1 /*
2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
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/slab.h>
24 #include <linux/syscore_ops.h>
25 #include <linux/interrupt.h>
26 #include <linux/msi.h>
27 #include <linux/amd-iommu.h>
28 #include <linux/export.h>
29 #include <asm/pci-direct.h>
30 #include <asm/iommu.h>
31 #include <asm/gart.h>
32 #include <asm/x86_init.h>
33 #include <asm/iommu_table.h>
34
35 #include "amd_iommu_proto.h"
36 #include "amd_iommu_types.h"
37
38 /*
39 * definitions for the ACPI scanning code
40 */
41 #define IVRS_HEADER_LENGTH 48
42
43 #define ACPI_IVHD_TYPE 0x10
44 #define ACPI_IVMD_TYPE_ALL 0x20
45 #define ACPI_IVMD_TYPE 0x21
46 #define ACPI_IVMD_TYPE_RANGE 0x22
47
48 #define IVHD_DEV_ALL 0x01
49 #define IVHD_DEV_SELECT 0x02
50 #define IVHD_DEV_SELECT_RANGE_START 0x03
51 #define IVHD_DEV_RANGE_END 0x04
52 #define IVHD_DEV_ALIAS 0x42
53 #define IVHD_DEV_ALIAS_RANGE 0x43
54 #define IVHD_DEV_EXT_SELECT 0x46
55 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
56
57 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
58 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
59 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
60 #define IVHD_FLAG_ISOC_EN_MASK 0x08
61
62 #define IVMD_FLAG_EXCL_RANGE 0x08
63 #define IVMD_FLAG_UNITY_MAP 0x01
64
65 #define ACPI_DEVFLAG_INITPASS 0x01
66 #define ACPI_DEVFLAG_EXTINT 0x02
67 #define ACPI_DEVFLAG_NMI 0x04
68 #define ACPI_DEVFLAG_SYSMGT1 0x10
69 #define ACPI_DEVFLAG_SYSMGT2 0x20
70 #define ACPI_DEVFLAG_LINT0 0x40
71 #define ACPI_DEVFLAG_LINT1 0x80
72 #define ACPI_DEVFLAG_ATSDIS 0x10000000
73
74 /*
75 * ACPI table definitions
76 *
77 * These data structures are laid over the table to parse the important values
78 * out of it.
79 */
80
81 /*
82 * structure describing one IOMMU in the ACPI table. Typically followed by one
83 * or more ivhd_entrys.
84 */
85 struct ivhd_header {
86 u8 type;
87 u8 flags;
88 u16 length;
89 u16 devid;
90 u16 cap_ptr;
91 u64 mmio_phys;
92 u16 pci_seg;
93 u16 info;
94 u32 reserved;
95 } __attribute__((packed));
96
97 /*
98 * A device entry describing which devices a specific IOMMU translates and
99 * which requestor ids they use.
100 */
101 struct ivhd_entry {
102 u8 type;
103 u16 devid;
104 u8 flags;
105 u32 ext;
106 } __attribute__((packed));
107
108 /*
109 * An AMD IOMMU memory definition structure. It defines things like exclusion
110 * ranges for devices and regions that should be unity mapped.
111 */
112 struct ivmd_header {
113 u8 type;
114 u8 flags;
115 u16 length;
116 u16 devid;
117 u16 aux;
118 u64 resv;
119 u64 range_start;
120 u64 range_length;
121 } __attribute__((packed));
122
123 bool amd_iommu_dump;
124
125 static int __initdata amd_iommu_detected;
126 static bool __initdata amd_iommu_disabled;
127
128 u16 amd_iommu_last_bdf; /* largest PCI device id we have
129 to handle */
130 LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
131 we find in ACPI */
132 bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
133
134 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
135 system */
136
137 /* Array to assign indices to IOMMUs*/
138 struct amd_iommu *amd_iommus[MAX_IOMMUS];
139 int amd_iommus_present;
140
141 /* IOMMUs have a non-present cache? */
142 bool amd_iommu_np_cache __read_mostly;
143 bool amd_iommu_iotlb_sup __read_mostly = true;
144
145 u32 amd_iommu_max_pasids __read_mostly = ~0;
146
147 bool amd_iommu_v2_present __read_mostly;
148
149 bool amd_iommu_force_isolation __read_mostly;
150
151 /*
152 * The ACPI table parsing functions set this variable on an error
153 */
154 static int __initdata amd_iommu_init_err;
155
156 /*
157 * List of protection domains - used during resume
158 */
159 LIST_HEAD(amd_iommu_pd_list);
160 spinlock_t amd_iommu_pd_lock;
161
162 /*
163 * Pointer to the device table which is shared by all AMD IOMMUs
164 * it is indexed by the PCI device id or the HT unit id and contains
165 * information about the domain the device belongs to as well as the
166 * page table root pointer.
167 */
168 struct dev_table_entry *amd_iommu_dev_table;
169
170 /*
171 * The alias table is a driver specific data structure which contains the
172 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
173 * More than one device can share the same requestor id.
174 */
175 u16 *amd_iommu_alias_table;
176
177 /*
178 * The rlookup table is used to find the IOMMU which is responsible
179 * for a specific device. It is also indexed by the PCI device id.
180 */
181 struct amd_iommu **amd_iommu_rlookup_table;
182
183 /*
184 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
185 * to know which ones are already in use.
186 */
187 unsigned long *amd_iommu_pd_alloc_bitmap;
188
189 static u32 dev_table_size; /* size of the device table */
190 static u32 alias_table_size; /* size of the alias table */
191 static u32 rlookup_table_size; /* size if the rlookup table */
192
193 /*
194 * This function flushes all internal caches of
195 * the IOMMU used by this driver.
196 */
197 extern void iommu_flush_all_caches(struct amd_iommu *iommu);
198
update_last_devid(u16 devid)199 static inline void update_last_devid(u16 devid)
200 {
201 if (devid > amd_iommu_last_bdf)
202 amd_iommu_last_bdf = devid;
203 }
204
tbl_size(int entry_size)205 static inline unsigned long tbl_size(int entry_size)
206 {
207 unsigned shift = PAGE_SHIFT +
208 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
209
210 return 1UL << shift;
211 }
212
213 /* Access to l1 and l2 indexed register spaces */
214
iommu_read_l1(struct amd_iommu * iommu,u16 l1,u8 address)215 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
216 {
217 u32 val;
218
219 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
220 pci_read_config_dword(iommu->dev, 0xfc, &val);
221 return val;
222 }
223
iommu_write_l1(struct amd_iommu * iommu,u16 l1,u8 address,u32 val)224 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
225 {
226 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
227 pci_write_config_dword(iommu->dev, 0xfc, val);
228 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
229 }
230
iommu_read_l2(struct amd_iommu * iommu,u8 address)231 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
232 {
233 u32 val;
234
235 pci_write_config_dword(iommu->dev, 0xf0, address);
236 pci_read_config_dword(iommu->dev, 0xf4, &val);
237 return val;
238 }
239
iommu_write_l2(struct amd_iommu * iommu,u8 address,u32 val)240 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
241 {
242 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
243 pci_write_config_dword(iommu->dev, 0xf4, val);
244 }
245
246 /****************************************************************************
247 *
248 * AMD IOMMU MMIO register space handling functions
249 *
250 * These functions are used to program the IOMMU device registers in
251 * MMIO space required for that driver.
252 *
253 ****************************************************************************/
254
255 /*
256 * This function set the exclusion range in the IOMMU. DMA accesses to the
257 * exclusion range are passed through untranslated
258 */
iommu_set_exclusion_range(struct amd_iommu * iommu)259 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
260 {
261 u64 start = iommu->exclusion_start & PAGE_MASK;
262 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
263 u64 entry;
264
265 if (!iommu->exclusion_start)
266 return;
267
268 entry = start | MMIO_EXCL_ENABLE_MASK;
269 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
270 &entry, sizeof(entry));
271
272 entry = limit;
273 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
274 &entry, sizeof(entry));
275 }
276
277 /* Programs the physical address of the device table into the IOMMU hardware */
iommu_set_device_table(struct amd_iommu * iommu)278 static void iommu_set_device_table(struct amd_iommu *iommu)
279 {
280 u64 entry;
281
282 BUG_ON(iommu->mmio_base == NULL);
283
284 entry = virt_to_phys(amd_iommu_dev_table);
285 entry |= (dev_table_size >> 12) - 1;
286 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
287 &entry, sizeof(entry));
288 }
289
290 /* Generic functions to enable/disable certain features of the IOMMU. */
iommu_feature_enable(struct amd_iommu * iommu,u8 bit)291 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
292 {
293 u32 ctrl;
294
295 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
296 ctrl |= (1 << bit);
297 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
298 }
299
iommu_feature_disable(struct amd_iommu * iommu,u8 bit)300 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
301 {
302 u32 ctrl;
303
304 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
305 ctrl &= ~(1 << bit);
306 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
307 }
308
iommu_set_inv_tlb_timeout(struct amd_iommu * iommu,int timeout)309 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
310 {
311 u32 ctrl;
312
313 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
314 ctrl &= ~CTRL_INV_TO_MASK;
315 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
316 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
317 }
318
319 /* Function to enable the hardware */
iommu_enable(struct amd_iommu * iommu)320 static void iommu_enable(struct amd_iommu *iommu)
321 {
322 static const char * const feat_str[] = {
323 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
324 "IA", "GA", "HE", "PC", NULL
325 };
326 int i;
327
328 printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx",
329 dev_name(&iommu->dev->dev), iommu->cap_ptr);
330
331 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
332 printk(KERN_CONT " extended features: ");
333 for (i = 0; feat_str[i]; ++i)
334 if (iommu_feature(iommu, (1ULL << i)))
335 printk(KERN_CONT " %s", feat_str[i]);
336 }
337 printk(KERN_CONT "\n");
338
339 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
340 }
341
iommu_disable(struct amd_iommu * iommu)342 static void iommu_disable(struct amd_iommu *iommu)
343 {
344 /* Disable command buffer */
345 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
346
347 /* Disable event logging and event interrupts */
348 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
349 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
350
351 /* Disable IOMMU hardware itself */
352 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
353 }
354
355 /*
356 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
357 * the system has one.
358 */
iommu_map_mmio_space(u64 address)359 static u8 * __init iommu_map_mmio_space(u64 address)
360 {
361 u8 *ret;
362
363 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
364 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
365 address);
366 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
367 return NULL;
368 }
369
370 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
371 if (ret != NULL)
372 return ret;
373
374 release_mem_region(address, MMIO_REGION_LENGTH);
375
376 return NULL;
377 }
378
iommu_unmap_mmio_space(struct amd_iommu * iommu)379 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
380 {
381 if (iommu->mmio_base)
382 iounmap(iommu->mmio_base);
383 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
384 }
385
386 /****************************************************************************
387 *
388 * The functions below belong to the first pass of AMD IOMMU ACPI table
389 * parsing. In this pass we try to find out the highest device id this
390 * code has to handle. Upon this information the size of the shared data
391 * structures is determined later.
392 *
393 ****************************************************************************/
394
395 /*
396 * This function calculates the length of a given IVHD entry
397 */
ivhd_entry_length(u8 * ivhd)398 static inline int ivhd_entry_length(u8 *ivhd)
399 {
400 return 0x04 << (*ivhd >> 6);
401 }
402
403 /*
404 * This function reads the last device id the IOMMU has to handle from the PCI
405 * capability header for this IOMMU
406 */
find_last_devid_on_pci(int bus,int dev,int fn,int cap_ptr)407 static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
408 {
409 u32 cap;
410
411 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
412 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
413
414 return 0;
415 }
416
417 /*
418 * After reading the highest device id from the IOMMU PCI capability header
419 * this function looks if there is a higher device id defined in the ACPI table
420 */
find_last_devid_from_ivhd(struct ivhd_header * h)421 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
422 {
423 u8 *p = (void *)h, *end = (void *)h;
424 struct ivhd_entry *dev;
425
426 p += sizeof(*h);
427 end += h->length;
428
429 find_last_devid_on_pci(PCI_BUS(h->devid),
430 PCI_SLOT(h->devid),
431 PCI_FUNC(h->devid),
432 h->cap_ptr);
433
434 while (p < end) {
435 dev = (struct ivhd_entry *)p;
436 switch (dev->type) {
437 case IVHD_DEV_SELECT:
438 case IVHD_DEV_RANGE_END:
439 case IVHD_DEV_ALIAS:
440 case IVHD_DEV_EXT_SELECT:
441 /* all the above subfield types refer to device ids */
442 update_last_devid(dev->devid);
443 break;
444 default:
445 break;
446 }
447 p += ivhd_entry_length(p);
448 }
449
450 WARN_ON(p != end);
451
452 return 0;
453 }
454
455 /*
456 * Iterate over all IVHD entries in the ACPI table and find the highest device
457 * id which we need to handle. This is the first of three functions which parse
458 * the ACPI table. So we check the checksum here.
459 */
find_last_devid_acpi(struct acpi_table_header * table)460 static int __init find_last_devid_acpi(struct acpi_table_header *table)
461 {
462 int i;
463 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
464 struct ivhd_header *h;
465
466 /*
467 * Validate checksum here so we don't need to do it when
468 * we actually parse the table
469 */
470 for (i = 0; i < table->length; ++i)
471 checksum += p[i];
472 if (checksum != 0) {
473 /* ACPI table corrupt */
474 amd_iommu_init_err = -ENODEV;
475 return 0;
476 }
477
478 p += IVRS_HEADER_LENGTH;
479
480 end += table->length;
481 while (p < end) {
482 h = (struct ivhd_header *)p;
483 switch (h->type) {
484 case ACPI_IVHD_TYPE:
485 find_last_devid_from_ivhd(h);
486 break;
487 default:
488 break;
489 }
490 p += h->length;
491 }
492 WARN_ON(p != end);
493
494 return 0;
495 }
496
497 /****************************************************************************
498 *
499 * The following functions belong the the code path which parses the ACPI table
500 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
501 * data structures, initialize the device/alias/rlookup table and also
502 * basically initialize the hardware.
503 *
504 ****************************************************************************/
505
506 /*
507 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
508 * write commands to that buffer later and the IOMMU will execute them
509 * asynchronously
510 */
alloc_command_buffer(struct amd_iommu * iommu)511 static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
512 {
513 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
514 get_order(CMD_BUFFER_SIZE));
515
516 if (cmd_buf == NULL)
517 return NULL;
518
519 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
520
521 return cmd_buf;
522 }
523
524 /*
525 * This function resets the command buffer if the IOMMU stopped fetching
526 * commands from it.
527 */
amd_iommu_reset_cmd_buffer(struct amd_iommu * iommu)528 void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
529 {
530 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
531
532 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
533 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
534
535 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
536 }
537
538 /*
539 * This function writes the command buffer address to the hardware and
540 * enables it.
541 */
iommu_enable_command_buffer(struct amd_iommu * iommu)542 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
543 {
544 u64 entry;
545
546 BUG_ON(iommu->cmd_buf == NULL);
547
548 entry = (u64)virt_to_phys(iommu->cmd_buf);
549 entry |= MMIO_CMD_SIZE_512;
550
551 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
552 &entry, sizeof(entry));
553
554 amd_iommu_reset_cmd_buffer(iommu);
555 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
556 }
557
free_command_buffer(struct amd_iommu * iommu)558 static void __init free_command_buffer(struct amd_iommu *iommu)
559 {
560 free_pages((unsigned long)iommu->cmd_buf,
561 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
562 }
563
564 /* allocates the memory where the IOMMU will log its events to */
alloc_event_buffer(struct amd_iommu * iommu)565 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
566 {
567 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
568 get_order(EVT_BUFFER_SIZE));
569
570 if (iommu->evt_buf == NULL)
571 return NULL;
572
573 iommu->evt_buf_size = EVT_BUFFER_SIZE;
574
575 return iommu->evt_buf;
576 }
577
iommu_enable_event_buffer(struct amd_iommu * iommu)578 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
579 {
580 u64 entry;
581
582 BUG_ON(iommu->evt_buf == NULL);
583
584 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
585
586 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
587 &entry, sizeof(entry));
588
589 /* set head and tail to zero manually */
590 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
591 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
592
593 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
594 }
595
free_event_buffer(struct amd_iommu * iommu)596 static void __init free_event_buffer(struct amd_iommu *iommu)
597 {
598 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
599 }
600
601 /* allocates the memory where the IOMMU will log its events to */
alloc_ppr_log(struct amd_iommu * iommu)602 static u8 * __init alloc_ppr_log(struct amd_iommu *iommu)
603 {
604 iommu->ppr_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
605 get_order(PPR_LOG_SIZE));
606
607 if (iommu->ppr_log == NULL)
608 return NULL;
609
610 return iommu->ppr_log;
611 }
612
iommu_enable_ppr_log(struct amd_iommu * iommu)613 static void iommu_enable_ppr_log(struct amd_iommu *iommu)
614 {
615 u64 entry;
616
617 if (iommu->ppr_log == NULL)
618 return;
619
620 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
621
622 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
623 &entry, sizeof(entry));
624
625 /* set head and tail to zero manually */
626 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
627 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
628
629 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
630 iommu_feature_enable(iommu, CONTROL_PPR_EN);
631 }
632
free_ppr_log(struct amd_iommu * iommu)633 static void __init free_ppr_log(struct amd_iommu *iommu)
634 {
635 if (iommu->ppr_log == NULL)
636 return;
637
638 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
639 }
640
iommu_enable_gt(struct amd_iommu * iommu)641 static void iommu_enable_gt(struct amd_iommu *iommu)
642 {
643 if (!iommu_feature(iommu, FEATURE_GT))
644 return;
645
646 iommu_feature_enable(iommu, CONTROL_GT_EN);
647 }
648
649 /* sets a specific bit in the device table entry. */
set_dev_entry_bit(u16 devid,u8 bit)650 static void set_dev_entry_bit(u16 devid, u8 bit)
651 {
652 int i = (bit >> 6) & 0x03;
653 int _bit = bit & 0x3f;
654
655 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
656 }
657
get_dev_entry_bit(u16 devid,u8 bit)658 static int get_dev_entry_bit(u16 devid, u8 bit)
659 {
660 int i = (bit >> 6) & 0x03;
661 int _bit = bit & 0x3f;
662
663 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
664 }
665
666
amd_iommu_apply_erratum_63(u16 devid)667 void amd_iommu_apply_erratum_63(u16 devid)
668 {
669 int sysmgt;
670
671 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
672 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
673
674 if (sysmgt == 0x01)
675 set_dev_entry_bit(devid, DEV_ENTRY_IW);
676 }
677
678 /* Writes the specific IOMMU for a device into the rlookup table */
set_iommu_for_device(struct amd_iommu * iommu,u16 devid)679 static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
680 {
681 amd_iommu_rlookup_table[devid] = iommu;
682 }
683
684 /*
685 * This function takes the device specific flags read from the ACPI
686 * table and sets up the device table entry with that information
687 */
set_dev_entry_from_acpi(struct amd_iommu * iommu,u16 devid,u32 flags,u32 ext_flags)688 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
689 u16 devid, u32 flags, u32 ext_flags)
690 {
691 if (flags & ACPI_DEVFLAG_INITPASS)
692 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
693 if (flags & ACPI_DEVFLAG_EXTINT)
694 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
695 if (flags & ACPI_DEVFLAG_NMI)
696 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
697 if (flags & ACPI_DEVFLAG_SYSMGT1)
698 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
699 if (flags & ACPI_DEVFLAG_SYSMGT2)
700 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
701 if (flags & ACPI_DEVFLAG_LINT0)
702 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
703 if (flags & ACPI_DEVFLAG_LINT1)
704 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
705
706 amd_iommu_apply_erratum_63(devid);
707
708 set_iommu_for_device(iommu, devid);
709 }
710
711 /*
712 * Reads the device exclusion range from ACPI and initialize IOMMU with
713 * it
714 */
set_device_exclusion_range(u16 devid,struct ivmd_header * m)715 static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
716 {
717 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
718
719 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
720 return;
721
722 if (iommu) {
723 /*
724 * We only can configure exclusion ranges per IOMMU, not
725 * per device. But we can enable the exclusion range per
726 * device. This is done here
727 */
728 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
729 iommu->exclusion_start = m->range_start;
730 iommu->exclusion_length = m->range_length;
731 }
732 }
733
734 /*
735 * This function reads some important data from the IOMMU PCI space and
736 * initializes the driver data structure with it. It reads the hardware
737 * capabilities and the first/last device entries
738 */
init_iommu_from_pci(struct amd_iommu * iommu)739 static void __init init_iommu_from_pci(struct amd_iommu *iommu)
740 {
741 int cap_ptr = iommu->cap_ptr;
742 u32 range, misc, low, high;
743 int i, j;
744
745 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
746 &iommu->cap);
747 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
748 &range);
749 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
750 &misc);
751
752 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
753 MMIO_GET_FD(range));
754 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
755 MMIO_GET_LD(range));
756 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
757
758 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
759 amd_iommu_iotlb_sup = false;
760
761 /* read extended feature bits */
762 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
763 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
764
765 iommu->features = ((u64)high << 32) | low;
766
767 if (iommu_feature(iommu, FEATURE_GT)) {
768 int glxval;
769 u32 pasids;
770 u64 shift;
771
772 shift = iommu->features & FEATURE_PASID_MASK;
773 shift >>= FEATURE_PASID_SHIFT;
774 pasids = (1 << shift);
775
776 amd_iommu_max_pasids = min(amd_iommu_max_pasids, pasids);
777
778 glxval = iommu->features & FEATURE_GLXVAL_MASK;
779 glxval >>= FEATURE_GLXVAL_SHIFT;
780
781 if (amd_iommu_max_glx_val == -1)
782 amd_iommu_max_glx_val = glxval;
783 else
784 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
785 }
786
787 if (iommu_feature(iommu, FEATURE_GT) &&
788 iommu_feature(iommu, FEATURE_PPR)) {
789 iommu->is_iommu_v2 = true;
790 amd_iommu_v2_present = true;
791 }
792
793 if (!is_rd890_iommu(iommu->dev))
794 return;
795
796 /*
797 * Some rd890 systems may not be fully reconfigured by the BIOS, so
798 * it's necessary for us to store this information so it can be
799 * reprogrammed on resume
800 */
801
802 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
803 &iommu->stored_addr_lo);
804 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
805 &iommu->stored_addr_hi);
806
807 /* Low bit locks writes to configuration space */
808 iommu->stored_addr_lo &= ~1;
809
810 for (i = 0; i < 6; i++)
811 for (j = 0; j < 0x12; j++)
812 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
813
814 for (i = 0; i < 0x83; i++)
815 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
816 }
817
818 /*
819 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
820 * initializes the hardware and our data structures with it.
821 */
init_iommu_from_acpi(struct amd_iommu * iommu,struct ivhd_header * h)822 static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
823 struct ivhd_header *h)
824 {
825 u8 *p = (u8 *)h;
826 u8 *end = p, flags = 0;
827 u16 devid = 0, devid_start = 0, devid_to = 0;
828 u32 dev_i, ext_flags = 0;
829 bool alias = false;
830 struct ivhd_entry *e;
831
832 /*
833 * First save the recommended feature enable bits from ACPI
834 */
835 iommu->acpi_flags = h->flags;
836
837 /*
838 * Done. Now parse the device entries
839 */
840 p += sizeof(struct ivhd_header);
841 end += h->length;
842
843
844 while (p < end) {
845 e = (struct ivhd_entry *)p;
846 switch (e->type) {
847 case IVHD_DEV_ALL:
848
849 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
850 " last device %02x:%02x.%x flags: %02x\n",
851 PCI_BUS(iommu->first_device),
852 PCI_SLOT(iommu->first_device),
853 PCI_FUNC(iommu->first_device),
854 PCI_BUS(iommu->last_device),
855 PCI_SLOT(iommu->last_device),
856 PCI_FUNC(iommu->last_device),
857 e->flags);
858
859 for (dev_i = iommu->first_device;
860 dev_i <= iommu->last_device; ++dev_i)
861 set_dev_entry_from_acpi(iommu, dev_i,
862 e->flags, 0);
863 break;
864 case IVHD_DEV_SELECT:
865
866 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
867 "flags: %02x\n",
868 PCI_BUS(e->devid),
869 PCI_SLOT(e->devid),
870 PCI_FUNC(e->devid),
871 e->flags);
872
873 devid = e->devid;
874 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
875 break;
876 case IVHD_DEV_SELECT_RANGE_START:
877
878 DUMP_printk(" DEV_SELECT_RANGE_START\t "
879 "devid: %02x:%02x.%x flags: %02x\n",
880 PCI_BUS(e->devid),
881 PCI_SLOT(e->devid),
882 PCI_FUNC(e->devid),
883 e->flags);
884
885 devid_start = e->devid;
886 flags = e->flags;
887 ext_flags = 0;
888 alias = false;
889 break;
890 case IVHD_DEV_ALIAS:
891
892 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
893 "flags: %02x devid_to: %02x:%02x.%x\n",
894 PCI_BUS(e->devid),
895 PCI_SLOT(e->devid),
896 PCI_FUNC(e->devid),
897 e->flags,
898 PCI_BUS(e->ext >> 8),
899 PCI_SLOT(e->ext >> 8),
900 PCI_FUNC(e->ext >> 8));
901
902 devid = e->devid;
903 devid_to = e->ext >> 8;
904 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
905 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
906 amd_iommu_alias_table[devid] = devid_to;
907 break;
908 case IVHD_DEV_ALIAS_RANGE:
909
910 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
911 "devid: %02x:%02x.%x flags: %02x "
912 "devid_to: %02x:%02x.%x\n",
913 PCI_BUS(e->devid),
914 PCI_SLOT(e->devid),
915 PCI_FUNC(e->devid),
916 e->flags,
917 PCI_BUS(e->ext >> 8),
918 PCI_SLOT(e->ext >> 8),
919 PCI_FUNC(e->ext >> 8));
920
921 devid_start = e->devid;
922 flags = e->flags;
923 devid_to = e->ext >> 8;
924 ext_flags = 0;
925 alias = true;
926 break;
927 case IVHD_DEV_EXT_SELECT:
928
929 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
930 "flags: %02x ext: %08x\n",
931 PCI_BUS(e->devid),
932 PCI_SLOT(e->devid),
933 PCI_FUNC(e->devid),
934 e->flags, e->ext);
935
936 devid = e->devid;
937 set_dev_entry_from_acpi(iommu, devid, e->flags,
938 e->ext);
939 break;
940 case IVHD_DEV_EXT_SELECT_RANGE:
941
942 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
943 "%02x:%02x.%x flags: %02x ext: %08x\n",
944 PCI_BUS(e->devid),
945 PCI_SLOT(e->devid),
946 PCI_FUNC(e->devid),
947 e->flags, e->ext);
948
949 devid_start = e->devid;
950 flags = e->flags;
951 ext_flags = e->ext;
952 alias = false;
953 break;
954 case IVHD_DEV_RANGE_END:
955
956 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
957 PCI_BUS(e->devid),
958 PCI_SLOT(e->devid),
959 PCI_FUNC(e->devid));
960
961 devid = e->devid;
962 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
963 if (alias) {
964 amd_iommu_alias_table[dev_i] = devid_to;
965 set_dev_entry_from_acpi(iommu,
966 devid_to, flags, ext_flags);
967 }
968 set_dev_entry_from_acpi(iommu, dev_i,
969 flags, ext_flags);
970 }
971 break;
972 default:
973 break;
974 }
975
976 p += ivhd_entry_length(p);
977 }
978 }
979
980 /* Initializes the device->iommu mapping for the driver */
init_iommu_devices(struct amd_iommu * iommu)981 static int __init init_iommu_devices(struct amd_iommu *iommu)
982 {
983 u32 i;
984
985 for (i = iommu->first_device; i <= iommu->last_device; ++i)
986 set_iommu_for_device(iommu, i);
987
988 return 0;
989 }
990
free_iommu_one(struct amd_iommu * iommu)991 static void __init free_iommu_one(struct amd_iommu *iommu)
992 {
993 free_command_buffer(iommu);
994 free_event_buffer(iommu);
995 free_ppr_log(iommu);
996 iommu_unmap_mmio_space(iommu);
997 }
998
free_iommu_all(void)999 static void __init free_iommu_all(void)
1000 {
1001 struct amd_iommu *iommu, *next;
1002
1003 for_each_iommu_safe(iommu, next) {
1004 list_del(&iommu->list);
1005 free_iommu_one(iommu);
1006 kfree(iommu);
1007 }
1008 }
1009
1010 /*
1011 * This function clues the initialization function for one IOMMU
1012 * together and also allocates the command buffer and programs the
1013 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1014 */
init_iommu_one(struct amd_iommu * iommu,struct ivhd_header * h)1015 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1016 {
1017 spin_lock_init(&iommu->lock);
1018
1019 /* Add IOMMU to internal data structures */
1020 list_add_tail(&iommu->list, &amd_iommu_list);
1021 iommu->index = amd_iommus_present++;
1022
1023 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1024 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1025 return -ENOSYS;
1026 }
1027
1028 /* Index is fine - add IOMMU to the array */
1029 amd_iommus[iommu->index] = iommu;
1030
1031 /*
1032 * Copy data from ACPI table entry to the iommu struct
1033 */
1034 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
1035 if (!iommu->dev)
1036 return 1;
1037
1038 iommu->cap_ptr = h->cap_ptr;
1039 iommu->pci_seg = h->pci_seg;
1040 iommu->mmio_phys = h->mmio_phys;
1041 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
1042 if (!iommu->mmio_base)
1043 return -ENOMEM;
1044
1045 iommu->cmd_buf = alloc_command_buffer(iommu);
1046 if (!iommu->cmd_buf)
1047 return -ENOMEM;
1048
1049 iommu->evt_buf = alloc_event_buffer(iommu);
1050 if (!iommu->evt_buf)
1051 return -ENOMEM;
1052
1053 iommu->int_enabled = false;
1054
1055 init_iommu_from_pci(iommu);
1056 init_iommu_from_acpi(iommu, h);
1057 init_iommu_devices(iommu);
1058
1059 if (iommu_feature(iommu, FEATURE_PPR)) {
1060 iommu->ppr_log = alloc_ppr_log(iommu);
1061 if (!iommu->ppr_log)
1062 return -ENOMEM;
1063 }
1064
1065 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1066 amd_iommu_np_cache = true;
1067
1068 return pci_enable_device(iommu->dev);
1069 }
1070
1071 /*
1072 * Iterates over all IOMMU entries in the ACPI table, allocates the
1073 * IOMMU structure and initializes it with init_iommu_one()
1074 */
init_iommu_all(struct acpi_table_header * table)1075 static int __init init_iommu_all(struct acpi_table_header *table)
1076 {
1077 u8 *p = (u8 *)table, *end = (u8 *)table;
1078 struct ivhd_header *h;
1079 struct amd_iommu *iommu;
1080 int ret;
1081
1082 end += table->length;
1083 p += IVRS_HEADER_LENGTH;
1084
1085 while (p < end) {
1086 h = (struct ivhd_header *)p;
1087 switch (*p) {
1088 case ACPI_IVHD_TYPE:
1089
1090 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1091 "seg: %d flags: %01x info %04x\n",
1092 PCI_BUS(h->devid), PCI_SLOT(h->devid),
1093 PCI_FUNC(h->devid), h->cap_ptr,
1094 h->pci_seg, h->flags, h->info);
1095 DUMP_printk(" mmio-addr: %016llx\n",
1096 h->mmio_phys);
1097
1098 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1099 if (iommu == NULL) {
1100 amd_iommu_init_err = -ENOMEM;
1101 return 0;
1102 }
1103
1104 ret = init_iommu_one(iommu, h);
1105 if (ret) {
1106 amd_iommu_init_err = ret;
1107 return 0;
1108 }
1109 break;
1110 default:
1111 break;
1112 }
1113 p += h->length;
1114
1115 }
1116 WARN_ON(p != end);
1117
1118 return 0;
1119 }
1120
1121 /****************************************************************************
1122 *
1123 * The following functions initialize the MSI interrupts for all IOMMUs
1124 * in the system. Its a bit challenging because there could be multiple
1125 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1126 * pci_dev.
1127 *
1128 ****************************************************************************/
1129
iommu_setup_msi(struct amd_iommu * iommu)1130 static int iommu_setup_msi(struct amd_iommu *iommu)
1131 {
1132 int r;
1133
1134 if (pci_enable_msi(iommu->dev))
1135 return 1;
1136
1137 r = request_threaded_irq(iommu->dev->irq,
1138 amd_iommu_int_handler,
1139 amd_iommu_int_thread,
1140 0, "AMD-Vi",
1141 iommu->dev);
1142
1143 if (r) {
1144 pci_disable_msi(iommu->dev);
1145 return 1;
1146 }
1147
1148 iommu->int_enabled = true;
1149 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1150
1151 if (iommu->ppr_log != NULL)
1152 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1153
1154 return 0;
1155 }
1156
iommu_init_msi(struct amd_iommu * iommu)1157 static int iommu_init_msi(struct amd_iommu *iommu)
1158 {
1159 if (iommu->int_enabled)
1160 return 0;
1161
1162 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
1163 return iommu_setup_msi(iommu);
1164
1165 return 1;
1166 }
1167
1168 /****************************************************************************
1169 *
1170 * The next functions belong to the third pass of parsing the ACPI
1171 * table. In this last pass the memory mapping requirements are
1172 * gathered (like exclusion and unity mapping reanges).
1173 *
1174 ****************************************************************************/
1175
free_unity_maps(void)1176 static void __init free_unity_maps(void)
1177 {
1178 struct unity_map_entry *entry, *next;
1179
1180 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1181 list_del(&entry->list);
1182 kfree(entry);
1183 }
1184 }
1185
1186 /* called when we find an exclusion range definition in ACPI */
init_exclusion_range(struct ivmd_header * m)1187 static int __init init_exclusion_range(struct ivmd_header *m)
1188 {
1189 int i;
1190
1191 switch (m->type) {
1192 case ACPI_IVMD_TYPE:
1193 set_device_exclusion_range(m->devid, m);
1194 break;
1195 case ACPI_IVMD_TYPE_ALL:
1196 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1197 set_device_exclusion_range(i, m);
1198 break;
1199 case ACPI_IVMD_TYPE_RANGE:
1200 for (i = m->devid; i <= m->aux; ++i)
1201 set_device_exclusion_range(i, m);
1202 break;
1203 default:
1204 break;
1205 }
1206
1207 return 0;
1208 }
1209
1210 /* called for unity map ACPI definition */
init_unity_map_range(struct ivmd_header * m)1211 static int __init init_unity_map_range(struct ivmd_header *m)
1212 {
1213 struct unity_map_entry *e = 0;
1214 char *s;
1215
1216 e = kzalloc(sizeof(*e), GFP_KERNEL);
1217 if (e == NULL)
1218 return -ENOMEM;
1219
1220 switch (m->type) {
1221 default:
1222 kfree(e);
1223 return 0;
1224 case ACPI_IVMD_TYPE:
1225 s = "IVMD_TYPEi\t\t\t";
1226 e->devid_start = e->devid_end = m->devid;
1227 break;
1228 case ACPI_IVMD_TYPE_ALL:
1229 s = "IVMD_TYPE_ALL\t\t";
1230 e->devid_start = 0;
1231 e->devid_end = amd_iommu_last_bdf;
1232 break;
1233 case ACPI_IVMD_TYPE_RANGE:
1234 s = "IVMD_TYPE_RANGE\t\t";
1235 e->devid_start = m->devid;
1236 e->devid_end = m->aux;
1237 break;
1238 }
1239 e->address_start = PAGE_ALIGN(m->range_start);
1240 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1241 e->prot = m->flags >> 1;
1242
1243 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1244 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1245 PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1246 PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1247 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1248 e->address_start, e->address_end, m->flags);
1249
1250 list_add_tail(&e->list, &amd_iommu_unity_map);
1251
1252 return 0;
1253 }
1254
1255 /* iterates over all memory definitions we find in the ACPI table */
init_memory_definitions(struct acpi_table_header * table)1256 static int __init init_memory_definitions(struct acpi_table_header *table)
1257 {
1258 u8 *p = (u8 *)table, *end = (u8 *)table;
1259 struct ivmd_header *m;
1260
1261 end += table->length;
1262 p += IVRS_HEADER_LENGTH;
1263
1264 while (p < end) {
1265 m = (struct ivmd_header *)p;
1266 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1267 init_exclusion_range(m);
1268 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1269 init_unity_map_range(m);
1270
1271 p += m->length;
1272 }
1273
1274 return 0;
1275 }
1276
1277 /*
1278 * Init the device table to not allow DMA access for devices and
1279 * suppress all page faults
1280 */
init_device_table(void)1281 static void init_device_table(void)
1282 {
1283 u32 devid;
1284
1285 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1286 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1287 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1288 }
1289 }
1290
iommu_init_flags(struct amd_iommu * iommu)1291 static void iommu_init_flags(struct amd_iommu *iommu)
1292 {
1293 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1294 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1295 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1296
1297 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1298 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1299 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1300
1301 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1302 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1303 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1304
1305 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1306 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1307 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1308
1309 /*
1310 * make IOMMU memory accesses cache coherent
1311 */
1312 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1313
1314 /* Set IOTLB invalidation timeout to 1s */
1315 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1316 }
1317
iommu_apply_resume_quirks(struct amd_iommu * iommu)1318 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1319 {
1320 int i, j;
1321 u32 ioc_feature_control;
1322 struct pci_dev *pdev = NULL;
1323
1324 /* RD890 BIOSes may not have completely reconfigured the iommu */
1325 if (!is_rd890_iommu(iommu->dev))
1326 return;
1327
1328 /*
1329 * First, we need to ensure that the iommu is enabled. This is
1330 * controlled by a register in the northbridge
1331 */
1332 pdev = pci_get_bus_and_slot(iommu->dev->bus->number, PCI_DEVFN(0, 0));
1333
1334 if (!pdev)
1335 return;
1336
1337 /* Select Northbridge indirect register 0x75 and enable writing */
1338 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1339 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1340
1341 /* Enable the iommu */
1342 if (!(ioc_feature_control & 0x1))
1343 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1344
1345 pci_dev_put(pdev);
1346
1347 /* Restore the iommu BAR */
1348 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1349 iommu->stored_addr_lo);
1350 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1351 iommu->stored_addr_hi);
1352
1353 /* Restore the l1 indirect regs for each of the 6 l1s */
1354 for (i = 0; i < 6; i++)
1355 for (j = 0; j < 0x12; j++)
1356 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1357
1358 /* Restore the l2 indirect regs */
1359 for (i = 0; i < 0x83; i++)
1360 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1361
1362 /* Lock PCI setup registers */
1363 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1364 iommu->stored_addr_lo | 1);
1365 }
1366
1367 /*
1368 * This function finally enables all IOMMUs found in the system after
1369 * they have been initialized
1370 */
enable_iommus(void)1371 static void enable_iommus(void)
1372 {
1373 struct amd_iommu *iommu;
1374
1375 for_each_iommu(iommu) {
1376 iommu_disable(iommu);
1377 iommu_init_flags(iommu);
1378 iommu_set_device_table(iommu);
1379 iommu_enable_command_buffer(iommu);
1380 iommu_enable_event_buffer(iommu);
1381 iommu_enable_ppr_log(iommu);
1382 iommu_enable_gt(iommu);
1383 iommu_set_exclusion_range(iommu);
1384 iommu_init_msi(iommu);
1385 iommu_enable(iommu);
1386 iommu_flush_all_caches(iommu);
1387 }
1388 }
1389
disable_iommus(void)1390 static void disable_iommus(void)
1391 {
1392 struct amd_iommu *iommu;
1393
1394 for_each_iommu(iommu)
1395 iommu_disable(iommu);
1396 }
1397
1398 /*
1399 * Suspend/Resume support
1400 * disable suspend until real resume implemented
1401 */
1402
amd_iommu_resume(void)1403 static void amd_iommu_resume(void)
1404 {
1405 struct amd_iommu *iommu;
1406
1407 for_each_iommu(iommu)
1408 iommu_apply_resume_quirks(iommu);
1409
1410 /* re-load the hardware */
1411 enable_iommus();
1412 }
1413
amd_iommu_suspend(void)1414 static int amd_iommu_suspend(void)
1415 {
1416 /* disable IOMMUs to go out of the way for BIOS */
1417 disable_iommus();
1418
1419 return 0;
1420 }
1421
1422 static struct syscore_ops amd_iommu_syscore_ops = {
1423 .suspend = amd_iommu_suspend,
1424 .resume = amd_iommu_resume,
1425 };
1426
1427 /*
1428 * This is the core init function for AMD IOMMU hardware in the system.
1429 * This function is called from the generic x86 DMA layer initialization
1430 * code.
1431 *
1432 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1433 * three times:
1434 *
1435 * 1 pass) Find the highest PCI device id the driver has to handle.
1436 * Upon this information the size of the data structures is
1437 * determined that needs to be allocated.
1438 *
1439 * 2 pass) Initialize the data structures just allocated with the
1440 * information in the ACPI table about available AMD IOMMUs
1441 * in the system. It also maps the PCI devices in the
1442 * system to specific IOMMUs
1443 *
1444 * 3 pass) After the basic data structures are allocated and
1445 * initialized we update them with information about memory
1446 * remapping requirements parsed out of the ACPI table in
1447 * this last pass.
1448 *
1449 * After that the hardware is initialized and ready to go. In the last
1450 * step we do some Linux specific things like registering the driver in
1451 * the dma_ops interface and initializing the suspend/resume support
1452 * functions. Finally it prints some information about AMD IOMMUs and
1453 * the driver state and enables the hardware.
1454 */
amd_iommu_init(void)1455 static int __init amd_iommu_init(void)
1456 {
1457 int i, ret = 0;
1458
1459 /*
1460 * First parse ACPI tables to find the largest Bus/Dev/Func
1461 * we need to handle. Upon this information the shared data
1462 * structures for the IOMMUs in the system will be allocated
1463 */
1464 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1465 return -ENODEV;
1466
1467 ret = amd_iommu_init_err;
1468 if (ret)
1469 goto out;
1470
1471 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1472 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1473 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1474
1475 ret = -ENOMEM;
1476
1477 /* Device table - directly used by all IOMMUs */
1478 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1479 get_order(dev_table_size));
1480 if (amd_iommu_dev_table == NULL)
1481 goto out;
1482
1483 /*
1484 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1485 * IOMMU see for that device
1486 */
1487 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1488 get_order(alias_table_size));
1489 if (amd_iommu_alias_table == NULL)
1490 goto free;
1491
1492 /* IOMMU rlookup table - find the IOMMU for a specific device */
1493 amd_iommu_rlookup_table = (void *)__get_free_pages(
1494 GFP_KERNEL | __GFP_ZERO,
1495 get_order(rlookup_table_size));
1496 if (amd_iommu_rlookup_table == NULL)
1497 goto free;
1498
1499 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1500 GFP_KERNEL | __GFP_ZERO,
1501 get_order(MAX_DOMAIN_ID/8));
1502 if (amd_iommu_pd_alloc_bitmap == NULL)
1503 goto free;
1504
1505 /* init the device table */
1506 init_device_table();
1507
1508 /*
1509 * let all alias entries point to itself
1510 */
1511 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1512 amd_iommu_alias_table[i] = i;
1513
1514 /*
1515 * never allocate domain 0 because its used as the non-allocated and
1516 * error value placeholder
1517 */
1518 amd_iommu_pd_alloc_bitmap[0] = 1;
1519
1520 spin_lock_init(&amd_iommu_pd_lock);
1521
1522 /*
1523 * now the data structures are allocated and basically initialized
1524 * start the real acpi table scan
1525 */
1526 ret = -ENODEV;
1527 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1528 goto free;
1529
1530 if (amd_iommu_init_err) {
1531 ret = amd_iommu_init_err;
1532 goto free;
1533 }
1534
1535 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1536 goto free;
1537
1538 if (amd_iommu_init_err) {
1539 ret = amd_iommu_init_err;
1540 goto free;
1541 }
1542
1543 ret = amd_iommu_init_devices();
1544 if (ret)
1545 goto free;
1546
1547 enable_iommus();
1548
1549 if (iommu_pass_through)
1550 ret = amd_iommu_init_passthrough();
1551 else
1552 ret = amd_iommu_init_dma_ops();
1553
1554 if (ret)
1555 goto free_disable;
1556
1557 amd_iommu_init_api();
1558
1559 amd_iommu_init_notifier();
1560
1561 register_syscore_ops(&amd_iommu_syscore_ops);
1562
1563 if (iommu_pass_through)
1564 goto out;
1565
1566 if (amd_iommu_unmap_flush)
1567 printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
1568 else
1569 printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n");
1570
1571 x86_platform.iommu_shutdown = disable_iommus;
1572 out:
1573 return ret;
1574
1575 free_disable:
1576 disable_iommus();
1577
1578 free:
1579 amd_iommu_uninit_devices();
1580
1581 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1582 get_order(MAX_DOMAIN_ID/8));
1583
1584 free_pages((unsigned long)amd_iommu_rlookup_table,
1585 get_order(rlookup_table_size));
1586
1587 free_pages((unsigned long)amd_iommu_alias_table,
1588 get_order(alias_table_size));
1589
1590 free_pages((unsigned long)amd_iommu_dev_table,
1591 get_order(dev_table_size));
1592
1593 free_iommu_all();
1594
1595 free_unity_maps();
1596
1597 #ifdef CONFIG_GART_IOMMU
1598 /*
1599 * We failed to initialize the AMD IOMMU - try fallback to GART
1600 * if possible.
1601 */
1602 gart_iommu_init();
1603
1604 #endif
1605
1606 goto out;
1607 }
1608
1609 /****************************************************************************
1610 *
1611 * Early detect code. This code runs at IOMMU detection time in the DMA
1612 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1613 * IOMMUs
1614 *
1615 ****************************************************************************/
early_amd_iommu_detect(struct acpi_table_header * table)1616 static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1617 {
1618 return 0;
1619 }
1620
amd_iommu_detect(void)1621 int __init amd_iommu_detect(void)
1622 {
1623 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
1624 return -ENODEV;
1625
1626 if (amd_iommu_disabled)
1627 return -ENODEV;
1628
1629 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1630 iommu_detected = 1;
1631 amd_iommu_detected = 1;
1632 x86_init.iommu.iommu_init = amd_iommu_init;
1633
1634 /* Make sure ACS will be enabled */
1635 pci_request_acs();
1636 return 1;
1637 }
1638 return -ENODEV;
1639 }
1640
1641 /****************************************************************************
1642 *
1643 * Parsing functions for the AMD IOMMU specific kernel command line
1644 * options.
1645 *
1646 ****************************************************************************/
1647
parse_amd_iommu_dump(char * str)1648 static int __init parse_amd_iommu_dump(char *str)
1649 {
1650 amd_iommu_dump = true;
1651
1652 return 1;
1653 }
1654
parse_amd_iommu_options(char * str)1655 static int __init parse_amd_iommu_options(char *str)
1656 {
1657 for (; *str; ++str) {
1658 if (strncmp(str, "fullflush", 9) == 0)
1659 amd_iommu_unmap_flush = true;
1660 if (strncmp(str, "off", 3) == 0)
1661 amd_iommu_disabled = true;
1662 if (strncmp(str, "force_isolation", 15) == 0)
1663 amd_iommu_force_isolation = true;
1664 }
1665
1666 return 1;
1667 }
1668
1669 __setup("amd_iommu_dump", parse_amd_iommu_dump);
1670 __setup("amd_iommu=", parse_amd_iommu_options);
1671
1672 IOMMU_INIT_FINISH(amd_iommu_detect,
1673 gart_iommu_hole_init,
1674 0,
1675 0);
1676
amd_iommu_v2_supported(void)1677 bool amd_iommu_v2_supported(void)
1678 {
1679 return amd_iommu_v2_present;
1680 }
1681 EXPORT_SYMBOL(amd_iommu_v2_supported);
1682