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