IVHD_DEV_SPECIAL: {
1595 			u8 handle, type;
1596 			const char *var;
1597 			u32 devid;
1598 			int ret;
1599 
1600 			handle = e->ext & 0xff;
1601 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8));
1602 			type   = (e->ext >> 24) & 0xff;
1603 
1604 			if (type == IVHD_SPECIAL_IOAPIC)
1605 				var = "IOAPIC";
1606 			else if (type == IVHD_SPECIAL_HPET)
1607 				var = "HPET";
1608 			else
1609 				var = "UNKNOWN";
1610 
1611 			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x, flags: %#02x\n",
1612 				    var, (int)handle,
1613 				    seg_id, PCI_BUS_NUM(devid),
1614 				    PCI_SLOT(devid),
1615 				    PCI_FUNC(devid),
1616 				    e->flags);
1617 
1618 			ret = add_special_device(type, handle, &devid, false);
1619 			if (ret)
1620 				return ret;
1621 
1622 			/*
1623 			 * add_special_device might update the devid in case a
1624 			 * command-line override is present. So call
1625 			 * set_dev_entry_from_acpi after add_special_device.
1626 			 */
1627 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1628 
1629 			break;
1630 		}
1631 		case IVHD_DEV_ACPI_HID: {
1632 			u32 devid;
1633 			u8 hid[ACPIHID_HID_LEN];
1634 			u8 uid[ACPIHID_UID_LEN];
1635 			int ret;
1636 
1637 			if (h->type != 0x40) {
1638 				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1639 				       e->type);
1640 				break;
1641 			}
1642 
1643 			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1644 			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1645 			hid[ACPIHID_HID_LEN - 1] = '\0';
1646 
1647 			if (!(*hid)) {
1648 				pr_err(FW_BUG "Invalid HID.\n");
1649 				break;
1650 			}
1651 
1652 			uid[0] = '\0';
1653 			switch (e->uidf) {
1654 			case UID_NOT_PRESENT:
1655 
1656 				if (e->uidl != 0)
1657 					pr_warn(FW_BUG "Invalid UID length.\n");
1658 
1659 				break;
1660 			case UID_IS_INTEGER:
1661 
1662 				sprintf(uid, "%d", e->uid);
1663 
1664 				break;
1665 			case UID_IS_CHARACTER:
1666 
1667 				memcpy(uid, &e->uid, e->uidl);
1668 				uid[e->uidl] = '\0';
1669 
1670 				break;
1671 			default:
1672 				break;
1673 			}
1674 
1675 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid);
1676 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x, flags: %#02x\n",
1677 				    hid, uid, seg_id,
1678 				    PCI_BUS_NUM(devid),
1679 				    PCI_SLOT(devid),
1680 				    PCI_FUNC(devid),
1681 				    e->flags);
1682 
1683 			flags = e->flags;
1684 
1685 			ret = add_acpi_hid_device(hid, uid, &devid, false);
1686 			if (ret)
1687 				return ret;
1688 
1689 			/*
1690 			 * add_special_device might update the devid in case a
1691 			 * command-line override is present. So call
1692 			 * set_dev_entry_from_acpi after add_special_device.
1693 			 */
1694 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1695 
1696 			break;
1697 		}
1698 		default:
1699 			break;
1700 		}
1701 
1702 		p += ivhd_entry_length(p);
1703 	}
1704 
1705 	return 0;
1706 }
1707 
1708 /* Allocate PCI segment data structure */
1709 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1710 					  struct acpi_table_header *ivrs_base)
1711 {
1712 	struct amd_iommu_pci_seg *pci_seg;
1713 	int last_bdf;
1714 
1715 	/*
1716 	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1717 	 * handle in this PCI segment. Upon this information the shared data
1718 	 * structures for the PCI segments in the system will be allocated.
1719 	 */
1720 	last_bdf = find_last_devid_acpi(ivrs_base, id);
1721 	if (last_bdf < 0)
1722 		return NULL;
1723 
1724 	pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1725 	if (pci_seg == NULL)
1726 		return NULL;
1727 
1728 	pci_seg->last_bdf = last_bdf;
1729 	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1730 	pci_seg->dev_table_size =
1731 		max(roundup_pow_of_two((last_bdf + 1) * DEV_TABLE_ENTRY_SIZE),
1732 		    SZ_4K);
1733 
1734 	pci_seg->id = id;
1735 	init_llist_head(&pci_seg->dev_data_list);
1736 	INIT_LIST_HEAD(&pci_seg->unity_map);
1737 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1738 
1739 	if (alloc_dev_table(pci_seg))
1740 		goto err_free_pci_seg;
1741 	if (alloc_alias_table(pci_seg))
1742 		goto err_free_dev_table;
1743 	if (alloc_rlookup_table(pci_seg))
1744 		goto err_free_alias_table;
1745 
1746 	return pci_seg;
1747 
1748 err_free_alias_table:
1749 	free_alias_table(pci_seg);
1750 err_free_dev_table:
1751 	free_dev_table(pci_seg);
1752 err_free_pci_seg:
1753 	list_del(&pci_seg->list);
1754 	kfree(pci_seg);
1755 	return NULL;
1756 }
1757 
1758 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1759 					struct acpi_table_header *ivrs_base)
1760 {
1761 	struct amd_iommu_pci_seg *pci_seg;
1762 
1763 	for_each_pci_segment(pci_seg) {
1764 		if (pci_seg->id == id)
1765 			return pci_seg;
1766 	}
1767 
1768 	return alloc_pci_segment(id, ivrs_base);
1769 }
1770 
1771 static void __init free_pci_segments(void)
1772 {
1773 	struct amd_iommu_pci_seg *pci_seg, *next;
1774 
1775 	for_each_pci_segment_safe(pci_seg, next) {
1776 		list_del(&pci_seg->list);
1777 		free_irq_lookup_table(pci_seg);
1778 		free_rlookup_table(pci_seg);
1779 		free_alias_table(pci_seg);
1780 		free_dev_table(pci_seg);
1781 		kfree(pci_seg);
1782 	}
1783 }
1784 
1785 static void __init free_sysfs(struct amd_iommu *iommu)
1786 {
1787 	if (iommu->iommu.dev) {
1788 		iommu_device_unregister(&iommu->iommu);
1789 		iommu_device_sysfs_remove(&iommu->iommu);
1790 	}
1791 }
1792 
1793 static void __init free_iommu_one(struct amd_iommu *iommu)
1794 {
1795 	free_sysfs(iommu);
1796 	free_iommu_buffers(iommu);
1797 	amd_iommu_free_ppr_log(iommu);
1798 	free_ga_log(iommu);
1799 	iommu_unmap_mmio_space(iommu);
1800 	amd_iommu_iopf_uninit(iommu);
1801 }
1802 
1803 static void __init free_iommu_all(void)
1804 {
1805 	struct amd_iommu *iommu, *next;
1806 
1807 	for_each_iommu_safe(iommu, next) {
1808 		list_del(&iommu->list);
1809 		free_iommu_one(iommu);
1810 		kfree(iommu);
1811 	}
1812 }
1813 
1814 /*
1815  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1816  * Workaround:
1817  *     BIOS should disable L2B micellaneous clock gating by setting
1818  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1819  */
1820 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1821 {
1822 	u32 value;
1823 
1824 	if ((boot_cpu_data.x86 != 0x15) ||
1825 	    (boot_cpu_data.x86_model < 0x10) ||
1826 	    (boot_cpu_data.x86_model > 0x1f))
1827 		return;
1828 
1829 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1830 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1831 
1832 	if (value & BIT(2))
1833 		return;
1834 
1835 	/* Select NB indirect register 0x90 and enable writing */
1836 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1837 
1838 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1839 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1840 
1841 	/* Clear the enable writing bit */
1842 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1843 }
1844 
1845 /*
1846  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1847  * Workaround:
1848  *     BIOS should enable ATS write permission check by setting
1849  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1850  */
1851 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1852 {
1853 	u32 value;
1854 
1855 	if ((boot_cpu_data.x86 != 0x15) ||
1856 	    (boot_cpu_data.x86_model < 0x30) ||
1857 	    (boot_cpu_data.x86_model > 0x3f))
1858 		return;
1859 
1860 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1861 	value = iommu_read_l2(iommu, 0x47);
1862 
1863 	if (value & BIT(0))
1864 		return;
1865 
1866 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1867 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1868 
1869 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1870 }
1871 
1872 /*
1873  * This function glues the initialization function for one IOMMU
1874  * together and also allocates the command buffer and programs the
1875  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1876  */
1877 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1878 				 struct acpi_table_header *ivrs_base)
1879 {
1880 	struct amd_iommu_pci_seg *pci_seg;
1881 
1882 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1883 	if (pci_seg == NULL)
1884 		return -ENOMEM;
1885 	iommu->pci_seg = pci_seg;
1886 
1887 	raw_spin_lock_init(&iommu->lock);
1888 	iommu->cmd_sem_val = 0;
1889 
1890 	/* Add IOMMU to internal data structures */
1891 	list_add_tail(&iommu->list, &amd_iommu_list);
1892 	iommu->index = amd_iommus_present++;
1893 
1894 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1895 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1896 		return -ENOSYS;
1897 	}
1898 
1899 	/*
1900 	 * Copy data from ACPI table entry to the iommu struct
1901 	 */
1902 	iommu->devid   = h->devid;
1903 	iommu->cap_ptr = h->cap_ptr;
1904 	iommu->mmio_phys = h->mmio_phys;
1905 
1906 	switch (h->type) {
1907 	case 0x10:
1908 		/* Check if IVHD EFR contains proper max banks/counters */
1909 		if ((h->efr_attr != 0) &&
1910 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1911 		    ((h->efr_attr & (0x3F << 17)) != 0))
1912 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1913 		else
1914 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1915 
1916 		/* GAM requires GA mode. */
1917 		if ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0)
1918 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1919 		break;
1920 	case 0x11:
1921 	case 0x40:
1922 		if (h->efr_reg & (1 << 9))
1923 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1924 		else
1925 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1926 
1927 		/* XT and GAM require GA mode. */
1928 		if ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0) {
1929 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1930 			break;
1931 		}
1932 
1933 		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1934 			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1935 
1936 		if (h->efr_attr & BIT(IOMMU_IVHD_ATTR_HATDIS_SHIFT)) {
1937 			pr_warn_once("Host Address Translation is not supported.\n");
1938 			amd_iommu_hatdis = true;
1939 		}
1940 
1941 		early_iommu_features_init(iommu, h);
1942 
1943 		break;
1944 	default:
1945 		return -EINVAL;
1946 	}
1947 
1948 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1949 						iommu->mmio_phys_end);
1950 	if (!iommu->mmio_base)
1951 		return -ENOMEM;
1952 
1953 	return init_iommu_from_acpi(iommu, h);
1954 }
1955 
1956 static int __init init_iommu_one_late(struct amd_iommu *iommu)
1957 {
1958 	int ret;
1959 
1960 	ret = alloc_iommu_buffers(iommu);
1961 	if (ret)
1962 		return ret;
1963 
1964 	iommu->int_enabled = false;
1965 
1966 	init_translation_status(iommu);
1967 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1968 		iommu_disable(iommu);
1969 		clear_translation_pre_enabled(iommu);
1970 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1971 			iommu->index);
1972 	}
1973 	if (amd_iommu_pre_enabled)
1974 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1975 
1976 	if (amd_iommu_irq_remap) {
1977 		ret = amd_iommu_create_irq_domain(iommu);
1978 		if (ret)
1979 			return ret;
1980 	}
1981 
1982 	/*
1983 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1984 	 * table tells us so, but this is a lie!
1985 	 */
1986 	iommu->pci_seg->rlookup_table[iommu->devid] = NULL;
1987 
1988 	return 0;
1989 }
1990 
1991 /**
1992  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1993  * @ivrs: Pointer to the IVRS header
1994  *
1995  * This function search through all IVDB of the maximum supported IVHD
1996  */
1997 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1998 {
1999 	u8 *base = (u8 *)ivrs;
2000 	struct ivhd_header *ivhd = (struct ivhd_header *)
2001 					(base + IVRS_HEADER_LENGTH);
2002 	u8 last_type = ivhd->type;
2003 	u16 devid = ivhd->devid;
2004 
2005 	while (((u8 *)ivhd - base < ivrs->length) &&
2006 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
2007 		u8 *p = (u8 *) ivhd;
2008 
2009 		if (ivhd->devid == devid)
2010 			last_type = ivhd->type;
2011 		ivhd = (struct ivhd_header *)(p + ivhd->length);
2012 	}
2013 
2014 	return last_type;
2015 }
2016 
2017 /*
2018  * Iterates over all IOMMU entries in the ACPI table, allocates the
2019  * IOMMU structure and initializes it with init_iommu_one()
2020  */
2021 static int __init init_iommu_all(struct acpi_table_header *table)
2022 {
2023 	u8 *p = (u8 *)table, *end = (u8 *)table;
2024 	struct ivhd_header *h;
2025 	struct amd_iommu *iommu;
2026 	int ret;
2027 
2028 	end += table->length;
2029 	p += IVRS_HEADER_LENGTH;
2030 
2031 	/* Phase 1: Process all IVHD blocks */
2032 	while (p < end) {
2033 		h = (struct ivhd_header *)p;
2034 		if (*p == amd_iommu_target_ivhd_type) {
2035 
2036 			DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x "
2037 				    "flags: %01x info %04x\n",
2038 				    h->pci_seg, PCI_BUS_NUM(h->devid),
2039 				    PCI_SLOT(h->devid), PCI_FUNC(h->devid),
2040 				    h->cap_ptr, h->flags, h->info);
2041 			DUMP_printk("       mmio-addr: %016llx\n",
2042 				    h->mmio_phys);
2043 
2044 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
2045 			if (iommu == NULL)
2046 				return -ENOMEM;
2047 
2048 			ret = init_iommu_one(iommu, h, table);
2049 			if (ret)
2050 				return ret;
2051 		}
2052 		p += h->length;
2053 
2054 	}
2055 	WARN_ON(p != end);
2056 
2057 	/* Phase 2 : Early feature support check */
2058 	get_global_efr();
2059 
2060 	/* Phase 3 : Enabling IOMMU features */
2061 	for_each_iommu(iommu) {
2062 		ret = init_iommu_one_late(iommu);
2063 		if (ret)
2064 			return ret;
2065 	}
2066 
2067 	return 0;
2068 }
2069 
2070 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
2071 {
2072 	u64 val;
2073 	struct pci_dev *pdev = iommu->dev;
2074 
2075 	if (!check_feature(FEATURE_PC))
2076 		return;
2077 
2078 	amd_iommu_pc_present = true;
2079 
2080 	pci_info(pdev, "IOMMU performance counters supported\n");
2081 
2082 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
2083 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
2084 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
2085 
2086 	return;
2087 }
2088 
2089 static ssize_t amd_iommu_show_cap(struct device *dev,
2090 				  struct device_attribute *attr,
2091 				  char *buf)
2092 {
2093 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
2094 	return sysfs_emit(buf, "%x\n", iommu->cap);
2095 }
2096 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
2097 
2098 static ssize_t amd_iommu_show_features(struct device *dev,
2099 				       struct device_attribute *attr,
2100 				       char *buf)
2101 {
2102 	return sysfs_emit(buf, "%llx:%llx\n", amd_iommu_efr, amd_iommu_efr2);
2103 }
2104 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
2105 
2106 static struct attribute *amd_iommu_attrs[] = {
2107 	&dev_attr_cap.attr,
2108 	&dev_attr_features.attr,
2109 	NULL,
2110 };
2111 
2112 static struct attribute_group amd_iommu_group = {
2113 	.name = "amd-iommu",
2114 	.attrs = amd_iommu_attrs,
2115 };
2116 
2117 static const struct attribute_group *amd_iommu_groups[] = {
2118 	&amd_iommu_group,
2119 	NULL,
2120 };
2121 
2122 /*
2123  * Note: IVHD 0x11 and 0x40 also contains exact copy
2124  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
2125  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
2126  */
2127 static void __init late_iommu_features_init(struct amd_iommu *iommu)
2128 {
2129 	u64 features, features2;
2130 
2131 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
2132 		return;
2133 
2134 	/* read extended feature bits */
2135 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
2136 	features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2);
2137 
2138 	if (!amd_iommu_efr) {
2139 		amd_iommu_efr = features;
2140 		amd_iommu_efr2 = features2;
2141 		return;
2142 	}
2143 
2144 	/*
2145 	 * Sanity check and warn if EFR values from
2146 	 * IVHD and MMIO conflict.
2147 	 */
2148 	if (features != amd_iommu_efr ||
2149 	    features2 != amd_iommu_efr2) {
2150 		pr_warn(FW_WARN
2151 			"EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n",
2152 			features, amd_iommu_efr,
2153 			features2, amd_iommu_efr2);
2154 	}
2155 }
2156 
2157 static int __init iommu_init_pci(struct amd_iommu *iommu)
2158 {
2159 	int cap_ptr = iommu->cap_ptr;
2160 	int ret;
2161 
2162 	iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2163 						 PCI_BUS_NUM(iommu->devid),
2164 						 iommu->devid & 0xff);
2165 	if (!iommu->dev)
2166 		return -ENODEV;
2167 
2168 	/* ACPI _PRT won't have an IRQ for IOMMU */
2169 	iommu->dev->irq_managed = 1;
2170 
2171 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2172 			      &iommu->cap);
2173 
2174 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2175 		amd_iommu_iotlb_sup = false;
2176 
2177 	late_iommu_features_init(iommu);
2178 
2179 	if (check_feature(FEATURE_GT)) {
2180 		int glxval;
2181 		u64 pasmax;
2182 
2183 		pasmax = FIELD_GET(FEATURE_PASMAX, amd_iommu_efr);
2184 		iommu->iommu.max_pasids = (1 << (pasmax + 1)) - 1;
2185 
2186 		BUG_ON(iommu->iommu.max_pasids & ~PASID_MASK);
2187 
2188 		glxval = FIELD_GET(FEATURE_GLX, amd_iommu_efr);
2189 
2190 		if (amd_iommu_max_glx_val == -1)
2191 			amd_iommu_max_glx_val = glxval;
2192 		else
2193 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2194 
2195 		iommu_enable_gt(iommu);
2196 	}
2197 
2198 	if (check_feature(FEATURE_PPR) && amd_iommu_alloc_ppr_log(iommu))
2199 		return -ENOMEM;
2200 
2201 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2202 		pr_info("Using strict mode due to virtualization\n");
2203 		iommu_set_dma_strict();
2204 		amd_iommu_np_cache = true;
2205 	}
2206 
2207 	init_iommu_perf_ctr(iommu);
2208 
2209 	if (is_rd890_iommu(iommu->dev)) {
2210 		int i, j;
2211 
2212 		iommu->root_pdev =
2213 			pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2214 						    iommu->dev->bus->number,
2215 						    PCI_DEVFN(0, 0));
2216 
2217 		/*
2218 		 * Some rd890 systems may not be fully reconfigured by the
2219 		 * BIOS, so it's necessary for us to store this information so
2220 		 * it can be reprogrammed on resume
2221 		 */
2222 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2223 				&iommu->stored_addr_lo);
2224 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2225 				&iommu->stored_addr_hi);
2226 
2227 		/* Low bit locks writes to configuration space */
2228 		iommu->stored_addr_lo &= ~1;
2229 
2230 		for (i = 0; i < 6; i++)
2231 			for (j = 0; j < 0x12; j++)
2232 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2233 
2234 		for (i = 0; i < 0x83; i++)
2235 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2236 	}
2237 
2238 	amd_iommu_erratum_746_workaround(iommu);
2239 	amd_iommu_ats_write_check_workaround(iommu);
2240 
2241 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2242 			       amd_iommu_groups, "ivhd%d", iommu->index);
2243 	if (ret)
2244 		return ret;
2245 
2246 	/*
2247 	 * Allocate per IOMMU IOPF queue here so that in attach device path,
2248 	 * PRI capable device can be added to IOPF queue
2249 	 */
2250 	if (amd_iommu_gt_ppr_supported()) {
2251 		ret = amd_iommu_iopf_init(iommu);
2252 		if (ret)
2253 			return ret;
2254 	}
2255 
2256 	ret = iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2257 	if (ret || amd_iommu_pgtable == PD_MODE_NONE) {
2258 		/*
2259 		 * Remove sysfs if DMA translation is not supported by the
2260 		 * IOMMU. Do not return an error to enable IRQ remapping
2261 		 * in state_next(), DTE[V, TV] must eventually be set to 0.
2262 		 */
2263 		iommu_device_sysfs_remove(&iommu->iommu);
2264 	}
2265 
2266 	return pci_enable_device(iommu->dev);
2267 }
2268 
2269 static void print_iommu_info(void)
2270 {
2271 	int i;
2272 	static const char * const feat_str[] = {
2273 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2274 		"IA", "GA", "HE", "PC"
2275 	};
2276 
2277 	if (amd_iommu_efr) {
2278 		pr_info("Extended features (%#llx, %#llx):", amd_iommu_efr, amd_iommu_efr2);
2279 
2280 		for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2281 			if (check_feature(1ULL << i))
2282 				pr_cont(" %s", feat_str[i]);
2283 		}
2284 
2285 		if (check_feature(FEATURE_GAM_VAPIC))
2286 			pr_cont(" GA_vAPIC");
2287 
2288 		if (check_feature(FEATURE_SNP))
2289 			pr_cont(" SNP");
2290 
2291 		if (check_feature2(FEATURE_SEVSNPIO_SUP))
2292 			pr_cont(" SEV-TIO");
2293 
2294 		pr_cont("\n");
2295 	}
2296 
2297 	if (irq_remapping_enabled) {
2298 		pr_info("Interrupt remapping enabled\n");
2299 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2300 			pr_info("X2APIC enabled\n");
2301 	}
2302 	if (amd_iommu_pgtable == PD_MODE_V2) {
2303 		pr_info("V2 page table enabled (Paging mode : %d level)\n",
2304 			amd_iommu_gpt_level);
2305 	}
2306 }
2307 
2308 static int __init amd_iommu_init_pci(void)
2309 {
2310 	struct amd_iommu *iommu;
2311 	struct amd_iommu_pci_seg *pci_seg;
2312 	int ret;
2313 
2314 	/* Init global identity domain before registering IOMMU */
2315 	amd_iommu_init_identity_domain();
2316 
2317 	for_each_iommu(iommu) {
2318 		ret = iommu_init_pci(iommu);
2319 		if (ret) {
2320 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2321 			       iommu->index, ret);
2322 			goto out;
2323 		}
2324 		/* Need to setup range after PCI init */
2325 		iommu_set_cwwb_range(iommu);
2326 	}
2327 
2328 	/*
2329 	 * Order is important here to make sure any unity map requirements are
2330 	 * fulfilled. The unity mappings are created and written to the device
2331 	 * table during the iommu_init_pci() call.
2332 	 *
2333 	 * After that we call init_device_table_dma() to make sure any
2334 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2335 	 * of all IOMMUs to make sure the changes to the device table are
2336 	 * active.
2337 	 */
2338 	for_each_pci_segment(pci_seg)
2339 		init_device_table_dma(pci_seg);
2340 
2341 	for_each_iommu(iommu)
2342 		amd_iommu_flush_all_caches(iommu);
2343 
2344 	print_iommu_info();
2345 
2346 out:
2347 	return ret;
2348 }
2349 
2350 /****************************************************************************
2351  *
2352  * The following functions initialize the MSI interrupts for all IOMMUs
2353  * in the system. It's a bit challenging because there could be multiple
2354  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2355  * pci_dev.
2356  *
2357  ****************************************************************************/
2358 
2359 static int iommu_setup_msi(struct amd_iommu *iommu)
2360 {
2361 	int r;
2362 
2363 	r = pci_enable_msi(iommu->dev);
2364 	if (r)
2365 		return r;
2366 
2367 	r = request_threaded_irq(iommu->dev->irq, NULL, amd_iommu_int_thread,
2368 				 IRQF_ONESHOT, "AMD-Vi", iommu);
2369 	if (r) {
2370 		pci_disable_msi(iommu->dev);
2371 		return r;
2372 	}
2373 
2374 	return 0;
2375 }
2376 
2377 union intcapxt {
2378 	u64	capxt;
2379 	struct {
2380 		u64	reserved_0		:  2,
2381 			dest_mode_logical	:  1,
2382 			reserved_1		:  5,
2383 			destid_0_23		: 24,
2384 			vector			:  8,
2385 			reserved_2		: 16,
2386 			destid_24_31		:  8;
2387 	};
2388 } __attribute__ ((packed));
2389 
2390 
2391 static struct irq_chip intcapxt_controller;
2392 
2393 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2394 				       struct irq_data *irqd, bool reserve)
2395 {
2396 	return 0;
2397 }
2398 
2399 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2400 					  struct irq_data *irqd)
2401 {
2402 }
2403 
2404 
2405 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2406 				    unsigned int nr_irqs, void *arg)
2407 {
2408 	struct irq_alloc_info *info = arg;
2409 	int i, ret;
2410 
2411 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2412 		return -EINVAL;
2413 
2414 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2415 	if (ret < 0)
2416 		return ret;
2417 
2418 	for (i = virq; i < virq + nr_irqs; i++) {
2419 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2420 
2421 		irqd->chip = &intcapxt_controller;
2422 		irqd->hwirq = info->hwirq;
2423 		irqd->chip_data = info->data;
2424 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2425 	}
2426 
2427 	return ret;
2428 }
2429 
2430 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2431 				    unsigned int nr_irqs)
2432 {
2433 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2434 }
2435 
2436 
2437 static void intcapxt_unmask_irq(struct irq_data *irqd)
2438 {
2439 	struct amd_iommu *iommu = irqd->chip_data;
2440 	struct irq_cfg *cfg = irqd_cfg(irqd);
2441 	union intcapxt xt;
2442 
2443 	xt.capxt = 0ULL;
2444 	xt.dest_mode_logical = apic->dest_mode_logical;
2445 	xt.vector = cfg->vector;
2446 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2447 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2448 
2449 	writeq(xt.capxt, iommu->mmio_base + irqd->hwirq);
2450 }
2451 
2452 static void intcapxt_mask_irq(struct irq_data *irqd)
2453 {
2454 	struct amd_iommu *iommu = irqd->chip_data;
2455 
2456 	writeq(0, iommu->mmio_base + irqd->hwirq);
2457 }
2458 
2459 
2460 static int intcapxt_set_affinity(struct irq_data *irqd,
2461 				 const struct cpumask *mask, bool force)
2462 {
2463 	struct irq_data *parent = irqd->parent_data;
2464 	int ret;
2465 
2466 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2467 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2468 		return ret;
2469 	return 0;
2470 }
2471 
2472 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2473 {
2474 	return on ? -EOPNOTSUPP : 0;
2475 }
2476 
2477 static struct irq_chip intcapxt_controller = {
2478 	.name			= "IOMMU-MSI",
2479 	.irq_unmask		= intcapxt_unmask_irq,
2480 	.irq_mask		= intcapxt_mask_irq,
2481 	.irq_ack		= irq_chip_ack_parent,
2482 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2483 	.irq_set_affinity       = intcapxt_set_affinity,
2484 	.irq_set_wake		= intcapxt_set_wake,
2485 	.flags			= IRQCHIP_MASK_ON_SUSPEND | IRQCHIP_MOVE_DEFERRED,
2486 };
2487 
2488 static const struct irq_domain_ops intcapxt_domain_ops = {
2489 	.alloc			= intcapxt_irqdomain_alloc,
2490 	.free			= intcapxt_irqdomain_free,
2491 	.activate		= intcapxt_irqdomain_activate,
2492 	.deactivate		= intcapxt_irqdomain_deactivate,
2493 };
2494 
2495 
2496 static struct irq_domain *iommu_irqdomain;
2497 
2498 static struct irq_domain *iommu_get_irqdomain(void)
2499 {
2500 	struct fwnode_handle *fn;
2501 
2502 	/* No need for locking here (yet) as the init is single-threaded */
2503 	if (iommu_irqdomain)
2504 		return iommu_irqdomain;
2505 
2506 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2507 	if (!fn)
2508 		return NULL;
2509 
2510 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2511 						      fn, &intcapxt_domain_ops,
2512 						      NULL);
2513 	if (!iommu_irqdomain)
2514 		irq_domain_free_fwnode(fn);
2515 
2516 	return iommu_irqdomain;
2517 }
2518 
2519 static int __iommu_setup_intcapxt(struct amd_iommu *iommu, const char *devname,
2520 				  int hwirq, irq_handler_t thread_fn)
2521 {
2522 	struct irq_domain *domain;
2523 	struct irq_alloc_info info;
2524 	int irq, ret;
2525 	int node = dev_to_node(&iommu->dev->dev);
2526 
2527 	domain = iommu_get_irqdomain();
2528 	if (!domain)
2529 		return -ENXIO;
2530 
2531 	init_irq_alloc_info(&info, NULL);
2532 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2533 	info.data = iommu;
2534 	info.hwirq = hwirq;
2535 
2536 	irq = irq_domain_alloc_irqs(domain, 1, node, &info);
2537 	if (irq < 0) {
2538 		irq_domain_remove(domain);
2539 		return irq;
2540 	}
2541 
2542 	ret = request_threaded_irq(irq, NULL, thread_fn, IRQF_ONESHOT, devname,
2543 				   iommu);
2544 	if (ret) {
2545 		irq_domain_free_irqs(irq, 1);
2546 		irq_domain_remove(domain);
2547 		return ret;
2548 	}
2549 
2550 	return 0;
2551 }
2552 
2553 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2554 {
2555 	int ret;
2556 
2557 	snprintf(iommu->evt_irq_name, sizeof(iommu->evt_irq_name),
2558 		 "AMD-Vi%d-Evt", iommu->index);
2559 	ret = __iommu_setup_intcapxt(iommu, iommu->evt_irq_name,
2560 				     MMIO_INTCAPXT_EVT_OFFSET,
2561 				     amd_iommu_int_thread_evtlog);
2562 	if (ret)
2563 		return ret;
2564 
2565 	snprintf(iommu->ppr_irq_name, sizeof(iommu->ppr_irq_name),
2566 		 "AMD-Vi%d-PPR", iommu->index);
2567 	ret = __iommu_setup_intcapxt(iommu, iommu->ppr_irq_name,
2568 				     MMIO_INTCAPXT_PPR_OFFSET,
2569 				     amd_iommu_int_thread_pprlog);
2570 	if (ret)
2571 		return ret;
2572 
2573 #ifdef CONFIG_IRQ_REMAP
2574 	snprintf(iommu->ga_irq_name, sizeof(iommu->ga_irq_name),
2575 		 "AMD-Vi%d-GA", iommu->index);
2576 	ret = __iommu_setup_intcapxt(iommu, iommu->ga_irq_name,
2577 				     MMIO_INTCAPXT_GALOG_OFFSET,
2578 				     amd_iommu_int_thread_galog);
2579 #endif
2580 
2581 	return ret;
2582 }
2583 
2584 static int iommu_init_irq(struct amd_iommu *iommu)
2585 {
2586 	int ret;
2587 
2588 	if (iommu->int_enabled)
2589 		goto enable_faults;
2590 
2591 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2592 		ret = iommu_setup_intcapxt(iommu);
2593 	else if (iommu->dev->msi_cap)
2594 		ret = iommu_setup_msi(iommu);
2595 	else
2596 		ret = -ENODEV;
2597 
2598 	if (ret)
2599 		return ret;
2600 
2601 	iommu->int_enabled = true;
2602 enable_faults:
2603 
2604 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2605 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2606 
2607 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2608 
2609 	return 0;
2610 }
2611 
2612 /****************************************************************************
2613  *
2614  * The next functions belong to the third pass of parsing the ACPI
2615  * table. In this last pass the memory mapping requirements are
2616  * gathered (like exclusion and unity mapping ranges).
2617  *
2618  ****************************************************************************/
2619 
2620 static void __init free_unity_maps(void)
2621 {
2622 	struct unity_map_entry *entry, *next;
2623 	struct amd_iommu_pci_seg *p, *pci_seg;
2624 
2625 	for_each_pci_segment_safe(pci_seg, p) {
2626 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2627 			list_del(&entry->list);
2628 			kfree(entry);
2629 		}
2630 	}
2631 }
2632 
2633 /* called for unity map ACPI definition */
2634 static int __init init_unity_map_range(struct ivmd_header *m,
2635 				       struct acpi_table_header *ivrs_base)
2636 {
2637 	struct unity_map_entry *e = NULL;
2638 	struct amd_iommu_pci_seg *pci_seg;
2639 	char *s;
2640 
2641 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2642 	if (pci_seg == NULL)
2643 		return -ENOMEM;
2644 
2645 	e = kzalloc(sizeof(*e), GFP_KERNEL);
2646 	if (e == NULL)
2647 		return -ENOMEM;
2648 
2649 	switch (m->type) {
2650 	default:
2651 		kfree(e);
2652 		return 0;
2653 	case ACPI_IVMD_TYPE:
2654 		s = "IVMD_TYPEi\t\t\t";
2655 		e->devid_start = e->devid_end = m->devid;
2656 		break;
2657 	case ACPI_IVMD_TYPE_ALL:
2658 		s = "IVMD_TYPE_ALL\t\t";
2659 		e->devid_start = 0;
2660 		e->devid_end = pci_seg->last_bdf;
2661 		break;
2662 	case ACPI_IVMD_TYPE_RANGE:
2663 		s = "IVMD_TYPE_RANGE\t\t";
2664 		e->devid_start = m->devid;
2665 		e->devid_end = m->aux;
2666 		break;
2667 	}
2668 	e->address_start = PAGE_ALIGN(m->range_start);
2669 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2670 	e->prot = m->flags >> 1;
2671 
2672 	/*
2673 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2674 	 * since some buggy BIOSes might lead to the overwritten exclusion
2675 	 * range (exclusion_start and exclusion_length members). This
2676 	 * happens when there are multiple exclusion ranges (IVMD entries)
2677 	 * defined in ACPI table.
2678 	 */
2679 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2680 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2681 
2682 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2683 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2684 		    " flags: %x\n", s, m->pci_seg,
2685 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2686 		    PCI_FUNC(e->devid_start), m->pci_seg,
2687 		    PCI_BUS_NUM(e->devid_end),
2688 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2689 		    e->address_start, e->address_end, m->flags);
2690 
2691 	list_add_tail(&e->list, &pci_seg->unity_map);
2692 
2693 	return 0;
2694 }
2695 
2696 /* iterates over all memory definitions we find in the ACPI table */
2697 static int __init init_memory_definitions(struct acpi_table_header *table)
2698 {
2699 	u8 *p = (u8 *)table, *end = (u8 *)table;
2700 	struct ivmd_header *m;
2701 
2702 	end += table->length;
2703 	p += IVRS_HEADER_LENGTH;
2704 
2705 	while (p < end) {
2706 		m = (struct ivmd_header *)p;
2707 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2708 			init_unity_map_range(m, table);
2709 
2710 		p += m->length;
2711 	}
2712 
2713 	return 0;
2714 }
2715 
2716 /*
2717  * Init the device table to not allow DMA access for devices
2718  */
2719 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2720 {
2721 	u32 devid;
2722 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2723 
2724 	if (!dev_table || amd_iommu_pgtable == PD_MODE_NONE)
2725 		return;
2726 
2727 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2728 		set_dte_bit(&dev_table[devid], DEV_ENTRY_VALID);
2729 		if (!amd_iommu_snp_en)
2730 			set_dte_bit(&dev_table[devid], DEV_ENTRY_TRANSLATION);
2731 	}
2732 }
2733 
2734 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2735 {
2736 	u32 devid;
2737 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2738 
2739 	if (dev_table == NULL)
2740 		return;
2741 
2742 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2743 		dev_table[devid].data[0] = 0ULL;
2744 		dev_table[devid].data[1] = 0ULL;
2745 	}
2746 }
2747 
2748 static void init_device_table(void)
2749 {
2750 	struct amd_iommu_pci_seg *pci_seg;
2751 	u32 devid;
2752 
2753 	if (!amd_iommu_irq_remap)
2754 		return;
2755 
2756 	for_each_pci_segment(pci_seg) {
2757 		for (devid = 0; devid <= pci_seg->last_bdf; ++devid)
2758 			set_dte_bit(&pci_seg->dev_table[devid], DEV_ENTRY_IRQ_TBL_EN);
2759 	}
2760 }
2761 
2762 static void iommu_init_flags(struct amd_iommu *iommu)
2763 {
2764 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2765 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2766 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2767 
2768 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2769 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2770 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2771 
2772 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2773 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2774 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2775 
2776 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2777 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2778 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2779 
2780 	/*
2781 	 * make IOMMU memory accesses cache coherent
2782 	 */
2783 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2784 
2785 	/* Set IOTLB invalidation timeout to 1s */
2786 	iommu_feature_set(iommu, CTRL_INV_TO_1S, CTRL_INV_TO_MASK, CONTROL_INV_TIMEOUT);
2787 
2788 	/* Enable Enhanced Peripheral Page Request Handling */
2789 	if (check_feature(FEATURE_EPHSUP))
2790 		iommu_feature_enable(iommu, CONTROL_EPH_EN);
2791 }
2792 
2793 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2794 {
2795 	int i, j;
2796 	u32 ioc_feature_control;
2797 	struct pci_dev *pdev = iommu->root_pdev;
2798 
2799 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2800 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2801 		return;
2802 
2803 	/*
2804 	 * First, we need to ensure that the iommu is enabled. This is
2805 	 * controlled by a register in the northbridge
2806 	 */
2807 
2808 	/* Select Northbridge indirect register 0x75 and enable writing */
2809 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2810 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2811 
2812 	/* Enable the iommu */
2813 	if (!(ioc_feature_control & 0x1))
2814 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2815 
2816 	/* Restore the iommu BAR */
2817 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2818 			       iommu->stored_addr_lo);
2819 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2820 			       iommu->stored_addr_hi);
2821 
2822 	/* Restore the l1 indirect regs for each of the 6 l1s */
2823 	for (i = 0; i < 6; i++)
2824 		for (j = 0; j < 0x12; j++)
2825 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2826 
2827 	/* Restore the l2 indirect regs */
2828 	for (i = 0; i < 0x83; i++)
2829 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2830 
2831 	/* Lock PCI setup registers */
2832 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2833 			       iommu->stored_addr_lo | 1);
2834 }
2835 
2836 static void iommu_enable_ga(struct amd_iommu *iommu)
2837 {
2838 #ifdef CONFIG_IRQ_REMAP
2839 	switch (amd_iommu_guest_ir) {
2840 	case AMD_IOMMU_GUEST_IR_VAPIC:
2841 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2842 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2843 		iommu->irte_ops = &irte_128_ops;
2844 		break;
2845 	default:
2846 		iommu->irte_ops = &irte_32_ops;
2847 		break;
2848 	}
2849 #endif
2850 }
2851 
2852 static void iommu_disable_irtcachedis(struct amd_iommu *iommu)
2853 {
2854 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
2855 }
2856 
2857 static void iommu_enable_irtcachedis(struct amd_iommu *iommu)
2858 {
2859 	u64 ctrl;
2860 
2861 	if (!amd_iommu_irtcachedis)
2862 		return;
2863 
2864 	/*
2865 	 * Note:
2866 	 * The support for IRTCacheDis feature is dertermined by
2867 	 * checking if the bit is writable.
2868 	 */
2869 	iommu_feature_enable(iommu, CONTROL_IRTCACHEDIS);
2870 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
2871 	ctrl &= (1ULL << CONTROL_IRTCACHEDIS);
2872 	if (ctrl)
2873 		iommu->irtcachedis_enabled = true;
2874 	pr_info("iommu%d (%#06x) : IRT cache is %s\n",
2875 		iommu->index, iommu->devid,
2876 		iommu->irtcachedis_enabled ? "disabled" : "enabled");
2877 }
2878 
2879 static void iommu_enable_2k_int(struct amd_iommu *iommu)
2880 {
2881 	if (!FEATURE_NUM_INT_REMAP_SUP_2K(amd_iommu_efr2))
2882 		return;
2883 
2884 	iommu_feature_set(iommu,
2885 			  CONTROL_NUM_INT_REMAP_MODE_2K,
2886 			  CONTROL_NUM_INT_REMAP_MODE_MASK,
2887 			  CONTROL_NUM_INT_REMAP_MODE);
2888 }
2889 
2890 static void early_enable_iommu(struct amd_iommu *iommu)
2891 {
2892 	iommu_disable(iommu);
2893 	iommu_init_flags(iommu);
2894 	iommu_set_device_table(iommu);
2895 	iommu_enable_command_buffer(iommu);
2896 	iommu_enable_event_buffer(iommu);
2897 	iommu_set_exclusion_range(iommu);
2898 	iommu_enable_gt(iommu);
2899 	iommu_enable_ga(iommu);
2900 	iommu_enable_xt(iommu);
2901 	iommu_enable_irtcachedis(iommu);
2902 	iommu_enable_2k_int(iommu);
2903 	iommu_enable(iommu);
2904 	amd_iommu_flush_all_caches(iommu);
2905 }
2906 
2907 /*
2908  * This function finally enables all IOMMUs found in the system after
2909  * they have been initialized.
2910  *
2911  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to reuse
2912  * the old content of device table entries. Not this case or reuse failed,
2913  * just continue as normal kernel does.
2914  */
2915 static void early_enable_iommus(void)
2916 {
2917 	struct amd_iommu *iommu;
2918 	struct amd_iommu_pci_seg *pci_seg;
2919 
2920 	if (!reuse_device_table()) {
2921 		/*
2922 		 * If come here because of failure in reusing device table from old
2923 		 * kernel with all IOMMUs enabled, print error message and try to
2924 		 * free allocated old_dev_tbl_cpy.
2925 		 */
2926 		if (amd_iommu_pre_enabled) {
2927 			pr_err("Failed to reuse DEV table from previous kernel.\n");
2928 			/*
2929 			 * Bail out early if unable to remap/reuse DEV table from
2930 			 * previous kernel if SNP enabled as IOMMU commands will
2931 			 * time out without DEV table and cause kdump boot panic.
2932 			 */
2933 			BUG_ON(check_feature(FEATURE_SNP));
2934 		}
2935 
2936 		for_each_pci_segment(pci_seg) {
2937 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2938 				memunmap((void *)pci_seg->old_dev_tbl_cpy);
2939 				pci_seg->old_dev_tbl_cpy = NULL;
2940 			}
2941 		}
2942 
2943 		for_each_iommu(iommu) {
2944 			clear_translation_pre_enabled(iommu);
2945 			early_enable_iommu(iommu);
2946 		}
2947 	} else {
2948 		pr_info("Reused DEV table from previous kernel.\n");
2949 
2950 		for_each_pci_segment(pci_seg) {
2951 			iommu_free_pages(pci_seg->dev_table);
2952 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2953 		}
2954 
2955 		for_each_iommu(iommu) {
2956 			iommu_disable_command_buffer(iommu);
2957 			iommu_disable_event_buffer(iommu);
2958 			iommu_disable_irtcachedis(iommu);
2959 			iommu_enable_command_buffer(iommu);
2960 			iommu_enable_event_buffer(iommu);
2961 			iommu_enable_ga(iommu);
2962 			iommu_enable_xt(iommu);
2963 			iommu_enable_irtcachedis(iommu);
2964 			iommu_enable_2k_int(iommu);
2965 			iommu_set_device_table(iommu);
2966 			amd_iommu_flush_all_caches(iommu);
2967 		}
2968 	}
2969 }
2970 
2971 static void enable_iommus_ppr(void)
2972 {
2973 	struct amd_iommu *iommu;
2974 
2975 	if (!amd_iommu_gt_ppr_supported())
2976 		return;
2977 
2978 	for_each_iommu(iommu)
2979 		amd_iommu_enable_ppr_log(iommu);
2980 }
2981 
2982 static void enable_iommus_vapic(void)
2983 {
2984 #ifdef CONFIG_IRQ_REMAP
2985 	u32 status, i;
2986 	struct amd_iommu *iommu;
2987 
2988 	for_each_iommu(iommu) {
2989 		/*
2990 		 * Disable GALog if already running. It could have been enabled
2991 		 * in the previous boot before kdump.
2992 		 */
2993 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2994 		if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2995 			continue;
2996 
2997 		iommu_feature_disable(iommu, CONTROL_GALOG_EN);
2998 		iommu_feature_disable(iommu, CONTROL_GAINT_EN);
2999 
3000 		/*
3001 		 * Need to set and poll check the GALOGRun bit to zero before
3002 		 * we can set/ modify GA Log registers safely.
3003 		 */
3004 		for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
3005 			status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
3006 			if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
3007 				break;
3008 			udelay(10);
3009 		}
3010 
3011 		if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
3012 			return;
3013 	}
3014 
3015 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
3016 	    !check_feature(FEATURE_GAM_VAPIC)) {
3017 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3018 		return;
3019 	}
3020 
3021 	if (amd_iommu_snp_en &&
3022 	    !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) {
3023 		pr_warn("Force to disable Virtual APIC due to SNP\n");
3024 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3025 		return;
3026 	}
3027 
3028 	/* Enabling GAM and SNPAVIC support */
3029 	for_each_iommu(iommu) {
3030 		if (iommu_init_ga_log(iommu) ||
3031 		    iommu_ga_log_enable(iommu))
3032 			return;
3033 
3034 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
3035 		if (amd_iommu_snp_en)
3036 			iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN);
3037 	}
3038 
3039 	amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
3040 	pr_info("Virtual APIC enabled\n");
3041 #endif
3042 }
3043 
3044 static void disable_iommus(void)
3045 {
3046 	struct amd_iommu *iommu;
3047 
3048 	for_each_iommu(iommu)
3049 		iommu_disable(iommu);
3050 
3051 #ifdef CONFIG_IRQ_REMAP
3052 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
3053 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
3054 #endif
3055 }
3056 
3057 /*
3058  * Suspend/Resume support
3059  * disable suspend until real resume implemented
3060  */
3061 
3062 static void amd_iommu_resume(void *data)
3063 {
3064 	struct amd_iommu *iommu;
3065 
3066 	for_each_iommu(iommu)
3067 		iommu_apply_resume_quirks(iommu);
3068 
3069 	/* re-load the hardware */
3070 	for_each_iommu(iommu)
3071 		early_enable_iommu(iommu);
3072 
3073 	amd_iommu_enable_interrupts();
3074 }
3075 
3076 static int amd_iommu_suspend(void *data)
3077 {
3078 	/* disable IOMMUs to go out of the way for BIOS */
3079 	disable_iommus();
3080 
3081 	return 0;
3082 }
3083 
3084 static const struct syscore_ops amd_iommu_syscore_ops = {
3085 	.suspend = amd_iommu_suspend,
3086 	.resume = amd_iommu_resume,
3087 };
3088 
3089 static struct syscore amd_iommu_syscore = {
3090 	.ops = &amd_iommu_syscore_ops,
3091 };
3092 
3093 static void __init free_iommu_resources(void)
3094 {
3095 	free_iommu_all();
3096 	free_pci_segments();
3097 }
3098 
3099 /* SB IOAPIC is always on this device in AMD systems */
3100 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
3101 
3102 static bool __init check_ioapic_information(void)
3103 {
3104 	const char *fw_bug = FW_BUG;
3105 	bool ret, has_sb_ioapic;
3106 	int idx;
3107 
3108 	has_sb_ioapic = false;
3109 	ret           = false;
3110 
3111 	/*
3112 	 * If we have map overrides on the kernel command line the
3113 	 * messages in this function might not describe firmware bugs
3114 	 * anymore - so be careful
3115 	 */
3116 	if (cmdline_maps)
3117 		fw_bug = "";
3118 
3119 	for (idx = 0; idx < nr_ioapics; idx++) {
3120 		int devid, id = mpc_ioapic_id(idx);
3121 
3122 		devid = get_ioapic_devid(id);
3123 		if (devid < 0) {
3124 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
3125 				fw_bug, id);
3126 			ret = false;
3127 		} else if (devid == IOAPIC_SB_DEVID) {
3128 			has_sb_ioapic = true;
3129 			ret           = true;
3130 		}
3131 	}
3132 
3133 	if (!has_sb_ioapic) {
3134 		/*
3135 		 * We expect the SB IOAPIC to be listed in the IVRS
3136 		 * table. The system timer is connected to the SB IOAPIC
3137 		 * and if we don't have it in the list the system will
3138 		 * panic at boot time.  This situation usually happens
3139 		 * when the BIOS is buggy and provides us the wrong
3140 		 * device id for the IOAPIC in the system.
3141 		 */
3142 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
3143 	}
3144 
3145 	if (!ret)
3146 		pr_err("Disabling interrupt remapping\n");
3147 
3148 	return ret;
3149 }
3150 
3151 static void __init free_dma_resources(void)
3152 {
3153 	amd_iommu_pdom_id_destroy();
3154 	free_unity_maps();
3155 }
3156 
3157 static void __init ivinfo_init(void *ivrs)
3158 {
3159 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
3160 }
3161 
3162 /*
3163  * This is the hardware init function for AMD IOMMU in the system.
3164  * This function is called either from amd_iommu_init or from the interrupt
3165  * remapping setup code.
3166  *
3167  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
3168  * four times:
3169  *
3170  *	1 pass) Discover the most comprehensive IVHD type to use.
3171  *
3172  *	2 pass) Find the highest PCI device id the driver has to handle.
3173  *		Upon this information the size of the data structures is
3174  *		determined that needs to be allocated.
3175  *
3176  *	3 pass) Initialize the data structures just allocated with the
3177  *		information in the ACPI table about available AMD IOMMUs
3178  *		in the system. It also maps the PCI devices in the
3179  *		system to specific IOMMUs
3180  *
3181  *	4 pass) After the basic data structures are allocated and
3182  *		initialized we update them with information about memory
3183  *		remapping requirements parsed out of the ACPI table in
3184  *		this last pass.
3185  *
3186  * After everything is set up the IOMMUs are enabled and the necessary
3187  * hotplug and suspend notifiers are registered.
3188  */
3189 static int __init early_amd_iommu_init(void)
3190 {
3191 	struct acpi_table_header *ivrs_base;
3192 	int ret;
3193 	acpi_status status;
3194 	u8 efr_hats;
3195 
3196 	if (!amd_iommu_detected)
3197 		return -ENODEV;
3198 
3199 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3200 	if (status == AE_NOT_FOUND)
3201 		return -ENODEV;
3202 	else if (ACPI_FAILURE(status)) {
3203 		const char *err = acpi_format_exception(status);
3204 		pr_err("IVRS table error: %s\n", err);
3205 		return -EINVAL;
3206 	}
3207 
3208 	if (!boot_cpu_has(X86_FEATURE_CX16)) {
3209 		pr_err("Failed to initialize. The CMPXCHG16B feature is required.\n");
3210 		ret = -EINVAL;
3211 		goto out;
3212 	}
3213 
3214 	/*
3215 	 * Validate checksum here so we don't need to do it when
3216 	 * we actually parse the table
3217 	 */
3218 	ret = check_ivrs_checksum(ivrs_base);
3219 	if (ret)
3220 		goto out;
3221 
3222 	ivinfo_init(ivrs_base);
3223 
3224 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
3225 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
3226 
3227 	/*
3228 	 * now the data structures are allocated and basically initialized
3229 	 * start the real acpi table scan
3230 	 */
3231 	ret = init_iommu_all(ivrs_base);
3232 	if (ret)
3233 		goto out;
3234 
3235 	/* 5 level guest page table */
3236 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3237 	    FIELD_GET(FEATURE_GATS, amd_iommu_efr) == GUEST_PGTABLE_5_LEVEL)
3238 		amd_iommu_gpt_level = PAGE_MODE_5_LEVEL;
3239 
3240 	efr_hats = FIELD_GET(FEATURE_HATS, amd_iommu_efr);
3241 	if (efr_hats != 0x3) {
3242 		/*
3243 		 * efr[HATS] bits specify the maximum host translation level
3244 		 * supported, with LEVEL 4 being initial max level.
3245 		 */
3246 		amd_iommu_hpt_level = efr_hats + PAGE_MODE_4_LEVEL;
3247 	} else {
3248 		pr_warn_once(FW_BUG "Disable host address translation due to invalid translation level (%#x).\n",
3249 			     efr_hats);
3250 		amd_iommu_hatdis = true;
3251 	}
3252 
3253 	if (amd_iommu_pgtable == PD_MODE_V2) {
3254 		if (!amd_iommu_v2_pgtbl_supported()) {
3255 			pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n");
3256 			amd_iommu_pgtable = PD_MODE_V1;
3257 		}
3258 	}
3259 
3260 	if (amd_iommu_hatdis) {
3261 		/*
3262 		 * Host (v1) page table is not available. Attempt to use
3263 		 * Guest (v2) page table.
3264 		 */
3265 		if (amd_iommu_v2_pgtbl_supported())
3266 			amd_iommu_pgtable = PD_MODE_V2;
3267 		else
3268 			amd_iommu_pgtable = PD_MODE_NONE;
3269 	}
3270 
3271 	/* Disable any previously enabled IOMMUs */
3272 	if (!is_kdump_kernel() || amd_iommu_disabled)
3273 		disable_iommus();
3274 
3275 	if (amd_iommu_irq_remap)
3276 		amd_iommu_irq_remap = check_ioapic_information();
3277 
3278 	if (amd_iommu_irq_remap) {
3279 		struct amd_iommu_pci_seg *pci_seg;
3280 		ret = -ENOMEM;
3281 		for_each_pci_segment(pci_seg) {
3282 			if (alloc_irq_lookup_table(pci_seg))
3283 				goto out;
3284 		}
3285 	}
3286 
3287 	ret = init_memory_definitions(ivrs_base);
3288 	if (ret)
3289 		goto out;
3290 
3291 	/* init the device table */
3292 	init_device_table();
3293 
3294 out:
3295 	/* Don't leak any ACPI memory */
3296 	acpi_put_table(ivrs_base);
3297 
3298 	return ret;
3299 }
3300 
3301 static int amd_iommu_enable_interrupts(void)
3302 {
3303 	struct amd_iommu *iommu;
3304 	int ret = 0;
3305 
3306 	for_each_iommu(iommu) {
3307 		ret = iommu_init_irq(iommu);
3308 		if (ret)
3309 			goto out;
3310 	}
3311 
3312 	/*
3313 	 * Interrupt handler is ready to process interrupts. Enable
3314 	 * PPR and GA log interrupt for all IOMMUs.
3315 	 */
3316 	enable_iommus_vapic();
3317 	enable_iommus_ppr();
3318 
3319 out:
3320 	return ret;
3321 }
3322 
3323 static bool __init detect_ivrs(void)
3324 {
3325 	struct acpi_table_header *ivrs_base;
3326 	acpi_status status;
3327 	int i;
3328 
3329 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3330 	if (status == AE_NOT_FOUND)
3331 		return false;
3332 	else if (ACPI_FAILURE(status)) {
3333 		const char *err = acpi_format_exception(status);
3334 		pr_err("IVRS table error: %s\n", err);
3335 		return false;
3336 	}
3337 
3338 	acpi_put_table(ivrs_base);
3339 
3340 	if (amd_iommu_force_enable)
3341 		goto out;
3342 
3343 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3344 	for (i = 0; i < 32; i++) {
3345 		u32 pci_id;
3346 
3347 		pci_id = read_pci_config(0, i, 0, 0);
3348 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3349 			pr_info("Disable IOMMU on Stoney Ridge\n");
3350 			return false;
3351 		}
3352 	}
3353 
3354 out:
3355 	/* Make sure ACS will be enabled during PCI probe */
3356 	pci_request_acs();
3357 
3358 	return true;
3359 }
3360 
3361 static __init void iommu_snp_enable(void)
3362 {
3363 #ifdef CONFIG_KVM_AMD_SEV
3364 	if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3365 		return;
3366 	/*
3367 	 * The SNP support requires that IOMMU must be enabled, and is
3368 	 * configured with V1 page table (DTE[Mode] = 0 is not supported).
3369 	 */
3370 	if (no_iommu || iommu_default_passthrough()) {
3371 		pr_warn("SNP: IOMMU disabled or configured in passthrough mode, SNP cannot be supported.\n");
3372 		goto disable_snp;
3373 	}
3374 
3375 	if (amd_iommu_pgtable != PD_MODE_V1) {
3376 		pr_warn("SNP: IOMMU is configured with V2 page table mode, SNP cannot be supported.\n");
3377 		goto disable_snp;
3378 	}
3379 
3380 	amd_iommu_snp_en = check_feature(FEATURE_SNP);
3381 	if (!amd_iommu_snp_en) {
3382 		pr_warn("SNP: IOMMU SNP feature not enabled, SNP cannot be supported.\n");
3383 		goto disable_snp;
3384 	}
3385 
3386 	/*
3387 	 * Enable host SNP support once SNP support is checked on IOMMU.
3388 	 */
3389 	if (snp_rmptable_init()) {
3390 		pr_warn("SNP: RMP initialization failed, SNP cannot be supported.\n");
3391 		goto disable_snp;
3392 	}
3393 
3394 	pr_info("IOMMU SNP support enabled.\n");
3395 	return;
3396 
3397 disable_snp:
3398 	cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3399 #endif
3400 }
3401 
3402 /****************************************************************************
3403  *
3404  * AMD IOMMU Initialization State Machine
3405  *
3406  ****************************************************************************/
3407 
3408 static int __init state_next(void)
3409 {
3410 	int ret = 0;
3411 
3412 	switch (init_state) {
3413 	case IOMMU_START_STATE:
3414 		if (!detect_ivrs()) {
3415 			init_state	= IOMMU_NOT_FOUND;
3416 			ret		= -ENODEV;
3417 		} else {
3418 			init_state	= IOMMU_IVRS_DETECTED;
3419 		}
3420 		break;
3421 	case IOMMU_IVRS_DETECTED:
3422 		if (amd_iommu_disabled) {
3423 			init_state = IOMMU_CMDLINE_DISABLED;
3424 			ret = -EINVAL;
3425 		} else {
3426 			ret = early_amd_iommu_init();
3427 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3428 		}
3429 		break;
3430 	case IOMMU_ACPI_FINISHED:
3431 		early_enable_iommus();
3432 		x86_platform.iommu_shutdown = disable_iommus;
3433 		init_state = IOMMU_ENABLED;
3434 		break;
3435 	case IOMMU_ENABLED:
3436 		register_syscore(&amd_iommu_syscore);
3437 		iommu_snp_enable();
3438 		ret = amd_iommu_init_pci();
3439 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3440 		break;
3441 	case IOMMU_PCI_INIT:
3442 		ret = amd_iommu_enable_interrupts();
3443 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3444 		break;
3445 	case IOMMU_INTERRUPTS_EN:
3446 		init_state = IOMMU_INITIALIZED;
3447 		break;
3448 	case IOMMU_INITIALIZED:
3449 		/* Nothing to do */
3450 		break;
3451 	case IOMMU_NOT_FOUND:
3452 	case IOMMU_INIT_ERROR:
3453 	case IOMMU_CMDLINE_DISABLED:
3454 		/* Error states => do nothing */
3455 		ret = -EINVAL;
3456 		break;
3457 	default:
3458 		/* Unknown state */
3459 		BUG();
3460 	}
3461 
3462 	if (ret) {
3463 		free_dma_resources();
3464 		if (!irq_remapping_enabled) {
3465 			disable_iommus();
3466 			free_iommu_resources();
3467 		} else {
3468 			struct amd_iommu *iommu;
3469 			struct amd_iommu_pci_seg *pci_seg;
3470 
3471 			for_each_pci_segment(pci_seg)
3472 				uninit_device_table_dma(pci_seg);
3473 
3474 			for_each_iommu(iommu)
3475 				amd_iommu_flush_all_caches(iommu);
3476 		}
3477 	}
3478 	return ret;
3479 }
3480 
3481 static int __init iommu_go_to_state(enum iommu_init_state state)
3482 {
3483 	int ret = -EINVAL;
3484 
3485 	while (init_state != state) {
3486 		if (init_state == IOMMU_NOT_FOUND         ||
3487 		    init_state == IOMMU_INIT_ERROR        ||
3488 		    init_state == IOMMU_CMDLINE_DISABLED)
3489 			break;
3490 		ret = state_next();
3491 	}
3492 
3493 	/*
3494 	 * SNP platform initilazation requires IOMMUs to be fully configured.
3495 	 * If the SNP support on IOMMUs has NOT been checked, simply mark SNP
3496 	 * as unsupported. If the SNP support on IOMMUs has been checked and
3497 	 * host SNP support enabled but RMP enforcement has not been enabled
3498 	 * in IOMMUs, then the system is in a half-baked state, but can limp
3499 	 * along as all memory should be Hypervisor-Owned in the RMP. WARN,
3500 	 * but leave SNP as "supported" to avoid confusing the kernel.
3501 	 */
3502 	if (ret && cc_platform_has(CC_ATTR_HOST_SEV_SNP) &&
3503 	    !WARN_ON_ONCE(amd_iommu_snp_en))
3504 		cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3505 
3506 	return ret;
3507 }
3508 
3509 #ifdef CONFIG_IRQ_REMAP
3510 int __init amd_iommu_prepare(void)
3511 {
3512 	int ret;
3513 
3514 	amd_iommu_irq_remap = true;
3515 
3516 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3517 	if (ret) {
3518 		amd_iommu_irq_remap = false;
3519 		return ret;
3520 	}
3521 
3522 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3523 }
3524 
3525 int __init amd_iommu_enable(void)
3526 {
3527 	int ret;
3528 
3529 	ret = iommu_go_to_state(IOMMU_ENABLED);
3530 	if (ret)
3531 		return ret;
3532 
3533 	irq_remapping_enabled = 1;
3534 	return amd_iommu_xt_mode;
3535 }
3536 
3537 void amd_iommu_disable(void)
3538 {
3539 	amd_iommu_suspend(NULL);
3540 }
3541 
3542 int amd_iommu_reenable(int mode)
3543 {
3544 	amd_iommu_resume(NULL);
3545 
3546 	return 0;
3547 }
3548 
3549 int amd_iommu_enable_faulting(unsigned int cpu)
3550 {
3551 	/* We enable MSI later when PCI is initialized */
3552 	return 0;
3553 }
3554 #endif
3555 
3556 /*
3557  * This is the core init function for AMD IOMMU hardware in the system.
3558  * This function is called from the generic x86 DMA layer initialization
3559  * code.
3560  */
3561 static int __init amd_iommu_init(void)
3562 {
3563 	int ret;
3564 
3565 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3566 #ifdef CONFIG_GART_IOMMU
3567 	if (ret && list_empty(&amd_iommu_list)) {
3568 		/*
3569 		 * We failed to initialize the AMD IOMMU - try fallback
3570 		 * to GART if possible.
3571 		 */
3572 		gart_iommu_init();
3573 	}
3574 #endif
3575 
3576 	if (!ret)
3577 		amd_iommu_debugfs_setup();
3578 
3579 	return ret;
3580 }
3581 
3582 static bool amd_iommu_sme_check(void)
3583 {
3584 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3585 	    (boot_cpu_data.x86 != 0x17))
3586 		return true;
3587 
3588 	/* For Fam17h, a specific level of support is required */
3589 	if (boot_cpu_data.microcode >= 0x08001205)
3590 		return true;
3591 
3592 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3593 	    (boot_cpu_data.microcode <= 0x080011ff))
3594 		return true;
3595 
3596 	pr_notice("IOMMU not currently supported when SME is active\n");
3597 
3598 	return false;
3599 }
3600 
3601 /****************************************************************************
3602  *
3603  * Early detect code. This code runs at IOMMU detection time in the DMA
3604  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3605  * IOMMUs
3606  *
3607  ****************************************************************************/
3608 void __init amd_iommu_detect(void)
3609 {
3610 	int ret;
3611 
3612 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3613 		goto disable_snp;
3614 
3615 	if (!amd_iommu_sme_check())
3616 		goto disable_snp;
3617 
3618 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3619 	if (ret)
3620 		goto disable_snp;
3621 
3622 	amd_iommu_detected = true;
3623 	iommu_detected = 1;
3624 	x86_init.iommu.iommu_init = amd_iommu_init;
3625 	return;
3626 
3627 disable_snp:
3628 	if (cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3629 		cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3630 }
3631 
3632 /****************************************************************************
3633  *
3634  * Parsing functions for the AMD IOMMU specific kernel command line
3635  * options.
3636  *
3637  ****************************************************************************/
3638 
3639 static int __init parse_amd_iommu_dump(char *str)
3640 {
3641 	amd_iommu_dump = true;
3642 
3643 	return 1;
3644 }
3645 
3646 static int __init parse_amd_iommu_intr(char *str)
3647 {
3648 	for (; *str; ++str) {
3649 		if (strncmp(str, "legacy", 6) == 0) {
3650 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3651 			break;
3652 		}
3653 		if (strncmp(str, "vapic", 5) == 0) {
3654 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3655 			break;
3656 		}
3657 	}
3658 	return 1;
3659 }
3660 
3661 static int __init parse_amd_iommu_options(char *str)
3662 {
3663 	if (!str)
3664 		return -EINVAL;
3665 
3666 	while (*str) {
3667 		if (strncmp(str, "fullflush", 9) == 0) {
3668 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3669 			iommu_set_dma_strict();
3670 		} else if (strncmp(str, "force_enable", 12) == 0) {
3671 			amd_iommu_force_enable = true;
3672 		} else if (strncmp(str, "off", 3) == 0) {
3673 			amd_iommu_disabled = true;
3674 		} else if (strncmp(str, "force_isolation", 15) == 0) {
3675 			amd_iommu_force_isolation = true;
3676 		} else if (strncmp(str, "pgtbl_v1", 8) == 0) {
3677 			amd_iommu_pgtable = PD_MODE_V1;
3678 		} else if (strncmp(str, "pgtbl_v2", 8) == 0) {
3679 			amd_iommu_pgtable = PD_MODE_V2;
3680 		} else if (strncmp(str, "irtcachedis", 11) == 0) {
3681 			amd_iommu_irtcachedis = true;
3682 		} else if (strncmp(str, "nohugepages", 11) == 0) {
3683 			pr_info("Restricting V1 page-sizes to 4KiB");
3684 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_4K;
3685 		} else if (strncmp(str, "v2_pgsizes_only", 15) == 0) {
3686 			pr_info("Restricting V1 page-sizes to 4KiB/2MiB/1GiB");
3687 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_V2;
3688 		} else {
3689 			pr_notice("Unknown option - '%s'\n", str);
3690 		}
3691 
3692 		str += strcspn(str, ",");
3693 		while (*str == ',')
3694 			str++;
3695 	}
3696 
3697 	return 1;
3698 }
3699 
3700 static int __init parse_ivrs_ioapic(char *str)
3701 {
3702 	u32 seg = 0, bus, dev, fn;
3703 	int id, i;
3704 	u32 devid;
3705 
3706 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3707 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3708 		goto found;
3709 
3710 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3711 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3712 		pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n",
3713 			str, id, seg, bus, dev, fn);
3714 		goto found;
3715 	}
3716 
3717 	pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3718 	return 1;
3719 
3720 found:
3721 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3722 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3723 			str);
3724 		return 1;
3725 	}
3726 
3727 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3728 
3729 	cmdline_maps			= true;
3730 	i				= early_ioapic_map_size++;
3731 	early_ioapic_map[i].id		= id;
3732 	early_ioapic_map[i].devid	= devid;
3733 	early_ioapic_map[i].cmd_line	= true;
3734 
3735 	return 1;
3736 }
3737 
3738 static int __init parse_ivrs_hpet(char *str)
3739 {
3740 	u32 seg = 0, bus, dev, fn;
3741 	int id, i;
3742 	u32 devid;
3743 
3744 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3745 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3746 		goto found;
3747 
3748 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3749 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3750 		pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n",
3751 			str, id, seg, bus, dev, fn);
3752 		goto found;
3753 	}
3754 
3755 	pr_err("Invalid command line: ivrs_hpet%s\n", str);
3756 	return 1;
3757 
3758 found:
3759 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3760 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3761 			str);
3762 		return 1;
3763 	}
3764 
3765 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3766 
3767 	cmdline_maps			= true;
3768 	i				= early_hpet_map_size++;
3769 	early_hpet_map[i].id		= id;
3770 	early_hpet_map[i].devid		= devid;
3771 	early_hpet_map[i].cmd_line	= true;
3772 
3773 	return 1;
3774 }
3775 
3776 #define ACPIID_LEN (ACPIHID_UID_LEN + ACPIHID_HID_LEN)
3777 
3778 static int __init parse_ivrs_acpihid(char *str)
3779 {
3780 	u32 seg = 0, bus, dev, fn;
3781 	char *hid, *uid, *p, *addr;
3782 	char acpiid[ACPIID_LEN + 1] = { }; /* size with NULL terminator */
3783 	int i;
3784 
3785 	addr = strchr(str, '@');
3786 	if (!addr) {
3787 		addr = strchr(str, '=');
3788 		if (!addr)
3789 			goto not_found;
3790 
3791 		++addr;
3792 
3793 		if (strlen(addr) > ACPIID_LEN)
3794 			goto not_found;
3795 
3796 		if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 ||
3797 		    sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) {
3798 			pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n",
3799 				str, acpiid, seg, bus, dev, fn);
3800 			goto found;
3801 		}
3802 		goto not_found;
3803 	}
3804 
3805 	/* We have the '@', make it the terminator to get just the acpiid */
3806 	*addr++ = 0;
3807 
3808 	if (strlen(str) > ACPIID_LEN)
3809 		goto not_found;
3810 
3811 	if (sscanf(str, "=%s", acpiid) != 1)
3812 		goto not_found;
3813 
3814 	if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 ||
3815 	    sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4)
3816 		goto found;
3817 
3818 not_found:
3819 	pr_err("Invalid command line: ivrs_acpihid%s\n", str);
3820 	return 1;
3821 
3822 found:
3823 	p = acpiid;
3824 	hid = strsep(&p, ":");
3825 	uid = p;
3826 
3827 	if (!hid || !(*hid) || !uid) {
3828 		pr_err("Invalid command line: hid or uid\n");
3829 		return 1;
3830 	}
3831 
3832 	/*
3833 	 * Ignore leading zeroes after ':', so e.g., AMDI0095:00
3834 	 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match
3835 	 */
3836 	while (*uid == '0' && *(uid + 1))
3837 		uid++;
3838 
3839 	if (strlen(hid) >= ACPIHID_HID_LEN) {
3840 		pr_err("Invalid command line: hid is too long\n");
3841 		return 1;
3842 	} else if (strlen(uid) >= ACPIHID_UID_LEN) {
3843 		pr_err("Invalid command line: uid is too long\n");
3844 		return 1;
3845 	}
3846 
3847 	i = early_acpihid_map_size++;
3848 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3849 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3850 	early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3851 	early_acpihid_map[i].cmd_line	= true;
3852 
3853 	return 1;
3854 }
3855 
3856 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3857 __setup("amd_iommu=",		parse_amd_iommu_options);
3858 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3859 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3860 __setup("ivrs_hpet",		parse_ivrs_hpet);
3861 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3862 
3863 bool amd_iommu_pasid_supported(void)
3864 {
3865 	/* CPU page table size should match IOMMU guest page table size */
3866 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3867 	    amd_iommu_gpt_level != PAGE_MODE_5_LEVEL)
3868 		return false;
3869 
3870 	/*
3871 	 * Since DTE[Mode]=0 is prohibited on SNP-enabled system
3872 	 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without
3873 	 * setting up IOMMUv1 page table.
3874 	 */
3875 	return amd_iommu_gt_ppr_supported() && !amd_iommu_snp_en;
3876 }
3877 
3878 struct amd_iommu *get_amd_iommu(unsigned int idx)
3879 {
3880 	unsigned int i = 0;
3881 	struct amd_iommu *iommu;
3882 
3883 	for_each_iommu(iommu)
3884 		if (i++ == idx)
3885 			return iommu;
3886 	return NULL;
3887 }
3888 
3889 /****************************************************************************
3890  *
3891  * IOMMU EFR Performance Counter support functionality. This code allows
3892  * access to the IOMMU PC functionality.
3893  *
3894  ****************************************************************************/
3895 
3896 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3897 {
3898 	struct amd_iommu *iommu = get_amd_iommu(idx);
3899 
3900 	if (iommu)
3901 		return iommu->max_banks;
3902 
3903 	return 0;
3904 }
3905 
3906 bool amd_iommu_pc_supported(void)
3907 {
3908 	return amd_iommu_pc_present;
3909 }
3910 
3911 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3912 {
3913 	struct amd_iommu *iommu = get_amd_iommu(idx);
3914 
3915 	if (iommu)
3916 		return iommu->max_counters;
3917 
3918 	return 0;
3919 }
3920 
3921 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3922 				u8 fxn, u64 *value, bool is_write)
3923 {
3924 	u32 offset;
3925 	u32 max_offset_lim;
3926 
3927 	/* Make sure the IOMMU PC resource is available */
3928 	if (!amd_iommu_pc_present)
3929 		return -ENODEV;
3930 
3931 	/* Check for valid iommu and pc register indexing */
3932 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3933 		return -ENODEV;
3934 
3935 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3936 
3937 	/* Limit the offset to the hw defined mmio region aperture */
3938 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3939 				(iommu->max_counters << 8) | 0x28);
3940 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3941 	    (offset > max_offset_lim))
3942 		return -EINVAL;
3943 
3944 	if (is_write) {
3945 		u64 val = *value & GENMASK_ULL(47, 0);
3946 
3947 		writel((u32)val, iommu->mmio_base + offset);
3948 		writel((val >> 32), iommu->mmio_base + offset + 4);
3949 	} else {
3950 		*value = readl(iommu->mmio_base + offset + 4);
3951 		*value <<= 32;
3952 		*value |= readl(iommu->mmio_base + offset);
3953 		*value &= GENMASK_ULL(47, 0);
3954 	}
3955 
3956 	return 0;
3957 }
3958 
3959 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3960 {
3961 	if (!iommu)
3962 		return -EINVAL;
3963 
3964 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3965 }
3966 
3967 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3968 {
3969 	if (!iommu)
3970 		return -EINVAL;
3971 
3972 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3973 }
3974 
3975 #ifdef CONFIG_KVM_AMD_SEV
3976 static int iommu_page_make_shared(void *page)
3977 {
3978 	unsigned long paddr, pfn;
3979 
3980 	paddr = iommu_virt_to_phys(page);
3981 	/* Cbit maybe set in the paddr */
3982 	pfn = __sme_clr(paddr) >> PAGE_SHIFT;
3983 
3984 	if (!(pfn % PTRS_PER_PMD)) {
3985 		int ret, level;
3986 		bool assigned;
3987 
3988 		ret = snp_lookup_rmpentry(pfn, &assigned, &level);
3989 		if (ret) {
3990 			pr_warn("IOMMU PFN %lx RMP lookup failed, ret %d\n", pfn, ret);
3991 			return ret;
3992 		}
3993 
3994 		if (!assigned) {
3995 			pr_warn("IOMMU PFN %lx not assigned in RMP table\n", pfn);
3996 			return -EINVAL;
3997 		}
3998 
3999 		if (level > PG_LEVEL_4K) {
4000 			ret = psmash(pfn);
4001 			if (!ret)
4002 				goto done;
4003 
4004 			pr_warn("PSMASH failed for IOMMU PFN %lx huge RMP entry, ret: %d, level: %d\n",
4005 				pfn, ret, level);
4006 			return ret;
4007 		}
4008 	}
4009 
4010 done:
4011 	return rmp_make_shared(pfn, PG_LEVEL_4K);
4012 }
4013 
4014 static int iommu_make_shared(void *va, size_t size)
4015 {
4016 	void *page;
4017 	int ret;
4018 
4019 	if (!va)
4020 		return 0;
4021 
4022 	for (page = va; page < (va + size); page += PAGE_SIZE) {
4023 		ret = iommu_page_make_shared(page);
4024 		if (ret)
4025 			return ret;
4026 	}
4027 
4028 	return 0;
4029 }
4030 
4031 int amd_iommu_snp_disable(void)
4032 {
4033 	struct amd_iommu *iommu;
4034 	int ret;
4035 
4036 	if (!amd_iommu_snp_en)
4037 		return 0;
4038 
4039 	for_each_iommu(iommu) {
4040 		ret = iommu_make_shared(iommu->evt_buf, EVT_BUFFER_SIZE);
4041 		if (ret)
4042 			return ret;
4043 
4044 		ret = iommu_make_shared(iommu->ppr_log, PPR_LOG_SIZE);
4045 		if (ret)
4046 			return ret;
4047 
4048 		ret = iommu_make_shared((void *)iommu->cmd_sem, PAGE_SIZE);
4049 		if (ret)
4050 			return ret;
4051 	}
4052 
4053 	return 0;
4054 }
4055 EXPORT_SYMBOL_GPL(amd_iommu_snp_disable);
4056 
4057 bool amd_iommu_sev_tio_supported(void)
4058 {
4059 	return check_feature2(FEATURE_SEVSNPIO_SUP);
4060 }
4061 EXPORT_SYMBOL_GPL(amd_iommu_sev_tio_supported);
4062 #endif
4063