1 /*
2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
4 * Leo Duran <leo.duran@amd.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20 #include <linux/pci.h>
21 #include <linux/acpi.h>
22 #include <linux/list.h>
23 #include <linux/bitmap.h>
24 #include <linux/slab.h>
25 #include <linux/syscore_ops.h>
26 #include <linux/interrupt.h>
27 #include <linux/msi.h>
28 #include <linux/amd-iommu.h>
29 #include <linux/export.h>
30 #include <linux/iommu.h>
31 #include <linux/kmemleak.h>
32 #include <linux/mem_encrypt.h>
33 #include <asm/pci-direct.h>
34 #include <asm/iommu.h>
35 #include <asm/gart.h>
36 #include <asm/x86_init.h>
37 #include <asm/iommu_table.h>
38 #include <asm/io_apic.h>
39 #include <asm/irq_remapping.h>
40
41 #include <linux/crash_dump.h>
42 #include "amd_iommu.h"
43 #include "amd_iommu_proto.h"
44 #include "amd_iommu_types.h"
45 #include "irq_remapping.h"
46
47 /*
48 * definitions for the ACPI scanning code
49 */
50 #define IVRS_HEADER_LENGTH 48
51
52 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
53 #define ACPI_IVMD_TYPE_ALL 0x20
54 #define ACPI_IVMD_TYPE 0x21
55 #define ACPI_IVMD_TYPE_RANGE 0x22
56
57 #define IVHD_DEV_ALL 0x01
58 #define IVHD_DEV_SELECT 0x02
59 #define IVHD_DEV_SELECT_RANGE_START 0x03
60 #define IVHD_DEV_RANGE_END 0x04
61 #define IVHD_DEV_ALIAS 0x42
62 #define IVHD_DEV_ALIAS_RANGE 0x43
63 #define IVHD_DEV_EXT_SELECT 0x46
64 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
65 #define IVHD_DEV_SPECIAL 0x48
66 #define IVHD_DEV_ACPI_HID 0xf0
67
68 #define UID_NOT_PRESENT 0
69 #define UID_IS_INTEGER 1
70 #define UID_IS_CHARACTER 2
71
72 #define IVHD_SPECIAL_IOAPIC 1
73 #define IVHD_SPECIAL_HPET 2
74
75 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
76 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
77 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
78 #define IVHD_FLAG_ISOC_EN_MASK 0x08
79
80 #define IVMD_FLAG_EXCL_RANGE 0x08
81 #define IVMD_FLAG_UNITY_MAP 0x01
82
83 #define ACPI_DEVFLAG_INITPASS 0x01
84 #define ACPI_DEVFLAG_EXTINT 0x02
85 #define ACPI_DEVFLAG_NMI 0x04
86 #define ACPI_DEVFLAG_SYSMGT1 0x10
87 #define ACPI_DEVFLAG_SYSMGT2 0x20
88 #define ACPI_DEVFLAG_LINT0 0x40
89 #define ACPI_DEVFLAG_LINT1 0x80
90 #define ACPI_DEVFLAG_ATSDIS 0x10000000
91
92 #define LOOP_TIMEOUT 100000
93 /*
94 * ACPI table definitions
95 *
96 * These data structures are laid over the table to parse the important values
97 * out of it.
98 */
99
100 extern const struct iommu_ops amd_iommu_ops;
101
102 /*
103 * structure describing one IOMMU in the ACPI table. Typically followed by one
104 * or more ivhd_entrys.
105 */
106 struct ivhd_header {
107 u8 type;
108 u8 flags;
109 u16 length;
110 u16 devid;
111 u16 cap_ptr;
112 u64 mmio_phys;
113 u16 pci_seg;
114 u16 info;
115 u32 efr_attr;
116
117 /* Following only valid on IVHD type 11h and 40h */
118 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
119 u64 res;
120 } __attribute__((packed));
121
122 /*
123 * A device entry describing which devices a specific IOMMU translates and
124 * which requestor ids they use.
125 */
126 struct ivhd_entry {
127 u8 type;
128 u16 devid;
129 u8 flags;
130 u32 ext;
131 u32 hidh;
132 u64 cid;
133 u8 uidf;
134 u8 uidl;
135 u8 uid;
136 } __attribute__((packed));
137
138 /*
139 * An AMD IOMMU memory definition structure. It defines things like exclusion
140 * ranges for devices and regions that should be unity mapped.
141 */
142 struct ivmd_header {
143 u8 type;
144 u8 flags;
145 u16 length;
146 u16 devid;
147 u16 aux;
148 u64 resv;
149 u64 range_start;
150 u64 range_length;
151 } __attribute__((packed));
152
153 bool amd_iommu_dump;
154 bool amd_iommu_irq_remap __read_mostly;
155
156 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
157 static int amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
158
159 static bool amd_iommu_detected;
160 static bool __initdata amd_iommu_disabled;
161 static int amd_iommu_target_ivhd_type;
162
163 u16 amd_iommu_last_bdf; /* largest PCI device id we have
164 to handle */
165 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
166 we find in ACPI */
167 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
168
169 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
170 system */
171
172 /* Array to assign indices to IOMMUs*/
173 struct amd_iommu *amd_iommus[MAX_IOMMUS];
174
175 /* Number of IOMMUs present in the system */
176 static int amd_iommus_present;
177
178 /* IOMMUs have a non-present cache? */
179 bool amd_iommu_np_cache __read_mostly;
180 bool amd_iommu_iotlb_sup __read_mostly = true;
181
182 u32 amd_iommu_max_pasid __read_mostly = ~0;
183
184 bool amd_iommu_v2_present __read_mostly;
185 static bool amd_iommu_pc_present __read_mostly;
186
187 bool amd_iommu_force_isolation __read_mostly;
188
189 /*
190 * List of protection domains - used during resume
191 */
192 LIST_HEAD(amd_iommu_pd_list);
193 spinlock_t amd_iommu_pd_lock;
194
195 /*
196 * Pointer to the device table which is shared by all AMD IOMMUs
197 * it is indexed by the PCI device id or the HT unit id and contains
198 * information about the domain the device belongs to as well as the
199 * page table root pointer.
200 */
201 struct dev_table_entry *amd_iommu_dev_table;
202 /*
203 * Pointer to a device table which the content of old device table
204 * will be copied to. It's only be used in kdump kernel.
205 */
206 static struct dev_table_entry *old_dev_tbl_cpy;
207
208 /*
209 * The alias table is a driver specific data structure which contains the
210 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
211 * More than one device can share the same requestor id.
212 */
213 u16 *amd_iommu_alias_table;
214
215 /*
216 * The rlookup table is used to find the IOMMU which is responsible
217 * for a specific device. It is also indexed by the PCI device id.
218 */
219 struct amd_iommu **amd_iommu_rlookup_table;
220 EXPORT_SYMBOL(amd_iommu_rlookup_table);
221
222 /*
223 * This table is used to find the irq remapping table for a given device id
224 * quickly.
225 */
226 struct irq_remap_table **irq_lookup_table;
227
228 /*
229 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
230 * to know which ones are already in use.
231 */
232 unsigned long *amd_iommu_pd_alloc_bitmap;
233
234 static u32 dev_table_size; /* size of the device table */
235 static u32 alias_table_size; /* size of the alias table */
236 static u32 rlookup_table_size; /* size if the rlookup table */
237
238 enum iommu_init_state {
239 IOMMU_START_STATE,
240 IOMMU_IVRS_DETECTED,
241 IOMMU_ACPI_FINISHED,
242 IOMMU_ENABLED,
243 IOMMU_PCI_INIT,
244 IOMMU_INTERRUPTS_EN,
245 IOMMU_DMA_OPS,
246 IOMMU_INITIALIZED,
247 IOMMU_NOT_FOUND,
248 IOMMU_INIT_ERROR,
249 IOMMU_CMDLINE_DISABLED,
250 };
251
252 /* Early ioapic and hpet maps from kernel command line */
253 #define EARLY_MAP_SIZE 4
254 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
255 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
256 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
257
258 static int __initdata early_ioapic_map_size;
259 static int __initdata early_hpet_map_size;
260 static int __initdata early_acpihid_map_size;
261
262 static bool __initdata cmdline_maps;
263
264 static enum iommu_init_state init_state = IOMMU_START_STATE;
265
266 static int amd_iommu_enable_interrupts(void);
267 static int __init iommu_go_to_state(enum iommu_init_state state);
268 static void init_device_table_dma(void);
269
270 static bool amd_iommu_pre_enabled = true;
271
translation_pre_enabled(struct amd_iommu * iommu)272 bool translation_pre_enabled(struct amd_iommu *iommu)
273 {
274 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
275 }
276 EXPORT_SYMBOL(translation_pre_enabled);
277
clear_translation_pre_enabled(struct amd_iommu * iommu)278 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
279 {
280 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
281 }
282
init_translation_status(struct amd_iommu * iommu)283 static void init_translation_status(struct amd_iommu *iommu)
284 {
285 u64 ctrl;
286
287 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
288 if (ctrl & (1<<CONTROL_IOMMU_EN))
289 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
290 }
291
update_last_devid(u16 devid)292 static inline void update_last_devid(u16 devid)
293 {
294 if (devid > amd_iommu_last_bdf)
295 amd_iommu_last_bdf = devid;
296 }
297
tbl_size(int entry_size)298 static inline unsigned long tbl_size(int entry_size)
299 {
300 unsigned shift = PAGE_SHIFT +
301 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
302
303 return 1UL << shift;
304 }
305
amd_iommu_get_num_iommus(void)306 int amd_iommu_get_num_iommus(void)
307 {
308 return amd_iommus_present;
309 }
310
311 /* Access to l1 and l2 indexed register spaces */
312
iommu_read_l1(struct amd_iommu * iommu,u16 l1,u8 address)313 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
314 {
315 u32 val;
316
317 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
318 pci_read_config_dword(iommu->dev, 0xfc, &val);
319 return val;
320 }
321
iommu_write_l1(struct amd_iommu * iommu,u16 l1,u8 address,u32 val)322 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
323 {
324 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
325 pci_write_config_dword(iommu->dev, 0xfc, val);
326 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
327 }
328
iommu_read_l2(struct amd_iommu * iommu,u8 address)329 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
330 {
331 u32 val;
332
333 pci_write_config_dword(iommu->dev, 0xf0, address);
334 pci_read_config_dword(iommu->dev, 0xf4, &val);
335 return val;
336 }
337
iommu_write_l2(struct amd_iommu * iommu,u8 address,u32 val)338 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
339 {
340 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
341 pci_write_config_dword(iommu->dev, 0xf4, val);
342 }
343
344 /****************************************************************************
345 *
346 * AMD IOMMU MMIO register space handling functions
347 *
348 * These functions are used to program the IOMMU device registers in
349 * MMIO space required for that driver.
350 *
351 ****************************************************************************/
352
353 /*
354 * This function set the exclusion range in the IOMMU. DMA accesses to the
355 * exclusion range are passed through untranslated
356 */
iommu_set_exclusion_range(struct amd_iommu * iommu)357 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
358 {
359 u64 start = iommu->exclusion_start & PAGE_MASK;
360 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
361 u64 entry;
362
363 if (!iommu->exclusion_start)
364 return;
365
366 entry = start | MMIO_EXCL_ENABLE_MASK;
367 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
368 &entry, sizeof(entry));
369
370 entry = limit;
371 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
372 &entry, sizeof(entry));
373 }
374
375 /* Programs the physical address of the device table into the IOMMU hardware */
iommu_set_device_table(struct amd_iommu * iommu)376 static void iommu_set_device_table(struct amd_iommu *iommu)
377 {
378 u64 entry;
379
380 BUG_ON(iommu->mmio_base == NULL);
381
382 entry = iommu_virt_to_phys(amd_iommu_dev_table);
383 entry |= (dev_table_size >> 12) - 1;
384 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
385 &entry, sizeof(entry));
386 }
387
388 /* Generic functions to enable/disable certain features of the IOMMU. */
iommu_feature_enable(struct amd_iommu * iommu,u8 bit)389 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
390 {
391 u64 ctrl;
392
393 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
394 ctrl |= (1ULL << bit);
395 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
396 }
397
iommu_feature_disable(struct amd_iommu * iommu,u8 bit)398 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
399 {
400 u64 ctrl;
401
402 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
403 ctrl &= ~(1ULL << bit);
404 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
405 }
406
iommu_set_inv_tlb_timeout(struct amd_iommu * iommu,int timeout)407 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
408 {
409 u64 ctrl;
410
411 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
412 ctrl &= ~CTRL_INV_TO_MASK;
413 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
414 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
415 }
416
417 /* Function to enable the hardware */
iommu_enable(struct amd_iommu * iommu)418 static void iommu_enable(struct amd_iommu *iommu)
419 {
420 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
421 }
422
iommu_disable(struct amd_iommu * iommu)423 static void iommu_disable(struct amd_iommu *iommu)
424 {
425 if (!iommu->mmio_base)
426 return;
427
428 /* Disable command buffer */
429 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
430
431 /* Disable event logging and event interrupts */
432 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
433 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
434
435 /* Disable IOMMU GA_LOG */
436 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
437 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
438
439 /* Disable IOMMU hardware itself */
440 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
441 }
442
443 /*
444 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
445 * the system has one.
446 */
iommu_map_mmio_space(u64 address,u64 end)447 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
448 {
449 if (!request_mem_region(address, end, "amd_iommu")) {
450 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
451 address, end);
452 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
453 return NULL;
454 }
455
456 return (u8 __iomem *)ioremap_nocache(address, end);
457 }
458
iommu_unmap_mmio_space(struct amd_iommu * iommu)459 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
460 {
461 if (iommu->mmio_base)
462 iounmap(iommu->mmio_base);
463 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
464 }
465
get_ivhd_header_size(struct ivhd_header * h)466 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
467 {
468 u32 size = 0;
469
470 switch (h->type) {
471 case 0x10:
472 size = 24;
473 break;
474 case 0x11:
475 case 0x40:
476 size = 40;
477 break;
478 }
479 return size;
480 }
481
482 /****************************************************************************
483 *
484 * The functions below belong to the first pass of AMD IOMMU ACPI table
485 * parsing. In this pass we try to find out the highest device id this
486 * code has to handle. Upon this information the size of the shared data
487 * structures is determined later.
488 *
489 ****************************************************************************/
490
491 /*
492 * This function calculates the length of a given IVHD entry
493 */
ivhd_entry_length(u8 * ivhd)494 static inline int ivhd_entry_length(u8 *ivhd)
495 {
496 u32 type = ((struct ivhd_entry *)ivhd)->type;
497
498 if (type < 0x80) {
499 return 0x04 << (*ivhd >> 6);
500 } else if (type == IVHD_DEV_ACPI_HID) {
501 /* For ACPI_HID, offset 21 is uid len */
502 return *((u8 *)ivhd + 21) + 22;
503 }
504 return 0;
505 }
506
507 /*
508 * After reading the highest device id from the IOMMU PCI capability header
509 * this function looks if there is a higher device id defined in the ACPI table
510 */
find_last_devid_from_ivhd(struct ivhd_header * h)511 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
512 {
513 u8 *p = (void *)h, *end = (void *)h;
514 struct ivhd_entry *dev;
515
516 u32 ivhd_size = get_ivhd_header_size(h);
517
518 if (!ivhd_size) {
519 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
520 return -EINVAL;
521 }
522
523 p += ivhd_size;
524 end += h->length;
525
526 while (p < end) {
527 dev = (struct ivhd_entry *)p;
528 switch (dev->type) {
529 case IVHD_DEV_ALL:
530 /* Use maximum BDF value for DEV_ALL */
531 update_last_devid(0xffff);
532 break;
533 case IVHD_DEV_SELECT:
534 case IVHD_DEV_RANGE_END:
535 case IVHD_DEV_ALIAS:
536 case IVHD_DEV_EXT_SELECT:
537 /* all the above subfield types refer to device ids */
538 update_last_devid(dev->devid);
539 break;
540 default:
541 break;
542 }
543 p += ivhd_entry_length(p);
544 }
545
546 WARN_ON(p != end);
547
548 return 0;
549 }
550
check_ivrs_checksum(struct acpi_table_header * table)551 static int __init check_ivrs_checksum(struct acpi_table_header *table)
552 {
553 int i;
554 u8 checksum = 0, *p = (u8 *)table;
555
556 for (i = 0; i < table->length; ++i)
557 checksum += p[i];
558 if (checksum != 0) {
559 /* ACPI table corrupt */
560 pr_err(FW_BUG "AMD-Vi: IVRS invalid checksum\n");
561 return -ENODEV;
562 }
563
564 return 0;
565 }
566
567 /*
568 * Iterate over all IVHD entries in the ACPI table and find the highest device
569 * id which we need to handle. This is the first of three functions which parse
570 * the ACPI table. So we check the checksum here.
571 */
find_last_devid_acpi(struct acpi_table_header * table)572 static int __init find_last_devid_acpi(struct acpi_table_header *table)
573 {
574 u8 *p = (u8 *)table, *end = (u8 *)table;
575 struct ivhd_header *h;
576
577 p += IVRS_HEADER_LENGTH;
578
579 end += table->length;
580 while (p < end) {
581 h = (struct ivhd_header *)p;
582 if (h->type == amd_iommu_target_ivhd_type) {
583 int ret = find_last_devid_from_ivhd(h);
584
585 if (ret)
586 return ret;
587 }
588 p += h->length;
589 }
590 WARN_ON(p != end);
591
592 return 0;
593 }
594
595 /****************************************************************************
596 *
597 * The following functions belong to the code path which parses the ACPI table
598 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
599 * data structures, initialize the device/alias/rlookup table and also
600 * basically initialize the hardware.
601 *
602 ****************************************************************************/
603
604 /*
605 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
606 * write commands to that buffer later and the IOMMU will execute them
607 * asynchronously
608 */
alloc_command_buffer(struct amd_iommu * iommu)609 static int __init alloc_command_buffer(struct amd_iommu *iommu)
610 {
611 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
612 get_order(CMD_BUFFER_SIZE));
613
614 return iommu->cmd_buf ? 0 : -ENOMEM;
615 }
616
617 /*
618 * This function resets the command buffer if the IOMMU stopped fetching
619 * commands from it.
620 */
amd_iommu_reset_cmd_buffer(struct amd_iommu * iommu)621 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
622 {
623 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
624
625 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
626 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
627 iommu->cmd_buf_head = 0;
628 iommu->cmd_buf_tail = 0;
629
630 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
631 }
632
633 /*
634 * This function writes the command buffer address to the hardware and
635 * enables it.
636 */
iommu_enable_command_buffer(struct amd_iommu * iommu)637 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
638 {
639 u64 entry;
640
641 BUG_ON(iommu->cmd_buf == NULL);
642
643 entry = iommu_virt_to_phys(iommu->cmd_buf);
644 entry |= MMIO_CMD_SIZE_512;
645
646 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
647 &entry, sizeof(entry));
648
649 amd_iommu_reset_cmd_buffer(iommu);
650 }
651
652 /*
653 * This function disables the command buffer
654 */
iommu_disable_command_buffer(struct amd_iommu * iommu)655 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
656 {
657 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
658 }
659
free_command_buffer(struct amd_iommu * iommu)660 static void __init free_command_buffer(struct amd_iommu *iommu)
661 {
662 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
663 }
664
665 /* allocates the memory where the IOMMU will log its events to */
alloc_event_buffer(struct amd_iommu * iommu)666 static int __init alloc_event_buffer(struct amd_iommu *iommu)
667 {
668 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
669 get_order(EVT_BUFFER_SIZE));
670
671 return iommu->evt_buf ? 0 : -ENOMEM;
672 }
673
iommu_enable_event_buffer(struct amd_iommu * iommu)674 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
675 {
676 u64 entry;
677
678 BUG_ON(iommu->evt_buf == NULL);
679
680 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
681
682 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
683 &entry, sizeof(entry));
684
685 /* set head and tail to zero manually */
686 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
687 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
688
689 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
690 }
691
692 /*
693 * This function disables the event log buffer
694 */
iommu_disable_event_buffer(struct amd_iommu * iommu)695 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
696 {
697 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
698 }
699
free_event_buffer(struct amd_iommu * iommu)700 static void __init free_event_buffer(struct amd_iommu *iommu)
701 {
702 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
703 }
704
705 /* allocates the memory where the IOMMU will log its events to */
alloc_ppr_log(struct amd_iommu * iommu)706 static int __init alloc_ppr_log(struct amd_iommu *iommu)
707 {
708 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
709 get_order(PPR_LOG_SIZE));
710
711 return iommu->ppr_log ? 0 : -ENOMEM;
712 }
713
iommu_enable_ppr_log(struct amd_iommu * iommu)714 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
715 {
716 u64 entry;
717
718 if (iommu->ppr_log == NULL)
719 return;
720
721 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
722
723 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
724 &entry, sizeof(entry));
725
726 /* set head and tail to zero manually */
727 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
728 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
729
730 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
731 iommu_feature_enable(iommu, CONTROL_PPR_EN);
732 }
733
free_ppr_log(struct amd_iommu * iommu)734 static void __init free_ppr_log(struct amd_iommu *iommu)
735 {
736 if (iommu->ppr_log == NULL)
737 return;
738
739 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
740 }
741
free_ga_log(struct amd_iommu * iommu)742 static void free_ga_log(struct amd_iommu *iommu)
743 {
744 #ifdef CONFIG_IRQ_REMAP
745 if (iommu->ga_log)
746 free_pages((unsigned long)iommu->ga_log,
747 get_order(GA_LOG_SIZE));
748 if (iommu->ga_log_tail)
749 free_pages((unsigned long)iommu->ga_log_tail,
750 get_order(8));
751 #endif
752 }
753
iommu_ga_log_enable(struct amd_iommu * iommu)754 static int iommu_ga_log_enable(struct amd_iommu *iommu)
755 {
756 #ifdef CONFIG_IRQ_REMAP
757 u32 status, i;
758
759 if (!iommu->ga_log)
760 return -EINVAL;
761
762 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
763
764 /* Check if already running */
765 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
766 return 0;
767
768 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
769 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
770
771 for (i = 0; i < LOOP_TIMEOUT; ++i) {
772 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
773 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
774 break;
775 }
776
777 if (i >= LOOP_TIMEOUT)
778 return -EINVAL;
779 #endif /* CONFIG_IRQ_REMAP */
780 return 0;
781 }
782
783 #ifdef CONFIG_IRQ_REMAP
iommu_init_ga_log(struct amd_iommu * iommu)784 static int iommu_init_ga_log(struct amd_iommu *iommu)
785 {
786 u64 entry;
787
788 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
789 return 0;
790
791 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
792 get_order(GA_LOG_SIZE));
793 if (!iommu->ga_log)
794 goto err_out;
795
796 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
797 get_order(8));
798 if (!iommu->ga_log_tail)
799 goto err_out;
800
801 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
802 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
803 &entry, sizeof(entry));
804 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
805 (BIT_ULL(52)-1)) & ~7ULL;
806 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
807 &entry, sizeof(entry));
808 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
809 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
810
811 return 0;
812 err_out:
813 free_ga_log(iommu);
814 return -EINVAL;
815 }
816 #endif /* CONFIG_IRQ_REMAP */
817
iommu_init_ga(struct amd_iommu * iommu)818 static int iommu_init_ga(struct amd_iommu *iommu)
819 {
820 int ret = 0;
821
822 #ifdef CONFIG_IRQ_REMAP
823 /* Note: We have already checked GASup from IVRS table.
824 * Now, we need to make sure that GAMSup is set.
825 */
826 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
827 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
828 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
829
830 ret = iommu_init_ga_log(iommu);
831 #endif /* CONFIG_IRQ_REMAP */
832
833 return ret;
834 }
835
iommu_enable_xt(struct amd_iommu * iommu)836 static void iommu_enable_xt(struct amd_iommu *iommu)
837 {
838 #ifdef CONFIG_IRQ_REMAP
839 /*
840 * XT mode (32-bit APIC destination ID) requires
841 * GA mode (128-bit IRTE support) as a prerequisite.
842 */
843 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
844 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
845 iommu_feature_enable(iommu, CONTROL_XT_EN);
846 #endif /* CONFIG_IRQ_REMAP */
847 }
848
iommu_enable_gt(struct amd_iommu * iommu)849 static void iommu_enable_gt(struct amd_iommu *iommu)
850 {
851 if (!iommu_feature(iommu, FEATURE_GT))
852 return;
853
854 iommu_feature_enable(iommu, CONTROL_GT_EN);
855 }
856
857 /* sets a specific bit in the device table entry. */
set_dev_entry_bit(u16 devid,u8 bit)858 static void set_dev_entry_bit(u16 devid, u8 bit)
859 {
860 int i = (bit >> 6) & 0x03;
861 int _bit = bit & 0x3f;
862
863 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
864 }
865
get_dev_entry_bit(u16 devid,u8 bit)866 static int get_dev_entry_bit(u16 devid, u8 bit)
867 {
868 int i = (bit >> 6) & 0x03;
869 int _bit = bit & 0x3f;
870
871 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
872 }
873
874
copy_device_table(void)875 static bool copy_device_table(void)
876 {
877 u64 int_ctl, int_tab_len, entry = 0, last_entry = 0;
878 struct dev_table_entry *old_devtb = NULL;
879 u32 lo, hi, devid, old_devtb_size;
880 phys_addr_t old_devtb_phys;
881 struct amd_iommu *iommu;
882 u16 dom_id, dte_v, irq_v;
883 gfp_t gfp_flag;
884 u64 tmp;
885
886 if (!amd_iommu_pre_enabled)
887 return false;
888
889 pr_warn("Translation is already enabled - trying to copy translation structures\n");
890 for_each_iommu(iommu) {
891 /* All IOMMUs should use the same device table with the same size */
892 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
893 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
894 entry = (((u64) hi) << 32) + lo;
895 if (last_entry && last_entry != entry) {
896 pr_err("IOMMU:%d should use the same dev table as others!\n",
897 iommu->index);
898 return false;
899 }
900 last_entry = entry;
901
902 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
903 if (old_devtb_size != dev_table_size) {
904 pr_err("The device table size of IOMMU:%d is not expected!\n",
905 iommu->index);
906 return false;
907 }
908 }
909
910 old_devtb_phys = entry & PAGE_MASK;
911 if (old_devtb_phys >= 0x100000000ULL) {
912 pr_err("The address of old device table is above 4G, not trustworthy!\n");
913 return false;
914 }
915 old_devtb = memremap(old_devtb_phys, dev_table_size, MEMREMAP_WB);
916 if (!old_devtb)
917 return false;
918
919 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32;
920 old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag,
921 get_order(dev_table_size));
922 if (old_dev_tbl_cpy == NULL) {
923 pr_err("Failed to allocate memory for copying old device table!\n");
924 return false;
925 }
926
927 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
928 old_dev_tbl_cpy[devid] = old_devtb[devid];
929 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
930 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
931
932 if (dte_v && dom_id) {
933 old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
934 old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
935 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
936 /* If gcr3 table existed, mask it out */
937 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
938 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
939 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
940 old_dev_tbl_cpy[devid].data[1] &= ~tmp;
941 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
942 tmp |= DTE_FLAG_GV;
943 old_dev_tbl_cpy[devid].data[0] &= ~tmp;
944 }
945 }
946
947 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
948 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
949 int_tab_len = old_devtb[devid].data[2] & DTE_IRQ_TABLE_LEN_MASK;
950 if (irq_v && (int_ctl || int_tab_len)) {
951 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
952 (int_tab_len != DTE_IRQ_TABLE_LEN)) {
953 pr_err("Wrong old irq remapping flag: %#x\n", devid);
954 return false;
955 }
956
957 old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
958 }
959 }
960 memunmap(old_devtb);
961
962 return true;
963 }
964
amd_iommu_apply_erratum_63(u16 devid)965 void amd_iommu_apply_erratum_63(u16 devid)
966 {
967 int sysmgt;
968
969 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
970 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
971
972 if (sysmgt == 0x01)
973 set_dev_entry_bit(devid, DEV_ENTRY_IW);
974 }
975
976 /* Writes the specific IOMMU for a device into the rlookup table */
set_iommu_for_device(struct amd_iommu * iommu,u16 devid)977 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
978 {
979 amd_iommu_rlookup_table[devid] = iommu;
980 }
981
982 /*
983 * This function takes the device specific flags read from the ACPI
984 * table and sets up the device table entry with that information
985 */
set_dev_entry_from_acpi(struct amd_iommu * iommu,u16 devid,u32 flags,u32 ext_flags)986 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
987 u16 devid, u32 flags, u32 ext_flags)
988 {
989 if (flags & ACPI_DEVFLAG_INITPASS)
990 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
991 if (flags & ACPI_DEVFLAG_EXTINT)
992 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
993 if (flags & ACPI_DEVFLAG_NMI)
994 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
995 if (flags & ACPI_DEVFLAG_SYSMGT1)
996 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
997 if (flags & ACPI_DEVFLAG_SYSMGT2)
998 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
999 if (flags & ACPI_DEVFLAG_LINT0)
1000 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
1001 if (flags & ACPI_DEVFLAG_LINT1)
1002 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
1003
1004 amd_iommu_apply_erratum_63(devid);
1005
1006 set_iommu_for_device(iommu, devid);
1007 }
1008
add_special_device(u8 type,u8 id,u16 * devid,bool cmd_line)1009 int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
1010 {
1011 struct devid_map *entry;
1012 struct list_head *list;
1013
1014 if (type == IVHD_SPECIAL_IOAPIC)
1015 list = &ioapic_map;
1016 else if (type == IVHD_SPECIAL_HPET)
1017 list = &hpet_map;
1018 else
1019 return -EINVAL;
1020
1021 list_for_each_entry(entry, list, list) {
1022 if (!(entry->id == id && entry->cmd_line))
1023 continue;
1024
1025 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
1026 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1027
1028 *devid = entry->devid;
1029
1030 return 0;
1031 }
1032
1033 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1034 if (!entry)
1035 return -ENOMEM;
1036
1037 entry->id = id;
1038 entry->devid = *devid;
1039 entry->cmd_line = cmd_line;
1040
1041 list_add_tail(&entry->list, list);
1042
1043 return 0;
1044 }
1045
add_acpi_hid_device(u8 * hid,u8 * uid,u16 * devid,bool cmd_line)1046 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
1047 bool cmd_line)
1048 {
1049 struct acpihid_map_entry *entry;
1050 struct list_head *list = &acpihid_map;
1051
1052 list_for_each_entry(entry, list, list) {
1053 if (strcmp(entry->hid, hid) ||
1054 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1055 !entry->cmd_line)
1056 continue;
1057
1058 pr_info("AMD-Vi: Command-line override for hid:%s uid:%s\n",
1059 hid, uid);
1060 *devid = entry->devid;
1061 return 0;
1062 }
1063
1064 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1065 if (!entry)
1066 return -ENOMEM;
1067
1068 memcpy(entry->uid, uid, strlen(uid));
1069 memcpy(entry->hid, hid, strlen(hid));
1070 entry->devid = *devid;
1071 entry->cmd_line = cmd_line;
1072 entry->root_devid = (entry->devid & (~0x7));
1073
1074 pr_info("AMD-Vi:%s, add hid:%s, uid:%s, rdevid:%d\n",
1075 entry->cmd_line ? "cmd" : "ivrs",
1076 entry->hid, entry->uid, entry->root_devid);
1077
1078 list_add_tail(&entry->list, list);
1079 return 0;
1080 }
1081
add_early_maps(void)1082 static int __init add_early_maps(void)
1083 {
1084 int i, ret;
1085
1086 for (i = 0; i < early_ioapic_map_size; ++i) {
1087 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1088 early_ioapic_map[i].id,
1089 &early_ioapic_map[i].devid,
1090 early_ioapic_map[i].cmd_line);
1091 if (ret)
1092 return ret;
1093 }
1094
1095 for (i = 0; i < early_hpet_map_size; ++i) {
1096 ret = add_special_device(IVHD_SPECIAL_HPET,
1097 early_hpet_map[i].id,
1098 &early_hpet_map[i].devid,
1099 early_hpet_map[i].cmd_line);
1100 if (ret)
1101 return ret;
1102 }
1103
1104 for (i = 0; i < early_acpihid_map_size; ++i) {
1105 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1106 early_acpihid_map[i].uid,
1107 &early_acpihid_map[i].devid,
1108 early_acpihid_map[i].cmd_line);
1109 if (ret)
1110 return ret;
1111 }
1112
1113 return 0;
1114 }
1115
1116 /*
1117 * Reads the device exclusion range from ACPI and initializes the IOMMU with
1118 * it
1119 */
set_device_exclusion_range(u16 devid,struct ivmd_header * m)1120 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
1121 {
1122 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
1123
1124 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
1125 return;
1126
1127 if (iommu) {
1128 /*
1129 * We only can configure exclusion ranges per IOMMU, not
1130 * per device. But we can enable the exclusion range per
1131 * device. This is done here
1132 */
1133 set_dev_entry_bit(devid, DEV_ENTRY_EX);
1134 iommu->exclusion_start = m->range_start;
1135 iommu->exclusion_length = m->range_length;
1136 }
1137 }
1138
1139 /*
1140 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1141 * initializes the hardware and our data structures with it.
1142 */
init_iommu_from_acpi(struct amd_iommu * iommu,struct ivhd_header * h)1143 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1144 struct ivhd_header *h)
1145 {
1146 u8 *p = (u8 *)h;
1147 u8 *end = p, flags = 0;
1148 u16 devid = 0, devid_start = 0, devid_to = 0;
1149 u32 dev_i, ext_flags = 0;
1150 bool alias = false;
1151 struct ivhd_entry *e;
1152 u32 ivhd_size;
1153 int ret;
1154
1155
1156 ret = add_early_maps();
1157 if (ret)
1158 return ret;
1159
1160 amd_iommu_apply_ivrs_quirks();
1161
1162 /*
1163 * First save the recommended feature enable bits from ACPI
1164 */
1165 iommu->acpi_flags = h->flags;
1166
1167 /*
1168 * Done. Now parse the device entries
1169 */
1170 ivhd_size = get_ivhd_header_size(h);
1171 if (!ivhd_size) {
1172 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
1173 return -EINVAL;
1174 }
1175
1176 p += ivhd_size;
1177
1178 end += h->length;
1179
1180
1181 while (p < end) {
1182 e = (struct ivhd_entry *)p;
1183 switch (e->type) {
1184 case IVHD_DEV_ALL:
1185
1186 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1187
1188 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1189 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1190 break;
1191 case IVHD_DEV_SELECT:
1192
1193 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1194 "flags: %02x\n",
1195 PCI_BUS_NUM(e->devid),
1196 PCI_SLOT(e->devid),
1197 PCI_FUNC(e->devid),
1198 e->flags);
1199
1200 devid = e->devid;
1201 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1202 break;
1203 case IVHD_DEV_SELECT_RANGE_START:
1204
1205 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1206 "devid: %02x:%02x.%x flags: %02x\n",
1207 PCI_BUS_NUM(e->devid),
1208 PCI_SLOT(e->devid),
1209 PCI_FUNC(e->devid),
1210 e->flags);
1211
1212 devid_start = e->devid;
1213 flags = e->flags;
1214 ext_flags = 0;
1215 alias = false;
1216 break;
1217 case IVHD_DEV_ALIAS:
1218
1219 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1220 "flags: %02x devid_to: %02x:%02x.%x\n",
1221 PCI_BUS_NUM(e->devid),
1222 PCI_SLOT(e->devid),
1223 PCI_FUNC(e->devid),
1224 e->flags,
1225 PCI_BUS_NUM(e->ext >> 8),
1226 PCI_SLOT(e->ext >> 8),
1227 PCI_FUNC(e->ext >> 8));
1228
1229 devid = e->devid;
1230 devid_to = e->ext >> 8;
1231 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1232 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1233 amd_iommu_alias_table[devid] = devid_to;
1234 break;
1235 case IVHD_DEV_ALIAS_RANGE:
1236
1237 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1238 "devid: %02x:%02x.%x flags: %02x "
1239 "devid_to: %02x:%02x.%x\n",
1240 PCI_BUS_NUM(e->devid),
1241 PCI_SLOT(e->devid),
1242 PCI_FUNC(e->devid),
1243 e->flags,
1244 PCI_BUS_NUM(e->ext >> 8),
1245 PCI_SLOT(e->ext >> 8),
1246 PCI_FUNC(e->ext >> 8));
1247
1248 devid_start = e->devid;
1249 flags = e->flags;
1250 devid_to = e->ext >> 8;
1251 ext_flags = 0;
1252 alias = true;
1253 break;
1254 case IVHD_DEV_EXT_SELECT:
1255
1256 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1257 "flags: %02x ext: %08x\n",
1258 PCI_BUS_NUM(e->devid),
1259 PCI_SLOT(e->devid),
1260 PCI_FUNC(e->devid),
1261 e->flags, e->ext);
1262
1263 devid = e->devid;
1264 set_dev_entry_from_acpi(iommu, devid, e->flags,
1265 e->ext);
1266 break;
1267 case IVHD_DEV_EXT_SELECT_RANGE:
1268
1269 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1270 "%02x:%02x.%x flags: %02x ext: %08x\n",
1271 PCI_BUS_NUM(e->devid),
1272 PCI_SLOT(e->devid),
1273 PCI_FUNC(e->devid),
1274 e->flags, e->ext);
1275
1276 devid_start = e->devid;
1277 flags = e->flags;
1278 ext_flags = e->ext;
1279 alias = false;
1280 break;
1281 case IVHD_DEV_RANGE_END:
1282
1283 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1284 PCI_BUS_NUM(e->devid),
1285 PCI_SLOT(e->devid),
1286 PCI_FUNC(e->devid));
1287
1288 devid = e->devid;
1289 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1290 if (alias) {
1291 amd_iommu_alias_table[dev_i] = devid_to;
1292 set_dev_entry_from_acpi(iommu,
1293 devid_to, flags, ext_flags);
1294 }
1295 set_dev_entry_from_acpi(iommu, dev_i,
1296 flags, ext_flags);
1297 }
1298 break;
1299 case IVHD_DEV_SPECIAL: {
1300 u8 handle, type;
1301 const char *var;
1302 u16 devid;
1303 int ret;
1304
1305 handle = e->ext & 0xff;
1306 devid = (e->ext >> 8) & 0xffff;
1307 type = (e->ext >> 24) & 0xff;
1308
1309 if (type == IVHD_SPECIAL_IOAPIC)
1310 var = "IOAPIC";
1311 else if (type == IVHD_SPECIAL_HPET)
1312 var = "HPET";
1313 else
1314 var = "UNKNOWN";
1315
1316 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1317 var, (int)handle,
1318 PCI_BUS_NUM(devid),
1319 PCI_SLOT(devid),
1320 PCI_FUNC(devid));
1321
1322 ret = add_special_device(type, handle, &devid, false);
1323 if (ret)
1324 return ret;
1325
1326 /*
1327 * add_special_device might update the devid in case a
1328 * command-line override is present. So call
1329 * set_dev_entry_from_acpi after add_special_device.
1330 */
1331 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1332
1333 break;
1334 }
1335 case IVHD_DEV_ACPI_HID: {
1336 u16 devid;
1337 u8 hid[ACPIHID_HID_LEN];
1338 u8 uid[ACPIHID_UID_LEN];
1339 int ret;
1340
1341 if (h->type != 0x40) {
1342 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1343 e->type);
1344 break;
1345 }
1346
1347 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1348 hid[ACPIHID_HID_LEN - 1] = '\0';
1349
1350 if (!(*hid)) {
1351 pr_err(FW_BUG "Invalid HID.\n");
1352 break;
1353 }
1354
1355 uid[0] = '\0';
1356 switch (e->uidf) {
1357 case UID_NOT_PRESENT:
1358
1359 if (e->uidl != 0)
1360 pr_warn(FW_BUG "Invalid UID length.\n");
1361
1362 break;
1363 case UID_IS_INTEGER:
1364
1365 sprintf(uid, "%d", e->uid);
1366
1367 break;
1368 case UID_IS_CHARACTER:
1369
1370 memcpy(uid, &e->uid, e->uidl);
1371 uid[e->uidl] = '\0';
1372
1373 break;
1374 default:
1375 break;
1376 }
1377
1378 devid = e->devid;
1379 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1380 hid, uid,
1381 PCI_BUS_NUM(devid),
1382 PCI_SLOT(devid),
1383 PCI_FUNC(devid));
1384
1385 flags = e->flags;
1386
1387 ret = add_acpi_hid_device(hid, uid, &devid, false);
1388 if (ret)
1389 return ret;
1390
1391 /*
1392 * add_special_device might update the devid in case a
1393 * command-line override is present. So call
1394 * set_dev_entry_from_acpi after add_special_device.
1395 */
1396 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1397
1398 break;
1399 }
1400 default:
1401 break;
1402 }
1403
1404 p += ivhd_entry_length(p);
1405 }
1406
1407 return 0;
1408 }
1409
free_iommu_one(struct amd_iommu * iommu)1410 static void __init free_iommu_one(struct amd_iommu *iommu)
1411 {
1412 free_command_buffer(iommu);
1413 free_event_buffer(iommu);
1414 free_ppr_log(iommu);
1415 free_ga_log(iommu);
1416 iommu_unmap_mmio_space(iommu);
1417 }
1418
free_iommu_all(void)1419 static void __init free_iommu_all(void)
1420 {
1421 struct amd_iommu *iommu, *next;
1422
1423 for_each_iommu_safe(iommu, next) {
1424 list_del(&iommu->list);
1425 free_iommu_one(iommu);
1426 kfree(iommu);
1427 }
1428 }
1429
1430 /*
1431 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1432 * Workaround:
1433 * BIOS should disable L2B micellaneous clock gating by setting
1434 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1435 */
amd_iommu_erratum_746_workaround(struct amd_iommu * iommu)1436 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1437 {
1438 u32 value;
1439
1440 if ((boot_cpu_data.x86 != 0x15) ||
1441 (boot_cpu_data.x86_model < 0x10) ||
1442 (boot_cpu_data.x86_model > 0x1f))
1443 return;
1444
1445 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1446 pci_read_config_dword(iommu->dev, 0xf4, &value);
1447
1448 if (value & BIT(2))
1449 return;
1450
1451 /* Select NB indirect register 0x90 and enable writing */
1452 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1453
1454 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1455 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1456 dev_name(&iommu->dev->dev));
1457
1458 /* Clear the enable writing bit */
1459 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1460 }
1461
1462 /*
1463 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1464 * Workaround:
1465 * BIOS should enable ATS write permission check by setting
1466 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1467 */
amd_iommu_ats_write_check_workaround(struct amd_iommu * iommu)1468 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1469 {
1470 u32 value;
1471
1472 if ((boot_cpu_data.x86 != 0x15) ||
1473 (boot_cpu_data.x86_model < 0x30) ||
1474 (boot_cpu_data.x86_model > 0x3f))
1475 return;
1476
1477 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1478 value = iommu_read_l2(iommu, 0x47);
1479
1480 if (value & BIT(0))
1481 return;
1482
1483 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1484 iommu_write_l2(iommu, 0x47, value | BIT(0));
1485
1486 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1487 dev_name(&iommu->dev->dev));
1488 }
1489
1490 /*
1491 * This function clues the initialization function for one IOMMU
1492 * together and also allocates the command buffer and programs the
1493 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1494 */
init_iommu_one(struct amd_iommu * iommu,struct ivhd_header * h)1495 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1496 {
1497 int ret;
1498
1499 raw_spin_lock_init(&iommu->lock);
1500
1501 /* Add IOMMU to internal data structures */
1502 list_add_tail(&iommu->list, &amd_iommu_list);
1503 iommu->index = amd_iommus_present++;
1504
1505 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1506 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1507 return -ENOSYS;
1508 }
1509
1510 /* Index is fine - add IOMMU to the array */
1511 amd_iommus[iommu->index] = iommu;
1512
1513 /*
1514 * Copy data from ACPI table entry to the iommu struct
1515 */
1516 iommu->devid = h->devid;
1517 iommu->cap_ptr = h->cap_ptr;
1518 iommu->pci_seg = h->pci_seg;
1519 iommu->mmio_phys = h->mmio_phys;
1520
1521 switch (h->type) {
1522 case 0x10:
1523 /* Check if IVHD EFR contains proper max banks/counters */
1524 if ((h->efr_attr != 0) &&
1525 ((h->efr_attr & (0xF << 13)) != 0) &&
1526 ((h->efr_attr & (0x3F << 17)) != 0))
1527 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1528 else
1529 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1530 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1531 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1532 if (((h->efr_attr & (0x1 << IOMMU_FEAT_XTSUP_SHIFT)) == 0))
1533 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1534 break;
1535 case 0x11:
1536 case 0x40:
1537 if (h->efr_reg & (1 << 9))
1538 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1539 else
1540 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1541 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1542 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1543 if (((h->efr_reg & (0x1 << IOMMU_EFR_XTSUP_SHIFT)) == 0))
1544 amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
1545 break;
1546 default:
1547 return -EINVAL;
1548 }
1549
1550 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1551 iommu->mmio_phys_end);
1552 if (!iommu->mmio_base)
1553 return -ENOMEM;
1554
1555 if (alloc_command_buffer(iommu))
1556 return -ENOMEM;
1557
1558 if (alloc_event_buffer(iommu))
1559 return -ENOMEM;
1560
1561 iommu->int_enabled = false;
1562
1563 init_translation_status(iommu);
1564 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1565 iommu_disable(iommu);
1566 clear_translation_pre_enabled(iommu);
1567 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1568 iommu->index);
1569 }
1570 if (amd_iommu_pre_enabled)
1571 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1572
1573 ret = init_iommu_from_acpi(iommu, h);
1574 if (ret)
1575 return ret;
1576
1577 ret = amd_iommu_create_irq_domain(iommu);
1578 if (ret)
1579 return ret;
1580
1581 /*
1582 * Make sure IOMMU is not considered to translate itself. The IVRS
1583 * table tells us so, but this is a lie!
1584 */
1585 amd_iommu_rlookup_table[iommu->devid] = NULL;
1586
1587 return 0;
1588 }
1589
1590 /**
1591 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1592 * @ivrs Pointer to the IVRS header
1593 *
1594 * This function search through all IVDB of the maximum supported IVHD
1595 */
get_highest_supported_ivhd_type(struct acpi_table_header * ivrs)1596 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1597 {
1598 u8 *base = (u8 *)ivrs;
1599 struct ivhd_header *ivhd = (struct ivhd_header *)
1600 (base + IVRS_HEADER_LENGTH);
1601 u8 last_type = ivhd->type;
1602 u16 devid = ivhd->devid;
1603
1604 while (((u8 *)ivhd - base < ivrs->length) &&
1605 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1606 u8 *p = (u8 *) ivhd;
1607
1608 if (ivhd->devid == devid)
1609 last_type = ivhd->type;
1610 ivhd = (struct ivhd_header *)(p + ivhd->length);
1611 }
1612
1613 return last_type;
1614 }
1615
1616 /*
1617 * Iterates over all IOMMU entries in the ACPI table, allocates the
1618 * IOMMU structure and initializes it with init_iommu_one()
1619 */
init_iommu_all(struct acpi_table_header * table)1620 static int __init init_iommu_all(struct acpi_table_header *table)
1621 {
1622 u8 *p = (u8 *)table, *end = (u8 *)table;
1623 struct ivhd_header *h;
1624 struct amd_iommu *iommu;
1625 int ret;
1626
1627 end += table->length;
1628 p += IVRS_HEADER_LENGTH;
1629
1630 while (p < end) {
1631 h = (struct ivhd_header *)p;
1632 if (*p == amd_iommu_target_ivhd_type) {
1633
1634 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1635 "seg: %d flags: %01x info %04x\n",
1636 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1637 PCI_FUNC(h->devid), h->cap_ptr,
1638 h->pci_seg, h->flags, h->info);
1639 DUMP_printk(" mmio-addr: %016llx\n",
1640 h->mmio_phys);
1641
1642 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1643 if (iommu == NULL)
1644 return -ENOMEM;
1645
1646 ret = init_iommu_one(iommu, h);
1647 if (ret)
1648 return ret;
1649 }
1650 p += h->length;
1651
1652 }
1653 WARN_ON(p != end);
1654
1655 return 0;
1656 }
1657
1658 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
1659 u8 fxn, u64 *value, bool is_write);
1660
init_iommu_perf_ctr(struct amd_iommu * iommu)1661 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1662 {
1663 u64 val = 0xabcd, val2 = 0;
1664
1665 if (!iommu_feature(iommu, FEATURE_PC))
1666 return;
1667
1668 amd_iommu_pc_present = true;
1669
1670 /* Check if the performance counters can be written to */
1671 if ((iommu_pc_get_set_reg(iommu, 0, 0, 0, &val, true)) ||
1672 (iommu_pc_get_set_reg(iommu, 0, 0, 0, &val2, false)) ||
1673 (val != val2)) {
1674 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1675 amd_iommu_pc_present = false;
1676 return;
1677 }
1678
1679 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1680
1681 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1682 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1683 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1684 }
1685
amd_iommu_show_cap(struct device * dev,struct device_attribute * attr,char * buf)1686 static ssize_t amd_iommu_show_cap(struct device *dev,
1687 struct device_attribute *attr,
1688 char *buf)
1689 {
1690 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1691 return sprintf(buf, "%x\n", iommu->cap);
1692 }
1693 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1694
amd_iommu_show_features(struct device * dev,struct device_attribute * attr,char * buf)1695 static ssize_t amd_iommu_show_features(struct device *dev,
1696 struct device_attribute *attr,
1697 char *buf)
1698 {
1699 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1700 return sprintf(buf, "%llx\n", iommu->features);
1701 }
1702 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1703
1704 static struct attribute *amd_iommu_attrs[] = {
1705 &dev_attr_cap.attr,
1706 &dev_attr_features.attr,
1707 NULL,
1708 };
1709
1710 static struct attribute_group amd_iommu_group = {
1711 .name = "amd-iommu",
1712 .attrs = amd_iommu_attrs,
1713 };
1714
1715 static const struct attribute_group *amd_iommu_groups[] = {
1716 &amd_iommu_group,
1717 NULL,
1718 };
1719
iommu_init_pci(struct amd_iommu * iommu)1720 static int __init iommu_init_pci(struct amd_iommu *iommu)
1721 {
1722 int cap_ptr = iommu->cap_ptr;
1723 u32 range, misc, low, high;
1724 int ret;
1725
1726 iommu->dev = pci_get_domain_bus_and_slot(0, PCI_BUS_NUM(iommu->devid),
1727 iommu->devid & 0xff);
1728 if (!iommu->dev)
1729 return -ENODEV;
1730
1731 /* Prevent binding other PCI device drivers to IOMMU devices */
1732 iommu->dev->match_driver = false;
1733
1734 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1735 &iommu->cap);
1736 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1737 &range);
1738 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1739 &misc);
1740
1741 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1742 amd_iommu_iotlb_sup = false;
1743
1744 /* read extended feature bits */
1745 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1746 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1747
1748 iommu->features = ((u64)high << 32) | low;
1749
1750 if (iommu_feature(iommu, FEATURE_GT)) {
1751 int glxval;
1752 u32 max_pasid;
1753 u64 pasmax;
1754
1755 pasmax = iommu->features & FEATURE_PASID_MASK;
1756 pasmax >>= FEATURE_PASID_SHIFT;
1757 max_pasid = (1 << (pasmax + 1)) - 1;
1758
1759 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1760
1761 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1762
1763 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1764 glxval >>= FEATURE_GLXVAL_SHIFT;
1765
1766 if (amd_iommu_max_glx_val == -1)
1767 amd_iommu_max_glx_val = glxval;
1768 else
1769 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1770 }
1771
1772 if (iommu_feature(iommu, FEATURE_GT) &&
1773 iommu_feature(iommu, FEATURE_PPR)) {
1774 iommu->is_iommu_v2 = true;
1775 amd_iommu_v2_present = true;
1776 }
1777
1778 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1779 return -ENOMEM;
1780
1781 ret = iommu_init_ga(iommu);
1782 if (ret)
1783 return ret;
1784
1785 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1786 amd_iommu_np_cache = true;
1787
1788 init_iommu_perf_ctr(iommu);
1789
1790 if (is_rd890_iommu(iommu->dev)) {
1791 int i, j;
1792
1793 iommu->root_pdev =
1794 pci_get_domain_bus_and_slot(0, iommu->dev->bus->number,
1795 PCI_DEVFN(0, 0));
1796
1797 /*
1798 * Some rd890 systems may not be fully reconfigured by the
1799 * BIOS, so it's necessary for us to store this information so
1800 * it can be reprogrammed on resume
1801 */
1802 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1803 &iommu->stored_addr_lo);
1804 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1805 &iommu->stored_addr_hi);
1806
1807 /* Low bit locks writes to configuration space */
1808 iommu->stored_addr_lo &= ~1;
1809
1810 for (i = 0; i < 6; i++)
1811 for (j = 0; j < 0x12; j++)
1812 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1813
1814 for (i = 0; i < 0x83; i++)
1815 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1816 }
1817
1818 amd_iommu_erratum_746_workaround(iommu);
1819 amd_iommu_ats_write_check_workaround(iommu);
1820
1821 iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
1822 amd_iommu_groups, "ivhd%d", iommu->index);
1823 iommu_device_set_ops(&iommu->iommu, &amd_iommu_ops);
1824 iommu_device_register(&iommu->iommu);
1825
1826 return pci_enable_device(iommu->dev);
1827 }
1828
print_iommu_info(void)1829 static void print_iommu_info(void)
1830 {
1831 static const char * const feat_str[] = {
1832 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1833 "IA", "GA", "HE", "PC"
1834 };
1835 struct amd_iommu *iommu;
1836
1837 for_each_iommu(iommu) {
1838 int i;
1839
1840 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1841 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1842
1843 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1844 pr_info("AMD-Vi: Extended features (%#llx):\n",
1845 iommu->features);
1846 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1847 if (iommu_feature(iommu, (1ULL << i)))
1848 pr_cont(" %s", feat_str[i]);
1849 }
1850
1851 if (iommu->features & FEATURE_GAM_VAPIC)
1852 pr_cont(" GA_vAPIC");
1853
1854 pr_cont("\n");
1855 }
1856 }
1857 if (irq_remapping_enabled) {
1858 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1859 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1860 pr_info("AMD-Vi: virtual APIC enabled\n");
1861 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
1862 pr_info("AMD-Vi: X2APIC enabled\n");
1863 }
1864 }
1865
amd_iommu_init_pci(void)1866 static int __init amd_iommu_init_pci(void)
1867 {
1868 struct amd_iommu *iommu;
1869 int ret = 0;
1870
1871 for_each_iommu(iommu) {
1872 ret = iommu_init_pci(iommu);
1873 if (ret)
1874 break;
1875 }
1876
1877 /*
1878 * Order is important here to make sure any unity map requirements are
1879 * fulfilled. The unity mappings are created and written to the device
1880 * table during the amd_iommu_init_api() call.
1881 *
1882 * After that we call init_device_table_dma() to make sure any
1883 * uninitialized DTE will block DMA, and in the end we flush the caches
1884 * of all IOMMUs to make sure the changes to the device table are
1885 * active.
1886 */
1887 ret = amd_iommu_init_api();
1888
1889 init_device_table_dma();
1890
1891 for_each_iommu(iommu)
1892 iommu_flush_all_caches(iommu);
1893
1894 if (!ret)
1895 print_iommu_info();
1896
1897 return ret;
1898 }
1899
1900 /****************************************************************************
1901 *
1902 * The following functions initialize the MSI interrupts for all IOMMUs
1903 * in the system. It's a bit challenging because there could be multiple
1904 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1905 * pci_dev.
1906 *
1907 ****************************************************************************/
1908
iommu_setup_msi(struct amd_iommu * iommu)1909 static int iommu_setup_msi(struct amd_iommu *iommu)
1910 {
1911 int r;
1912
1913 r = pci_enable_msi(iommu->dev);
1914 if (r)
1915 return r;
1916
1917 r = request_threaded_irq(iommu->dev->irq,
1918 amd_iommu_int_handler,
1919 amd_iommu_int_thread,
1920 0, "AMD-Vi",
1921 iommu);
1922
1923 if (r) {
1924 pci_disable_msi(iommu->dev);
1925 return r;
1926 }
1927
1928 iommu->int_enabled = true;
1929
1930 return 0;
1931 }
1932
iommu_init_msi(struct amd_iommu * iommu)1933 static int iommu_init_msi(struct amd_iommu *iommu)
1934 {
1935 int ret;
1936
1937 if (iommu->int_enabled)
1938 goto enable_faults;
1939
1940 if (iommu->dev->msi_cap)
1941 ret = iommu_setup_msi(iommu);
1942 else
1943 ret = -ENODEV;
1944
1945 if (ret)
1946 return ret;
1947
1948 enable_faults:
1949 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1950
1951 if (iommu->ppr_log != NULL)
1952 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1953
1954 iommu_ga_log_enable(iommu);
1955
1956 return 0;
1957 }
1958
1959 /****************************************************************************
1960 *
1961 * The next functions belong to the third pass of parsing the ACPI
1962 * table. In this last pass the memory mapping requirements are
1963 * gathered (like exclusion and unity mapping ranges).
1964 *
1965 ****************************************************************************/
1966
free_unity_maps(void)1967 static void __init free_unity_maps(void)
1968 {
1969 struct unity_map_entry *entry, *next;
1970
1971 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1972 list_del(&entry->list);
1973 kfree(entry);
1974 }
1975 }
1976
1977 /* called when we find an exclusion range definition in ACPI */
init_exclusion_range(struct ivmd_header * m)1978 static int __init init_exclusion_range(struct ivmd_header *m)
1979 {
1980 int i;
1981
1982 switch (m->type) {
1983 case ACPI_IVMD_TYPE:
1984 set_device_exclusion_range(m->devid, m);
1985 break;
1986 case ACPI_IVMD_TYPE_ALL:
1987 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1988 set_device_exclusion_range(i, m);
1989 break;
1990 case ACPI_IVMD_TYPE_RANGE:
1991 for (i = m->devid; i <= m->aux; ++i)
1992 set_device_exclusion_range(i, m);
1993 break;
1994 default:
1995 break;
1996 }
1997
1998 return 0;
1999 }
2000
2001 /* called for unity map ACPI definition */
init_unity_map_range(struct ivmd_header * m)2002 static int __init init_unity_map_range(struct ivmd_header *m)
2003 {
2004 struct unity_map_entry *e = NULL;
2005 char *s;
2006
2007 e = kzalloc(sizeof(*e), GFP_KERNEL);
2008 if (e == NULL)
2009 return -ENOMEM;
2010
2011 if (m->flags & IVMD_FLAG_EXCL_RANGE)
2012 init_exclusion_range(m);
2013
2014 switch (m->type) {
2015 default:
2016 kfree(e);
2017 return 0;
2018 case ACPI_IVMD_TYPE:
2019 s = "IVMD_TYPEi\t\t\t";
2020 e->devid_start = e->devid_end = m->devid;
2021 break;
2022 case ACPI_IVMD_TYPE_ALL:
2023 s = "IVMD_TYPE_ALL\t\t";
2024 e->devid_start = 0;
2025 e->devid_end = amd_iommu_last_bdf;
2026 break;
2027 case ACPI_IVMD_TYPE_RANGE:
2028 s = "IVMD_TYPE_RANGE\t\t";
2029 e->devid_start = m->devid;
2030 e->devid_end = m->aux;
2031 break;
2032 }
2033 e->address_start = PAGE_ALIGN(m->range_start);
2034 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2035 e->prot = m->flags >> 1;
2036
2037 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
2038 " range_start: %016llx range_end: %016llx flags: %x\n", s,
2039 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2040 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
2041 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2042 e->address_start, e->address_end, m->flags);
2043
2044 list_add_tail(&e->list, &amd_iommu_unity_map);
2045
2046 return 0;
2047 }
2048
2049 /* iterates over all memory definitions we find in the ACPI table */
init_memory_definitions(struct acpi_table_header * table)2050 static int __init init_memory_definitions(struct acpi_table_header *table)
2051 {
2052 u8 *p = (u8 *)table, *end = (u8 *)table;
2053 struct ivmd_header *m;
2054
2055 end += table->length;
2056 p += IVRS_HEADER_LENGTH;
2057
2058 while (p < end) {
2059 m = (struct ivmd_header *)p;
2060 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2061 init_unity_map_range(m);
2062
2063 p += m->length;
2064 }
2065
2066 return 0;
2067 }
2068
2069 /*
2070 * Init the device table to not allow DMA access for devices
2071 */
init_device_table_dma(void)2072 static void init_device_table_dma(void)
2073 {
2074 u32 devid;
2075
2076 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2077 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
2078 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
2079 }
2080 }
2081
uninit_device_table_dma(void)2082 static void __init uninit_device_table_dma(void)
2083 {
2084 u32 devid;
2085
2086 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
2087 amd_iommu_dev_table[devid].data[0] = 0ULL;
2088 amd_iommu_dev_table[devid].data[1] = 0ULL;
2089 }
2090 }
2091
init_device_table(void)2092 static void init_device_table(void)
2093 {
2094 u32 devid;
2095
2096 if (!amd_iommu_irq_remap)
2097 return;
2098
2099 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
2100 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
2101 }
2102
iommu_init_flags(struct amd_iommu * iommu)2103 static void iommu_init_flags(struct amd_iommu *iommu)
2104 {
2105 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2106 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2107 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2108
2109 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2110 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2111 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2112
2113 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2114 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2115 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2116
2117 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2118 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2119 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2120
2121 /*
2122 * make IOMMU memory accesses cache coherent
2123 */
2124 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2125
2126 /* Set IOTLB invalidation timeout to 1s */
2127 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2128 }
2129
iommu_apply_resume_quirks(struct amd_iommu * iommu)2130 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2131 {
2132 int i, j;
2133 u32 ioc_feature_control;
2134 struct pci_dev *pdev = iommu->root_pdev;
2135
2136 /* RD890 BIOSes may not have completely reconfigured the iommu */
2137 if (!is_rd890_iommu(iommu->dev) || !pdev)
2138 return;
2139
2140 /*
2141 * First, we need to ensure that the iommu is enabled. This is
2142 * controlled by a register in the northbridge
2143 */
2144
2145 /* Select Northbridge indirect register 0x75 and enable writing */
2146 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2147 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2148
2149 /* Enable the iommu */
2150 if (!(ioc_feature_control & 0x1))
2151 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2152
2153 /* Restore the iommu BAR */
2154 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2155 iommu->stored_addr_lo);
2156 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2157 iommu->stored_addr_hi);
2158
2159 /* Restore the l1 indirect regs for each of the 6 l1s */
2160 for (i = 0; i < 6; i++)
2161 for (j = 0; j < 0x12; j++)
2162 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2163
2164 /* Restore the l2 indirect regs */
2165 for (i = 0; i < 0x83; i++)
2166 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2167
2168 /* Lock PCI setup registers */
2169 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2170 iommu->stored_addr_lo | 1);
2171 }
2172
iommu_enable_ga(struct amd_iommu * iommu)2173 static void iommu_enable_ga(struct amd_iommu *iommu)
2174 {
2175 #ifdef CONFIG_IRQ_REMAP
2176 switch (amd_iommu_guest_ir) {
2177 case AMD_IOMMU_GUEST_IR_VAPIC:
2178 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2179 /* Fall through */
2180 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2181 iommu_feature_enable(iommu, CONTROL_GA_EN);
2182 iommu->irte_ops = &irte_128_ops;
2183 break;
2184 default:
2185 iommu->irte_ops = &irte_32_ops;
2186 break;
2187 }
2188 #endif
2189 }
2190
early_enable_iommu(struct amd_iommu * iommu)2191 static void early_enable_iommu(struct amd_iommu *iommu)
2192 {
2193 iommu_disable(iommu);
2194 iommu_init_flags(iommu);
2195 iommu_set_device_table(iommu);
2196 iommu_enable_command_buffer(iommu);
2197 iommu_enable_event_buffer(iommu);
2198 iommu_set_exclusion_range(iommu);
2199 iommu_enable_ga(iommu);
2200 iommu_enable_xt(iommu);
2201 iommu_enable(iommu);
2202 iommu_flush_all_caches(iommu);
2203 }
2204
2205 /*
2206 * This function finally enables all IOMMUs found in the system after
2207 * they have been initialized.
2208 *
2209 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2210 * the old content of device table entries. Not this case or copy failed,
2211 * just continue as normal kernel does.
2212 */
early_enable_iommus(void)2213 static void early_enable_iommus(void)
2214 {
2215 struct amd_iommu *iommu;
2216
2217
2218 if (!copy_device_table()) {
2219 /*
2220 * If come here because of failure in copying device table from old
2221 * kernel with all IOMMUs enabled, print error message and try to
2222 * free allocated old_dev_tbl_cpy.
2223 */
2224 if (amd_iommu_pre_enabled)
2225 pr_err("Failed to copy DEV table from previous kernel.\n");
2226 if (old_dev_tbl_cpy != NULL)
2227 free_pages((unsigned long)old_dev_tbl_cpy,
2228 get_order(dev_table_size));
2229
2230 for_each_iommu(iommu) {
2231 clear_translation_pre_enabled(iommu);
2232 early_enable_iommu(iommu);
2233 }
2234 } else {
2235 pr_info("Copied DEV table from previous kernel.\n");
2236 free_pages((unsigned long)amd_iommu_dev_table,
2237 get_order(dev_table_size));
2238 amd_iommu_dev_table = old_dev_tbl_cpy;
2239 for_each_iommu(iommu) {
2240 iommu_disable_command_buffer(iommu);
2241 iommu_disable_event_buffer(iommu);
2242 iommu_enable_command_buffer(iommu);
2243 iommu_enable_event_buffer(iommu);
2244 iommu_enable_ga(iommu);
2245 iommu_enable_xt(iommu);
2246 iommu_set_device_table(iommu);
2247 iommu_flush_all_caches(iommu);
2248 }
2249 }
2250
2251 #ifdef CONFIG_IRQ_REMAP
2252 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2253 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2254 #endif
2255 }
2256
enable_iommus_v2(void)2257 static void enable_iommus_v2(void)
2258 {
2259 struct amd_iommu *iommu;
2260
2261 for_each_iommu(iommu) {
2262 iommu_enable_ppr_log(iommu);
2263 iommu_enable_gt(iommu);
2264 }
2265 }
2266
enable_iommus(void)2267 static void enable_iommus(void)
2268 {
2269 early_enable_iommus();
2270
2271 enable_iommus_v2();
2272 }
2273
disable_iommus(void)2274 static void disable_iommus(void)
2275 {
2276 struct amd_iommu *iommu;
2277
2278 for_each_iommu(iommu)
2279 iommu_disable(iommu);
2280
2281 #ifdef CONFIG_IRQ_REMAP
2282 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2283 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2284 #endif
2285 }
2286
2287 /*
2288 * Suspend/Resume support
2289 * disable suspend until real resume implemented
2290 */
2291
amd_iommu_resume(void)2292 static void amd_iommu_resume(void)
2293 {
2294 struct amd_iommu *iommu;
2295
2296 for_each_iommu(iommu)
2297 iommu_apply_resume_quirks(iommu);
2298
2299 /* re-load the hardware */
2300 enable_iommus();
2301
2302 amd_iommu_enable_interrupts();
2303 }
2304
amd_iommu_suspend(void)2305 static int amd_iommu_suspend(void)
2306 {
2307 /* disable IOMMUs to go out of the way for BIOS */
2308 disable_iommus();
2309
2310 return 0;
2311 }
2312
2313 static struct syscore_ops amd_iommu_syscore_ops = {
2314 .suspend = amd_iommu_suspend,
2315 .resume = amd_iommu_resume,
2316 };
2317
free_iommu_resources(void)2318 static void __init free_iommu_resources(void)
2319 {
2320 kmemleak_free(irq_lookup_table);
2321 free_pages((unsigned long)irq_lookup_table,
2322 get_order(rlookup_table_size));
2323 irq_lookup_table = NULL;
2324
2325 kmem_cache_destroy(amd_iommu_irq_cache);
2326 amd_iommu_irq_cache = NULL;
2327
2328 free_pages((unsigned long)amd_iommu_rlookup_table,
2329 get_order(rlookup_table_size));
2330 amd_iommu_rlookup_table = NULL;
2331
2332 free_pages((unsigned long)amd_iommu_alias_table,
2333 get_order(alias_table_size));
2334 amd_iommu_alias_table = NULL;
2335
2336 free_pages((unsigned long)amd_iommu_dev_table,
2337 get_order(dev_table_size));
2338 amd_iommu_dev_table = NULL;
2339
2340 free_iommu_all();
2341
2342 #ifdef CONFIG_GART_IOMMU
2343 /*
2344 * We failed to initialize the AMD IOMMU - try fallback to GART
2345 * if possible.
2346 */
2347 gart_iommu_init();
2348
2349 #endif
2350 }
2351
2352 /* SB IOAPIC is always on this device in AMD systems */
2353 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2354
check_ioapic_information(void)2355 static bool __init check_ioapic_information(void)
2356 {
2357 const char *fw_bug = FW_BUG;
2358 bool ret, has_sb_ioapic;
2359 int idx;
2360
2361 has_sb_ioapic = false;
2362 ret = false;
2363
2364 /*
2365 * If we have map overrides on the kernel command line the
2366 * messages in this function might not describe firmware bugs
2367 * anymore - so be careful
2368 */
2369 if (cmdline_maps)
2370 fw_bug = "";
2371
2372 for (idx = 0; idx < nr_ioapics; idx++) {
2373 int devid, id = mpc_ioapic_id(idx);
2374
2375 devid = get_ioapic_devid(id);
2376 if (devid < 0) {
2377 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
2378 fw_bug, id);
2379 ret = false;
2380 } else if (devid == IOAPIC_SB_DEVID) {
2381 has_sb_ioapic = true;
2382 ret = true;
2383 }
2384 }
2385
2386 if (!has_sb_ioapic) {
2387 /*
2388 * We expect the SB IOAPIC to be listed in the IVRS
2389 * table. The system timer is connected to the SB IOAPIC
2390 * and if we don't have it in the list the system will
2391 * panic at boot time. This situation usually happens
2392 * when the BIOS is buggy and provides us the wrong
2393 * device id for the IOAPIC in the system.
2394 */
2395 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
2396 }
2397
2398 if (!ret)
2399 pr_err("AMD-Vi: Disabling interrupt remapping\n");
2400
2401 return ret;
2402 }
2403
free_dma_resources(void)2404 static void __init free_dma_resources(void)
2405 {
2406 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2407 get_order(MAX_DOMAIN_ID/8));
2408 amd_iommu_pd_alloc_bitmap = NULL;
2409
2410 free_unity_maps();
2411 }
2412
2413 /*
2414 * This is the hardware init function for AMD IOMMU in the system.
2415 * This function is called either from amd_iommu_init or from the interrupt
2416 * remapping setup code.
2417 *
2418 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2419 * four times:
2420 *
2421 * 1 pass) Discover the most comprehensive IVHD type to use.
2422 *
2423 * 2 pass) Find the highest PCI device id the driver has to handle.
2424 * Upon this information the size of the data structures is
2425 * determined that needs to be allocated.
2426 *
2427 * 3 pass) Initialize the data structures just allocated with the
2428 * information in the ACPI table about available AMD IOMMUs
2429 * in the system. It also maps the PCI devices in the
2430 * system to specific IOMMUs
2431 *
2432 * 4 pass) After the basic data structures are allocated and
2433 * initialized we update them with information about memory
2434 * remapping requirements parsed out of the ACPI table in
2435 * this last pass.
2436 *
2437 * After everything is set up the IOMMUs are enabled and the necessary
2438 * hotplug and suspend notifiers are registered.
2439 */
early_amd_iommu_init(void)2440 static int __init early_amd_iommu_init(void)
2441 {
2442 struct acpi_table_header *ivrs_base;
2443 acpi_status status;
2444 int i, remap_cache_sz, ret = 0;
2445
2446 if (!amd_iommu_detected)
2447 return -ENODEV;
2448
2449 status = acpi_get_table("IVRS", 0, &ivrs_base);
2450 if (status == AE_NOT_FOUND)
2451 return -ENODEV;
2452 else if (ACPI_FAILURE(status)) {
2453 const char *err = acpi_format_exception(status);
2454 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2455 return -EINVAL;
2456 }
2457
2458 /*
2459 * Validate checksum here so we don't need to do it when
2460 * we actually parse the table
2461 */
2462 ret = check_ivrs_checksum(ivrs_base);
2463 if (ret)
2464 goto out;
2465
2466 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2467 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2468
2469 /*
2470 * First parse ACPI tables to find the largest Bus/Dev/Func
2471 * we need to handle. Upon this information the shared data
2472 * structures for the IOMMUs in the system will be allocated
2473 */
2474 ret = find_last_devid_acpi(ivrs_base);
2475 if (ret)
2476 goto out;
2477
2478 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2479 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2480 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2481
2482 /* Device table - directly used by all IOMMUs */
2483 ret = -ENOMEM;
2484 amd_iommu_dev_table = (void *)__get_free_pages(
2485 GFP_KERNEL | __GFP_ZERO | GFP_DMA32,
2486 get_order(dev_table_size));
2487 if (amd_iommu_dev_table == NULL)
2488 goto out;
2489
2490 /*
2491 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2492 * IOMMU see for that device
2493 */
2494 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2495 get_order(alias_table_size));
2496 if (amd_iommu_alias_table == NULL)
2497 goto out;
2498
2499 /* IOMMU rlookup table - find the IOMMU for a specific device */
2500 amd_iommu_rlookup_table = (void *)__get_free_pages(
2501 GFP_KERNEL | __GFP_ZERO,
2502 get_order(rlookup_table_size));
2503 if (amd_iommu_rlookup_table == NULL)
2504 goto out;
2505
2506 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2507 GFP_KERNEL | __GFP_ZERO,
2508 get_order(MAX_DOMAIN_ID/8));
2509 if (amd_iommu_pd_alloc_bitmap == NULL)
2510 goto out;
2511
2512 /*
2513 * let all alias entries point to itself
2514 */
2515 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2516 amd_iommu_alias_table[i] = i;
2517
2518 /*
2519 * never allocate domain 0 because its used as the non-allocated and
2520 * error value placeholder
2521 */
2522 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2523
2524 spin_lock_init(&amd_iommu_pd_lock);
2525
2526 /*
2527 * now the data structures are allocated and basically initialized
2528 * start the real acpi table scan
2529 */
2530 ret = init_iommu_all(ivrs_base);
2531 if (ret)
2532 goto out;
2533
2534 /* Disable any previously enabled IOMMUs */
2535 if (!is_kdump_kernel() || amd_iommu_disabled)
2536 disable_iommus();
2537
2538 if (amd_iommu_irq_remap)
2539 amd_iommu_irq_remap = check_ioapic_information();
2540
2541 if (amd_iommu_irq_remap) {
2542 /*
2543 * Interrupt remapping enabled, create kmem_cache for the
2544 * remapping tables.
2545 */
2546 ret = -ENOMEM;
2547 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2548 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2549 else
2550 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2551 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2552 remap_cache_sz,
2553 IRQ_TABLE_ALIGNMENT,
2554 0, NULL);
2555 if (!amd_iommu_irq_cache)
2556 goto out;
2557
2558 irq_lookup_table = (void *)__get_free_pages(
2559 GFP_KERNEL | __GFP_ZERO,
2560 get_order(rlookup_table_size));
2561 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2562 1, GFP_KERNEL);
2563 if (!irq_lookup_table)
2564 goto out;
2565 }
2566
2567 ret = init_memory_definitions(ivrs_base);
2568 if (ret)
2569 goto out;
2570
2571 /* init the device table */
2572 init_device_table();
2573
2574 out:
2575 /* Don't leak any ACPI memory */
2576 acpi_put_table(ivrs_base);
2577 ivrs_base = NULL;
2578
2579 return ret;
2580 }
2581
amd_iommu_enable_interrupts(void)2582 static int amd_iommu_enable_interrupts(void)
2583 {
2584 struct amd_iommu *iommu;
2585 int ret = 0;
2586
2587 for_each_iommu(iommu) {
2588 ret = iommu_init_msi(iommu);
2589 if (ret)
2590 goto out;
2591 }
2592
2593 out:
2594 return ret;
2595 }
2596
detect_ivrs(void)2597 static bool detect_ivrs(void)
2598 {
2599 struct acpi_table_header *ivrs_base;
2600 acpi_status status;
2601
2602 status = acpi_get_table("IVRS", 0, &ivrs_base);
2603 if (status == AE_NOT_FOUND)
2604 return false;
2605 else if (ACPI_FAILURE(status)) {
2606 const char *err = acpi_format_exception(status);
2607 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2608 return false;
2609 }
2610
2611 acpi_put_table(ivrs_base);
2612
2613 /* Make sure ACS will be enabled during PCI probe */
2614 pci_request_acs();
2615
2616 return true;
2617 }
2618
2619 /****************************************************************************
2620 *
2621 * AMD IOMMU Initialization State Machine
2622 *
2623 ****************************************************************************/
2624
state_next(void)2625 static int __init state_next(void)
2626 {
2627 int ret = 0;
2628
2629 switch (init_state) {
2630 case IOMMU_START_STATE:
2631 if (!detect_ivrs()) {
2632 init_state = IOMMU_NOT_FOUND;
2633 ret = -ENODEV;
2634 } else {
2635 init_state = IOMMU_IVRS_DETECTED;
2636 }
2637 break;
2638 case IOMMU_IVRS_DETECTED:
2639 ret = early_amd_iommu_init();
2640 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2641 if (init_state == IOMMU_ACPI_FINISHED && amd_iommu_disabled) {
2642 pr_info("AMD-Vi: AMD IOMMU disabled on kernel command-line\n");
2643 free_dma_resources();
2644 free_iommu_resources();
2645 init_state = IOMMU_CMDLINE_DISABLED;
2646 ret = -EINVAL;
2647 }
2648 break;
2649 case IOMMU_ACPI_FINISHED:
2650 early_enable_iommus();
2651 x86_platform.iommu_shutdown = disable_iommus;
2652 init_state = IOMMU_ENABLED;
2653 break;
2654 case IOMMU_ENABLED:
2655 register_syscore_ops(&amd_iommu_syscore_ops);
2656 ret = amd_iommu_init_pci();
2657 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2658 enable_iommus_v2();
2659 break;
2660 case IOMMU_PCI_INIT:
2661 ret = amd_iommu_enable_interrupts();
2662 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2663 break;
2664 case IOMMU_INTERRUPTS_EN:
2665 ret = amd_iommu_init_dma_ops();
2666 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2667 break;
2668 case IOMMU_DMA_OPS:
2669 init_state = IOMMU_INITIALIZED;
2670 break;
2671 case IOMMU_INITIALIZED:
2672 /* Nothing to do */
2673 break;
2674 case IOMMU_NOT_FOUND:
2675 case IOMMU_INIT_ERROR:
2676 case IOMMU_CMDLINE_DISABLED:
2677 /* Error states => do nothing */
2678 ret = -EINVAL;
2679 break;
2680 default:
2681 /* Unknown state */
2682 BUG();
2683 }
2684
2685 return ret;
2686 }
2687
iommu_go_to_state(enum iommu_init_state state)2688 static int __init iommu_go_to_state(enum iommu_init_state state)
2689 {
2690 int ret = -EINVAL;
2691
2692 while (init_state != state) {
2693 if (init_state == IOMMU_NOT_FOUND ||
2694 init_state == IOMMU_INIT_ERROR ||
2695 init_state == IOMMU_CMDLINE_DISABLED)
2696 break;
2697 ret = state_next();
2698 }
2699
2700 return ret;
2701 }
2702
2703 #ifdef CONFIG_IRQ_REMAP
amd_iommu_prepare(void)2704 int __init amd_iommu_prepare(void)
2705 {
2706 int ret;
2707
2708 amd_iommu_irq_remap = true;
2709
2710 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2711 if (ret)
2712 return ret;
2713 return amd_iommu_irq_remap ? 0 : -ENODEV;
2714 }
2715
amd_iommu_enable(void)2716 int __init amd_iommu_enable(void)
2717 {
2718 int ret;
2719
2720 ret = iommu_go_to_state(IOMMU_ENABLED);
2721 if (ret)
2722 return ret;
2723
2724 irq_remapping_enabled = 1;
2725 return amd_iommu_xt_mode;
2726 }
2727
amd_iommu_disable(void)2728 void amd_iommu_disable(void)
2729 {
2730 amd_iommu_suspend();
2731 }
2732
amd_iommu_reenable(int mode)2733 int amd_iommu_reenable(int mode)
2734 {
2735 amd_iommu_resume();
2736
2737 return 0;
2738 }
2739
amd_iommu_enable_faulting(void)2740 int __init amd_iommu_enable_faulting(void)
2741 {
2742 /* We enable MSI later when PCI is initialized */
2743 return 0;
2744 }
2745 #endif
2746
2747 /*
2748 * This is the core init function for AMD IOMMU hardware in the system.
2749 * This function is called from the generic x86 DMA layer initialization
2750 * code.
2751 */
amd_iommu_init(void)2752 static int __init amd_iommu_init(void)
2753 {
2754 struct amd_iommu *iommu;
2755 int ret;
2756
2757 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2758 if (ret) {
2759 free_dma_resources();
2760 if (!irq_remapping_enabled) {
2761 disable_iommus();
2762 free_iommu_resources();
2763 } else {
2764 uninit_device_table_dma();
2765 for_each_iommu(iommu)
2766 iommu_flush_all_caches(iommu);
2767 }
2768 }
2769
2770 for_each_iommu(iommu)
2771 amd_iommu_debugfs_setup(iommu);
2772
2773 return ret;
2774 }
2775
amd_iommu_sme_check(void)2776 static bool amd_iommu_sme_check(void)
2777 {
2778 if (!sme_active() || (boot_cpu_data.x86 != 0x17))
2779 return true;
2780
2781 /* For Fam17h, a specific level of support is required */
2782 if (boot_cpu_data.microcode >= 0x08001205)
2783 return true;
2784
2785 if ((boot_cpu_data.microcode >= 0x08001126) &&
2786 (boot_cpu_data.microcode <= 0x080011ff))
2787 return true;
2788
2789 pr_notice("AMD-Vi: IOMMU not currently supported when SME is active\n");
2790
2791 return false;
2792 }
2793
2794 /****************************************************************************
2795 *
2796 * Early detect code. This code runs at IOMMU detection time in the DMA
2797 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2798 * IOMMUs
2799 *
2800 ****************************************************************************/
amd_iommu_detect(void)2801 int __init amd_iommu_detect(void)
2802 {
2803 int ret;
2804
2805 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2806 return -ENODEV;
2807
2808 if (!amd_iommu_sme_check())
2809 return -ENODEV;
2810
2811 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2812 if (ret)
2813 return ret;
2814
2815 amd_iommu_detected = true;
2816 iommu_detected = 1;
2817 x86_init.iommu.iommu_init = amd_iommu_init;
2818
2819 return 1;
2820 }
2821
2822 /****************************************************************************
2823 *
2824 * Parsing functions for the AMD IOMMU specific kernel command line
2825 * options.
2826 *
2827 ****************************************************************************/
2828
parse_amd_iommu_dump(char * str)2829 static int __init parse_amd_iommu_dump(char *str)
2830 {
2831 amd_iommu_dump = true;
2832
2833 return 1;
2834 }
2835
parse_amd_iommu_intr(char * str)2836 static int __init parse_amd_iommu_intr(char *str)
2837 {
2838 for (; *str; ++str) {
2839 if (strncmp(str, "legacy", 6) == 0) {
2840 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2841 break;
2842 }
2843 if (strncmp(str, "vapic", 5) == 0) {
2844 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2845 break;
2846 }
2847 }
2848 return 1;
2849 }
2850
parse_amd_iommu_options(char * str)2851 static int __init parse_amd_iommu_options(char *str)
2852 {
2853 for (; *str; ++str) {
2854 if (strncmp(str, "fullflush", 9) == 0)
2855 amd_iommu_unmap_flush = true;
2856 if (strncmp(str, "off", 3) == 0)
2857 amd_iommu_disabled = true;
2858 if (strncmp(str, "force_isolation", 15) == 0)
2859 amd_iommu_force_isolation = true;
2860 }
2861
2862 return 1;
2863 }
2864
parse_ivrs_ioapic(char * str)2865 static int __init parse_ivrs_ioapic(char *str)
2866 {
2867 unsigned int bus, dev, fn;
2868 int ret, id, i;
2869 u16 devid;
2870
2871 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2872
2873 if (ret != 4) {
2874 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2875 return 1;
2876 }
2877
2878 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2879 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2880 str);
2881 return 1;
2882 }
2883
2884 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2885
2886 cmdline_maps = true;
2887 i = early_ioapic_map_size++;
2888 early_ioapic_map[i].id = id;
2889 early_ioapic_map[i].devid = devid;
2890 early_ioapic_map[i].cmd_line = true;
2891
2892 return 1;
2893 }
2894
parse_ivrs_hpet(char * str)2895 static int __init parse_ivrs_hpet(char *str)
2896 {
2897 unsigned int bus, dev, fn;
2898 int ret, id, i;
2899 u16 devid;
2900
2901 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2902
2903 if (ret != 4) {
2904 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2905 return 1;
2906 }
2907
2908 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2909 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2910 str);
2911 return 1;
2912 }
2913
2914 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2915
2916 cmdline_maps = true;
2917 i = early_hpet_map_size++;
2918 early_hpet_map[i].id = id;
2919 early_hpet_map[i].devid = devid;
2920 early_hpet_map[i].cmd_line = true;
2921
2922 return 1;
2923 }
2924
parse_ivrs_acpihid(char * str)2925 static int __init parse_ivrs_acpihid(char *str)
2926 {
2927 u32 bus, dev, fn;
2928 char *hid, *uid, *p;
2929 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
2930 int ret, i;
2931
2932 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
2933 if (ret != 4) {
2934 pr_err("AMD-Vi: Invalid command line: ivrs_acpihid(%s)\n", str);
2935 return 1;
2936 }
2937
2938 p = acpiid;
2939 hid = strsep(&p, ":");
2940 uid = p;
2941
2942 if (!hid || !(*hid) || !uid) {
2943 pr_err("AMD-Vi: Invalid command line: hid or uid\n");
2944 return 1;
2945 }
2946
2947 i = early_acpihid_map_size++;
2948 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
2949 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
2950 early_acpihid_map[i].devid =
2951 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2952 early_acpihid_map[i].cmd_line = true;
2953
2954 return 1;
2955 }
2956
2957 __setup("amd_iommu_dump", parse_amd_iommu_dump);
2958 __setup("amd_iommu=", parse_amd_iommu_options);
2959 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
2960 __setup("ivrs_ioapic", parse_ivrs_ioapic);
2961 __setup("ivrs_hpet", parse_ivrs_hpet);
2962 __setup("ivrs_acpihid", parse_ivrs_acpihid);
2963
2964 IOMMU_INIT_FINISH(amd_iommu_detect,
2965 gart_iommu_hole_init,
2966 NULL,
2967 NULL);
2968
amd_iommu_v2_supported(void)2969 bool amd_iommu_v2_supported(void)
2970 {
2971 return amd_iommu_v2_present;
2972 }
2973 EXPORT_SYMBOL(amd_iommu_v2_supported);
2974
get_amd_iommu(unsigned int idx)2975 struct amd_iommu *get_amd_iommu(unsigned int idx)
2976 {
2977 unsigned int i = 0;
2978 struct amd_iommu *iommu;
2979
2980 for_each_iommu(iommu)
2981 if (i++ == idx)
2982 return iommu;
2983 return NULL;
2984 }
2985 EXPORT_SYMBOL(get_amd_iommu);
2986
2987 /****************************************************************************
2988 *
2989 * IOMMU EFR Performance Counter support functionality. This code allows
2990 * access to the IOMMU PC functionality.
2991 *
2992 ****************************************************************************/
2993
amd_iommu_pc_get_max_banks(unsigned int idx)2994 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
2995 {
2996 struct amd_iommu *iommu = get_amd_iommu(idx);
2997
2998 if (iommu)
2999 return iommu->max_banks;
3000
3001 return 0;
3002 }
3003 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
3004
amd_iommu_pc_supported(void)3005 bool amd_iommu_pc_supported(void)
3006 {
3007 return amd_iommu_pc_present;
3008 }
3009 EXPORT_SYMBOL(amd_iommu_pc_supported);
3010
amd_iommu_pc_get_max_counters(unsigned int idx)3011 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3012 {
3013 struct amd_iommu *iommu = get_amd_iommu(idx);
3014
3015 if (iommu)
3016 return iommu->max_counters;
3017
3018 return 0;
3019 }
3020 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
3021
iommu_pc_get_set_reg(struct amd_iommu * iommu,u8 bank,u8 cntr,u8 fxn,u64 * value,bool is_write)3022 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3023 u8 fxn, u64 *value, bool is_write)
3024 {
3025 u32 offset;
3026 u32 max_offset_lim;
3027
3028 /* Make sure the IOMMU PC resource is available */
3029 if (!amd_iommu_pc_present)
3030 return -ENODEV;
3031
3032 /* Check for valid iommu and pc register indexing */
3033 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3034 return -ENODEV;
3035
3036 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3037
3038 /* Limit the offset to the hw defined mmio region aperture */
3039 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3040 (iommu->max_counters << 8) | 0x28);
3041 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3042 (offset > max_offset_lim))
3043 return -EINVAL;
3044
3045 if (is_write) {
3046 u64 val = *value & GENMASK_ULL(47, 0);
3047
3048 writel((u32)val, iommu->mmio_base + offset);
3049 writel((val >> 32), iommu->mmio_base + offset + 4);
3050 } else {
3051 *value = readl(iommu->mmio_base + offset + 4);
3052 *value <<= 32;
3053 *value |= readl(iommu->mmio_base + offset);
3054 *value &= GENMASK_ULL(47, 0);
3055 }
3056
3057 return 0;
3058 }
3059
amd_iommu_pc_get_reg(struct amd_iommu * iommu,u8 bank,u8 cntr,u8 fxn,u64 * value)3060 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3061 {
3062 if (!iommu)
3063 return -EINVAL;
3064
3065 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3066 }
3067 EXPORT_SYMBOL(amd_iommu_pc_get_reg);
3068
amd_iommu_pc_set_reg(struct amd_iommu * iommu,u8 bank,u8 cntr,u8 fxn,u64 * value)3069 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3070 {
3071 if (!iommu)
3072 return -EINVAL;
3073
3074 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3075 }
3076 EXPORT_SYMBOL(amd_iommu_pc_set_reg);
3077