xref: /linux/drivers/iommu/amd/init.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
4  * Author: Joerg Roedel <jroedel@suse.de>
5  *         Leo Duran <leo.duran@amd.com>
6  */
7 
8 #define pr_fmt(fmt)     "AMD-Vi: " fmt
9 #define dev_fmt(fmt)    pr_fmt(fmt)
10 
11 #include <linux/pci.h>
12 #include <linux/acpi.h>
13 #include <linux/list.h>
14 #include <linux/bitmap.h>
15 #include <linux/slab.h>
16 #include <linux/syscore_ops.h>
17 #include <linux/interrupt.h>
18 #include <linux/msi.h>
19 #include <linux/irq.h>
20 #include <linux/amd-iommu.h>
21 #include <linux/export.h>
22 #include <linux/kmemleak.h>
23 #include <linux/cc_platform.h>
24 #include <linux/iopoll.h>
25 #include <asm/pci-direct.h>
26 #include <asm/iommu.h>
27 #include <asm/apic.h>
28 #include <asm/gart.h>
29 #include <asm/x86_init.h>
30 #include <asm/io_apic.h>
31 #include <asm/irq_remapping.h>
32 #include <asm/set_memory.h>
33 #include <asm/sev.h>
34 
35 #include <linux/crash_dump.h>
36 
37 #include "amd_iommu.h"
38 #include "../irq_remapping.h"
39 #include "../iommu-pages.h"
40 
41 /*
42  * definitions for the ACPI scanning code
43  */
44 #define IVRS_HEADER_LENGTH 48
45 
46 #define ACPI_IVHD_TYPE_MAX_SUPPORTED	0x40
47 #define ACPI_IVMD_TYPE_ALL              0x20
48 #define ACPI_IVMD_TYPE                  0x21
49 #define ACPI_IVMD_TYPE_RANGE            0x22
50 
51 #define IVHD_DEV_ALL                    0x01
52 #define IVHD_DEV_SELECT                 0x02
53 #define IVHD_DEV_SELECT_RANGE_START     0x03
54 #define IVHD_DEV_RANGE_END              0x04
55 #define IVHD_DEV_ALIAS                  0x42
56 #define IVHD_DEV_ALIAS_RANGE            0x43
57 #define IVHD_DEV_EXT_SELECT             0x46
58 #define IVHD_DEV_EXT_SELECT_RANGE       0x47
59 #define IVHD_DEV_SPECIAL		0x48
60 #define IVHD_DEV_ACPI_HID		0xf0
61 
62 #define UID_NOT_PRESENT                 0
63 #define UID_IS_INTEGER                  1
64 #define UID_IS_CHARACTER                2
65 
66 #define IVHD_SPECIAL_IOAPIC		1
67 #define IVHD_SPECIAL_HPET		2
68 
69 #define IVHD_FLAG_HT_TUN_EN_MASK        0x01
70 #define IVHD_FLAG_PASSPW_EN_MASK        0x02
71 #define IVHD_FLAG_RESPASSPW_EN_MASK     0x04
72 #define IVHD_FLAG_ISOC_EN_MASK          0x08
73 
74 #define IVMD_FLAG_EXCL_RANGE            0x08
75 #define IVMD_FLAG_IW                    0x04
76 #define IVMD_FLAG_IR                    0x02
77 #define IVMD_FLAG_UNITY_MAP             0x01
78 
79 #define ACPI_DEVFLAG_INITPASS           0x01
80 #define ACPI_DEVFLAG_EXTINT             0x02
81 #define ACPI_DEVFLAG_NMI                0x04
82 #define ACPI_DEVFLAG_SYSMGT1            0x10
83 #define ACPI_DEVFLAG_SYSMGT2            0x20
84 #define ACPI_DEVFLAG_LINT0              0x40
85 #define ACPI_DEVFLAG_LINT1              0x80
86 #define ACPI_DEVFLAG_ATSDIS             0x10000000
87 
88 #define IVRS_GET_SBDF_ID(seg, bus, dev, fn)	(((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \
89 						 | ((dev & 0x1f) << 3) | (fn & 0x7))
90 
91 /*
92  * ACPI table definitions
93  *
94  * These data structures are laid over the table to parse the important values
95  * out of it.
96  */
97 
98 /*
99  * structure describing one IOMMU in the ACPI table. Typically followed by one
100  * or more ivhd_entrys.
101  */
102 struct ivhd_header {
103 	u8 type;
104 	u8 flags;
105 	u16 length;
106 	u16 devid;
107 	u16 cap_ptr;
108 	u64 mmio_phys;
109 	u16 pci_seg;
110 	u16 info;
111 	u32 efr_attr;
112 
113 	/* Following only valid on IVHD type 11h and 40h */
114 	u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
115 	u64 efr_reg2;
116 } __attribute__((packed));
117 
118 /*
119  * A device entry describing which devices a specific IOMMU translates and
120  * which requestor ids they use.
121  */
122 struct ivhd_entry {
123 	u8 type;
124 	u16 devid;
125 	u8 flags;
126 	struct_group(ext_hid,
127 		u32 ext;
128 		u32 hidh;
129 	);
130 	u64 cid;
131 	u8 uidf;
132 	u8 uidl;
133 	u8 uid;
134 } __attribute__((packed));
135 
136 /*
137  * An AMD IOMMU memory definition structure. It defines things like exclusion
138  * ranges for devices and regions that should be unity mapped.
139  */
140 struct ivmd_header {
141 	u8 type;
142 	u8 flags;
143 	u16 length;
144 	u16 devid;
145 	u16 aux;
146 	u16 pci_seg;
147 	u8  resv[6];
148 	u64 range_start;
149 	u64 range_length;
150 } __attribute__((packed));
151 
152 bool amd_iommu_dump;
153 bool amd_iommu_irq_remap __read_mostly;
154 
155 enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1;
156 /* Guest page table level */
157 int amd_iommu_gpt_level = PAGE_MODE_4_LEVEL;
158 
159 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
160 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
161 
162 static bool amd_iommu_detected;
163 static bool amd_iommu_disabled __initdata;
164 static bool amd_iommu_force_enable __initdata;
165 static bool amd_iommu_irtcachedis;
166 static int amd_iommu_target_ivhd_type;
167 
168 /* Global EFR and EFR2 registers */
169 u64 amd_iommu_efr;
170 u64 amd_iommu_efr2;
171 
172 /* SNP is enabled on the system? */
173 bool amd_iommu_snp_en;
174 EXPORT_SYMBOL(amd_iommu_snp_en);
175 
176 LIST_HEAD(amd_iommu_pci_seg_list);	/* list of all PCI segments */
177 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
178 					   system */
179 
180 /* Array to assign indices to IOMMUs*/
181 struct amd_iommu *amd_iommus[MAX_IOMMUS];
182 
183 /* Number of IOMMUs present in the system */
184 static int amd_iommus_present;
185 
186 /* IOMMUs have a non-present cache? */
187 bool amd_iommu_np_cache __read_mostly;
188 bool amd_iommu_iotlb_sup __read_mostly = true;
189 
190 static bool amd_iommu_pc_present __read_mostly;
191 bool amdr_ivrs_remap_support __read_mostly;
192 
193 bool amd_iommu_force_isolation __read_mostly;
194 
195 unsigned long amd_iommu_pgsize_bitmap __ro_after_init = AMD_IOMMU_PGSIZES;
196 
197 /*
198  * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
199  * to know which ones are already in use.
200  */
201 unsigned long *amd_iommu_pd_alloc_bitmap;
202 
203 enum iommu_init_state {
204 	IOMMU_START_STATE,
205 	IOMMU_IVRS_DETECTED,
206 	IOMMU_ACPI_FINISHED,
207 	IOMMU_ENABLED,
208 	IOMMU_PCI_INIT,
209 	IOMMU_INTERRUPTS_EN,
210 	IOMMU_INITIALIZED,
211 	IOMMU_NOT_FOUND,
212 	IOMMU_INIT_ERROR,
213 	IOMMU_CMDLINE_DISABLED,
214 };
215 
216 /* Early ioapic and hpet maps from kernel command line */
217 #define EARLY_MAP_SIZE		4
218 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
219 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
220 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
221 
222 static int __initdata early_ioapic_map_size;
223 static int __initdata early_hpet_map_size;
224 static int __initdata early_acpihid_map_size;
225 
226 static bool __initdata cmdline_maps;
227 
228 static enum iommu_init_state init_state = IOMMU_START_STATE;
229 
230 static int amd_iommu_enable_interrupts(void);
231 static int __init iommu_go_to_state(enum iommu_init_state state);
232 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
233 
234 static bool amd_iommu_pre_enabled = true;
235 
236 static u32 amd_iommu_ivinfo __initdata;
237 
238 bool translation_pre_enabled(struct amd_iommu *iommu)
239 {
240 	return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
241 }
242 
243 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
244 {
245 	iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
246 }
247 
248 static void init_translation_status(struct amd_iommu *iommu)
249 {
250 	u64 ctrl;
251 
252 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
253 	if (ctrl & (1<<CONTROL_IOMMU_EN))
254 		iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
255 }
256 
257 static inline unsigned long tbl_size(int entry_size, int last_bdf)
258 {
259 	unsigned shift = PAGE_SHIFT +
260 			 get_order((last_bdf + 1) * entry_size);
261 
262 	return 1UL << shift;
263 }
264 
265 int amd_iommu_get_num_iommus(void)
266 {
267 	return amd_iommus_present;
268 }
269 
270 /*
271  * Iterate through all the IOMMUs to get common EFR
272  * masks among all IOMMUs and warn if found inconsistency.
273  */
274 static __init void get_global_efr(void)
275 {
276 	struct amd_iommu *iommu;
277 
278 	for_each_iommu(iommu) {
279 		u64 tmp = iommu->features;
280 		u64 tmp2 = iommu->features2;
281 
282 		if (list_is_first(&iommu->list, &amd_iommu_list)) {
283 			amd_iommu_efr = tmp;
284 			amd_iommu_efr2 = tmp2;
285 			continue;
286 		}
287 
288 		if (amd_iommu_efr == tmp &&
289 		    amd_iommu_efr2 == tmp2)
290 			continue;
291 
292 		pr_err(FW_BUG
293 		       "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n",
294 		       tmp, tmp2, amd_iommu_efr, amd_iommu_efr2,
295 		       iommu->index, iommu->pci_seg->id,
296 		       PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid),
297 		       PCI_FUNC(iommu->devid));
298 
299 		amd_iommu_efr &= tmp;
300 		amd_iommu_efr2 &= tmp2;
301 	}
302 
303 	pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2);
304 }
305 
306 /*
307  * For IVHD type 0x11/0x40, EFR is also available via IVHD.
308  * Default to IVHD EFR since it is available sooner
309  * (i.e. before PCI init).
310  */
311 static void __init early_iommu_features_init(struct amd_iommu *iommu,
312 					     struct ivhd_header *h)
313 {
314 	if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) {
315 		iommu->features = h->efr_reg;
316 		iommu->features2 = h->efr_reg2;
317 	}
318 	if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
319 		amdr_ivrs_remap_support = true;
320 }
321 
322 /* Access to l1 and l2 indexed register spaces */
323 
324 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
325 {
326 	u32 val;
327 
328 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
329 	pci_read_config_dword(iommu->dev, 0xfc, &val);
330 	return val;
331 }
332 
333 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
334 {
335 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
336 	pci_write_config_dword(iommu->dev, 0xfc, val);
337 	pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
338 }
339 
340 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
341 {
342 	u32 val;
343 
344 	pci_write_config_dword(iommu->dev, 0xf0, address);
345 	pci_read_config_dword(iommu->dev, 0xf4, &val);
346 	return val;
347 }
348 
349 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
350 {
351 	pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
352 	pci_write_config_dword(iommu->dev, 0xf4, val);
353 }
354 
355 /****************************************************************************
356  *
357  * AMD IOMMU MMIO register space handling functions
358  *
359  * These functions are used to program the IOMMU device registers in
360  * MMIO space required for that driver.
361  *
362  ****************************************************************************/
363 
364 /*
365  * This function set the exclusion range in the IOMMU. DMA accesses to the
366  * exclusion range are passed through untranslated
367  */
368 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
369 {
370 	u64 start = iommu->exclusion_start & PAGE_MASK;
371 	u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
372 	u64 entry;
373 
374 	if (!iommu->exclusion_start)
375 		return;
376 
377 	entry = start | MMIO_EXCL_ENABLE_MASK;
378 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
379 			&entry, sizeof(entry));
380 
381 	entry = limit;
382 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
383 			&entry, sizeof(entry));
384 }
385 
386 static void iommu_set_cwwb_range(struct amd_iommu *iommu)
387 {
388 	u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
389 	u64 entry = start & PM_ADDR_MASK;
390 
391 	if (!check_feature(FEATURE_SNP))
392 		return;
393 
394 	/* Note:
395 	 * Re-purpose Exclusion base/limit registers for Completion wait
396 	 * write-back base/limit.
397 	 */
398 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
399 		    &entry, sizeof(entry));
400 
401 	/* Note:
402 	 * Default to 4 Kbytes, which can be specified by setting base
403 	 * address equal to the limit address.
404 	 */
405 	memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
406 		    &entry, sizeof(entry));
407 }
408 
409 /* Programs the physical address of the device table into the IOMMU hardware */
410 static void iommu_set_device_table(struct amd_iommu *iommu)
411 {
412 	u64 entry;
413 	u32 dev_table_size = iommu->pci_seg->dev_table_size;
414 	void *dev_table = (void *)get_dev_table(iommu);
415 
416 	BUG_ON(iommu->mmio_base == NULL);
417 
418 	entry = iommu_virt_to_phys(dev_table);
419 	entry |= (dev_table_size >> 12) - 1;
420 	memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
421 			&entry, sizeof(entry));
422 }
423 
424 /* Generic functions to enable/disable certain features of the IOMMU. */
425 void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
426 {
427 	u64 ctrl;
428 
429 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
430 	ctrl |= (1ULL << bit);
431 	writeq(ctrl, iommu->mmio_base +  MMIO_CONTROL_OFFSET);
432 }
433 
434 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
435 {
436 	u64 ctrl;
437 
438 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
439 	ctrl &= ~(1ULL << bit);
440 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
441 }
442 
443 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
444 {
445 	u64 ctrl;
446 
447 	ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
448 	ctrl &= ~CTRL_INV_TO_MASK;
449 	ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
450 	writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
451 }
452 
453 /* Function to enable the hardware */
454 static void iommu_enable(struct amd_iommu *iommu)
455 {
456 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
457 }
458 
459 static void iommu_disable(struct amd_iommu *iommu)
460 {
461 	if (!iommu->mmio_base)
462 		return;
463 
464 	/* Disable command buffer */
465 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
466 
467 	/* Disable event logging and event interrupts */
468 	iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
469 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
470 
471 	/* Disable IOMMU GA_LOG */
472 	iommu_feature_disable(iommu, CONTROL_GALOG_EN);
473 	iommu_feature_disable(iommu, CONTROL_GAINT_EN);
474 
475 	/* Disable IOMMU PPR logging */
476 	iommu_feature_disable(iommu, CONTROL_PPRLOG_EN);
477 	iommu_feature_disable(iommu, CONTROL_PPRINT_EN);
478 
479 	/* Disable IOMMU hardware itself */
480 	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
481 
482 	/* Clear IRTE cache disabling bit */
483 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
484 }
485 
486 /*
487  * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
488  * the system has one.
489  */
490 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
491 {
492 	if (!request_mem_region(address, end, "amd_iommu")) {
493 		pr_err("Can not reserve memory region %llx-%llx for mmio\n",
494 			address, end);
495 		pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
496 		return NULL;
497 	}
498 
499 	return (u8 __iomem *)ioremap(address, end);
500 }
501 
502 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
503 {
504 	if (iommu->mmio_base)
505 		iounmap(iommu->mmio_base);
506 	release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
507 }
508 
509 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
510 {
511 	u32 size = 0;
512 
513 	switch (h->type) {
514 	case 0x10:
515 		size = 24;
516 		break;
517 	case 0x11:
518 	case 0x40:
519 		size = 40;
520 		break;
521 	}
522 	return size;
523 }
524 
525 /****************************************************************************
526  *
527  * The functions below belong to the first pass of AMD IOMMU ACPI table
528  * parsing. In this pass we try to find out the highest device id this
529  * code has to handle. Upon this information the size of the shared data
530  * structures is determined later.
531  *
532  ****************************************************************************/
533 
534 /*
535  * This function calculates the length of a given IVHD entry
536  */
537 static inline int ivhd_entry_length(u8 *ivhd)
538 {
539 	u32 type = ((struct ivhd_entry *)ivhd)->type;
540 
541 	if (type < 0x80) {
542 		return 0x04 << (*ivhd >> 6);
543 	} else if (type == IVHD_DEV_ACPI_HID) {
544 		/* For ACPI_HID, offset 21 is uid len */
545 		return *((u8 *)ivhd + 21) + 22;
546 	}
547 	return 0;
548 }
549 
550 /*
551  * After reading the highest device id from the IOMMU PCI capability header
552  * this function looks if there is a higher device id defined in the ACPI table
553  */
554 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
555 {
556 	u8 *p = (void *)h, *end = (void *)h;
557 	struct ivhd_entry *dev;
558 	int last_devid = -EINVAL;
559 
560 	u32 ivhd_size = get_ivhd_header_size(h);
561 
562 	if (!ivhd_size) {
563 		pr_err("Unsupported IVHD type %#x\n", h->type);
564 		return -EINVAL;
565 	}
566 
567 	p += ivhd_size;
568 	end += h->length;
569 
570 	while (p < end) {
571 		dev = (struct ivhd_entry *)p;
572 		switch (dev->type) {
573 		case IVHD_DEV_ALL:
574 			/* Use maximum BDF value for DEV_ALL */
575 			return 0xffff;
576 		case IVHD_DEV_SELECT:
577 		case IVHD_DEV_RANGE_END:
578 		case IVHD_DEV_ALIAS:
579 		case IVHD_DEV_EXT_SELECT:
580 			/* all the above subfield types refer to device ids */
581 			if (dev->devid > last_devid)
582 				last_devid = dev->devid;
583 			break;
584 		default:
585 			break;
586 		}
587 		p += ivhd_entry_length(p);
588 	}
589 
590 	WARN_ON(p != end);
591 
592 	return last_devid;
593 }
594 
595 static int __init check_ivrs_checksum(struct acpi_table_header *table)
596 {
597 	int i;
598 	u8 checksum = 0, *p = (u8 *)table;
599 
600 	for (i = 0; i < table->length; ++i)
601 		checksum += p[i];
602 	if (checksum != 0) {
603 		/* ACPI table corrupt */
604 		pr_err(FW_BUG "IVRS invalid checksum\n");
605 		return -ENODEV;
606 	}
607 
608 	return 0;
609 }
610 
611 /*
612  * Iterate over all IVHD entries in the ACPI table and find the highest device
613  * id which we need to handle. This is the first of three functions which parse
614  * the ACPI table. So we check the checksum here.
615  */
616 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
617 {
618 	u8 *p = (u8 *)table, *end = (u8 *)table;
619 	struct ivhd_header *h;
620 	int last_devid, last_bdf = 0;
621 
622 	p += IVRS_HEADER_LENGTH;
623 
624 	end += table->length;
625 	while (p < end) {
626 		h = (struct ivhd_header *)p;
627 		if (h->pci_seg == pci_seg &&
628 		    h->type == amd_iommu_target_ivhd_type) {
629 			last_devid = find_last_devid_from_ivhd(h);
630 
631 			if (last_devid < 0)
632 				return -EINVAL;
633 			if (last_devid > last_bdf)
634 				last_bdf = last_devid;
635 		}
636 		p += h->length;
637 	}
638 	WARN_ON(p != end);
639 
640 	return last_bdf;
641 }
642 
643 /****************************************************************************
644  *
645  * The following functions belong to the code path which parses the ACPI table
646  * the second time. In this ACPI parsing iteration we allocate IOMMU specific
647  * data structures, initialize the per PCI segment device/alias/rlookup table
648  * and also basically initialize the hardware.
649  *
650  ****************************************************************************/
651 
652 /* Allocate per PCI segment device table */
653 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
654 {
655 	pci_seg->dev_table = iommu_alloc_pages(GFP_KERNEL | GFP_DMA32,
656 					       get_order(pci_seg->dev_table_size));
657 	if (!pci_seg->dev_table)
658 		return -ENOMEM;
659 
660 	return 0;
661 }
662 
663 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
664 {
665 	iommu_free_pages(pci_seg->dev_table,
666 			 get_order(pci_seg->dev_table_size));
667 	pci_seg->dev_table = NULL;
668 }
669 
670 /* Allocate per PCI segment IOMMU rlookup table. */
671 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
672 {
673 	pci_seg->rlookup_table = iommu_alloc_pages(GFP_KERNEL,
674 						   get_order(pci_seg->rlookup_table_size));
675 	if (pci_seg->rlookup_table == NULL)
676 		return -ENOMEM;
677 
678 	return 0;
679 }
680 
681 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
682 {
683 	iommu_free_pages(pci_seg->rlookup_table,
684 			 get_order(pci_seg->rlookup_table_size));
685 	pci_seg->rlookup_table = NULL;
686 }
687 
688 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
689 {
690 	pci_seg->irq_lookup_table = iommu_alloc_pages(GFP_KERNEL,
691 						      get_order(pci_seg->rlookup_table_size));
692 	kmemleak_alloc(pci_seg->irq_lookup_table,
693 		       pci_seg->rlookup_table_size, 1, GFP_KERNEL);
694 	if (pci_seg->irq_lookup_table == NULL)
695 		return -ENOMEM;
696 
697 	return 0;
698 }
699 
700 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
701 {
702 	kmemleak_free(pci_seg->irq_lookup_table);
703 	iommu_free_pages(pci_seg->irq_lookup_table,
704 			 get_order(pci_seg->rlookup_table_size));
705 	pci_seg->irq_lookup_table = NULL;
706 }
707 
708 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
709 {
710 	int i;
711 
712 	pci_seg->alias_table = iommu_alloc_pages(GFP_KERNEL,
713 						 get_order(pci_seg->alias_table_size));
714 	if (!pci_seg->alias_table)
715 		return -ENOMEM;
716 
717 	/*
718 	 * let all alias entries point to itself
719 	 */
720 	for (i = 0; i <= pci_seg->last_bdf; ++i)
721 		pci_seg->alias_table[i] = i;
722 
723 	return 0;
724 }
725 
726 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
727 {
728 	iommu_free_pages(pci_seg->alias_table,
729 			 get_order(pci_seg->alias_table_size));
730 	pci_seg->alias_table = NULL;
731 }
732 
733 /*
734  * Allocates the command buffer. This buffer is per AMD IOMMU. We can
735  * write commands to that buffer later and the IOMMU will execute them
736  * asynchronously
737  */
738 static int __init alloc_command_buffer(struct amd_iommu *iommu)
739 {
740 	iommu->cmd_buf = iommu_alloc_pages(GFP_KERNEL,
741 					   get_order(CMD_BUFFER_SIZE));
742 
743 	return iommu->cmd_buf ? 0 : -ENOMEM;
744 }
745 
746 /*
747  * Interrupt handler has processed all pending events and adjusted head
748  * and tail pointer. Reset overflow mask and restart logging again.
749  */
750 void amd_iommu_restart_log(struct amd_iommu *iommu, const char *evt_type,
751 			   u8 cntrl_intr, u8 cntrl_log,
752 			   u32 status_run_mask, u32 status_overflow_mask)
753 {
754 	u32 status;
755 
756 	status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
757 	if (status & status_run_mask)
758 		return;
759 
760 	pr_info_ratelimited("IOMMU %s log restarting\n", evt_type);
761 
762 	iommu_feature_disable(iommu, cntrl_log);
763 	iommu_feature_disable(iommu, cntrl_intr);
764 
765 	writel(status_overflow_mask, iommu->mmio_base + MMIO_STATUS_OFFSET);
766 
767 	iommu_feature_enable(iommu, cntrl_intr);
768 	iommu_feature_enable(iommu, cntrl_log);
769 }
770 
771 /*
772  * This function restarts event logging in case the IOMMU experienced
773  * an event log buffer overflow.
774  */
775 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
776 {
777 	amd_iommu_restart_log(iommu, "Event", CONTROL_EVT_INT_EN,
778 			      CONTROL_EVT_LOG_EN, MMIO_STATUS_EVT_RUN_MASK,
779 			      MMIO_STATUS_EVT_OVERFLOW_MASK);
780 }
781 
782 /*
783  * This function restarts event logging in case the IOMMU experienced
784  * GA log overflow.
785  */
786 void amd_iommu_restart_ga_log(struct amd_iommu *iommu)
787 {
788 	amd_iommu_restart_log(iommu, "GA", CONTROL_GAINT_EN,
789 			      CONTROL_GALOG_EN, MMIO_STATUS_GALOG_RUN_MASK,
790 			      MMIO_STATUS_GALOG_OVERFLOW_MASK);
791 }
792 
793 /*
794  * This function resets the command buffer if the IOMMU stopped fetching
795  * commands from it.
796  */
797 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
798 {
799 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
800 
801 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
802 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
803 	iommu->cmd_buf_head = 0;
804 	iommu->cmd_buf_tail = 0;
805 
806 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
807 }
808 
809 /*
810  * This function writes the command buffer address to the hardware and
811  * enables it.
812  */
813 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
814 {
815 	u64 entry;
816 
817 	BUG_ON(iommu->cmd_buf == NULL);
818 
819 	entry = iommu_virt_to_phys(iommu->cmd_buf);
820 	entry |= MMIO_CMD_SIZE_512;
821 
822 	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
823 		    &entry, sizeof(entry));
824 
825 	amd_iommu_reset_cmd_buffer(iommu);
826 }
827 
828 /*
829  * This function disables the command buffer
830  */
831 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
832 {
833 	iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
834 }
835 
836 static void __init free_command_buffer(struct amd_iommu *iommu)
837 {
838 	iommu_free_pages(iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
839 }
840 
841 void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
842 				  size_t size)
843 {
844 	int order = get_order(size);
845 	void *buf = iommu_alloc_pages(gfp, order);
846 
847 	if (buf &&
848 	    check_feature(FEATURE_SNP) &&
849 	    set_memory_4k((unsigned long)buf, (1 << order))) {
850 		iommu_free_pages(buf, order);
851 		buf = NULL;
852 	}
853 
854 	return buf;
855 }
856 
857 /* allocates the memory where the IOMMU will log its events to */
858 static int __init alloc_event_buffer(struct amd_iommu *iommu)
859 {
860 	iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL,
861 					      EVT_BUFFER_SIZE);
862 
863 	return iommu->evt_buf ? 0 : -ENOMEM;
864 }
865 
866 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
867 {
868 	u64 entry;
869 
870 	BUG_ON(iommu->evt_buf == NULL);
871 
872 	entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
873 
874 	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
875 		    &entry, sizeof(entry));
876 
877 	/* set head and tail to zero manually */
878 	writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
879 	writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
880 
881 	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
882 }
883 
884 /*
885  * This function disables the event log buffer
886  */
887 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
888 {
889 	iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
890 }
891 
892 static void __init free_event_buffer(struct amd_iommu *iommu)
893 {
894 	iommu_free_pages(iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
895 }
896 
897 static void free_ga_log(struct amd_iommu *iommu)
898 {
899 #ifdef CONFIG_IRQ_REMAP
900 	iommu_free_pages(iommu->ga_log, get_order(GA_LOG_SIZE));
901 	iommu_free_pages(iommu->ga_log_tail, get_order(8));
902 #endif
903 }
904 
905 #ifdef CONFIG_IRQ_REMAP
906 static int iommu_ga_log_enable(struct amd_iommu *iommu)
907 {
908 	u32 status, i;
909 	u64 entry;
910 
911 	if (!iommu->ga_log)
912 		return -EINVAL;
913 
914 	entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
915 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
916 		    &entry, sizeof(entry));
917 	entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
918 		 (BIT_ULL(52)-1)) & ~7ULL;
919 	memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
920 		    &entry, sizeof(entry));
921 	writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
922 	writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
923 
924 
925 	iommu_feature_enable(iommu, CONTROL_GAINT_EN);
926 	iommu_feature_enable(iommu, CONTROL_GALOG_EN);
927 
928 	for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
929 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
930 		if (status & (MMIO_STATUS_GALOG_RUN_MASK))
931 			break;
932 		udelay(10);
933 	}
934 
935 	if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
936 		return -EINVAL;
937 
938 	return 0;
939 }
940 
941 static int iommu_init_ga_log(struct amd_iommu *iommu)
942 {
943 	if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
944 		return 0;
945 
946 	iommu->ga_log = iommu_alloc_pages(GFP_KERNEL, get_order(GA_LOG_SIZE));
947 	if (!iommu->ga_log)
948 		goto err_out;
949 
950 	iommu->ga_log_tail = iommu_alloc_pages(GFP_KERNEL, get_order(8));
951 	if (!iommu->ga_log_tail)
952 		goto err_out;
953 
954 	return 0;
955 err_out:
956 	free_ga_log(iommu);
957 	return -EINVAL;
958 }
959 #endif /* CONFIG_IRQ_REMAP */
960 
961 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
962 {
963 	iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL, 1);
964 
965 	return iommu->cmd_sem ? 0 : -ENOMEM;
966 }
967 
968 static void __init free_cwwb_sem(struct amd_iommu *iommu)
969 {
970 	if (iommu->cmd_sem)
971 		iommu_free_page((void *)iommu->cmd_sem);
972 }
973 
974 static void iommu_enable_xt(struct amd_iommu *iommu)
975 {
976 #ifdef CONFIG_IRQ_REMAP
977 	/*
978 	 * XT mode (32-bit APIC destination ID) requires
979 	 * GA mode (128-bit IRTE support) as a prerequisite.
980 	 */
981 	if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
982 	    amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
983 		iommu_feature_enable(iommu, CONTROL_XT_EN);
984 #endif /* CONFIG_IRQ_REMAP */
985 }
986 
987 static void iommu_enable_gt(struct amd_iommu *iommu)
988 {
989 	if (!check_feature(FEATURE_GT))
990 		return;
991 
992 	iommu_feature_enable(iommu, CONTROL_GT_EN);
993 }
994 
995 /* sets a specific bit in the device table entry. */
996 static void __set_dev_entry_bit(struct dev_table_entry *dev_table,
997 				u16 devid, u8 bit)
998 {
999 	int i = (bit >> 6) & 0x03;
1000 	int _bit = bit & 0x3f;
1001 
1002 	dev_table[devid].data[i] |= (1UL << _bit);
1003 }
1004 
1005 static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1006 {
1007 	struct dev_table_entry *dev_table = get_dev_table(iommu);
1008 
1009 	return __set_dev_entry_bit(dev_table, devid, bit);
1010 }
1011 
1012 static int __get_dev_entry_bit(struct dev_table_entry *dev_table,
1013 			       u16 devid, u8 bit)
1014 {
1015 	int i = (bit >> 6) & 0x03;
1016 	int _bit = bit & 0x3f;
1017 
1018 	return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1019 }
1020 
1021 static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1022 {
1023 	struct dev_table_entry *dev_table = get_dev_table(iommu);
1024 
1025 	return __get_dev_entry_bit(dev_table, devid, bit);
1026 }
1027 
1028 static bool __copy_device_table(struct amd_iommu *iommu)
1029 {
1030 	u64 int_ctl, int_tab_len, entry = 0;
1031 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1032 	struct dev_table_entry *old_devtb = NULL;
1033 	u32 lo, hi, devid, old_devtb_size;
1034 	phys_addr_t old_devtb_phys;
1035 	u16 dom_id, dte_v, irq_v;
1036 	u64 tmp;
1037 
1038 	/* Each IOMMU use separate device table with the same size */
1039 	lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1040 	hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1041 	entry = (((u64) hi) << 32) + lo;
1042 
1043 	old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1044 	if (old_devtb_size != pci_seg->dev_table_size) {
1045 		pr_err("The device table size of IOMMU:%d is not expected!\n",
1046 			iommu->index);
1047 		return false;
1048 	}
1049 
1050 	/*
1051 	 * When SME is enabled in the first kernel, the entry includes the
1052 	 * memory encryption mask(sme_me_mask), we must remove the memory
1053 	 * encryption mask to obtain the true physical address in kdump kernel.
1054 	 */
1055 	old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1056 
1057 	if (old_devtb_phys >= 0x100000000ULL) {
1058 		pr_err("The address of old device table is above 4G, not trustworthy!\n");
1059 		return false;
1060 	}
1061 	old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1062 		    ? (__force void *)ioremap_encrypted(old_devtb_phys,
1063 							pci_seg->dev_table_size)
1064 		    : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB);
1065 
1066 	if (!old_devtb)
1067 		return false;
1068 
1069 	pci_seg->old_dev_tbl_cpy = iommu_alloc_pages(GFP_KERNEL | GFP_DMA32,
1070 						     get_order(pci_seg->dev_table_size));
1071 	if (pci_seg->old_dev_tbl_cpy == NULL) {
1072 		pr_err("Failed to allocate memory for copying old device table!\n");
1073 		memunmap(old_devtb);
1074 		return false;
1075 	}
1076 
1077 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
1078 		pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1079 		dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1080 		dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1081 
1082 		if (dte_v && dom_id) {
1083 			pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1084 			pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1085 			__set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
1086 			/* If gcr3 table existed, mask it out */
1087 			if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1088 				tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1089 				tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1090 				pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1091 				tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1092 				tmp |= DTE_FLAG_GV;
1093 				pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1094 			}
1095 		}
1096 
1097 		irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1098 		int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1099 		int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1100 		if (irq_v && (int_ctl || int_tab_len)) {
1101 			if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1102 			    (int_tab_len != DTE_INTTABLEN)) {
1103 				pr_err("Wrong old irq remapping flag: %#x\n", devid);
1104 				memunmap(old_devtb);
1105 				return false;
1106 			}
1107 
1108 			pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1109 		}
1110 	}
1111 	memunmap(old_devtb);
1112 
1113 	return true;
1114 }
1115 
1116 static bool copy_device_table(void)
1117 {
1118 	struct amd_iommu *iommu;
1119 	struct amd_iommu_pci_seg *pci_seg;
1120 
1121 	if (!amd_iommu_pre_enabled)
1122 		return false;
1123 
1124 	pr_warn("Translation is already enabled - trying to copy translation structures\n");
1125 
1126 	/*
1127 	 * All IOMMUs within PCI segment shares common device table.
1128 	 * Hence copy device table only once per PCI segment.
1129 	 */
1130 	for_each_pci_segment(pci_seg) {
1131 		for_each_iommu(iommu) {
1132 			if (pci_seg->id != iommu->pci_seg->id)
1133 				continue;
1134 			if (!__copy_device_table(iommu))
1135 				return false;
1136 			break;
1137 		}
1138 	}
1139 
1140 	return true;
1141 }
1142 
1143 void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid)
1144 {
1145 	int sysmgt;
1146 
1147 	sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) |
1148 		 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1);
1149 
1150 	if (sysmgt == 0x01)
1151 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW);
1152 }
1153 
1154 /*
1155  * This function takes the device specific flags read from the ACPI
1156  * table and sets up the device table entry with that information
1157  */
1158 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1159 					   u16 devid, u32 flags, u32 ext_flags)
1160 {
1161 	if (flags & ACPI_DEVFLAG_INITPASS)
1162 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS);
1163 	if (flags & ACPI_DEVFLAG_EXTINT)
1164 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS);
1165 	if (flags & ACPI_DEVFLAG_NMI)
1166 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS);
1167 	if (flags & ACPI_DEVFLAG_SYSMGT1)
1168 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1);
1169 	if (flags & ACPI_DEVFLAG_SYSMGT2)
1170 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2);
1171 	if (flags & ACPI_DEVFLAG_LINT0)
1172 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS);
1173 	if (flags & ACPI_DEVFLAG_LINT1)
1174 		set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS);
1175 
1176 	amd_iommu_apply_erratum_63(iommu, devid);
1177 
1178 	amd_iommu_set_rlookup_table(iommu, devid);
1179 }
1180 
1181 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line)
1182 {
1183 	struct devid_map *entry;
1184 	struct list_head *list;
1185 
1186 	if (type == IVHD_SPECIAL_IOAPIC)
1187 		list = &ioapic_map;
1188 	else if (type == IVHD_SPECIAL_HPET)
1189 		list = &hpet_map;
1190 	else
1191 		return -EINVAL;
1192 
1193 	list_for_each_entry(entry, list, list) {
1194 		if (!(entry->id == id && entry->cmd_line))
1195 			continue;
1196 
1197 		pr_info("Command-line override present for %s id %d - ignoring\n",
1198 			type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1199 
1200 		*devid = entry->devid;
1201 
1202 		return 0;
1203 	}
1204 
1205 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1206 	if (!entry)
1207 		return -ENOMEM;
1208 
1209 	entry->id	= id;
1210 	entry->devid	= *devid;
1211 	entry->cmd_line	= cmd_line;
1212 
1213 	list_add_tail(&entry->list, list);
1214 
1215 	return 0;
1216 }
1217 
1218 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid,
1219 				      bool cmd_line)
1220 {
1221 	struct acpihid_map_entry *entry;
1222 	struct list_head *list = &acpihid_map;
1223 
1224 	list_for_each_entry(entry, list, list) {
1225 		if (strcmp(entry->hid, hid) ||
1226 		    (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1227 		    !entry->cmd_line)
1228 			continue;
1229 
1230 		pr_info("Command-line override for hid:%s uid:%s\n",
1231 			hid, uid);
1232 		*devid = entry->devid;
1233 		return 0;
1234 	}
1235 
1236 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1237 	if (!entry)
1238 		return -ENOMEM;
1239 
1240 	memcpy(entry->uid, uid, strlen(uid));
1241 	memcpy(entry->hid, hid, strlen(hid));
1242 	entry->devid = *devid;
1243 	entry->cmd_line	= cmd_line;
1244 	entry->root_devid = (entry->devid & (~0x7));
1245 
1246 	pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1247 		entry->cmd_line ? "cmd" : "ivrs",
1248 		entry->hid, entry->uid, entry->root_devid);
1249 
1250 	list_add_tail(&entry->list, list);
1251 	return 0;
1252 }
1253 
1254 static int __init add_early_maps(void)
1255 {
1256 	int i, ret;
1257 
1258 	for (i = 0; i < early_ioapic_map_size; ++i) {
1259 		ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1260 					 early_ioapic_map[i].id,
1261 					 &early_ioapic_map[i].devid,
1262 					 early_ioapic_map[i].cmd_line);
1263 		if (ret)
1264 			return ret;
1265 	}
1266 
1267 	for (i = 0; i < early_hpet_map_size; ++i) {
1268 		ret = add_special_device(IVHD_SPECIAL_HPET,
1269 					 early_hpet_map[i].id,
1270 					 &early_hpet_map[i].devid,
1271 					 early_hpet_map[i].cmd_line);
1272 		if (ret)
1273 			return ret;
1274 	}
1275 
1276 	for (i = 0; i < early_acpihid_map_size; ++i) {
1277 		ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1278 					  early_acpihid_map[i].uid,
1279 					  &early_acpihid_map[i].devid,
1280 					  early_acpihid_map[i].cmd_line);
1281 		if (ret)
1282 			return ret;
1283 	}
1284 
1285 	return 0;
1286 }
1287 
1288 /*
1289  * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1290  * initializes the hardware and our data structures with it.
1291  */
1292 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1293 					struct ivhd_header *h)
1294 {
1295 	u8 *p = (u8 *)h;
1296 	u8 *end = p, flags = 0;
1297 	u16 devid = 0, devid_start = 0, devid_to = 0, seg_id;
1298 	u32 dev_i, ext_flags = 0;
1299 	bool alias = false;
1300 	struct ivhd_entry *e;
1301 	struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1302 	u32 ivhd_size;
1303 	int ret;
1304 
1305 
1306 	ret = add_early_maps();
1307 	if (ret)
1308 		return ret;
1309 
1310 	amd_iommu_apply_ivrs_quirks();
1311 
1312 	/*
1313 	 * First save the recommended feature enable bits from ACPI
1314 	 */
1315 	iommu->acpi_flags = h->flags;
1316 
1317 	/*
1318 	 * Done. Now parse the device entries
1319 	 */
1320 	ivhd_size = get_ivhd_header_size(h);
1321 	if (!ivhd_size) {
1322 		pr_err("Unsupported IVHD type %#x\n", h->type);
1323 		return -EINVAL;
1324 	}
1325 
1326 	p += ivhd_size;
1327 
1328 	end += h->length;
1329 
1330 
1331 	while (p < end) {
1332 		e = (struct ivhd_entry *)p;
1333 		seg_id = pci_seg->id;
1334 
1335 		switch (e->type) {
1336 		case IVHD_DEV_ALL:
1337 
1338 			DUMP_printk("  DEV_ALL\t\t\tflags: %02x\n", e->flags);
1339 
1340 			for (dev_i = 0; dev_i <= pci_seg->last_bdf; ++dev_i)
1341 				set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1342 			break;
1343 		case IVHD_DEV_SELECT:
1344 
1345 			DUMP_printk("  DEV_SELECT\t\t\t devid: %04x:%02x:%02x.%x "
1346 				    "flags: %02x\n",
1347 				    seg_id, PCI_BUS_NUM(e->devid),
1348 				    PCI_SLOT(e->devid),
1349 				    PCI_FUNC(e->devid),
1350 				    e->flags);
1351 
1352 			devid = e->devid;
1353 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1354 			break;
1355 		case IVHD_DEV_SELECT_RANGE_START:
1356 
1357 			DUMP_printk("  DEV_SELECT_RANGE_START\t "
1358 				    "devid: %04x:%02x:%02x.%x flags: %02x\n",
1359 				    seg_id, PCI_BUS_NUM(e->devid),
1360 				    PCI_SLOT(e->devid),
1361 				    PCI_FUNC(e->devid),
1362 				    e->flags);
1363 
1364 			devid_start = e->devid;
1365 			flags = e->flags;
1366 			ext_flags = 0;
1367 			alias = false;
1368 			break;
1369 		case IVHD_DEV_ALIAS:
1370 
1371 			DUMP_printk("  DEV_ALIAS\t\t\t devid: %04x:%02x:%02x.%x "
1372 				    "flags: %02x devid_to: %02x:%02x.%x\n",
1373 				    seg_id, PCI_BUS_NUM(e->devid),
1374 				    PCI_SLOT(e->devid),
1375 				    PCI_FUNC(e->devid),
1376 				    e->flags,
1377 				    PCI_BUS_NUM(e->ext >> 8),
1378 				    PCI_SLOT(e->ext >> 8),
1379 				    PCI_FUNC(e->ext >> 8));
1380 
1381 			devid = e->devid;
1382 			devid_to = e->ext >> 8;
1383 			set_dev_entry_from_acpi(iommu, devid   , e->flags, 0);
1384 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1385 			pci_seg->alias_table[devid] = devid_to;
1386 			break;
1387 		case IVHD_DEV_ALIAS_RANGE:
1388 
1389 			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
1390 				    "devid: %04x:%02x:%02x.%x flags: %02x "
1391 				    "devid_to: %04x:%02x:%02x.%x\n",
1392 				    seg_id, PCI_BUS_NUM(e->devid),
1393 				    PCI_SLOT(e->devid),
1394 				    PCI_FUNC(e->devid),
1395 				    e->flags,
1396 				    seg_id, PCI_BUS_NUM(e->ext >> 8),
1397 				    PCI_SLOT(e->ext >> 8),
1398 				    PCI_FUNC(e->ext >> 8));
1399 
1400 			devid_start = e->devid;
1401 			flags = e->flags;
1402 			devid_to = e->ext >> 8;
1403 			ext_flags = 0;
1404 			alias = true;
1405 			break;
1406 		case IVHD_DEV_EXT_SELECT:
1407 
1408 			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %04x:%02x:%02x.%x "
1409 				    "flags: %02x ext: %08x\n",
1410 				    seg_id, PCI_BUS_NUM(e->devid),
1411 				    PCI_SLOT(e->devid),
1412 				    PCI_FUNC(e->devid),
1413 				    e->flags, e->ext);
1414 
1415 			devid = e->devid;
1416 			set_dev_entry_from_acpi(iommu, devid, e->flags,
1417 						e->ext);
1418 			break;
1419 		case IVHD_DEV_EXT_SELECT_RANGE:
1420 
1421 			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
1422 				    "%04x:%02x:%02x.%x flags: %02x ext: %08x\n",
1423 				    seg_id, PCI_BUS_NUM(e->devid),
1424 				    PCI_SLOT(e->devid),
1425 				    PCI_FUNC(e->devid),
1426 				    e->flags, e->ext);
1427 
1428 			devid_start = e->devid;
1429 			flags = e->flags;
1430 			ext_flags = e->ext;
1431 			alias = false;
1432 			break;
1433 		case IVHD_DEV_RANGE_END:
1434 
1435 			DUMP_printk("  DEV_RANGE_END\t\t devid: %04x:%02x:%02x.%x\n",
1436 				    seg_id, PCI_BUS_NUM(e->devid),
1437 				    PCI_SLOT(e->devid),
1438 				    PCI_FUNC(e->devid));
1439 
1440 			devid = e->devid;
1441 			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1442 				if (alias) {
1443 					pci_seg->alias_table[dev_i] = devid_to;
1444 					set_dev_entry_from_acpi(iommu,
1445 						devid_to, flags, ext_flags);
1446 				}
1447 				set_dev_entry_from_acpi(iommu, dev_i,
1448 							flags, ext_flags);
1449 			}
1450 			break;
1451 		case IVHD_DEV_SPECIAL: {
1452 			u8 handle, type;
1453 			const char *var;
1454 			u32 devid;
1455 			int ret;
1456 
1457 			handle = e->ext & 0xff;
1458 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8));
1459 			type   = (e->ext >> 24) & 0xff;
1460 
1461 			if (type == IVHD_SPECIAL_IOAPIC)
1462 				var = "IOAPIC";
1463 			else if (type == IVHD_SPECIAL_HPET)
1464 				var = "HPET";
1465 			else
1466 				var = "UNKNOWN";
1467 
1468 			DUMP_printk("  DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x\n",
1469 				    var, (int)handle,
1470 				    seg_id, PCI_BUS_NUM(devid),
1471 				    PCI_SLOT(devid),
1472 				    PCI_FUNC(devid));
1473 
1474 			ret = add_special_device(type, handle, &devid, false);
1475 			if (ret)
1476 				return ret;
1477 
1478 			/*
1479 			 * add_special_device might update the devid in case a
1480 			 * command-line override is present. So call
1481 			 * set_dev_entry_from_acpi after add_special_device.
1482 			 */
1483 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1484 
1485 			break;
1486 		}
1487 		case IVHD_DEV_ACPI_HID: {
1488 			u32 devid;
1489 			u8 hid[ACPIHID_HID_LEN];
1490 			u8 uid[ACPIHID_UID_LEN];
1491 			int ret;
1492 
1493 			if (h->type != 0x40) {
1494 				pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1495 				       e->type);
1496 				break;
1497 			}
1498 
1499 			BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1500 			memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1501 			hid[ACPIHID_HID_LEN - 1] = '\0';
1502 
1503 			if (!(*hid)) {
1504 				pr_err(FW_BUG "Invalid HID.\n");
1505 				break;
1506 			}
1507 
1508 			uid[0] = '\0';
1509 			switch (e->uidf) {
1510 			case UID_NOT_PRESENT:
1511 
1512 				if (e->uidl != 0)
1513 					pr_warn(FW_BUG "Invalid UID length.\n");
1514 
1515 				break;
1516 			case UID_IS_INTEGER:
1517 
1518 				sprintf(uid, "%d", e->uid);
1519 
1520 				break;
1521 			case UID_IS_CHARACTER:
1522 
1523 				memcpy(uid, &e->uid, e->uidl);
1524 				uid[e->uidl] = '\0';
1525 
1526 				break;
1527 			default:
1528 				break;
1529 			}
1530 
1531 			devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid);
1532 			DUMP_printk("  DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x\n",
1533 				    hid, uid, seg_id,
1534 				    PCI_BUS_NUM(devid),
1535 				    PCI_SLOT(devid),
1536 				    PCI_FUNC(devid));
1537 
1538 			flags = e->flags;
1539 
1540 			ret = add_acpi_hid_device(hid, uid, &devid, false);
1541 			if (ret)
1542 				return ret;
1543 
1544 			/*
1545 			 * add_special_device might update the devid in case a
1546 			 * command-line override is present. So call
1547 			 * set_dev_entry_from_acpi after add_special_device.
1548 			 */
1549 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1550 
1551 			break;
1552 		}
1553 		default:
1554 			break;
1555 		}
1556 
1557 		p += ivhd_entry_length(p);
1558 	}
1559 
1560 	return 0;
1561 }
1562 
1563 /* Allocate PCI segment data structure */
1564 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1565 					  struct acpi_table_header *ivrs_base)
1566 {
1567 	struct amd_iommu_pci_seg *pci_seg;
1568 	int last_bdf;
1569 
1570 	/*
1571 	 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1572 	 * handle in this PCI segment. Upon this information the shared data
1573 	 * structures for the PCI segments in the system will be allocated.
1574 	 */
1575 	last_bdf = find_last_devid_acpi(ivrs_base, id);
1576 	if (last_bdf < 0)
1577 		return NULL;
1578 
1579 	pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1580 	if (pci_seg == NULL)
1581 		return NULL;
1582 
1583 	pci_seg->last_bdf = last_bdf;
1584 	DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1585 	pci_seg->dev_table_size     = tbl_size(DEV_TABLE_ENTRY_SIZE, last_bdf);
1586 	pci_seg->alias_table_size   = tbl_size(ALIAS_TABLE_ENTRY_SIZE, last_bdf);
1587 	pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE, last_bdf);
1588 
1589 	pci_seg->id = id;
1590 	init_llist_head(&pci_seg->dev_data_list);
1591 	INIT_LIST_HEAD(&pci_seg->unity_map);
1592 	list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1593 
1594 	if (alloc_dev_table(pci_seg))
1595 		return NULL;
1596 	if (alloc_alias_table(pci_seg))
1597 		return NULL;
1598 	if (alloc_rlookup_table(pci_seg))
1599 		return NULL;
1600 
1601 	return pci_seg;
1602 }
1603 
1604 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1605 					struct acpi_table_header *ivrs_base)
1606 {
1607 	struct amd_iommu_pci_seg *pci_seg;
1608 
1609 	for_each_pci_segment(pci_seg) {
1610 		if (pci_seg->id == id)
1611 			return pci_seg;
1612 	}
1613 
1614 	return alloc_pci_segment(id, ivrs_base);
1615 }
1616 
1617 static void __init free_pci_segments(void)
1618 {
1619 	struct amd_iommu_pci_seg *pci_seg, *next;
1620 
1621 	for_each_pci_segment_safe(pci_seg, next) {
1622 		list_del(&pci_seg->list);
1623 		free_irq_lookup_table(pci_seg);
1624 		free_rlookup_table(pci_seg);
1625 		free_alias_table(pci_seg);
1626 		free_dev_table(pci_seg);
1627 		kfree(pci_seg);
1628 	}
1629 }
1630 
1631 static void __init free_sysfs(struct amd_iommu *iommu)
1632 {
1633 	if (iommu->iommu.dev) {
1634 		iommu_device_unregister(&iommu->iommu);
1635 		iommu_device_sysfs_remove(&iommu->iommu);
1636 	}
1637 }
1638 
1639 static void __init free_iommu_one(struct amd_iommu *iommu)
1640 {
1641 	free_sysfs(iommu);
1642 	free_cwwb_sem(iommu);
1643 	free_command_buffer(iommu);
1644 	free_event_buffer(iommu);
1645 	amd_iommu_free_ppr_log(iommu);
1646 	free_ga_log(iommu);
1647 	iommu_unmap_mmio_space(iommu);
1648 	amd_iommu_iopf_uninit(iommu);
1649 }
1650 
1651 static void __init free_iommu_all(void)
1652 {
1653 	struct amd_iommu *iommu, *next;
1654 
1655 	for_each_iommu_safe(iommu, next) {
1656 		list_del(&iommu->list);
1657 		free_iommu_one(iommu);
1658 		kfree(iommu);
1659 	}
1660 }
1661 
1662 /*
1663  * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1664  * Workaround:
1665  *     BIOS should disable L2B micellaneous clock gating by setting
1666  *     L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1667  */
1668 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1669 {
1670 	u32 value;
1671 
1672 	if ((boot_cpu_data.x86 != 0x15) ||
1673 	    (boot_cpu_data.x86_model < 0x10) ||
1674 	    (boot_cpu_data.x86_model > 0x1f))
1675 		return;
1676 
1677 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1678 	pci_read_config_dword(iommu->dev, 0xf4, &value);
1679 
1680 	if (value & BIT(2))
1681 		return;
1682 
1683 	/* Select NB indirect register 0x90 and enable writing */
1684 	pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1685 
1686 	pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1687 	pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1688 
1689 	/* Clear the enable writing bit */
1690 	pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1691 }
1692 
1693 /*
1694  * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1695  * Workaround:
1696  *     BIOS should enable ATS write permission check by setting
1697  *     L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1698  */
1699 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1700 {
1701 	u32 value;
1702 
1703 	if ((boot_cpu_data.x86 != 0x15) ||
1704 	    (boot_cpu_data.x86_model < 0x30) ||
1705 	    (boot_cpu_data.x86_model > 0x3f))
1706 		return;
1707 
1708 	/* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1709 	value = iommu_read_l2(iommu, 0x47);
1710 
1711 	if (value & BIT(0))
1712 		return;
1713 
1714 	/* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1715 	iommu_write_l2(iommu, 0x47, value | BIT(0));
1716 
1717 	pci_info(iommu->dev, "Applying ATS write check workaround\n");
1718 }
1719 
1720 /*
1721  * This function glues the initialization function for one IOMMU
1722  * together and also allocates the command buffer and programs the
1723  * hardware. It does NOT enable the IOMMU. This is done afterwards.
1724  */
1725 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1726 				 struct acpi_table_header *ivrs_base)
1727 {
1728 	struct amd_iommu_pci_seg *pci_seg;
1729 
1730 	pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1731 	if (pci_seg == NULL)
1732 		return -ENOMEM;
1733 	iommu->pci_seg = pci_seg;
1734 
1735 	raw_spin_lock_init(&iommu->lock);
1736 	atomic64_set(&iommu->cmd_sem_val, 0);
1737 
1738 	/* Add IOMMU to internal data structures */
1739 	list_add_tail(&iommu->list, &amd_iommu_list);
1740 	iommu->index = amd_iommus_present++;
1741 
1742 	if (unlikely(iommu->index >= MAX_IOMMUS)) {
1743 		WARN(1, "System has more IOMMUs than supported by this driver\n");
1744 		return -ENOSYS;
1745 	}
1746 
1747 	/* Index is fine - add IOMMU to the array */
1748 	amd_iommus[iommu->index] = iommu;
1749 
1750 	/*
1751 	 * Copy data from ACPI table entry to the iommu struct
1752 	 */
1753 	iommu->devid   = h->devid;
1754 	iommu->cap_ptr = h->cap_ptr;
1755 	iommu->mmio_phys = h->mmio_phys;
1756 
1757 	switch (h->type) {
1758 	case 0x10:
1759 		/* Check if IVHD EFR contains proper max banks/counters */
1760 		if ((h->efr_attr != 0) &&
1761 		    ((h->efr_attr & (0xF << 13)) != 0) &&
1762 		    ((h->efr_attr & (0x3F << 17)) != 0))
1763 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1764 		else
1765 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1766 
1767 		/*
1768 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1769 		 * GAM also requires GA mode. Therefore, we need to
1770 		 * check cmpxchg16b support before enabling it.
1771 		 */
1772 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1773 		    ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1774 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1775 		break;
1776 	case 0x11:
1777 	case 0x40:
1778 		if (h->efr_reg & (1 << 9))
1779 			iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1780 		else
1781 			iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1782 
1783 		/*
1784 		 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1785 		 * XT, GAM also requires GA mode. Therefore, we need to
1786 		 * check cmpxchg16b support before enabling them.
1787 		 */
1788 		if (!boot_cpu_has(X86_FEATURE_CX16) ||
1789 		    ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1790 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1791 			break;
1792 		}
1793 
1794 		if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1795 			amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1796 
1797 		early_iommu_features_init(iommu, h);
1798 
1799 		break;
1800 	default:
1801 		return -EINVAL;
1802 	}
1803 
1804 	iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1805 						iommu->mmio_phys_end);
1806 	if (!iommu->mmio_base)
1807 		return -ENOMEM;
1808 
1809 	return init_iommu_from_acpi(iommu, h);
1810 }
1811 
1812 static int __init init_iommu_one_late(struct amd_iommu *iommu)
1813 {
1814 	int ret;
1815 
1816 	if (alloc_cwwb_sem(iommu))
1817 		return -ENOMEM;
1818 
1819 	if (alloc_command_buffer(iommu))
1820 		return -ENOMEM;
1821 
1822 	if (alloc_event_buffer(iommu))
1823 		return -ENOMEM;
1824 
1825 	iommu->int_enabled = false;
1826 
1827 	init_translation_status(iommu);
1828 	if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1829 		iommu_disable(iommu);
1830 		clear_translation_pre_enabled(iommu);
1831 		pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1832 			iommu->index);
1833 	}
1834 	if (amd_iommu_pre_enabled)
1835 		amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1836 
1837 	if (amd_iommu_irq_remap) {
1838 		ret = amd_iommu_create_irq_domain(iommu);
1839 		if (ret)
1840 			return ret;
1841 	}
1842 
1843 	/*
1844 	 * Make sure IOMMU is not considered to translate itself. The IVRS
1845 	 * table tells us so, but this is a lie!
1846 	 */
1847 	iommu->pci_seg->rlookup_table[iommu->devid] = NULL;
1848 
1849 	return 0;
1850 }
1851 
1852 /**
1853  * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1854  * @ivrs: Pointer to the IVRS header
1855  *
1856  * This function search through all IVDB of the maximum supported IVHD
1857  */
1858 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1859 {
1860 	u8 *base = (u8 *)ivrs;
1861 	struct ivhd_header *ivhd = (struct ivhd_header *)
1862 					(base + IVRS_HEADER_LENGTH);
1863 	u8 last_type = ivhd->type;
1864 	u16 devid = ivhd->devid;
1865 
1866 	while (((u8 *)ivhd - base < ivrs->length) &&
1867 	       (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1868 		u8 *p = (u8 *) ivhd;
1869 
1870 		if (ivhd->devid == devid)
1871 			last_type = ivhd->type;
1872 		ivhd = (struct ivhd_header *)(p + ivhd->length);
1873 	}
1874 
1875 	return last_type;
1876 }
1877 
1878 /*
1879  * Iterates over all IOMMU entries in the ACPI table, allocates the
1880  * IOMMU structure and initializes it with init_iommu_one()
1881  */
1882 static int __init init_iommu_all(struct acpi_table_header *table)
1883 {
1884 	u8 *p = (u8 *)table, *end = (u8 *)table;
1885 	struct ivhd_header *h;
1886 	struct amd_iommu *iommu;
1887 	int ret;
1888 
1889 	end += table->length;
1890 	p += IVRS_HEADER_LENGTH;
1891 
1892 	/* Phase 1: Process all IVHD blocks */
1893 	while (p < end) {
1894 		h = (struct ivhd_header *)p;
1895 		if (*p == amd_iommu_target_ivhd_type) {
1896 
1897 			DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x "
1898 				    "flags: %01x info %04x\n",
1899 				    h->pci_seg, PCI_BUS_NUM(h->devid),
1900 				    PCI_SLOT(h->devid), PCI_FUNC(h->devid),
1901 				    h->cap_ptr, h->flags, h->info);
1902 			DUMP_printk("       mmio-addr: %016llx\n",
1903 				    h->mmio_phys);
1904 
1905 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1906 			if (iommu == NULL)
1907 				return -ENOMEM;
1908 
1909 			ret = init_iommu_one(iommu, h, table);
1910 			if (ret)
1911 				return ret;
1912 		}
1913 		p += h->length;
1914 
1915 	}
1916 	WARN_ON(p != end);
1917 
1918 	/* Phase 2 : Early feature support check */
1919 	get_global_efr();
1920 
1921 	/* Phase 3 : Enabling IOMMU features */
1922 	for_each_iommu(iommu) {
1923 		ret = init_iommu_one_late(iommu);
1924 		if (ret)
1925 			return ret;
1926 	}
1927 
1928 	return 0;
1929 }
1930 
1931 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1932 {
1933 	u64 val;
1934 	struct pci_dev *pdev = iommu->dev;
1935 
1936 	if (!check_feature(FEATURE_PC))
1937 		return;
1938 
1939 	amd_iommu_pc_present = true;
1940 
1941 	pci_info(pdev, "IOMMU performance counters supported\n");
1942 
1943 	val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1944 	iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1945 	iommu->max_counters = (u8) ((val >> 7) & 0xf);
1946 
1947 	return;
1948 }
1949 
1950 static ssize_t amd_iommu_show_cap(struct device *dev,
1951 				  struct device_attribute *attr,
1952 				  char *buf)
1953 {
1954 	struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1955 	return sysfs_emit(buf, "%x\n", iommu->cap);
1956 }
1957 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1958 
1959 static ssize_t amd_iommu_show_features(struct device *dev,
1960 				       struct device_attribute *attr,
1961 				       char *buf)
1962 {
1963 	return sysfs_emit(buf, "%llx:%llx\n", amd_iommu_efr, amd_iommu_efr2);
1964 }
1965 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1966 
1967 static struct attribute *amd_iommu_attrs[] = {
1968 	&dev_attr_cap.attr,
1969 	&dev_attr_features.attr,
1970 	NULL,
1971 };
1972 
1973 static struct attribute_group amd_iommu_group = {
1974 	.name = "amd-iommu",
1975 	.attrs = amd_iommu_attrs,
1976 };
1977 
1978 static const struct attribute_group *amd_iommu_groups[] = {
1979 	&amd_iommu_group,
1980 	NULL,
1981 };
1982 
1983 /*
1984  * Note: IVHD 0x11 and 0x40 also contains exact copy
1985  * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1986  * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1987  */
1988 static void __init late_iommu_features_init(struct amd_iommu *iommu)
1989 {
1990 	u64 features, features2;
1991 
1992 	if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1993 		return;
1994 
1995 	/* read extended feature bits */
1996 	features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1997 	features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2);
1998 
1999 	if (!amd_iommu_efr) {
2000 		amd_iommu_efr = features;
2001 		amd_iommu_efr2 = features2;
2002 		return;
2003 	}
2004 
2005 	/*
2006 	 * Sanity check and warn if EFR values from
2007 	 * IVHD and MMIO conflict.
2008 	 */
2009 	if (features != amd_iommu_efr ||
2010 	    features2 != amd_iommu_efr2) {
2011 		pr_warn(FW_WARN
2012 			"EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n",
2013 			features, amd_iommu_efr,
2014 			features2, amd_iommu_efr2);
2015 	}
2016 }
2017 
2018 static int __init iommu_init_pci(struct amd_iommu *iommu)
2019 {
2020 	int cap_ptr = iommu->cap_ptr;
2021 	int ret;
2022 
2023 	iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2024 						 PCI_BUS_NUM(iommu->devid),
2025 						 iommu->devid & 0xff);
2026 	if (!iommu->dev)
2027 		return -ENODEV;
2028 
2029 	/* Prevent binding other PCI device drivers to IOMMU devices */
2030 	iommu->dev->match_driver = false;
2031 
2032 	/* ACPI _PRT won't have an IRQ for IOMMU */
2033 	iommu->dev->irq_managed = 1;
2034 
2035 	pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2036 			      &iommu->cap);
2037 
2038 	if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2039 		amd_iommu_iotlb_sup = false;
2040 
2041 	late_iommu_features_init(iommu);
2042 
2043 	if (check_feature(FEATURE_GT)) {
2044 		int glxval;
2045 		u64 pasmax;
2046 
2047 		pasmax = FIELD_GET(FEATURE_PASMAX, amd_iommu_efr);
2048 		iommu->iommu.max_pasids = (1 << (pasmax + 1)) - 1;
2049 
2050 		BUG_ON(iommu->iommu.max_pasids & ~PASID_MASK);
2051 
2052 		glxval = FIELD_GET(FEATURE_GLX, amd_iommu_efr);
2053 
2054 		if (amd_iommu_max_glx_val == -1)
2055 			amd_iommu_max_glx_val = glxval;
2056 		else
2057 			amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2058 
2059 		iommu_enable_gt(iommu);
2060 	}
2061 
2062 	if (check_feature(FEATURE_PPR) && amd_iommu_alloc_ppr_log(iommu))
2063 		return -ENOMEM;
2064 
2065 	if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2066 		pr_info("Using strict mode due to virtualization\n");
2067 		iommu_set_dma_strict();
2068 		amd_iommu_np_cache = true;
2069 	}
2070 
2071 	init_iommu_perf_ctr(iommu);
2072 
2073 	if (amd_iommu_pgtable == AMD_IOMMU_V2) {
2074 		if (!check_feature(FEATURE_GIOSUP) ||
2075 		    !check_feature(FEATURE_GT)) {
2076 			pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n");
2077 			amd_iommu_pgtable = AMD_IOMMU_V1;
2078 		}
2079 	}
2080 
2081 	if (is_rd890_iommu(iommu->dev)) {
2082 		int i, j;
2083 
2084 		iommu->root_pdev =
2085 			pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2086 						    iommu->dev->bus->number,
2087 						    PCI_DEVFN(0, 0));
2088 
2089 		/*
2090 		 * Some rd890 systems may not be fully reconfigured by the
2091 		 * BIOS, so it's necessary for us to store this information so
2092 		 * it can be reprogrammed on resume
2093 		 */
2094 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2095 				&iommu->stored_addr_lo);
2096 		pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2097 				&iommu->stored_addr_hi);
2098 
2099 		/* Low bit locks writes to configuration space */
2100 		iommu->stored_addr_lo &= ~1;
2101 
2102 		for (i = 0; i < 6; i++)
2103 			for (j = 0; j < 0x12; j++)
2104 				iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2105 
2106 		for (i = 0; i < 0x83; i++)
2107 			iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2108 	}
2109 
2110 	amd_iommu_erratum_746_workaround(iommu);
2111 	amd_iommu_ats_write_check_workaround(iommu);
2112 
2113 	ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2114 			       amd_iommu_groups, "ivhd%d", iommu->index);
2115 	if (ret)
2116 		return ret;
2117 
2118 	/*
2119 	 * Allocate per IOMMU IOPF queue here so that in attach device path,
2120 	 * PRI capable device can be added to IOPF queue
2121 	 */
2122 	if (amd_iommu_gt_ppr_supported()) {
2123 		ret = amd_iommu_iopf_init(iommu);
2124 		if (ret)
2125 			return ret;
2126 	}
2127 
2128 	iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2129 
2130 	return pci_enable_device(iommu->dev);
2131 }
2132 
2133 static void print_iommu_info(void)
2134 {
2135 	int i;
2136 	static const char * const feat_str[] = {
2137 		"PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2138 		"IA", "GA", "HE", "PC"
2139 	};
2140 
2141 	if (amd_iommu_efr) {
2142 		pr_info("Extended features (%#llx, %#llx):", amd_iommu_efr, amd_iommu_efr2);
2143 
2144 		for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2145 			if (check_feature(1ULL << i))
2146 				pr_cont(" %s", feat_str[i]);
2147 		}
2148 
2149 		if (check_feature(FEATURE_GAM_VAPIC))
2150 			pr_cont(" GA_vAPIC");
2151 
2152 		if (check_feature(FEATURE_SNP))
2153 			pr_cont(" SNP");
2154 
2155 		pr_cont("\n");
2156 	}
2157 
2158 	if (irq_remapping_enabled) {
2159 		pr_info("Interrupt remapping enabled\n");
2160 		if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2161 			pr_info("X2APIC enabled\n");
2162 	}
2163 	if (amd_iommu_pgtable == AMD_IOMMU_V2) {
2164 		pr_info("V2 page table enabled (Paging mode : %d level)\n",
2165 			amd_iommu_gpt_level);
2166 	}
2167 }
2168 
2169 static int __init amd_iommu_init_pci(void)
2170 {
2171 	struct amd_iommu *iommu;
2172 	struct amd_iommu_pci_seg *pci_seg;
2173 	int ret;
2174 
2175 	for_each_iommu(iommu) {
2176 		ret = iommu_init_pci(iommu);
2177 		if (ret) {
2178 			pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2179 			       iommu->index, ret);
2180 			goto out;
2181 		}
2182 		/* Need to setup range after PCI init */
2183 		iommu_set_cwwb_range(iommu);
2184 	}
2185 
2186 	/*
2187 	 * Order is important here to make sure any unity map requirements are
2188 	 * fulfilled. The unity mappings are created and written to the device
2189 	 * table during the iommu_init_pci() call.
2190 	 *
2191 	 * After that we call init_device_table_dma() to make sure any
2192 	 * uninitialized DTE will block DMA, and in the end we flush the caches
2193 	 * of all IOMMUs to make sure the changes to the device table are
2194 	 * active.
2195 	 */
2196 	for_each_pci_segment(pci_seg)
2197 		init_device_table_dma(pci_seg);
2198 
2199 	for_each_iommu(iommu)
2200 		amd_iommu_flush_all_caches(iommu);
2201 
2202 	print_iommu_info();
2203 
2204 out:
2205 	return ret;
2206 }
2207 
2208 /****************************************************************************
2209  *
2210  * The following functions initialize the MSI interrupts for all IOMMUs
2211  * in the system. It's a bit challenging because there could be multiple
2212  * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2213  * pci_dev.
2214  *
2215  ****************************************************************************/
2216 
2217 static int iommu_setup_msi(struct amd_iommu *iommu)
2218 {
2219 	int r;
2220 
2221 	r = pci_enable_msi(iommu->dev);
2222 	if (r)
2223 		return r;
2224 
2225 	r = request_threaded_irq(iommu->dev->irq,
2226 				 amd_iommu_int_handler,
2227 				 amd_iommu_int_thread,
2228 				 0, "AMD-Vi",
2229 				 iommu);
2230 
2231 	if (r) {
2232 		pci_disable_msi(iommu->dev);
2233 		return r;
2234 	}
2235 
2236 	return 0;
2237 }
2238 
2239 union intcapxt {
2240 	u64	capxt;
2241 	struct {
2242 		u64	reserved_0		:  2,
2243 			dest_mode_logical	:  1,
2244 			reserved_1		:  5,
2245 			destid_0_23		: 24,
2246 			vector			:  8,
2247 			reserved_2		: 16,
2248 			destid_24_31		:  8;
2249 	};
2250 } __attribute__ ((packed));
2251 
2252 
2253 static struct irq_chip intcapxt_controller;
2254 
2255 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2256 				       struct irq_data *irqd, bool reserve)
2257 {
2258 	return 0;
2259 }
2260 
2261 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2262 					  struct irq_data *irqd)
2263 {
2264 }
2265 
2266 
2267 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2268 				    unsigned int nr_irqs, void *arg)
2269 {
2270 	struct irq_alloc_info *info = arg;
2271 	int i, ret;
2272 
2273 	if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2274 		return -EINVAL;
2275 
2276 	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2277 	if (ret < 0)
2278 		return ret;
2279 
2280 	for (i = virq; i < virq + nr_irqs; i++) {
2281 		struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2282 
2283 		irqd->chip = &intcapxt_controller;
2284 		irqd->hwirq = info->hwirq;
2285 		irqd->chip_data = info->data;
2286 		__irq_set_handler(i, handle_edge_irq, 0, "edge");
2287 	}
2288 
2289 	return ret;
2290 }
2291 
2292 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2293 				    unsigned int nr_irqs)
2294 {
2295 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
2296 }
2297 
2298 
2299 static void intcapxt_unmask_irq(struct irq_data *irqd)
2300 {
2301 	struct amd_iommu *iommu = irqd->chip_data;
2302 	struct irq_cfg *cfg = irqd_cfg(irqd);
2303 	union intcapxt xt;
2304 
2305 	xt.capxt = 0ULL;
2306 	xt.dest_mode_logical = apic->dest_mode_logical;
2307 	xt.vector = cfg->vector;
2308 	xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2309 	xt.destid_24_31 = cfg->dest_apicid >> 24;
2310 
2311 	writeq(xt.capxt, iommu->mmio_base + irqd->hwirq);
2312 }
2313 
2314 static void intcapxt_mask_irq(struct irq_data *irqd)
2315 {
2316 	struct amd_iommu *iommu = irqd->chip_data;
2317 
2318 	writeq(0, iommu->mmio_base + irqd->hwirq);
2319 }
2320 
2321 
2322 static int intcapxt_set_affinity(struct irq_data *irqd,
2323 				 const struct cpumask *mask, bool force)
2324 {
2325 	struct irq_data *parent = irqd->parent_data;
2326 	int ret;
2327 
2328 	ret = parent->chip->irq_set_affinity(parent, mask, force);
2329 	if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2330 		return ret;
2331 	return 0;
2332 }
2333 
2334 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2335 {
2336 	return on ? -EOPNOTSUPP : 0;
2337 }
2338 
2339 static struct irq_chip intcapxt_controller = {
2340 	.name			= "IOMMU-MSI",
2341 	.irq_unmask		= intcapxt_unmask_irq,
2342 	.irq_mask		= intcapxt_mask_irq,
2343 	.irq_ack		= irq_chip_ack_parent,
2344 	.irq_retrigger		= irq_chip_retrigger_hierarchy,
2345 	.irq_set_affinity       = intcapxt_set_affinity,
2346 	.irq_set_wake		= intcapxt_set_wake,
2347 	.flags			= IRQCHIP_MASK_ON_SUSPEND,
2348 };
2349 
2350 static const struct irq_domain_ops intcapxt_domain_ops = {
2351 	.alloc			= intcapxt_irqdomain_alloc,
2352 	.free			= intcapxt_irqdomain_free,
2353 	.activate		= intcapxt_irqdomain_activate,
2354 	.deactivate		= intcapxt_irqdomain_deactivate,
2355 };
2356 
2357 
2358 static struct irq_domain *iommu_irqdomain;
2359 
2360 static struct irq_domain *iommu_get_irqdomain(void)
2361 {
2362 	struct fwnode_handle *fn;
2363 
2364 	/* No need for locking here (yet) as the init is single-threaded */
2365 	if (iommu_irqdomain)
2366 		return iommu_irqdomain;
2367 
2368 	fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2369 	if (!fn)
2370 		return NULL;
2371 
2372 	iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2373 						      fn, &intcapxt_domain_ops,
2374 						      NULL);
2375 	if (!iommu_irqdomain)
2376 		irq_domain_free_fwnode(fn);
2377 
2378 	return iommu_irqdomain;
2379 }
2380 
2381 static int __iommu_setup_intcapxt(struct amd_iommu *iommu, const char *devname,
2382 				  int hwirq, irq_handler_t thread_fn)
2383 {
2384 	struct irq_domain *domain;
2385 	struct irq_alloc_info info;
2386 	int irq, ret;
2387 	int node = dev_to_node(&iommu->dev->dev);
2388 
2389 	domain = iommu_get_irqdomain();
2390 	if (!domain)
2391 		return -ENXIO;
2392 
2393 	init_irq_alloc_info(&info, NULL);
2394 	info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2395 	info.data = iommu;
2396 	info.hwirq = hwirq;
2397 
2398 	irq = irq_domain_alloc_irqs(domain, 1, node, &info);
2399 	if (irq < 0) {
2400 		irq_domain_remove(domain);
2401 		return irq;
2402 	}
2403 
2404 	ret = request_threaded_irq(irq, amd_iommu_int_handler,
2405 				   thread_fn, 0, devname, iommu);
2406 	if (ret) {
2407 		irq_domain_free_irqs(irq, 1);
2408 		irq_domain_remove(domain);
2409 		return ret;
2410 	}
2411 
2412 	return 0;
2413 }
2414 
2415 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2416 {
2417 	int ret;
2418 
2419 	snprintf(iommu->evt_irq_name, sizeof(iommu->evt_irq_name),
2420 		 "AMD-Vi%d-Evt", iommu->index);
2421 	ret = __iommu_setup_intcapxt(iommu, iommu->evt_irq_name,
2422 				     MMIO_INTCAPXT_EVT_OFFSET,
2423 				     amd_iommu_int_thread_evtlog);
2424 	if (ret)
2425 		return ret;
2426 
2427 	snprintf(iommu->ppr_irq_name, sizeof(iommu->ppr_irq_name),
2428 		 "AMD-Vi%d-PPR", iommu->index);
2429 	ret = __iommu_setup_intcapxt(iommu, iommu->ppr_irq_name,
2430 				     MMIO_INTCAPXT_PPR_OFFSET,
2431 				     amd_iommu_int_thread_pprlog);
2432 	if (ret)
2433 		return ret;
2434 
2435 #ifdef CONFIG_IRQ_REMAP
2436 	snprintf(iommu->ga_irq_name, sizeof(iommu->ga_irq_name),
2437 		 "AMD-Vi%d-GA", iommu->index);
2438 	ret = __iommu_setup_intcapxt(iommu, iommu->ga_irq_name,
2439 				     MMIO_INTCAPXT_GALOG_OFFSET,
2440 				     amd_iommu_int_thread_galog);
2441 #endif
2442 
2443 	return ret;
2444 }
2445 
2446 static int iommu_init_irq(struct amd_iommu *iommu)
2447 {
2448 	int ret;
2449 
2450 	if (iommu->int_enabled)
2451 		goto enable_faults;
2452 
2453 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2454 		ret = iommu_setup_intcapxt(iommu);
2455 	else if (iommu->dev->msi_cap)
2456 		ret = iommu_setup_msi(iommu);
2457 	else
2458 		ret = -ENODEV;
2459 
2460 	if (ret)
2461 		return ret;
2462 
2463 	iommu->int_enabled = true;
2464 enable_faults:
2465 
2466 	if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2467 		iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2468 
2469 	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2470 
2471 	return 0;
2472 }
2473 
2474 /****************************************************************************
2475  *
2476  * The next functions belong to the third pass of parsing the ACPI
2477  * table. In this last pass the memory mapping requirements are
2478  * gathered (like exclusion and unity mapping ranges).
2479  *
2480  ****************************************************************************/
2481 
2482 static void __init free_unity_maps(void)
2483 {
2484 	struct unity_map_entry *entry, *next;
2485 	struct amd_iommu_pci_seg *p, *pci_seg;
2486 
2487 	for_each_pci_segment_safe(pci_seg, p) {
2488 		list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2489 			list_del(&entry->list);
2490 			kfree(entry);
2491 		}
2492 	}
2493 }
2494 
2495 /* called for unity map ACPI definition */
2496 static int __init init_unity_map_range(struct ivmd_header *m,
2497 				       struct acpi_table_header *ivrs_base)
2498 {
2499 	struct unity_map_entry *e = NULL;
2500 	struct amd_iommu_pci_seg *pci_seg;
2501 	char *s;
2502 
2503 	pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2504 	if (pci_seg == NULL)
2505 		return -ENOMEM;
2506 
2507 	e = kzalloc(sizeof(*e), GFP_KERNEL);
2508 	if (e == NULL)
2509 		return -ENOMEM;
2510 
2511 	switch (m->type) {
2512 	default:
2513 		kfree(e);
2514 		return 0;
2515 	case ACPI_IVMD_TYPE:
2516 		s = "IVMD_TYPEi\t\t\t";
2517 		e->devid_start = e->devid_end = m->devid;
2518 		break;
2519 	case ACPI_IVMD_TYPE_ALL:
2520 		s = "IVMD_TYPE_ALL\t\t";
2521 		e->devid_start = 0;
2522 		e->devid_end = pci_seg->last_bdf;
2523 		break;
2524 	case ACPI_IVMD_TYPE_RANGE:
2525 		s = "IVMD_TYPE_RANGE\t\t";
2526 		e->devid_start = m->devid;
2527 		e->devid_end = m->aux;
2528 		break;
2529 	}
2530 	e->address_start = PAGE_ALIGN(m->range_start);
2531 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2532 	e->prot = m->flags >> 1;
2533 
2534 	/*
2535 	 * Treat per-device exclusion ranges as r/w unity-mapped regions
2536 	 * since some buggy BIOSes might lead to the overwritten exclusion
2537 	 * range (exclusion_start and exclusion_length members). This
2538 	 * happens when there are multiple exclusion ranges (IVMD entries)
2539 	 * defined in ACPI table.
2540 	 */
2541 	if (m->flags & IVMD_FLAG_EXCL_RANGE)
2542 		e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2543 
2544 	DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2545 		    "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2546 		    " flags: %x\n", s, m->pci_seg,
2547 		    PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2548 		    PCI_FUNC(e->devid_start), m->pci_seg,
2549 		    PCI_BUS_NUM(e->devid_end),
2550 		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2551 		    e->address_start, e->address_end, m->flags);
2552 
2553 	list_add_tail(&e->list, &pci_seg->unity_map);
2554 
2555 	return 0;
2556 }
2557 
2558 /* iterates over all memory definitions we find in the ACPI table */
2559 static int __init init_memory_definitions(struct acpi_table_header *table)
2560 {
2561 	u8 *p = (u8 *)table, *end = (u8 *)table;
2562 	struct ivmd_header *m;
2563 
2564 	end += table->length;
2565 	p += IVRS_HEADER_LENGTH;
2566 
2567 	while (p < end) {
2568 		m = (struct ivmd_header *)p;
2569 		if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2570 			init_unity_map_range(m, table);
2571 
2572 		p += m->length;
2573 	}
2574 
2575 	return 0;
2576 }
2577 
2578 /*
2579  * Init the device table to not allow DMA access for devices
2580  */
2581 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2582 {
2583 	u32 devid;
2584 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2585 
2586 	if (dev_table == NULL)
2587 		return;
2588 
2589 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2590 		__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID);
2591 		if (!amd_iommu_snp_en)
2592 			__set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION);
2593 	}
2594 }
2595 
2596 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2597 {
2598 	u32 devid;
2599 	struct dev_table_entry *dev_table = pci_seg->dev_table;
2600 
2601 	if (dev_table == NULL)
2602 		return;
2603 
2604 	for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2605 		dev_table[devid].data[0] = 0ULL;
2606 		dev_table[devid].data[1] = 0ULL;
2607 	}
2608 }
2609 
2610 static void init_device_table(void)
2611 {
2612 	struct amd_iommu_pci_seg *pci_seg;
2613 	u32 devid;
2614 
2615 	if (!amd_iommu_irq_remap)
2616 		return;
2617 
2618 	for_each_pci_segment(pci_seg) {
2619 		for (devid = 0; devid <= pci_seg->last_bdf; ++devid)
2620 			__set_dev_entry_bit(pci_seg->dev_table,
2621 					    devid, DEV_ENTRY_IRQ_TBL_EN);
2622 	}
2623 }
2624 
2625 static void iommu_init_flags(struct amd_iommu *iommu)
2626 {
2627 	iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2628 		iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2629 		iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2630 
2631 	iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2632 		iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2633 		iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2634 
2635 	iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2636 		iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2637 		iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2638 
2639 	iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2640 		iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2641 		iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2642 
2643 	/*
2644 	 * make IOMMU memory accesses cache coherent
2645 	 */
2646 	iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2647 
2648 	/* Set IOTLB invalidation timeout to 1s */
2649 	iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2650 }
2651 
2652 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2653 {
2654 	int i, j;
2655 	u32 ioc_feature_control;
2656 	struct pci_dev *pdev = iommu->root_pdev;
2657 
2658 	/* RD890 BIOSes may not have completely reconfigured the iommu */
2659 	if (!is_rd890_iommu(iommu->dev) || !pdev)
2660 		return;
2661 
2662 	/*
2663 	 * First, we need to ensure that the iommu is enabled. This is
2664 	 * controlled by a register in the northbridge
2665 	 */
2666 
2667 	/* Select Northbridge indirect register 0x75 and enable writing */
2668 	pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2669 	pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2670 
2671 	/* Enable the iommu */
2672 	if (!(ioc_feature_control & 0x1))
2673 		pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2674 
2675 	/* Restore the iommu BAR */
2676 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2677 			       iommu->stored_addr_lo);
2678 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2679 			       iommu->stored_addr_hi);
2680 
2681 	/* Restore the l1 indirect regs for each of the 6 l1s */
2682 	for (i = 0; i < 6; i++)
2683 		for (j = 0; j < 0x12; j++)
2684 			iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2685 
2686 	/* Restore the l2 indirect regs */
2687 	for (i = 0; i < 0x83; i++)
2688 		iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2689 
2690 	/* Lock PCI setup registers */
2691 	pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2692 			       iommu->stored_addr_lo | 1);
2693 }
2694 
2695 static void iommu_enable_ga(struct amd_iommu *iommu)
2696 {
2697 #ifdef CONFIG_IRQ_REMAP
2698 	switch (amd_iommu_guest_ir) {
2699 	case AMD_IOMMU_GUEST_IR_VAPIC:
2700 	case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2701 		iommu_feature_enable(iommu, CONTROL_GA_EN);
2702 		iommu->irte_ops = &irte_128_ops;
2703 		break;
2704 	default:
2705 		iommu->irte_ops = &irte_32_ops;
2706 		break;
2707 	}
2708 #endif
2709 }
2710 
2711 static void iommu_disable_irtcachedis(struct amd_iommu *iommu)
2712 {
2713 	iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
2714 }
2715 
2716 static void iommu_enable_irtcachedis(struct amd_iommu *iommu)
2717 {
2718 	u64 ctrl;
2719 
2720 	if (!amd_iommu_irtcachedis)
2721 		return;
2722 
2723 	/*
2724 	 * Note:
2725 	 * The support for IRTCacheDis feature is dertermined by
2726 	 * checking if the bit is writable.
2727 	 */
2728 	iommu_feature_enable(iommu, CONTROL_IRTCACHEDIS);
2729 	ctrl = readq(iommu->mmio_base +  MMIO_CONTROL_OFFSET);
2730 	ctrl &= (1ULL << CONTROL_IRTCACHEDIS);
2731 	if (ctrl)
2732 		iommu->irtcachedis_enabled = true;
2733 	pr_info("iommu%d (%#06x) : IRT cache is %s\n",
2734 		iommu->index, iommu->devid,
2735 		iommu->irtcachedis_enabled ? "disabled" : "enabled");
2736 }
2737 
2738 static void early_enable_iommu(struct amd_iommu *iommu)
2739 {
2740 	iommu_disable(iommu);
2741 	iommu_init_flags(iommu);
2742 	iommu_set_device_table(iommu);
2743 	iommu_enable_command_buffer(iommu);
2744 	iommu_enable_event_buffer(iommu);
2745 	iommu_set_exclusion_range(iommu);
2746 	iommu_enable_gt(iommu);
2747 	iommu_enable_ga(iommu);
2748 	iommu_enable_xt(iommu);
2749 	iommu_enable_irtcachedis(iommu);
2750 	iommu_enable(iommu);
2751 	amd_iommu_flush_all_caches(iommu);
2752 }
2753 
2754 /*
2755  * This function finally enables all IOMMUs found in the system after
2756  * they have been initialized.
2757  *
2758  * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2759  * the old content of device table entries. Not this case or copy failed,
2760  * just continue as normal kernel does.
2761  */
2762 static void early_enable_iommus(void)
2763 {
2764 	struct amd_iommu *iommu;
2765 	struct amd_iommu_pci_seg *pci_seg;
2766 
2767 	if (!copy_device_table()) {
2768 		/*
2769 		 * If come here because of failure in copying device table from old
2770 		 * kernel with all IOMMUs enabled, print error message and try to
2771 		 * free allocated old_dev_tbl_cpy.
2772 		 */
2773 		if (amd_iommu_pre_enabled)
2774 			pr_err("Failed to copy DEV table from previous kernel.\n");
2775 
2776 		for_each_pci_segment(pci_seg) {
2777 			if (pci_seg->old_dev_tbl_cpy != NULL) {
2778 				iommu_free_pages(pci_seg->old_dev_tbl_cpy,
2779 						 get_order(pci_seg->dev_table_size));
2780 				pci_seg->old_dev_tbl_cpy = NULL;
2781 			}
2782 		}
2783 
2784 		for_each_iommu(iommu) {
2785 			clear_translation_pre_enabled(iommu);
2786 			early_enable_iommu(iommu);
2787 		}
2788 	} else {
2789 		pr_info("Copied DEV table from previous kernel.\n");
2790 
2791 		for_each_pci_segment(pci_seg) {
2792 			iommu_free_pages(pci_seg->dev_table,
2793 					 get_order(pci_seg->dev_table_size));
2794 			pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2795 		}
2796 
2797 		for_each_iommu(iommu) {
2798 			iommu_disable_command_buffer(iommu);
2799 			iommu_disable_event_buffer(iommu);
2800 			iommu_disable_irtcachedis(iommu);
2801 			iommu_enable_command_buffer(iommu);
2802 			iommu_enable_event_buffer(iommu);
2803 			iommu_enable_ga(iommu);
2804 			iommu_enable_xt(iommu);
2805 			iommu_enable_irtcachedis(iommu);
2806 			iommu_set_device_table(iommu);
2807 			amd_iommu_flush_all_caches(iommu);
2808 		}
2809 	}
2810 }
2811 
2812 static void enable_iommus_ppr(void)
2813 {
2814 	struct amd_iommu *iommu;
2815 
2816 	if (!amd_iommu_gt_ppr_supported())
2817 		return;
2818 
2819 	for_each_iommu(iommu)
2820 		amd_iommu_enable_ppr_log(iommu);
2821 }
2822 
2823 static void enable_iommus_vapic(void)
2824 {
2825 #ifdef CONFIG_IRQ_REMAP
2826 	u32 status, i;
2827 	struct amd_iommu *iommu;
2828 
2829 	for_each_iommu(iommu) {
2830 		/*
2831 		 * Disable GALog if already running. It could have been enabled
2832 		 * in the previous boot before kdump.
2833 		 */
2834 		status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2835 		if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2836 			continue;
2837 
2838 		iommu_feature_disable(iommu, CONTROL_GALOG_EN);
2839 		iommu_feature_disable(iommu, CONTROL_GAINT_EN);
2840 
2841 		/*
2842 		 * Need to set and poll check the GALOGRun bit to zero before
2843 		 * we can set/ modify GA Log registers safely.
2844 		 */
2845 		for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
2846 			status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2847 			if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2848 				break;
2849 			udelay(10);
2850 		}
2851 
2852 		if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
2853 			return;
2854 	}
2855 
2856 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2857 	    !check_feature(FEATURE_GAM_VAPIC)) {
2858 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2859 		return;
2860 	}
2861 
2862 	if (amd_iommu_snp_en &&
2863 	    !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) {
2864 		pr_warn("Force to disable Virtual APIC due to SNP\n");
2865 		amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2866 		return;
2867 	}
2868 
2869 	/* Enabling GAM and SNPAVIC support */
2870 	for_each_iommu(iommu) {
2871 		if (iommu_init_ga_log(iommu) ||
2872 		    iommu_ga_log_enable(iommu))
2873 			return;
2874 
2875 		iommu_feature_enable(iommu, CONTROL_GAM_EN);
2876 		if (amd_iommu_snp_en)
2877 			iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN);
2878 	}
2879 
2880 	amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2881 	pr_info("Virtual APIC enabled\n");
2882 #endif
2883 }
2884 
2885 static void enable_iommus(void)
2886 {
2887 	early_enable_iommus();
2888 }
2889 
2890 static void disable_iommus(void)
2891 {
2892 	struct amd_iommu *iommu;
2893 
2894 	for_each_iommu(iommu)
2895 		iommu_disable(iommu);
2896 
2897 #ifdef CONFIG_IRQ_REMAP
2898 	if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2899 		amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2900 #endif
2901 }
2902 
2903 /*
2904  * Suspend/Resume support
2905  * disable suspend until real resume implemented
2906  */
2907 
2908 static void amd_iommu_resume(void)
2909 {
2910 	struct amd_iommu *iommu;
2911 
2912 	for_each_iommu(iommu)
2913 		iommu_apply_resume_quirks(iommu);
2914 
2915 	/* re-load the hardware */
2916 	enable_iommus();
2917 
2918 	amd_iommu_enable_interrupts();
2919 }
2920 
2921 static int amd_iommu_suspend(void)
2922 {
2923 	/* disable IOMMUs to go out of the way for BIOS */
2924 	disable_iommus();
2925 
2926 	return 0;
2927 }
2928 
2929 static struct syscore_ops amd_iommu_syscore_ops = {
2930 	.suspend = amd_iommu_suspend,
2931 	.resume = amd_iommu_resume,
2932 };
2933 
2934 static void __init free_iommu_resources(void)
2935 {
2936 	kmem_cache_destroy(amd_iommu_irq_cache);
2937 	amd_iommu_irq_cache = NULL;
2938 
2939 	free_iommu_all();
2940 	free_pci_segments();
2941 }
2942 
2943 /* SB IOAPIC is always on this device in AMD systems */
2944 #define IOAPIC_SB_DEVID		((0x00 << 8) | PCI_DEVFN(0x14, 0))
2945 
2946 static bool __init check_ioapic_information(void)
2947 {
2948 	const char *fw_bug = FW_BUG;
2949 	bool ret, has_sb_ioapic;
2950 	int idx;
2951 
2952 	has_sb_ioapic = false;
2953 	ret           = false;
2954 
2955 	/*
2956 	 * If we have map overrides on the kernel command line the
2957 	 * messages in this function might not describe firmware bugs
2958 	 * anymore - so be careful
2959 	 */
2960 	if (cmdline_maps)
2961 		fw_bug = "";
2962 
2963 	for (idx = 0; idx < nr_ioapics; idx++) {
2964 		int devid, id = mpc_ioapic_id(idx);
2965 
2966 		devid = get_ioapic_devid(id);
2967 		if (devid < 0) {
2968 			pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2969 				fw_bug, id);
2970 			ret = false;
2971 		} else if (devid == IOAPIC_SB_DEVID) {
2972 			has_sb_ioapic = true;
2973 			ret           = true;
2974 		}
2975 	}
2976 
2977 	if (!has_sb_ioapic) {
2978 		/*
2979 		 * We expect the SB IOAPIC to be listed in the IVRS
2980 		 * table. The system timer is connected to the SB IOAPIC
2981 		 * and if we don't have it in the list the system will
2982 		 * panic at boot time.  This situation usually happens
2983 		 * when the BIOS is buggy and provides us the wrong
2984 		 * device id for the IOAPIC in the system.
2985 		 */
2986 		pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2987 	}
2988 
2989 	if (!ret)
2990 		pr_err("Disabling interrupt remapping\n");
2991 
2992 	return ret;
2993 }
2994 
2995 static void __init free_dma_resources(void)
2996 {
2997 	iommu_free_pages(amd_iommu_pd_alloc_bitmap,
2998 			 get_order(MAX_DOMAIN_ID / 8));
2999 	amd_iommu_pd_alloc_bitmap = NULL;
3000 
3001 	free_unity_maps();
3002 }
3003 
3004 static void __init ivinfo_init(void *ivrs)
3005 {
3006 	amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
3007 }
3008 
3009 /*
3010  * This is the hardware init function for AMD IOMMU in the system.
3011  * This function is called either from amd_iommu_init or from the interrupt
3012  * remapping setup code.
3013  *
3014  * This function basically parses the ACPI table for AMD IOMMU (IVRS)
3015  * four times:
3016  *
3017  *	1 pass) Discover the most comprehensive IVHD type to use.
3018  *
3019  *	2 pass) Find the highest PCI device id the driver has to handle.
3020  *		Upon this information the size of the data structures is
3021  *		determined that needs to be allocated.
3022  *
3023  *	3 pass) Initialize the data structures just allocated with the
3024  *		information in the ACPI table about available AMD IOMMUs
3025  *		in the system. It also maps the PCI devices in the
3026  *		system to specific IOMMUs
3027  *
3028  *	4 pass) After the basic data structures are allocated and
3029  *		initialized we update them with information about memory
3030  *		remapping requirements parsed out of the ACPI table in
3031  *		this last pass.
3032  *
3033  * After everything is set up the IOMMUs are enabled and the necessary
3034  * hotplug and suspend notifiers are registered.
3035  */
3036 static int __init early_amd_iommu_init(void)
3037 {
3038 	struct acpi_table_header *ivrs_base;
3039 	int remap_cache_sz, ret;
3040 	acpi_status status;
3041 
3042 	if (!amd_iommu_detected)
3043 		return -ENODEV;
3044 
3045 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3046 	if (status == AE_NOT_FOUND)
3047 		return -ENODEV;
3048 	else if (ACPI_FAILURE(status)) {
3049 		const char *err = acpi_format_exception(status);
3050 		pr_err("IVRS table error: %s\n", err);
3051 		return -EINVAL;
3052 	}
3053 
3054 	/*
3055 	 * Validate checksum here so we don't need to do it when
3056 	 * we actually parse the table
3057 	 */
3058 	ret = check_ivrs_checksum(ivrs_base);
3059 	if (ret)
3060 		goto out;
3061 
3062 	ivinfo_init(ivrs_base);
3063 
3064 	amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
3065 	DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
3066 
3067 	/* Device table - directly used by all IOMMUs */
3068 	ret = -ENOMEM;
3069 
3070 	amd_iommu_pd_alloc_bitmap = iommu_alloc_pages(GFP_KERNEL,
3071 						      get_order(MAX_DOMAIN_ID / 8));
3072 	if (amd_iommu_pd_alloc_bitmap == NULL)
3073 		goto out;
3074 
3075 	/*
3076 	 * never allocate domain 0 because its used as the non-allocated and
3077 	 * error value placeholder
3078 	 */
3079 	__set_bit(0, amd_iommu_pd_alloc_bitmap);
3080 
3081 	/*
3082 	 * now the data structures are allocated and basically initialized
3083 	 * start the real acpi table scan
3084 	 */
3085 	ret = init_iommu_all(ivrs_base);
3086 	if (ret)
3087 		goto out;
3088 
3089 	/* 5 level guest page table */
3090 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3091 	    FIELD_GET(FEATURE_GATS, amd_iommu_efr) == GUEST_PGTABLE_5_LEVEL)
3092 		amd_iommu_gpt_level = PAGE_MODE_5_LEVEL;
3093 
3094 	/* Disable any previously enabled IOMMUs */
3095 	if (!is_kdump_kernel() || amd_iommu_disabled)
3096 		disable_iommus();
3097 
3098 	if (amd_iommu_irq_remap)
3099 		amd_iommu_irq_remap = check_ioapic_information();
3100 
3101 	if (amd_iommu_irq_remap) {
3102 		struct amd_iommu_pci_seg *pci_seg;
3103 		/*
3104 		 * Interrupt remapping enabled, create kmem_cache for the
3105 		 * remapping tables.
3106 		 */
3107 		ret = -ENOMEM;
3108 		if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
3109 			remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
3110 		else
3111 			remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
3112 		amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
3113 							remap_cache_sz,
3114 							DTE_INTTAB_ALIGNMENT,
3115 							0, NULL);
3116 		if (!amd_iommu_irq_cache)
3117 			goto out;
3118 
3119 		for_each_pci_segment(pci_seg) {
3120 			if (alloc_irq_lookup_table(pci_seg))
3121 				goto out;
3122 		}
3123 	}
3124 
3125 	ret = init_memory_definitions(ivrs_base);
3126 	if (ret)
3127 		goto out;
3128 
3129 	/* init the device table */
3130 	init_device_table();
3131 
3132 out:
3133 	/* Don't leak any ACPI memory */
3134 	acpi_put_table(ivrs_base);
3135 
3136 	return ret;
3137 }
3138 
3139 static int amd_iommu_enable_interrupts(void)
3140 {
3141 	struct amd_iommu *iommu;
3142 	int ret = 0;
3143 
3144 	for_each_iommu(iommu) {
3145 		ret = iommu_init_irq(iommu);
3146 		if (ret)
3147 			goto out;
3148 	}
3149 
3150 	/*
3151 	 * Interrupt handler is ready to process interrupts. Enable
3152 	 * PPR and GA log interrupt for all IOMMUs.
3153 	 */
3154 	enable_iommus_vapic();
3155 	enable_iommus_ppr();
3156 
3157 out:
3158 	return ret;
3159 }
3160 
3161 static bool __init detect_ivrs(void)
3162 {
3163 	struct acpi_table_header *ivrs_base;
3164 	acpi_status status;
3165 	int i;
3166 
3167 	status = acpi_get_table("IVRS", 0, &ivrs_base);
3168 	if (status == AE_NOT_FOUND)
3169 		return false;
3170 	else if (ACPI_FAILURE(status)) {
3171 		const char *err = acpi_format_exception(status);
3172 		pr_err("IVRS table error: %s\n", err);
3173 		return false;
3174 	}
3175 
3176 	acpi_put_table(ivrs_base);
3177 
3178 	if (amd_iommu_force_enable)
3179 		goto out;
3180 
3181 	/* Don't use IOMMU if there is Stoney Ridge graphics */
3182 	for (i = 0; i < 32; i++) {
3183 		u32 pci_id;
3184 
3185 		pci_id = read_pci_config(0, i, 0, 0);
3186 		if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3187 			pr_info("Disable IOMMU on Stoney Ridge\n");
3188 			return false;
3189 		}
3190 	}
3191 
3192 out:
3193 	/* Make sure ACS will be enabled during PCI probe */
3194 	pci_request_acs();
3195 
3196 	return true;
3197 }
3198 
3199 static void iommu_snp_enable(void)
3200 {
3201 #ifdef CONFIG_KVM_AMD_SEV
3202 	if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3203 		return;
3204 	/*
3205 	 * The SNP support requires that IOMMU must be enabled, and is
3206 	 * configured with V1 page table (DTE[Mode] = 0 is not supported).
3207 	 */
3208 	if (no_iommu || iommu_default_passthrough()) {
3209 		pr_warn("SNP: IOMMU disabled or configured in passthrough mode, SNP cannot be supported.\n");
3210 		goto disable_snp;
3211 	}
3212 
3213 	if (amd_iommu_pgtable != AMD_IOMMU_V1) {
3214 		pr_warn("SNP: IOMMU is configured with V2 page table mode, SNP cannot be supported.\n");
3215 		goto disable_snp;
3216 	}
3217 
3218 	amd_iommu_snp_en = check_feature(FEATURE_SNP);
3219 	if (!amd_iommu_snp_en) {
3220 		pr_warn("SNP: IOMMU SNP feature not enabled, SNP cannot be supported.\n");
3221 		goto disable_snp;
3222 	}
3223 
3224 	pr_info("IOMMU SNP support enabled.\n");
3225 	return;
3226 
3227 disable_snp:
3228 	cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3229 #endif
3230 }
3231 
3232 /****************************************************************************
3233  *
3234  * AMD IOMMU Initialization State Machine
3235  *
3236  ****************************************************************************/
3237 
3238 static int __init state_next(void)
3239 {
3240 	int ret = 0;
3241 
3242 	switch (init_state) {
3243 	case IOMMU_START_STATE:
3244 		if (!detect_ivrs()) {
3245 			init_state	= IOMMU_NOT_FOUND;
3246 			ret		= -ENODEV;
3247 		} else {
3248 			init_state	= IOMMU_IVRS_DETECTED;
3249 		}
3250 		break;
3251 	case IOMMU_IVRS_DETECTED:
3252 		if (amd_iommu_disabled) {
3253 			init_state = IOMMU_CMDLINE_DISABLED;
3254 			ret = -EINVAL;
3255 		} else {
3256 			ret = early_amd_iommu_init();
3257 			init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3258 		}
3259 		break;
3260 	case IOMMU_ACPI_FINISHED:
3261 		early_enable_iommus();
3262 		x86_platform.iommu_shutdown = disable_iommus;
3263 		init_state = IOMMU_ENABLED;
3264 		break;
3265 	case IOMMU_ENABLED:
3266 		register_syscore_ops(&amd_iommu_syscore_ops);
3267 		iommu_snp_enable();
3268 		ret = amd_iommu_init_pci();
3269 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3270 		break;
3271 	case IOMMU_PCI_INIT:
3272 		ret = amd_iommu_enable_interrupts();
3273 		init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3274 		break;
3275 	case IOMMU_INTERRUPTS_EN:
3276 		init_state = IOMMU_INITIALIZED;
3277 		break;
3278 	case IOMMU_INITIALIZED:
3279 		/* Nothing to do */
3280 		break;
3281 	case IOMMU_NOT_FOUND:
3282 	case IOMMU_INIT_ERROR:
3283 	case IOMMU_CMDLINE_DISABLED:
3284 		/* Error states => do nothing */
3285 		ret = -EINVAL;
3286 		break;
3287 	default:
3288 		/* Unknown state */
3289 		BUG();
3290 	}
3291 
3292 	if (ret) {
3293 		free_dma_resources();
3294 		if (!irq_remapping_enabled) {
3295 			disable_iommus();
3296 			free_iommu_resources();
3297 		} else {
3298 			struct amd_iommu *iommu;
3299 			struct amd_iommu_pci_seg *pci_seg;
3300 
3301 			for_each_pci_segment(pci_seg)
3302 				uninit_device_table_dma(pci_seg);
3303 
3304 			for_each_iommu(iommu)
3305 				amd_iommu_flush_all_caches(iommu);
3306 		}
3307 	}
3308 	return ret;
3309 }
3310 
3311 static int __init iommu_go_to_state(enum iommu_init_state state)
3312 {
3313 	int ret = -EINVAL;
3314 
3315 	while (init_state != state) {
3316 		if (init_state == IOMMU_NOT_FOUND         ||
3317 		    init_state == IOMMU_INIT_ERROR        ||
3318 		    init_state == IOMMU_CMDLINE_DISABLED)
3319 			break;
3320 		ret = state_next();
3321 	}
3322 
3323 	return ret;
3324 }
3325 
3326 #ifdef CONFIG_IRQ_REMAP
3327 int __init amd_iommu_prepare(void)
3328 {
3329 	int ret;
3330 
3331 	amd_iommu_irq_remap = true;
3332 
3333 	ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3334 	if (ret) {
3335 		amd_iommu_irq_remap = false;
3336 		return ret;
3337 	}
3338 
3339 	return amd_iommu_irq_remap ? 0 : -ENODEV;
3340 }
3341 
3342 int __init amd_iommu_enable(void)
3343 {
3344 	int ret;
3345 
3346 	ret = iommu_go_to_state(IOMMU_ENABLED);
3347 	if (ret)
3348 		return ret;
3349 
3350 	irq_remapping_enabled = 1;
3351 	return amd_iommu_xt_mode;
3352 }
3353 
3354 void amd_iommu_disable(void)
3355 {
3356 	amd_iommu_suspend();
3357 }
3358 
3359 int amd_iommu_reenable(int mode)
3360 {
3361 	amd_iommu_resume();
3362 
3363 	return 0;
3364 }
3365 
3366 int amd_iommu_enable_faulting(unsigned int cpu)
3367 {
3368 	/* We enable MSI later when PCI is initialized */
3369 	return 0;
3370 }
3371 #endif
3372 
3373 /*
3374  * This is the core init function for AMD IOMMU hardware in the system.
3375  * This function is called from the generic x86 DMA layer initialization
3376  * code.
3377  */
3378 static int __init amd_iommu_init(void)
3379 {
3380 	struct amd_iommu *iommu;
3381 	int ret;
3382 
3383 	ret = iommu_go_to_state(IOMMU_INITIALIZED);
3384 #ifdef CONFIG_GART_IOMMU
3385 	if (ret && list_empty(&amd_iommu_list)) {
3386 		/*
3387 		 * We failed to initialize the AMD IOMMU - try fallback
3388 		 * to GART if possible.
3389 		 */
3390 		gart_iommu_init();
3391 	}
3392 #endif
3393 
3394 	for_each_iommu(iommu)
3395 		amd_iommu_debugfs_setup(iommu);
3396 
3397 	return ret;
3398 }
3399 
3400 static bool amd_iommu_sme_check(void)
3401 {
3402 	if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3403 	    (boot_cpu_data.x86 != 0x17))
3404 		return true;
3405 
3406 	/* For Fam17h, a specific level of support is required */
3407 	if (boot_cpu_data.microcode >= 0x08001205)
3408 		return true;
3409 
3410 	if ((boot_cpu_data.microcode >= 0x08001126) &&
3411 	    (boot_cpu_data.microcode <= 0x080011ff))
3412 		return true;
3413 
3414 	pr_notice("IOMMU not currently supported when SME is active\n");
3415 
3416 	return false;
3417 }
3418 
3419 /****************************************************************************
3420  *
3421  * Early detect code. This code runs at IOMMU detection time in the DMA
3422  * layer. It just looks if there is an IVRS ACPI table to detect AMD
3423  * IOMMUs
3424  *
3425  ****************************************************************************/
3426 int __init amd_iommu_detect(void)
3427 {
3428 	int ret;
3429 
3430 	if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3431 		return -ENODEV;
3432 
3433 	if (!amd_iommu_sme_check())
3434 		return -ENODEV;
3435 
3436 	ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3437 	if (ret)
3438 		return ret;
3439 
3440 	amd_iommu_detected = true;
3441 	iommu_detected = 1;
3442 	x86_init.iommu.iommu_init = amd_iommu_init;
3443 
3444 	return 1;
3445 }
3446 
3447 /****************************************************************************
3448  *
3449  * Parsing functions for the AMD IOMMU specific kernel command line
3450  * options.
3451  *
3452  ****************************************************************************/
3453 
3454 static int __init parse_amd_iommu_dump(char *str)
3455 {
3456 	amd_iommu_dump = true;
3457 
3458 	return 1;
3459 }
3460 
3461 static int __init parse_amd_iommu_intr(char *str)
3462 {
3463 	for (; *str; ++str) {
3464 		if (strncmp(str, "legacy", 6) == 0) {
3465 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3466 			break;
3467 		}
3468 		if (strncmp(str, "vapic", 5) == 0) {
3469 			amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3470 			break;
3471 		}
3472 	}
3473 	return 1;
3474 }
3475 
3476 static int __init parse_amd_iommu_options(char *str)
3477 {
3478 	if (!str)
3479 		return -EINVAL;
3480 
3481 	while (*str) {
3482 		if (strncmp(str, "fullflush", 9) == 0) {
3483 			pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3484 			iommu_set_dma_strict();
3485 		} else if (strncmp(str, "force_enable", 12) == 0) {
3486 			amd_iommu_force_enable = true;
3487 		} else if (strncmp(str, "off", 3) == 0) {
3488 			amd_iommu_disabled = true;
3489 		} else if (strncmp(str, "force_isolation", 15) == 0) {
3490 			amd_iommu_force_isolation = true;
3491 		} else if (strncmp(str, "pgtbl_v1", 8) == 0) {
3492 			amd_iommu_pgtable = AMD_IOMMU_V1;
3493 		} else if (strncmp(str, "pgtbl_v2", 8) == 0) {
3494 			amd_iommu_pgtable = AMD_IOMMU_V2;
3495 		} else if (strncmp(str, "irtcachedis", 11) == 0) {
3496 			amd_iommu_irtcachedis = true;
3497 		} else if (strncmp(str, "nohugepages", 11) == 0) {
3498 			pr_info("Restricting V1 page-sizes to 4KiB");
3499 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_4K;
3500 		} else if (strncmp(str, "v2_pgsizes_only", 15) == 0) {
3501 			pr_info("Restricting V1 page-sizes to 4KiB/2MiB/1GiB");
3502 			amd_iommu_pgsize_bitmap = AMD_IOMMU_PGSIZES_V2;
3503 		} else {
3504 			pr_notice("Unknown option - '%s'\n", str);
3505 		}
3506 
3507 		str += strcspn(str, ",");
3508 		while (*str == ',')
3509 			str++;
3510 	}
3511 
3512 	return 1;
3513 }
3514 
3515 static int __init parse_ivrs_ioapic(char *str)
3516 {
3517 	u32 seg = 0, bus, dev, fn;
3518 	int id, i;
3519 	u32 devid;
3520 
3521 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3522 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3523 		goto found;
3524 
3525 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3526 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3527 		pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n",
3528 			str, id, seg, bus, dev, fn);
3529 		goto found;
3530 	}
3531 
3532 	pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3533 	return 1;
3534 
3535 found:
3536 	if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3537 		pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3538 			str);
3539 		return 1;
3540 	}
3541 
3542 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3543 
3544 	cmdline_maps			= true;
3545 	i				= early_ioapic_map_size++;
3546 	early_ioapic_map[i].id		= id;
3547 	early_ioapic_map[i].devid	= devid;
3548 	early_ioapic_map[i].cmd_line	= true;
3549 
3550 	return 1;
3551 }
3552 
3553 static int __init parse_ivrs_hpet(char *str)
3554 {
3555 	u32 seg = 0, bus, dev, fn;
3556 	int id, i;
3557 	u32 devid;
3558 
3559 	if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3560 	    sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3561 		goto found;
3562 
3563 	if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3564 	    sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3565 		pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n",
3566 			str, id, seg, bus, dev, fn);
3567 		goto found;
3568 	}
3569 
3570 	pr_err("Invalid command line: ivrs_hpet%s\n", str);
3571 	return 1;
3572 
3573 found:
3574 	if (early_hpet_map_size == EARLY_MAP_SIZE) {
3575 		pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3576 			str);
3577 		return 1;
3578 	}
3579 
3580 	devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3581 
3582 	cmdline_maps			= true;
3583 	i				= early_hpet_map_size++;
3584 	early_hpet_map[i].id		= id;
3585 	early_hpet_map[i].devid		= devid;
3586 	early_hpet_map[i].cmd_line	= true;
3587 
3588 	return 1;
3589 }
3590 
3591 #define ACPIID_LEN (ACPIHID_UID_LEN + ACPIHID_HID_LEN)
3592 
3593 static int __init parse_ivrs_acpihid(char *str)
3594 {
3595 	u32 seg = 0, bus, dev, fn;
3596 	char *hid, *uid, *p, *addr;
3597 	char acpiid[ACPIID_LEN] = {0};
3598 	int i;
3599 
3600 	addr = strchr(str, '@');
3601 	if (!addr) {
3602 		addr = strchr(str, '=');
3603 		if (!addr)
3604 			goto not_found;
3605 
3606 		++addr;
3607 
3608 		if (strlen(addr) > ACPIID_LEN)
3609 			goto not_found;
3610 
3611 		if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 ||
3612 		    sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) {
3613 			pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n",
3614 				str, acpiid, seg, bus, dev, fn);
3615 			goto found;
3616 		}
3617 		goto not_found;
3618 	}
3619 
3620 	/* We have the '@', make it the terminator to get just the acpiid */
3621 	*addr++ = 0;
3622 
3623 	if (strlen(str) > ACPIID_LEN + 1)
3624 		goto not_found;
3625 
3626 	if (sscanf(str, "=%s", acpiid) != 1)
3627 		goto not_found;
3628 
3629 	if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 ||
3630 	    sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4)
3631 		goto found;
3632 
3633 not_found:
3634 	pr_err("Invalid command line: ivrs_acpihid%s\n", str);
3635 	return 1;
3636 
3637 found:
3638 	p = acpiid;
3639 	hid = strsep(&p, ":");
3640 	uid = p;
3641 
3642 	if (!hid || !(*hid) || !uid) {
3643 		pr_err("Invalid command line: hid or uid\n");
3644 		return 1;
3645 	}
3646 
3647 	/*
3648 	 * Ignore leading zeroes after ':', so e.g., AMDI0095:00
3649 	 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match
3650 	 */
3651 	while (*uid == '0' && *(uid + 1))
3652 		uid++;
3653 
3654 	i = early_acpihid_map_size++;
3655 	memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3656 	memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3657 	early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3658 	early_acpihid_map[i].cmd_line	= true;
3659 
3660 	return 1;
3661 }
3662 
3663 __setup("amd_iommu_dump",	parse_amd_iommu_dump);
3664 __setup("amd_iommu=",		parse_amd_iommu_options);
3665 __setup("amd_iommu_intr=",	parse_amd_iommu_intr);
3666 __setup("ivrs_ioapic",		parse_ivrs_ioapic);
3667 __setup("ivrs_hpet",		parse_ivrs_hpet);
3668 __setup("ivrs_acpihid",		parse_ivrs_acpihid);
3669 
3670 bool amd_iommu_pasid_supported(void)
3671 {
3672 	/* CPU page table size should match IOMMU guest page table size */
3673 	if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3674 	    amd_iommu_gpt_level != PAGE_MODE_5_LEVEL)
3675 		return false;
3676 
3677 	/*
3678 	 * Since DTE[Mode]=0 is prohibited on SNP-enabled system
3679 	 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without
3680 	 * setting up IOMMUv1 page table.
3681 	 */
3682 	return amd_iommu_gt_ppr_supported() && !amd_iommu_snp_en;
3683 }
3684 
3685 struct amd_iommu *get_amd_iommu(unsigned int idx)
3686 {
3687 	unsigned int i = 0;
3688 	struct amd_iommu *iommu;
3689 
3690 	for_each_iommu(iommu)
3691 		if (i++ == idx)
3692 			return iommu;
3693 	return NULL;
3694 }
3695 
3696 /****************************************************************************
3697  *
3698  * IOMMU EFR Performance Counter support functionality. This code allows
3699  * access to the IOMMU PC functionality.
3700  *
3701  ****************************************************************************/
3702 
3703 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3704 {
3705 	struct amd_iommu *iommu = get_amd_iommu(idx);
3706 
3707 	if (iommu)
3708 		return iommu->max_banks;
3709 
3710 	return 0;
3711 }
3712 
3713 bool amd_iommu_pc_supported(void)
3714 {
3715 	return amd_iommu_pc_present;
3716 }
3717 
3718 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3719 {
3720 	struct amd_iommu *iommu = get_amd_iommu(idx);
3721 
3722 	if (iommu)
3723 		return iommu->max_counters;
3724 
3725 	return 0;
3726 }
3727 
3728 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3729 				u8 fxn, u64 *value, bool is_write)
3730 {
3731 	u32 offset;
3732 	u32 max_offset_lim;
3733 
3734 	/* Make sure the IOMMU PC resource is available */
3735 	if (!amd_iommu_pc_present)
3736 		return -ENODEV;
3737 
3738 	/* Check for valid iommu and pc register indexing */
3739 	if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3740 		return -ENODEV;
3741 
3742 	offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3743 
3744 	/* Limit the offset to the hw defined mmio region aperture */
3745 	max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3746 				(iommu->max_counters << 8) | 0x28);
3747 	if ((offset < MMIO_CNTR_REG_OFFSET) ||
3748 	    (offset > max_offset_lim))
3749 		return -EINVAL;
3750 
3751 	if (is_write) {
3752 		u64 val = *value & GENMASK_ULL(47, 0);
3753 
3754 		writel((u32)val, iommu->mmio_base + offset);
3755 		writel((val >> 32), iommu->mmio_base + offset + 4);
3756 	} else {
3757 		*value = readl(iommu->mmio_base + offset + 4);
3758 		*value <<= 32;
3759 		*value |= readl(iommu->mmio_base + offset);
3760 		*value &= GENMASK_ULL(47, 0);
3761 	}
3762 
3763 	return 0;
3764 }
3765 
3766 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3767 {
3768 	if (!iommu)
3769 		return -EINVAL;
3770 
3771 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3772 }
3773 
3774 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3775 {
3776 	if (!iommu)
3777 		return -EINVAL;
3778 
3779 	return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3780 }
3781 
3782 #ifdef CONFIG_KVM_AMD_SEV
3783 static int iommu_page_make_shared(void *page)
3784 {
3785 	unsigned long paddr, pfn;
3786 
3787 	paddr = iommu_virt_to_phys(page);
3788 	/* Cbit maybe set in the paddr */
3789 	pfn = __sme_clr(paddr) >> PAGE_SHIFT;
3790 
3791 	if (!(pfn % PTRS_PER_PMD)) {
3792 		int ret, level;
3793 		bool assigned;
3794 
3795 		ret = snp_lookup_rmpentry(pfn, &assigned, &level);
3796 		if (ret) {
3797 			pr_warn("IOMMU PFN %lx RMP lookup failed, ret %d\n", pfn, ret);
3798 			return ret;
3799 		}
3800 
3801 		if (!assigned) {
3802 			pr_warn("IOMMU PFN %lx not assigned in RMP table\n", pfn);
3803 			return -EINVAL;
3804 		}
3805 
3806 		if (level > PG_LEVEL_4K) {
3807 			ret = psmash(pfn);
3808 			if (!ret)
3809 				goto done;
3810 
3811 			pr_warn("PSMASH failed for IOMMU PFN %lx huge RMP entry, ret: %d, level: %d\n",
3812 				pfn, ret, level);
3813 			return ret;
3814 		}
3815 	}
3816 
3817 done:
3818 	return rmp_make_shared(pfn, PG_LEVEL_4K);
3819 }
3820 
3821 static int iommu_make_shared(void *va, size_t size)
3822 {
3823 	void *page;
3824 	int ret;
3825 
3826 	if (!va)
3827 		return 0;
3828 
3829 	for (page = va; page < (va + size); page += PAGE_SIZE) {
3830 		ret = iommu_page_make_shared(page);
3831 		if (ret)
3832 			return ret;
3833 	}
3834 
3835 	return 0;
3836 }
3837 
3838 int amd_iommu_snp_disable(void)
3839 {
3840 	struct amd_iommu *iommu;
3841 	int ret;
3842 
3843 	if (!amd_iommu_snp_en)
3844 		return 0;
3845 
3846 	for_each_iommu(iommu) {
3847 		ret = iommu_make_shared(iommu->evt_buf, EVT_BUFFER_SIZE);
3848 		if (ret)
3849 			return ret;
3850 
3851 		ret = iommu_make_shared(iommu->ppr_log, PPR_LOG_SIZE);
3852 		if (ret)
3853 			return ret;
3854 
3855 		ret = iommu_make_shared((void *)iommu->cmd_sem, PAGE_SIZE);
3856 		if (ret)
3857 			return ret;
3858 	}
3859 
3860 	return 0;
3861 }
3862 EXPORT_SYMBOL_GPL(amd_iommu_snp_disable);
3863 #endif
3864