Loading...
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 <asm/pci-direct.h>
29#include <asm/iommu.h>
30#include <asm/gart.h>
31#include <asm/x86_init.h>
32#include <asm/iommu_table.h>
33
34#include "amd_iommu_proto.h"
35#include "amd_iommu_types.h"
36
37/*
38 * definitions for the ACPI scanning code
39 */
40#define IVRS_HEADER_LENGTH 48
41
42#define ACPI_IVHD_TYPE 0x10
43#define ACPI_IVMD_TYPE_ALL 0x20
44#define ACPI_IVMD_TYPE 0x21
45#define ACPI_IVMD_TYPE_RANGE 0x22
46
47#define IVHD_DEV_ALL 0x01
48#define IVHD_DEV_SELECT 0x02
49#define IVHD_DEV_SELECT_RANGE_START 0x03
50#define IVHD_DEV_RANGE_END 0x04
51#define IVHD_DEV_ALIAS 0x42
52#define IVHD_DEV_ALIAS_RANGE 0x43
53#define IVHD_DEV_EXT_SELECT 0x46
54#define IVHD_DEV_EXT_SELECT_RANGE 0x47
55
56#define IVHD_FLAG_HT_TUN_EN_MASK 0x01
57#define IVHD_FLAG_PASSPW_EN_MASK 0x02
58#define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
59#define IVHD_FLAG_ISOC_EN_MASK 0x08
60
61#define IVMD_FLAG_EXCL_RANGE 0x08
62#define IVMD_FLAG_UNITY_MAP 0x01
63
64#define ACPI_DEVFLAG_INITPASS 0x01
65#define ACPI_DEVFLAG_EXTINT 0x02
66#define ACPI_DEVFLAG_NMI 0x04
67#define ACPI_DEVFLAG_SYSMGT1 0x10
68#define ACPI_DEVFLAG_SYSMGT2 0x20
69#define ACPI_DEVFLAG_LINT0 0x40
70#define ACPI_DEVFLAG_LINT1 0x80
71#define ACPI_DEVFLAG_ATSDIS 0x10000000
72
73/*
74 * ACPI table definitions
75 *
76 * These data structures are laid over the table to parse the important values
77 * out of it.
78 */
79
80/*
81 * structure describing one IOMMU in the ACPI table. Typically followed by one
82 * or more ivhd_entrys.
83 */
84struct ivhd_header {
85 u8 type;
86 u8 flags;
87 u16 length;
88 u16 devid;
89 u16 cap_ptr;
90 u64 mmio_phys;
91 u16 pci_seg;
92 u16 info;
93 u32 reserved;
94} __attribute__((packed));
95
96/*
97 * A device entry describing which devices a specific IOMMU translates and
98 * which requestor ids they use.
99 */
100struct ivhd_entry {
101 u8 type;
102 u16 devid;
103 u8 flags;
104 u32 ext;
105} __attribute__((packed));
106
107/*
108 * An AMD IOMMU memory definition structure. It defines things like exclusion
109 * ranges for devices and regions that should be unity mapped.
110 */
111struct ivmd_header {
112 u8 type;
113 u8 flags;
114 u16 length;
115 u16 devid;
116 u16 aux;
117 u64 resv;
118 u64 range_start;
119 u64 range_length;
120} __attribute__((packed));
121
122bool amd_iommu_dump;
123
124static int __initdata amd_iommu_detected;
125static bool __initdata amd_iommu_disabled;
126
127u16 amd_iommu_last_bdf; /* largest PCI device id we have
128 to handle */
129LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
130 we find in ACPI */
131bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
132
133LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
134 system */
135
136/* Array to assign indices to IOMMUs*/
137struct amd_iommu *amd_iommus[MAX_IOMMUS];
138int amd_iommus_present;
139
140/* IOMMUs have a non-present cache? */
141bool amd_iommu_np_cache __read_mostly;
142bool amd_iommu_iotlb_sup __read_mostly = true;
143
144/*
145 * The ACPI table parsing functions set this variable on an error
146 */
147static int __initdata amd_iommu_init_err;
148
149/*
150 * List of protection domains - used during resume
151 */
152LIST_HEAD(amd_iommu_pd_list);
153spinlock_t amd_iommu_pd_lock;
154
155/*
156 * Pointer to the device table which is shared by all AMD IOMMUs
157 * it is indexed by the PCI device id or the HT unit id and contains
158 * information about the domain the device belongs to as well as the
159 * page table root pointer.
160 */
161struct dev_table_entry *amd_iommu_dev_table;
162
163/*
164 * The alias table is a driver specific data structure which contains the
165 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
166 * More than one device can share the same requestor id.
167 */
168u16 *amd_iommu_alias_table;
169
170/*
171 * The rlookup table is used to find the IOMMU which is responsible
172 * for a specific device. It is also indexed by the PCI device id.
173 */
174struct amd_iommu **amd_iommu_rlookup_table;
175
176/*
177 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
178 * to know which ones are already in use.
179 */
180unsigned long *amd_iommu_pd_alloc_bitmap;
181
182static u32 dev_table_size; /* size of the device table */
183static u32 alias_table_size; /* size of the alias table */
184static u32 rlookup_table_size; /* size if the rlookup table */
185
186/*
187 * This function flushes all internal caches of
188 * the IOMMU used by this driver.
189 */
190extern void iommu_flush_all_caches(struct amd_iommu *iommu);
191
192static inline void update_last_devid(u16 devid)
193{
194 if (devid > amd_iommu_last_bdf)
195 amd_iommu_last_bdf = devid;
196}
197
198static inline unsigned long tbl_size(int entry_size)
199{
200 unsigned shift = PAGE_SHIFT +
201 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
202
203 return 1UL << shift;
204}
205
206/* Access to l1 and l2 indexed register spaces */
207
208static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
209{
210 u32 val;
211
212 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
213 pci_read_config_dword(iommu->dev, 0xfc, &val);
214 return val;
215}
216
217static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
218{
219 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
220 pci_write_config_dword(iommu->dev, 0xfc, val);
221 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
222}
223
224static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
225{
226 u32 val;
227
228 pci_write_config_dword(iommu->dev, 0xf0, address);
229 pci_read_config_dword(iommu->dev, 0xf4, &val);
230 return val;
231}
232
233static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
234{
235 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
236 pci_write_config_dword(iommu->dev, 0xf4, val);
237}
238
239/****************************************************************************
240 *
241 * AMD IOMMU MMIO register space handling functions
242 *
243 * These functions are used to program the IOMMU device registers in
244 * MMIO space required for that driver.
245 *
246 ****************************************************************************/
247
248/*
249 * This function set the exclusion range in the IOMMU. DMA accesses to the
250 * exclusion range are passed through untranslated
251 */
252static void iommu_set_exclusion_range(struct amd_iommu *iommu)
253{
254 u64 start = iommu->exclusion_start & PAGE_MASK;
255 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
256 u64 entry;
257
258 if (!iommu->exclusion_start)
259 return;
260
261 entry = start | MMIO_EXCL_ENABLE_MASK;
262 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
263 &entry, sizeof(entry));
264
265 entry = limit;
266 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
267 &entry, sizeof(entry));
268}
269
270/* Programs the physical address of the device table into the IOMMU hardware */
271static void __init iommu_set_device_table(struct amd_iommu *iommu)
272{
273 u64 entry;
274
275 BUG_ON(iommu->mmio_base == NULL);
276
277 entry = virt_to_phys(amd_iommu_dev_table);
278 entry |= (dev_table_size >> 12) - 1;
279 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
280 &entry, sizeof(entry));
281}
282
283/* Generic functions to enable/disable certain features of the IOMMU. */
284static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
285{
286 u32 ctrl;
287
288 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
289 ctrl |= (1 << bit);
290 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
291}
292
293static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
294{
295 u32 ctrl;
296
297 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
298 ctrl &= ~(1 << bit);
299 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
300}
301
302/* Function to enable the hardware */
303static void iommu_enable(struct amd_iommu *iommu)
304{
305 static const char * const feat_str[] = {
306 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
307 "IA", "GA", "HE", "PC", NULL
308 };
309 int i;
310
311 printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx",
312 dev_name(&iommu->dev->dev), iommu->cap_ptr);
313
314 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
315 printk(KERN_CONT " extended features: ");
316 for (i = 0; feat_str[i]; ++i)
317 if (iommu_feature(iommu, (1ULL << i)))
318 printk(KERN_CONT " %s", feat_str[i]);
319 }
320 printk(KERN_CONT "\n");
321
322 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
323}
324
325static void iommu_disable(struct amd_iommu *iommu)
326{
327 /* Disable command buffer */
328 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
329
330 /* Disable event logging and event interrupts */
331 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
332 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
333
334 /* Disable IOMMU hardware itself */
335 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
336}
337
338/*
339 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
340 * the system has one.
341 */
342static u8 * __init iommu_map_mmio_space(u64 address)
343{
344 u8 *ret;
345
346 if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
347 pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
348 address);
349 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
350 return NULL;
351 }
352
353 ret = ioremap_nocache(address, MMIO_REGION_LENGTH);
354 if (ret != NULL)
355 return ret;
356
357 release_mem_region(address, MMIO_REGION_LENGTH);
358
359 return NULL;
360}
361
362static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
363{
364 if (iommu->mmio_base)
365 iounmap(iommu->mmio_base);
366 release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
367}
368
369/****************************************************************************
370 *
371 * The functions below belong to the first pass of AMD IOMMU ACPI table
372 * parsing. In this pass we try to find out the highest device id this
373 * code has to handle. Upon this information the size of the shared data
374 * structures is determined later.
375 *
376 ****************************************************************************/
377
378/*
379 * This function calculates the length of a given IVHD entry
380 */
381static inline int ivhd_entry_length(u8 *ivhd)
382{
383 return 0x04 << (*ivhd >> 6);
384}
385
386/*
387 * This function reads the last device id the IOMMU has to handle from the PCI
388 * capability header for this IOMMU
389 */
390static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
391{
392 u32 cap;
393
394 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
395 update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
396
397 return 0;
398}
399
400/*
401 * After reading the highest device id from the IOMMU PCI capability header
402 * this function looks if there is a higher device id defined in the ACPI table
403 */
404static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
405{
406 u8 *p = (void *)h, *end = (void *)h;
407 struct ivhd_entry *dev;
408
409 p += sizeof(*h);
410 end += h->length;
411
412 find_last_devid_on_pci(PCI_BUS(h->devid),
413 PCI_SLOT(h->devid),
414 PCI_FUNC(h->devid),
415 h->cap_ptr);
416
417 while (p < end) {
418 dev = (struct ivhd_entry *)p;
419 switch (dev->type) {
420 case IVHD_DEV_SELECT:
421 case IVHD_DEV_RANGE_END:
422 case IVHD_DEV_ALIAS:
423 case IVHD_DEV_EXT_SELECT:
424 /* all the above subfield types refer to device ids */
425 update_last_devid(dev->devid);
426 break;
427 default:
428 break;
429 }
430 p += ivhd_entry_length(p);
431 }
432
433 WARN_ON(p != end);
434
435 return 0;
436}
437
438/*
439 * Iterate over all IVHD entries in the ACPI table and find the highest device
440 * id which we need to handle. This is the first of three functions which parse
441 * the ACPI table. So we check the checksum here.
442 */
443static int __init find_last_devid_acpi(struct acpi_table_header *table)
444{
445 int i;
446 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
447 struct ivhd_header *h;
448
449 /*
450 * Validate checksum here so we don't need to do it when
451 * we actually parse the table
452 */
453 for (i = 0; i < table->length; ++i)
454 checksum += p[i];
455 if (checksum != 0) {
456 /* ACPI table corrupt */
457 amd_iommu_init_err = -ENODEV;
458 return 0;
459 }
460
461 p += IVRS_HEADER_LENGTH;
462
463 end += table->length;
464 while (p < end) {
465 h = (struct ivhd_header *)p;
466 switch (h->type) {
467 case ACPI_IVHD_TYPE:
468 find_last_devid_from_ivhd(h);
469 break;
470 default:
471 break;
472 }
473 p += h->length;
474 }
475 WARN_ON(p != end);
476
477 return 0;
478}
479
480/****************************************************************************
481 *
482 * The following functions belong the the code path which parses the ACPI table
483 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
484 * data structures, initialize the device/alias/rlookup table and also
485 * basically initialize the hardware.
486 *
487 ****************************************************************************/
488
489/*
490 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
491 * write commands to that buffer later and the IOMMU will execute them
492 * asynchronously
493 */
494static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
495{
496 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
497 get_order(CMD_BUFFER_SIZE));
498
499 if (cmd_buf == NULL)
500 return NULL;
501
502 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
503
504 return cmd_buf;
505}
506
507/*
508 * This function resets the command buffer if the IOMMU stopped fetching
509 * commands from it.
510 */
511void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
512{
513 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
514
515 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
516 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
517
518 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
519}
520
521/*
522 * This function writes the command buffer address to the hardware and
523 * enables it.
524 */
525static void iommu_enable_command_buffer(struct amd_iommu *iommu)
526{
527 u64 entry;
528
529 BUG_ON(iommu->cmd_buf == NULL);
530
531 entry = (u64)virt_to_phys(iommu->cmd_buf);
532 entry |= MMIO_CMD_SIZE_512;
533
534 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
535 &entry, sizeof(entry));
536
537 amd_iommu_reset_cmd_buffer(iommu);
538 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
539}
540
541static void __init free_command_buffer(struct amd_iommu *iommu)
542{
543 free_pages((unsigned long)iommu->cmd_buf,
544 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
545}
546
547/* allocates the memory where the IOMMU will log its events to */
548static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
549{
550 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
551 get_order(EVT_BUFFER_SIZE));
552
553 if (iommu->evt_buf == NULL)
554 return NULL;
555
556 iommu->evt_buf_size = EVT_BUFFER_SIZE;
557
558 return iommu->evt_buf;
559}
560
561static void iommu_enable_event_buffer(struct amd_iommu *iommu)
562{
563 u64 entry;
564
565 BUG_ON(iommu->evt_buf == NULL);
566
567 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
568
569 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
570 &entry, sizeof(entry));
571
572 /* set head and tail to zero manually */
573 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
574 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
575
576 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
577}
578
579static void __init free_event_buffer(struct amd_iommu *iommu)
580{
581 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
582}
583
584/* sets a specific bit in the device table entry. */
585static void set_dev_entry_bit(u16 devid, u8 bit)
586{
587 int i = (bit >> 5) & 0x07;
588 int _bit = bit & 0x1f;
589
590 amd_iommu_dev_table[devid].data[i] |= (1 << _bit);
591}
592
593static int get_dev_entry_bit(u16 devid, u8 bit)
594{
595 int i = (bit >> 5) & 0x07;
596 int _bit = bit & 0x1f;
597
598 return (amd_iommu_dev_table[devid].data[i] & (1 << _bit)) >> _bit;
599}
600
601
602void amd_iommu_apply_erratum_63(u16 devid)
603{
604 int sysmgt;
605
606 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
607 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
608
609 if (sysmgt == 0x01)
610 set_dev_entry_bit(devid, DEV_ENTRY_IW);
611}
612
613/* Writes the specific IOMMU for a device into the rlookup table */
614static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
615{
616 amd_iommu_rlookup_table[devid] = iommu;
617}
618
619/*
620 * This function takes the device specific flags read from the ACPI
621 * table and sets up the device table entry with that information
622 */
623static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
624 u16 devid, u32 flags, u32 ext_flags)
625{
626 if (flags & ACPI_DEVFLAG_INITPASS)
627 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
628 if (flags & ACPI_DEVFLAG_EXTINT)
629 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
630 if (flags & ACPI_DEVFLAG_NMI)
631 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
632 if (flags & ACPI_DEVFLAG_SYSMGT1)
633 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
634 if (flags & ACPI_DEVFLAG_SYSMGT2)
635 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
636 if (flags & ACPI_DEVFLAG_LINT0)
637 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
638 if (flags & ACPI_DEVFLAG_LINT1)
639 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
640
641 amd_iommu_apply_erratum_63(devid);
642
643 set_iommu_for_device(iommu, devid);
644}
645
646/*
647 * Reads the device exclusion range from ACPI and initialize IOMMU with
648 * it
649 */
650static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
651{
652 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
653
654 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
655 return;
656
657 if (iommu) {
658 /*
659 * We only can configure exclusion ranges per IOMMU, not
660 * per device. But we can enable the exclusion range per
661 * device. This is done here
662 */
663 set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
664 iommu->exclusion_start = m->range_start;
665 iommu->exclusion_length = m->range_length;
666 }
667}
668
669/*
670 * This function reads some important data from the IOMMU PCI space and
671 * initializes the driver data structure with it. It reads the hardware
672 * capabilities and the first/last device entries
673 */
674static void __init init_iommu_from_pci(struct amd_iommu *iommu)
675{
676 int cap_ptr = iommu->cap_ptr;
677 u32 range, misc, low, high;
678 int i, j;
679
680 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
681 &iommu->cap);
682 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
683 &range);
684 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
685 &misc);
686
687 iommu->first_device = calc_devid(MMIO_GET_BUS(range),
688 MMIO_GET_FD(range));
689 iommu->last_device = calc_devid(MMIO_GET_BUS(range),
690 MMIO_GET_LD(range));
691 iommu->evt_msi_num = MMIO_MSI_NUM(misc);
692
693 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
694 amd_iommu_iotlb_sup = false;
695
696 /* read extended feature bits */
697 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
698 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
699
700 iommu->features = ((u64)high << 32) | low;
701
702 if (!is_rd890_iommu(iommu->dev))
703 return;
704
705 /*
706 * Some rd890 systems may not be fully reconfigured by the BIOS, so
707 * it's necessary for us to store this information so it can be
708 * reprogrammed on resume
709 */
710
711 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
712 &iommu->stored_addr_lo);
713 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
714 &iommu->stored_addr_hi);
715
716 /* Low bit locks writes to configuration space */
717 iommu->stored_addr_lo &= ~1;
718
719 for (i = 0; i < 6; i++)
720 for (j = 0; j < 0x12; j++)
721 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
722
723 for (i = 0; i < 0x83; i++)
724 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
725}
726
727/*
728 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
729 * initializes the hardware and our data structures with it.
730 */
731static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
732 struct ivhd_header *h)
733{
734 u8 *p = (u8 *)h;
735 u8 *end = p, flags = 0;
736 u16 devid = 0, devid_start = 0, devid_to = 0;
737 u32 dev_i, ext_flags = 0;
738 bool alias = false;
739 struct ivhd_entry *e;
740
741 /*
742 * First save the recommended feature enable bits from ACPI
743 */
744 iommu->acpi_flags = h->flags;
745
746 /*
747 * Done. Now parse the device entries
748 */
749 p += sizeof(struct ivhd_header);
750 end += h->length;
751
752
753 while (p < end) {
754 e = (struct ivhd_entry *)p;
755 switch (e->type) {
756 case IVHD_DEV_ALL:
757
758 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
759 " last device %02x:%02x.%x flags: %02x\n",
760 PCI_BUS(iommu->first_device),
761 PCI_SLOT(iommu->first_device),
762 PCI_FUNC(iommu->first_device),
763 PCI_BUS(iommu->last_device),
764 PCI_SLOT(iommu->last_device),
765 PCI_FUNC(iommu->last_device),
766 e->flags);
767
768 for (dev_i = iommu->first_device;
769 dev_i <= iommu->last_device; ++dev_i)
770 set_dev_entry_from_acpi(iommu, dev_i,
771 e->flags, 0);
772 break;
773 case IVHD_DEV_SELECT:
774
775 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
776 "flags: %02x\n",
777 PCI_BUS(e->devid),
778 PCI_SLOT(e->devid),
779 PCI_FUNC(e->devid),
780 e->flags);
781
782 devid = e->devid;
783 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
784 break;
785 case IVHD_DEV_SELECT_RANGE_START:
786
787 DUMP_printk(" DEV_SELECT_RANGE_START\t "
788 "devid: %02x:%02x.%x flags: %02x\n",
789 PCI_BUS(e->devid),
790 PCI_SLOT(e->devid),
791 PCI_FUNC(e->devid),
792 e->flags);
793
794 devid_start = e->devid;
795 flags = e->flags;
796 ext_flags = 0;
797 alias = false;
798 break;
799 case IVHD_DEV_ALIAS:
800
801 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
802 "flags: %02x devid_to: %02x:%02x.%x\n",
803 PCI_BUS(e->devid),
804 PCI_SLOT(e->devid),
805 PCI_FUNC(e->devid),
806 e->flags,
807 PCI_BUS(e->ext >> 8),
808 PCI_SLOT(e->ext >> 8),
809 PCI_FUNC(e->ext >> 8));
810
811 devid = e->devid;
812 devid_to = e->ext >> 8;
813 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
814 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
815 amd_iommu_alias_table[devid] = devid_to;
816 break;
817 case IVHD_DEV_ALIAS_RANGE:
818
819 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
820 "devid: %02x:%02x.%x flags: %02x "
821 "devid_to: %02x:%02x.%x\n",
822 PCI_BUS(e->devid),
823 PCI_SLOT(e->devid),
824 PCI_FUNC(e->devid),
825 e->flags,
826 PCI_BUS(e->ext >> 8),
827 PCI_SLOT(e->ext >> 8),
828 PCI_FUNC(e->ext >> 8));
829
830 devid_start = e->devid;
831 flags = e->flags;
832 devid_to = e->ext >> 8;
833 ext_flags = 0;
834 alias = true;
835 break;
836 case IVHD_DEV_EXT_SELECT:
837
838 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
839 "flags: %02x ext: %08x\n",
840 PCI_BUS(e->devid),
841 PCI_SLOT(e->devid),
842 PCI_FUNC(e->devid),
843 e->flags, e->ext);
844
845 devid = e->devid;
846 set_dev_entry_from_acpi(iommu, devid, e->flags,
847 e->ext);
848 break;
849 case IVHD_DEV_EXT_SELECT_RANGE:
850
851 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
852 "%02x:%02x.%x flags: %02x ext: %08x\n",
853 PCI_BUS(e->devid),
854 PCI_SLOT(e->devid),
855 PCI_FUNC(e->devid),
856 e->flags, e->ext);
857
858 devid_start = e->devid;
859 flags = e->flags;
860 ext_flags = e->ext;
861 alias = false;
862 break;
863 case IVHD_DEV_RANGE_END:
864
865 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
866 PCI_BUS(e->devid),
867 PCI_SLOT(e->devid),
868 PCI_FUNC(e->devid));
869
870 devid = e->devid;
871 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
872 if (alias) {
873 amd_iommu_alias_table[dev_i] = devid_to;
874 set_dev_entry_from_acpi(iommu,
875 devid_to, flags, ext_flags);
876 }
877 set_dev_entry_from_acpi(iommu, dev_i,
878 flags, ext_flags);
879 }
880 break;
881 default:
882 break;
883 }
884
885 p += ivhd_entry_length(p);
886 }
887}
888
889/* Initializes the device->iommu mapping for the driver */
890static int __init init_iommu_devices(struct amd_iommu *iommu)
891{
892 u32 i;
893
894 for (i = iommu->first_device; i <= iommu->last_device; ++i)
895 set_iommu_for_device(iommu, i);
896
897 return 0;
898}
899
900static void __init free_iommu_one(struct amd_iommu *iommu)
901{
902 free_command_buffer(iommu);
903 free_event_buffer(iommu);
904 iommu_unmap_mmio_space(iommu);
905}
906
907static void __init free_iommu_all(void)
908{
909 struct amd_iommu *iommu, *next;
910
911 for_each_iommu_safe(iommu, next) {
912 list_del(&iommu->list);
913 free_iommu_one(iommu);
914 kfree(iommu);
915 }
916}
917
918/*
919 * This function clues the initialization function for one IOMMU
920 * together and also allocates the command buffer and programs the
921 * hardware. It does NOT enable the IOMMU. This is done afterwards.
922 */
923static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
924{
925 spin_lock_init(&iommu->lock);
926
927 /* Add IOMMU to internal data structures */
928 list_add_tail(&iommu->list, &amd_iommu_list);
929 iommu->index = amd_iommus_present++;
930
931 if (unlikely(iommu->index >= MAX_IOMMUS)) {
932 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
933 return -ENOSYS;
934 }
935
936 /* Index is fine - add IOMMU to the array */
937 amd_iommus[iommu->index] = iommu;
938
939 /*
940 * Copy data from ACPI table entry to the iommu struct
941 */
942 iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
943 if (!iommu->dev)
944 return 1;
945
946 iommu->cap_ptr = h->cap_ptr;
947 iommu->pci_seg = h->pci_seg;
948 iommu->mmio_phys = h->mmio_phys;
949 iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
950 if (!iommu->mmio_base)
951 return -ENOMEM;
952
953 iommu->cmd_buf = alloc_command_buffer(iommu);
954 if (!iommu->cmd_buf)
955 return -ENOMEM;
956
957 iommu->evt_buf = alloc_event_buffer(iommu);
958 if (!iommu->evt_buf)
959 return -ENOMEM;
960
961 iommu->int_enabled = false;
962
963 init_iommu_from_pci(iommu);
964 init_iommu_from_acpi(iommu, h);
965 init_iommu_devices(iommu);
966
967 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
968 amd_iommu_np_cache = true;
969
970 return pci_enable_device(iommu->dev);
971}
972
973/*
974 * Iterates over all IOMMU entries in the ACPI table, allocates the
975 * IOMMU structure and initializes it with init_iommu_one()
976 */
977static int __init init_iommu_all(struct acpi_table_header *table)
978{
979 u8 *p = (u8 *)table, *end = (u8 *)table;
980 struct ivhd_header *h;
981 struct amd_iommu *iommu;
982 int ret;
983
984 end += table->length;
985 p += IVRS_HEADER_LENGTH;
986
987 while (p < end) {
988 h = (struct ivhd_header *)p;
989 switch (*p) {
990 case ACPI_IVHD_TYPE:
991
992 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
993 "seg: %d flags: %01x info %04x\n",
994 PCI_BUS(h->devid), PCI_SLOT(h->devid),
995 PCI_FUNC(h->devid), h->cap_ptr,
996 h->pci_seg, h->flags, h->info);
997 DUMP_printk(" mmio-addr: %016llx\n",
998 h->mmio_phys);
999
1000 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1001 if (iommu == NULL) {
1002 amd_iommu_init_err = -ENOMEM;
1003 return 0;
1004 }
1005
1006 ret = init_iommu_one(iommu, h);
1007 if (ret) {
1008 amd_iommu_init_err = ret;
1009 return 0;
1010 }
1011 break;
1012 default:
1013 break;
1014 }
1015 p += h->length;
1016
1017 }
1018 WARN_ON(p != end);
1019
1020 return 0;
1021}
1022
1023/****************************************************************************
1024 *
1025 * The following functions initialize the MSI interrupts for all IOMMUs
1026 * in the system. Its a bit challenging because there could be multiple
1027 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1028 * pci_dev.
1029 *
1030 ****************************************************************************/
1031
1032static int iommu_setup_msi(struct amd_iommu *iommu)
1033{
1034 int r;
1035
1036 if (pci_enable_msi(iommu->dev))
1037 return 1;
1038
1039 r = request_threaded_irq(iommu->dev->irq,
1040 amd_iommu_int_handler,
1041 amd_iommu_int_thread,
1042 0, "AMD-Vi",
1043 iommu->dev);
1044
1045 if (r) {
1046 pci_disable_msi(iommu->dev);
1047 return 1;
1048 }
1049
1050 iommu->int_enabled = true;
1051 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1052
1053 return 0;
1054}
1055
1056static int iommu_init_msi(struct amd_iommu *iommu)
1057{
1058 if (iommu->int_enabled)
1059 return 0;
1060
1061 if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
1062 return iommu_setup_msi(iommu);
1063
1064 return 1;
1065}
1066
1067/****************************************************************************
1068 *
1069 * The next functions belong to the third pass of parsing the ACPI
1070 * table. In this last pass the memory mapping requirements are
1071 * gathered (like exclusion and unity mapping reanges).
1072 *
1073 ****************************************************************************/
1074
1075static void __init free_unity_maps(void)
1076{
1077 struct unity_map_entry *entry, *next;
1078
1079 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1080 list_del(&entry->list);
1081 kfree(entry);
1082 }
1083}
1084
1085/* called when we find an exclusion range definition in ACPI */
1086static int __init init_exclusion_range(struct ivmd_header *m)
1087{
1088 int i;
1089
1090 switch (m->type) {
1091 case ACPI_IVMD_TYPE:
1092 set_device_exclusion_range(m->devid, m);
1093 break;
1094 case ACPI_IVMD_TYPE_ALL:
1095 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1096 set_device_exclusion_range(i, m);
1097 break;
1098 case ACPI_IVMD_TYPE_RANGE:
1099 for (i = m->devid; i <= m->aux; ++i)
1100 set_device_exclusion_range(i, m);
1101 break;
1102 default:
1103 break;
1104 }
1105
1106 return 0;
1107}
1108
1109/* called for unity map ACPI definition */
1110static int __init init_unity_map_range(struct ivmd_header *m)
1111{
1112 struct unity_map_entry *e = 0;
1113 char *s;
1114
1115 e = kzalloc(sizeof(*e), GFP_KERNEL);
1116 if (e == NULL)
1117 return -ENOMEM;
1118
1119 switch (m->type) {
1120 default:
1121 kfree(e);
1122 return 0;
1123 case ACPI_IVMD_TYPE:
1124 s = "IVMD_TYPEi\t\t\t";
1125 e->devid_start = e->devid_end = m->devid;
1126 break;
1127 case ACPI_IVMD_TYPE_ALL:
1128 s = "IVMD_TYPE_ALL\t\t";
1129 e->devid_start = 0;
1130 e->devid_end = amd_iommu_last_bdf;
1131 break;
1132 case ACPI_IVMD_TYPE_RANGE:
1133 s = "IVMD_TYPE_RANGE\t\t";
1134 e->devid_start = m->devid;
1135 e->devid_end = m->aux;
1136 break;
1137 }
1138 e->address_start = PAGE_ALIGN(m->range_start);
1139 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1140 e->prot = m->flags >> 1;
1141
1142 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1143 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1144 PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1145 PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1146 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1147 e->address_start, e->address_end, m->flags);
1148
1149 list_add_tail(&e->list, &amd_iommu_unity_map);
1150
1151 return 0;
1152}
1153
1154/* iterates over all memory definitions we find in the ACPI table */
1155static int __init init_memory_definitions(struct acpi_table_header *table)
1156{
1157 u8 *p = (u8 *)table, *end = (u8 *)table;
1158 struct ivmd_header *m;
1159
1160 end += table->length;
1161 p += IVRS_HEADER_LENGTH;
1162
1163 while (p < end) {
1164 m = (struct ivmd_header *)p;
1165 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1166 init_exclusion_range(m);
1167 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1168 init_unity_map_range(m);
1169
1170 p += m->length;
1171 }
1172
1173 return 0;
1174}
1175
1176/*
1177 * Init the device table to not allow DMA access for devices and
1178 * suppress all page faults
1179 */
1180static void init_device_table(void)
1181{
1182 u32 devid;
1183
1184 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1185 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1186 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1187 }
1188}
1189
1190static void iommu_init_flags(struct amd_iommu *iommu)
1191{
1192 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1193 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1194 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1195
1196 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1197 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1198 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1199
1200 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1201 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1202 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1203
1204 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1205 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1206 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1207
1208 /*
1209 * make IOMMU memory accesses cache coherent
1210 */
1211 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1212}
1213
1214static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1215{
1216 int i, j;
1217 u32 ioc_feature_control;
1218 struct pci_dev *pdev = NULL;
1219
1220 /* RD890 BIOSes may not have completely reconfigured the iommu */
1221 if (!is_rd890_iommu(iommu->dev))
1222 return;
1223
1224 /*
1225 * First, we need to ensure that the iommu is enabled. This is
1226 * controlled by a register in the northbridge
1227 */
1228 pdev = pci_get_bus_and_slot(iommu->dev->bus->number, PCI_DEVFN(0, 0));
1229
1230 if (!pdev)
1231 return;
1232
1233 /* Select Northbridge indirect register 0x75 and enable writing */
1234 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1235 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1236
1237 /* Enable the iommu */
1238 if (!(ioc_feature_control & 0x1))
1239 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1240
1241 pci_dev_put(pdev);
1242
1243 /* Restore the iommu BAR */
1244 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1245 iommu->stored_addr_lo);
1246 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1247 iommu->stored_addr_hi);
1248
1249 /* Restore the l1 indirect regs for each of the 6 l1s */
1250 for (i = 0; i < 6; i++)
1251 for (j = 0; j < 0x12; j++)
1252 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1253
1254 /* Restore the l2 indirect regs */
1255 for (i = 0; i < 0x83; i++)
1256 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1257
1258 /* Lock PCI setup registers */
1259 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1260 iommu->stored_addr_lo | 1);
1261}
1262
1263/*
1264 * This function finally enables all IOMMUs found in the system after
1265 * they have been initialized
1266 */
1267static void enable_iommus(void)
1268{
1269 struct amd_iommu *iommu;
1270
1271 for_each_iommu(iommu) {
1272 iommu_disable(iommu);
1273 iommu_init_flags(iommu);
1274 iommu_set_device_table(iommu);
1275 iommu_enable_command_buffer(iommu);
1276 iommu_enable_event_buffer(iommu);
1277 iommu_set_exclusion_range(iommu);
1278 iommu_init_msi(iommu);
1279 iommu_enable(iommu);
1280 iommu_flush_all_caches(iommu);
1281 }
1282}
1283
1284static void disable_iommus(void)
1285{
1286 struct amd_iommu *iommu;
1287
1288 for_each_iommu(iommu)
1289 iommu_disable(iommu);
1290}
1291
1292/*
1293 * Suspend/Resume support
1294 * disable suspend until real resume implemented
1295 */
1296
1297static void amd_iommu_resume(void)
1298{
1299 struct amd_iommu *iommu;
1300
1301 for_each_iommu(iommu)
1302 iommu_apply_resume_quirks(iommu);
1303
1304 /* re-load the hardware */
1305 enable_iommus();
1306
1307 /*
1308 * we have to flush after the IOMMUs are enabled because a
1309 * disabled IOMMU will never execute the commands we send
1310 */
1311 for_each_iommu(iommu)
1312 iommu_flush_all_caches(iommu);
1313}
1314
1315static int amd_iommu_suspend(void)
1316{
1317 /* disable IOMMUs to go out of the way for BIOS */
1318 disable_iommus();
1319
1320 return 0;
1321}
1322
1323static struct syscore_ops amd_iommu_syscore_ops = {
1324 .suspend = amd_iommu_suspend,
1325 .resume = amd_iommu_resume,
1326};
1327
1328/*
1329 * This is the core init function for AMD IOMMU hardware in the system.
1330 * This function is called from the generic x86 DMA layer initialization
1331 * code.
1332 *
1333 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1334 * three times:
1335 *
1336 * 1 pass) Find the highest PCI device id the driver has to handle.
1337 * Upon this information the size of the data structures is
1338 * determined that needs to be allocated.
1339 *
1340 * 2 pass) Initialize the data structures just allocated with the
1341 * information in the ACPI table about available AMD IOMMUs
1342 * in the system. It also maps the PCI devices in the
1343 * system to specific IOMMUs
1344 *
1345 * 3 pass) After the basic data structures are allocated and
1346 * initialized we update them with information about memory
1347 * remapping requirements parsed out of the ACPI table in
1348 * this last pass.
1349 *
1350 * After that the hardware is initialized and ready to go. In the last
1351 * step we do some Linux specific things like registering the driver in
1352 * the dma_ops interface and initializing the suspend/resume support
1353 * functions. Finally it prints some information about AMD IOMMUs and
1354 * the driver state and enables the hardware.
1355 */
1356static int __init amd_iommu_init(void)
1357{
1358 int i, ret = 0;
1359
1360 /*
1361 * First parse ACPI tables to find the largest Bus/Dev/Func
1362 * we need to handle. Upon this information the shared data
1363 * structures for the IOMMUs in the system will be allocated
1364 */
1365 if (acpi_table_parse("IVRS", find_last_devid_acpi) != 0)
1366 return -ENODEV;
1367
1368 ret = amd_iommu_init_err;
1369 if (ret)
1370 goto out;
1371
1372 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1373 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1374 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1375
1376 ret = -ENOMEM;
1377
1378 /* Device table - directly used by all IOMMUs */
1379 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1380 get_order(dev_table_size));
1381 if (amd_iommu_dev_table == NULL)
1382 goto out;
1383
1384 /*
1385 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1386 * IOMMU see for that device
1387 */
1388 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1389 get_order(alias_table_size));
1390 if (amd_iommu_alias_table == NULL)
1391 goto free;
1392
1393 /* IOMMU rlookup table - find the IOMMU for a specific device */
1394 amd_iommu_rlookup_table = (void *)__get_free_pages(
1395 GFP_KERNEL | __GFP_ZERO,
1396 get_order(rlookup_table_size));
1397 if (amd_iommu_rlookup_table == NULL)
1398 goto free;
1399
1400 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1401 GFP_KERNEL | __GFP_ZERO,
1402 get_order(MAX_DOMAIN_ID/8));
1403 if (amd_iommu_pd_alloc_bitmap == NULL)
1404 goto free;
1405
1406 /* init the device table */
1407 init_device_table();
1408
1409 /*
1410 * let all alias entries point to itself
1411 */
1412 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1413 amd_iommu_alias_table[i] = i;
1414
1415 /*
1416 * never allocate domain 0 because its used as the non-allocated and
1417 * error value placeholder
1418 */
1419 amd_iommu_pd_alloc_bitmap[0] = 1;
1420
1421 spin_lock_init(&amd_iommu_pd_lock);
1422
1423 /*
1424 * now the data structures are allocated and basically initialized
1425 * start the real acpi table scan
1426 */
1427 ret = -ENODEV;
1428 if (acpi_table_parse("IVRS", init_iommu_all) != 0)
1429 goto free;
1430
1431 if (amd_iommu_init_err) {
1432 ret = amd_iommu_init_err;
1433 goto free;
1434 }
1435
1436 if (acpi_table_parse("IVRS", init_memory_definitions) != 0)
1437 goto free;
1438
1439 if (amd_iommu_init_err) {
1440 ret = amd_iommu_init_err;
1441 goto free;
1442 }
1443
1444 ret = amd_iommu_init_devices();
1445 if (ret)
1446 goto free;
1447
1448 enable_iommus();
1449
1450 if (iommu_pass_through)
1451 ret = amd_iommu_init_passthrough();
1452 else
1453 ret = amd_iommu_init_dma_ops();
1454
1455 if (ret)
1456 goto free_disable;
1457
1458 amd_iommu_init_api();
1459
1460 amd_iommu_init_notifier();
1461
1462 register_syscore_ops(&amd_iommu_syscore_ops);
1463
1464 if (iommu_pass_through)
1465 goto out;
1466
1467 if (amd_iommu_unmap_flush)
1468 printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n");
1469 else
1470 printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n");
1471
1472 x86_platform.iommu_shutdown = disable_iommus;
1473out:
1474 return ret;
1475
1476free_disable:
1477 disable_iommus();
1478
1479free:
1480 amd_iommu_uninit_devices();
1481
1482 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1483 get_order(MAX_DOMAIN_ID/8));
1484
1485 free_pages((unsigned long)amd_iommu_rlookup_table,
1486 get_order(rlookup_table_size));
1487
1488 free_pages((unsigned long)amd_iommu_alias_table,
1489 get_order(alias_table_size));
1490
1491 free_pages((unsigned long)amd_iommu_dev_table,
1492 get_order(dev_table_size));
1493
1494 free_iommu_all();
1495
1496 free_unity_maps();
1497
1498#ifdef CONFIG_GART_IOMMU
1499 /*
1500 * We failed to initialize the AMD IOMMU - try fallback to GART
1501 * if possible.
1502 */
1503 gart_iommu_init();
1504
1505#endif
1506
1507 goto out;
1508}
1509
1510/****************************************************************************
1511 *
1512 * Early detect code. This code runs at IOMMU detection time in the DMA
1513 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1514 * IOMMUs
1515 *
1516 ****************************************************************************/
1517static int __init early_amd_iommu_detect(struct acpi_table_header *table)
1518{
1519 return 0;
1520}
1521
1522int __init amd_iommu_detect(void)
1523{
1524 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
1525 return -ENODEV;
1526
1527 if (amd_iommu_disabled)
1528 return -ENODEV;
1529
1530 if (acpi_table_parse("IVRS", early_amd_iommu_detect) == 0) {
1531 iommu_detected = 1;
1532 amd_iommu_detected = 1;
1533 x86_init.iommu.iommu_init = amd_iommu_init;
1534
1535 /* Make sure ACS will be enabled */
1536 pci_request_acs();
1537 return 1;
1538 }
1539 return -ENODEV;
1540}
1541
1542/****************************************************************************
1543 *
1544 * Parsing functions for the AMD IOMMU specific kernel command line
1545 * options.
1546 *
1547 ****************************************************************************/
1548
1549static int __init parse_amd_iommu_dump(char *str)
1550{
1551 amd_iommu_dump = true;
1552
1553 return 1;
1554}
1555
1556static int __init parse_amd_iommu_options(char *str)
1557{
1558 for (; *str; ++str) {
1559 if (strncmp(str, "fullflush", 9) == 0)
1560 amd_iommu_unmap_flush = true;
1561 if (strncmp(str, "off", 3) == 0)
1562 amd_iommu_disabled = true;
1563 }
1564
1565 return 1;
1566}
1567
1568__setup("amd_iommu_dump", parse_amd_iommu_dump);
1569__setup("amd_iommu=", parse_amd_iommu_options);
1570
1571IOMMU_INIT_FINISH(amd_iommu_detect,
1572 gart_iommu_hole_init,
1573 0,
1574 0);
1/*
2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <jroedel@suse.de>
4 * Leo Duran <leo.duran@amd.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/pci.h>
21#include <linux/acpi.h>
22#include <linux/list.h>
23#include <linux/bitmap.h>
24#include <linux/slab.h>
25#include <linux/syscore_ops.h>
26#include <linux/interrupt.h>
27#include <linux/msi.h>
28#include <linux/amd-iommu.h>
29#include <linux/export.h>
30#include <linux/iommu.h>
31#include <linux/kmemleak.h>
32#include <asm/pci-direct.h>
33#include <asm/iommu.h>
34#include <asm/gart.h>
35#include <asm/x86_init.h>
36#include <asm/iommu_table.h>
37#include <asm/io_apic.h>
38#include <asm/irq_remapping.h>
39
40#include "amd_iommu_proto.h"
41#include "amd_iommu_types.h"
42#include "irq_remapping.h"
43
44/*
45 * definitions for the ACPI scanning code
46 */
47#define IVRS_HEADER_LENGTH 48
48
49#define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
50#define ACPI_IVMD_TYPE_ALL 0x20
51#define ACPI_IVMD_TYPE 0x21
52#define ACPI_IVMD_TYPE_RANGE 0x22
53
54#define IVHD_DEV_ALL 0x01
55#define IVHD_DEV_SELECT 0x02
56#define IVHD_DEV_SELECT_RANGE_START 0x03
57#define IVHD_DEV_RANGE_END 0x04
58#define IVHD_DEV_ALIAS 0x42
59#define IVHD_DEV_ALIAS_RANGE 0x43
60#define IVHD_DEV_EXT_SELECT 0x46
61#define IVHD_DEV_EXT_SELECT_RANGE 0x47
62#define IVHD_DEV_SPECIAL 0x48
63#define IVHD_DEV_ACPI_HID 0xf0
64
65#define UID_NOT_PRESENT 0
66#define UID_IS_INTEGER 1
67#define UID_IS_CHARACTER 2
68
69#define IVHD_SPECIAL_IOAPIC 1
70#define IVHD_SPECIAL_HPET 2
71
72#define IVHD_FLAG_HT_TUN_EN_MASK 0x01
73#define IVHD_FLAG_PASSPW_EN_MASK 0x02
74#define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
75#define IVHD_FLAG_ISOC_EN_MASK 0x08
76
77#define IVMD_FLAG_EXCL_RANGE 0x08
78#define IVMD_FLAG_UNITY_MAP 0x01
79
80#define ACPI_DEVFLAG_INITPASS 0x01
81#define ACPI_DEVFLAG_EXTINT 0x02
82#define ACPI_DEVFLAG_NMI 0x04
83#define ACPI_DEVFLAG_SYSMGT1 0x10
84#define ACPI_DEVFLAG_SYSMGT2 0x20
85#define ACPI_DEVFLAG_LINT0 0x40
86#define ACPI_DEVFLAG_LINT1 0x80
87#define ACPI_DEVFLAG_ATSDIS 0x10000000
88
89#define LOOP_TIMEOUT 100000
90/*
91 * ACPI table definitions
92 *
93 * These data structures are laid over the table to parse the important values
94 * out of it.
95 */
96
97/*
98 * structure describing one IOMMU in the ACPI table. Typically followed by one
99 * or more ivhd_entrys.
100 */
101struct ivhd_header {
102 u8 type;
103 u8 flags;
104 u16 length;
105 u16 devid;
106 u16 cap_ptr;
107 u64 mmio_phys;
108 u16 pci_seg;
109 u16 info;
110 u32 efr_attr;
111
112 /* Following only valid on IVHD type 11h and 40h */
113 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
114 u64 res;
115} __attribute__((packed));
116
117/*
118 * A device entry describing which devices a specific IOMMU translates and
119 * which requestor ids they use.
120 */
121struct ivhd_entry {
122 u8 type;
123 u16 devid;
124 u8 flags;
125 u32 ext;
126 u32 hidh;
127 u64 cid;
128 u8 uidf;
129 u8 uidl;
130 u8 uid;
131} __attribute__((packed));
132
133/*
134 * An AMD IOMMU memory definition structure. It defines things like exclusion
135 * ranges for devices and regions that should be unity mapped.
136 */
137struct ivmd_header {
138 u8 type;
139 u8 flags;
140 u16 length;
141 u16 devid;
142 u16 aux;
143 u64 resv;
144 u64 range_start;
145 u64 range_length;
146} __attribute__((packed));
147
148bool amd_iommu_dump;
149bool amd_iommu_irq_remap __read_mostly;
150
151int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
152
153static bool amd_iommu_detected;
154static bool __initdata amd_iommu_disabled;
155static int amd_iommu_target_ivhd_type;
156
157u16 amd_iommu_last_bdf; /* largest PCI device id we have
158 to handle */
159LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
160 we find in ACPI */
161bool amd_iommu_unmap_flush; /* if true, flush on every unmap */
162
163LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
164 system */
165
166/* Array to assign indices to IOMMUs*/
167struct amd_iommu *amd_iommus[MAX_IOMMUS];
168int amd_iommus_present;
169
170/* IOMMUs have a non-present cache? */
171bool amd_iommu_np_cache __read_mostly;
172bool amd_iommu_iotlb_sup __read_mostly = true;
173
174u32 amd_iommu_max_pasid __read_mostly = ~0;
175
176bool amd_iommu_v2_present __read_mostly;
177static bool amd_iommu_pc_present __read_mostly;
178
179bool amd_iommu_force_isolation __read_mostly;
180
181/*
182 * List of protection domains - used during resume
183 */
184LIST_HEAD(amd_iommu_pd_list);
185spinlock_t amd_iommu_pd_lock;
186
187/*
188 * Pointer to the device table which is shared by all AMD IOMMUs
189 * it is indexed by the PCI device id or the HT unit id and contains
190 * information about the domain the device belongs to as well as the
191 * page table root pointer.
192 */
193struct dev_table_entry *amd_iommu_dev_table;
194
195/*
196 * The alias table is a driver specific data structure which contains the
197 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
198 * More than one device can share the same requestor id.
199 */
200u16 *amd_iommu_alias_table;
201
202/*
203 * The rlookup table is used to find the IOMMU which is responsible
204 * for a specific device. It is also indexed by the PCI device id.
205 */
206struct amd_iommu **amd_iommu_rlookup_table;
207
208/*
209 * This table is used to find the irq remapping table for a given device id
210 * quickly.
211 */
212struct irq_remap_table **irq_lookup_table;
213
214/*
215 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
216 * to know which ones are already in use.
217 */
218unsigned long *amd_iommu_pd_alloc_bitmap;
219
220static u32 dev_table_size; /* size of the device table */
221static u32 alias_table_size; /* size of the alias table */
222static u32 rlookup_table_size; /* size if the rlookup table */
223
224enum iommu_init_state {
225 IOMMU_START_STATE,
226 IOMMU_IVRS_DETECTED,
227 IOMMU_ACPI_FINISHED,
228 IOMMU_ENABLED,
229 IOMMU_PCI_INIT,
230 IOMMU_INTERRUPTS_EN,
231 IOMMU_DMA_OPS,
232 IOMMU_INITIALIZED,
233 IOMMU_NOT_FOUND,
234 IOMMU_INIT_ERROR,
235};
236
237/* Early ioapic and hpet maps from kernel command line */
238#define EARLY_MAP_SIZE 4
239static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
240static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
241static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
242
243static int __initdata early_ioapic_map_size;
244static int __initdata early_hpet_map_size;
245static int __initdata early_acpihid_map_size;
246
247static bool __initdata cmdline_maps;
248
249static enum iommu_init_state init_state = IOMMU_START_STATE;
250
251static int amd_iommu_enable_interrupts(void);
252static int __init iommu_go_to_state(enum iommu_init_state state);
253static void init_device_table_dma(void);
254
255static int iommu_pc_get_set_reg_val(struct amd_iommu *iommu,
256 u8 bank, u8 cntr, u8 fxn,
257 u64 *value, bool is_write);
258
259static inline void update_last_devid(u16 devid)
260{
261 if (devid > amd_iommu_last_bdf)
262 amd_iommu_last_bdf = devid;
263}
264
265static inline unsigned long tbl_size(int entry_size)
266{
267 unsigned shift = PAGE_SHIFT +
268 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
269
270 return 1UL << shift;
271}
272
273/* Access to l1 and l2 indexed register spaces */
274
275static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
276{
277 u32 val;
278
279 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
280 pci_read_config_dword(iommu->dev, 0xfc, &val);
281 return val;
282}
283
284static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
285{
286 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
287 pci_write_config_dword(iommu->dev, 0xfc, val);
288 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
289}
290
291static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
292{
293 u32 val;
294
295 pci_write_config_dword(iommu->dev, 0xf0, address);
296 pci_read_config_dword(iommu->dev, 0xf4, &val);
297 return val;
298}
299
300static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
301{
302 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
303 pci_write_config_dword(iommu->dev, 0xf4, val);
304}
305
306/****************************************************************************
307 *
308 * AMD IOMMU MMIO register space handling functions
309 *
310 * These functions are used to program the IOMMU device registers in
311 * MMIO space required for that driver.
312 *
313 ****************************************************************************/
314
315/*
316 * This function set the exclusion range in the IOMMU. DMA accesses to the
317 * exclusion range are passed through untranslated
318 */
319static void iommu_set_exclusion_range(struct amd_iommu *iommu)
320{
321 u64 start = iommu->exclusion_start & PAGE_MASK;
322 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
323 u64 entry;
324
325 if (!iommu->exclusion_start)
326 return;
327
328 entry = start | MMIO_EXCL_ENABLE_MASK;
329 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
330 &entry, sizeof(entry));
331
332 entry = limit;
333 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
334 &entry, sizeof(entry));
335}
336
337/* Programs the physical address of the device table into the IOMMU hardware */
338static void iommu_set_device_table(struct amd_iommu *iommu)
339{
340 u64 entry;
341
342 BUG_ON(iommu->mmio_base == NULL);
343
344 entry = virt_to_phys(amd_iommu_dev_table);
345 entry |= (dev_table_size >> 12) - 1;
346 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
347 &entry, sizeof(entry));
348}
349
350/* Generic functions to enable/disable certain features of the IOMMU. */
351static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
352{
353 u32 ctrl;
354
355 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
356 ctrl |= (1 << bit);
357 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
358}
359
360static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
361{
362 u32 ctrl;
363
364 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
365 ctrl &= ~(1 << bit);
366 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
367}
368
369static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
370{
371 u32 ctrl;
372
373 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
374 ctrl &= ~CTRL_INV_TO_MASK;
375 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
376 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
377}
378
379/* Function to enable the hardware */
380static void iommu_enable(struct amd_iommu *iommu)
381{
382 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
383}
384
385static void iommu_disable(struct amd_iommu *iommu)
386{
387 /* Disable command buffer */
388 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
389
390 /* Disable event logging and event interrupts */
391 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
392 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
393
394 /* Disable IOMMU GA_LOG */
395 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
396 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
397
398 /* Disable IOMMU hardware itself */
399 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
400}
401
402/*
403 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
404 * the system has one.
405 */
406static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
407{
408 if (!request_mem_region(address, end, "amd_iommu")) {
409 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
410 address, end);
411 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
412 return NULL;
413 }
414
415 return (u8 __iomem *)ioremap_nocache(address, end);
416}
417
418static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
419{
420 if (iommu->mmio_base)
421 iounmap(iommu->mmio_base);
422 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
423}
424
425static inline u32 get_ivhd_header_size(struct ivhd_header *h)
426{
427 u32 size = 0;
428
429 switch (h->type) {
430 case 0x10:
431 size = 24;
432 break;
433 case 0x11:
434 case 0x40:
435 size = 40;
436 break;
437 }
438 return size;
439}
440
441/****************************************************************************
442 *
443 * The functions below belong to the first pass of AMD IOMMU ACPI table
444 * parsing. In this pass we try to find out the highest device id this
445 * code has to handle. Upon this information the size of the shared data
446 * structures is determined later.
447 *
448 ****************************************************************************/
449
450/*
451 * This function calculates the length of a given IVHD entry
452 */
453static inline int ivhd_entry_length(u8 *ivhd)
454{
455 u32 type = ((struct ivhd_entry *)ivhd)->type;
456
457 if (type < 0x80) {
458 return 0x04 << (*ivhd >> 6);
459 } else if (type == IVHD_DEV_ACPI_HID) {
460 /* For ACPI_HID, offset 21 is uid len */
461 return *((u8 *)ivhd + 21) + 22;
462 }
463 return 0;
464}
465
466/*
467 * After reading the highest device id from the IOMMU PCI capability header
468 * this function looks if there is a higher device id defined in the ACPI table
469 */
470static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
471{
472 u8 *p = (void *)h, *end = (void *)h;
473 struct ivhd_entry *dev;
474
475 u32 ivhd_size = get_ivhd_header_size(h);
476
477 if (!ivhd_size) {
478 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
479 return -EINVAL;
480 }
481
482 p += ivhd_size;
483 end += h->length;
484
485 while (p < end) {
486 dev = (struct ivhd_entry *)p;
487 switch (dev->type) {
488 case IVHD_DEV_ALL:
489 /* Use maximum BDF value for DEV_ALL */
490 update_last_devid(0xffff);
491 break;
492 case IVHD_DEV_SELECT:
493 case IVHD_DEV_RANGE_END:
494 case IVHD_DEV_ALIAS:
495 case IVHD_DEV_EXT_SELECT:
496 /* all the above subfield types refer to device ids */
497 update_last_devid(dev->devid);
498 break;
499 default:
500 break;
501 }
502 p += ivhd_entry_length(p);
503 }
504
505 WARN_ON(p != end);
506
507 return 0;
508}
509
510static int __init check_ivrs_checksum(struct acpi_table_header *table)
511{
512 int i;
513 u8 checksum = 0, *p = (u8 *)table;
514
515 for (i = 0; i < table->length; ++i)
516 checksum += p[i];
517 if (checksum != 0) {
518 /* ACPI table corrupt */
519 pr_err(FW_BUG "AMD-Vi: IVRS invalid checksum\n");
520 return -ENODEV;
521 }
522
523 return 0;
524}
525
526/*
527 * Iterate over all IVHD entries in the ACPI table and find the highest device
528 * id which we need to handle. This is the first of three functions which parse
529 * the ACPI table. So we check the checksum here.
530 */
531static int __init find_last_devid_acpi(struct acpi_table_header *table)
532{
533 u8 *p = (u8 *)table, *end = (u8 *)table;
534 struct ivhd_header *h;
535
536 p += IVRS_HEADER_LENGTH;
537
538 end += table->length;
539 while (p < end) {
540 h = (struct ivhd_header *)p;
541 if (h->type == amd_iommu_target_ivhd_type) {
542 int ret = find_last_devid_from_ivhd(h);
543
544 if (ret)
545 return ret;
546 }
547 p += h->length;
548 }
549 WARN_ON(p != end);
550
551 return 0;
552}
553
554/****************************************************************************
555 *
556 * The following functions belong to the code path which parses the ACPI table
557 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
558 * data structures, initialize the device/alias/rlookup table and also
559 * basically initialize the hardware.
560 *
561 ****************************************************************************/
562
563/*
564 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
565 * write commands to that buffer later and the IOMMU will execute them
566 * asynchronously
567 */
568static int __init alloc_command_buffer(struct amd_iommu *iommu)
569{
570 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
571 get_order(CMD_BUFFER_SIZE));
572
573 return iommu->cmd_buf ? 0 : -ENOMEM;
574}
575
576/*
577 * This function resets the command buffer if the IOMMU stopped fetching
578 * commands from it.
579 */
580void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
581{
582 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
583
584 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
585 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
586
587 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
588}
589
590/*
591 * This function writes the command buffer address to the hardware and
592 * enables it.
593 */
594static void iommu_enable_command_buffer(struct amd_iommu *iommu)
595{
596 u64 entry;
597
598 BUG_ON(iommu->cmd_buf == NULL);
599
600 entry = (u64)virt_to_phys(iommu->cmd_buf);
601 entry |= MMIO_CMD_SIZE_512;
602
603 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
604 &entry, sizeof(entry));
605
606 amd_iommu_reset_cmd_buffer(iommu);
607}
608
609static void __init free_command_buffer(struct amd_iommu *iommu)
610{
611 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
612}
613
614/* allocates the memory where the IOMMU will log its events to */
615static int __init alloc_event_buffer(struct amd_iommu *iommu)
616{
617 iommu->evt_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
618 get_order(EVT_BUFFER_SIZE));
619
620 return iommu->evt_buf ? 0 : -ENOMEM;
621}
622
623static void iommu_enable_event_buffer(struct amd_iommu *iommu)
624{
625 u64 entry;
626
627 BUG_ON(iommu->evt_buf == NULL);
628
629 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
630
631 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
632 &entry, sizeof(entry));
633
634 /* set head and tail to zero manually */
635 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
636 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
637
638 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
639}
640
641static void __init free_event_buffer(struct amd_iommu *iommu)
642{
643 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
644}
645
646/* allocates the memory where the IOMMU will log its events to */
647static int __init alloc_ppr_log(struct amd_iommu *iommu)
648{
649 iommu->ppr_log = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
650 get_order(PPR_LOG_SIZE));
651
652 return iommu->ppr_log ? 0 : -ENOMEM;
653}
654
655static void iommu_enable_ppr_log(struct amd_iommu *iommu)
656{
657 u64 entry;
658
659 if (iommu->ppr_log == NULL)
660 return;
661
662 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
663
664 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
665 &entry, sizeof(entry));
666
667 /* set head and tail to zero manually */
668 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
669 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
670
671 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
672 iommu_feature_enable(iommu, CONTROL_PPR_EN);
673}
674
675static void __init free_ppr_log(struct amd_iommu *iommu)
676{
677 if (iommu->ppr_log == NULL)
678 return;
679
680 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
681}
682
683static void free_ga_log(struct amd_iommu *iommu)
684{
685#ifdef CONFIG_IRQ_REMAP
686 if (iommu->ga_log)
687 free_pages((unsigned long)iommu->ga_log,
688 get_order(GA_LOG_SIZE));
689 if (iommu->ga_log_tail)
690 free_pages((unsigned long)iommu->ga_log_tail,
691 get_order(8));
692#endif
693}
694
695static int iommu_ga_log_enable(struct amd_iommu *iommu)
696{
697#ifdef CONFIG_IRQ_REMAP
698 u32 status, i;
699
700 if (!iommu->ga_log)
701 return -EINVAL;
702
703 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
704
705 /* Check if already running */
706 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
707 return 0;
708
709 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
710 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
711
712 for (i = 0; i < LOOP_TIMEOUT; ++i) {
713 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
714 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
715 break;
716 }
717
718 if (i >= LOOP_TIMEOUT)
719 return -EINVAL;
720#endif /* CONFIG_IRQ_REMAP */
721 return 0;
722}
723
724#ifdef CONFIG_IRQ_REMAP
725static int iommu_init_ga_log(struct amd_iommu *iommu)
726{
727 u64 entry;
728
729 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
730 return 0;
731
732 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
733 get_order(GA_LOG_SIZE));
734 if (!iommu->ga_log)
735 goto err_out;
736
737 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
738 get_order(8));
739 if (!iommu->ga_log_tail)
740 goto err_out;
741
742 entry = (u64)virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
743 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
744 &entry, sizeof(entry));
745 entry = ((u64)virt_to_phys(iommu->ga_log) & 0xFFFFFFFFFFFFFULL) & ~7ULL;
746 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
747 &entry, sizeof(entry));
748 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
749 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
750
751 return 0;
752err_out:
753 free_ga_log(iommu);
754 return -EINVAL;
755}
756#endif /* CONFIG_IRQ_REMAP */
757
758static int iommu_init_ga(struct amd_iommu *iommu)
759{
760 int ret = 0;
761
762#ifdef CONFIG_IRQ_REMAP
763 /* Note: We have already checked GASup from IVRS table.
764 * Now, we need to make sure that GAMSup is set.
765 */
766 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
767 !iommu_feature(iommu, FEATURE_GAM_VAPIC))
768 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
769
770 ret = iommu_init_ga_log(iommu);
771#endif /* CONFIG_IRQ_REMAP */
772
773 return ret;
774}
775
776static void iommu_enable_gt(struct amd_iommu *iommu)
777{
778 if (!iommu_feature(iommu, FEATURE_GT))
779 return;
780
781 iommu_feature_enable(iommu, CONTROL_GT_EN);
782}
783
784/* sets a specific bit in the device table entry. */
785static void set_dev_entry_bit(u16 devid, u8 bit)
786{
787 int i = (bit >> 6) & 0x03;
788 int _bit = bit & 0x3f;
789
790 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
791}
792
793static int get_dev_entry_bit(u16 devid, u8 bit)
794{
795 int i = (bit >> 6) & 0x03;
796 int _bit = bit & 0x3f;
797
798 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
799}
800
801
802void amd_iommu_apply_erratum_63(u16 devid)
803{
804 int sysmgt;
805
806 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
807 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
808
809 if (sysmgt == 0x01)
810 set_dev_entry_bit(devid, DEV_ENTRY_IW);
811}
812
813/* Writes the specific IOMMU for a device into the rlookup table */
814static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
815{
816 amd_iommu_rlookup_table[devid] = iommu;
817}
818
819/*
820 * This function takes the device specific flags read from the ACPI
821 * table and sets up the device table entry with that information
822 */
823static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
824 u16 devid, u32 flags, u32 ext_flags)
825{
826 if (flags & ACPI_DEVFLAG_INITPASS)
827 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
828 if (flags & ACPI_DEVFLAG_EXTINT)
829 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
830 if (flags & ACPI_DEVFLAG_NMI)
831 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
832 if (flags & ACPI_DEVFLAG_SYSMGT1)
833 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
834 if (flags & ACPI_DEVFLAG_SYSMGT2)
835 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
836 if (flags & ACPI_DEVFLAG_LINT0)
837 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
838 if (flags & ACPI_DEVFLAG_LINT1)
839 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
840
841 amd_iommu_apply_erratum_63(devid);
842
843 set_iommu_for_device(iommu, devid);
844}
845
846static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
847{
848 struct devid_map *entry;
849 struct list_head *list;
850
851 if (type == IVHD_SPECIAL_IOAPIC)
852 list = &ioapic_map;
853 else if (type == IVHD_SPECIAL_HPET)
854 list = &hpet_map;
855 else
856 return -EINVAL;
857
858 list_for_each_entry(entry, list, list) {
859 if (!(entry->id == id && entry->cmd_line))
860 continue;
861
862 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
863 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
864
865 *devid = entry->devid;
866
867 return 0;
868 }
869
870 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
871 if (!entry)
872 return -ENOMEM;
873
874 entry->id = id;
875 entry->devid = *devid;
876 entry->cmd_line = cmd_line;
877
878 list_add_tail(&entry->list, list);
879
880 return 0;
881}
882
883static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u16 *devid,
884 bool cmd_line)
885{
886 struct acpihid_map_entry *entry;
887 struct list_head *list = &acpihid_map;
888
889 list_for_each_entry(entry, list, list) {
890 if (strcmp(entry->hid, hid) ||
891 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
892 !entry->cmd_line)
893 continue;
894
895 pr_info("AMD-Vi: Command-line override for hid:%s uid:%s\n",
896 hid, uid);
897 *devid = entry->devid;
898 return 0;
899 }
900
901 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
902 if (!entry)
903 return -ENOMEM;
904
905 memcpy(entry->uid, uid, strlen(uid));
906 memcpy(entry->hid, hid, strlen(hid));
907 entry->devid = *devid;
908 entry->cmd_line = cmd_line;
909 entry->root_devid = (entry->devid & (~0x7));
910
911 pr_info("AMD-Vi:%s, add hid:%s, uid:%s, rdevid:%d\n",
912 entry->cmd_line ? "cmd" : "ivrs",
913 entry->hid, entry->uid, entry->root_devid);
914
915 list_add_tail(&entry->list, list);
916 return 0;
917}
918
919static int __init add_early_maps(void)
920{
921 int i, ret;
922
923 for (i = 0; i < early_ioapic_map_size; ++i) {
924 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
925 early_ioapic_map[i].id,
926 &early_ioapic_map[i].devid,
927 early_ioapic_map[i].cmd_line);
928 if (ret)
929 return ret;
930 }
931
932 for (i = 0; i < early_hpet_map_size; ++i) {
933 ret = add_special_device(IVHD_SPECIAL_HPET,
934 early_hpet_map[i].id,
935 &early_hpet_map[i].devid,
936 early_hpet_map[i].cmd_line);
937 if (ret)
938 return ret;
939 }
940
941 for (i = 0; i < early_acpihid_map_size; ++i) {
942 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
943 early_acpihid_map[i].uid,
944 &early_acpihid_map[i].devid,
945 early_acpihid_map[i].cmd_line);
946 if (ret)
947 return ret;
948 }
949
950 return 0;
951}
952
953/*
954 * Reads the device exclusion range from ACPI and initializes the IOMMU with
955 * it
956 */
957static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
958{
959 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
960
961 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
962 return;
963
964 if (iommu) {
965 /*
966 * We only can configure exclusion ranges per IOMMU, not
967 * per device. But we can enable the exclusion range per
968 * device. This is done here
969 */
970 set_dev_entry_bit(devid, DEV_ENTRY_EX);
971 iommu->exclusion_start = m->range_start;
972 iommu->exclusion_length = m->range_length;
973 }
974}
975
976/*
977 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
978 * initializes the hardware and our data structures with it.
979 */
980static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
981 struct ivhd_header *h)
982{
983 u8 *p = (u8 *)h;
984 u8 *end = p, flags = 0;
985 u16 devid = 0, devid_start = 0, devid_to = 0;
986 u32 dev_i, ext_flags = 0;
987 bool alias = false;
988 struct ivhd_entry *e;
989 u32 ivhd_size;
990 int ret;
991
992
993 ret = add_early_maps();
994 if (ret)
995 return ret;
996
997 /*
998 * First save the recommended feature enable bits from ACPI
999 */
1000 iommu->acpi_flags = h->flags;
1001
1002 /*
1003 * Done. Now parse the device entries
1004 */
1005 ivhd_size = get_ivhd_header_size(h);
1006 if (!ivhd_size) {
1007 pr_err("AMD-Vi: Unsupported IVHD type %#x\n", h->type);
1008 return -EINVAL;
1009 }
1010
1011 p += ivhd_size;
1012
1013 end += h->length;
1014
1015
1016 while (p < end) {
1017 e = (struct ivhd_entry *)p;
1018 switch (e->type) {
1019 case IVHD_DEV_ALL:
1020
1021 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1022
1023 for (dev_i = 0; dev_i <= amd_iommu_last_bdf; ++dev_i)
1024 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1025 break;
1026 case IVHD_DEV_SELECT:
1027
1028 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
1029 "flags: %02x\n",
1030 PCI_BUS_NUM(e->devid),
1031 PCI_SLOT(e->devid),
1032 PCI_FUNC(e->devid),
1033 e->flags);
1034
1035 devid = e->devid;
1036 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1037 break;
1038 case IVHD_DEV_SELECT_RANGE_START:
1039
1040 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1041 "devid: %02x:%02x.%x flags: %02x\n",
1042 PCI_BUS_NUM(e->devid),
1043 PCI_SLOT(e->devid),
1044 PCI_FUNC(e->devid),
1045 e->flags);
1046
1047 devid_start = e->devid;
1048 flags = e->flags;
1049 ext_flags = 0;
1050 alias = false;
1051 break;
1052 case IVHD_DEV_ALIAS:
1053
1054 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
1055 "flags: %02x devid_to: %02x:%02x.%x\n",
1056 PCI_BUS_NUM(e->devid),
1057 PCI_SLOT(e->devid),
1058 PCI_FUNC(e->devid),
1059 e->flags,
1060 PCI_BUS_NUM(e->ext >> 8),
1061 PCI_SLOT(e->ext >> 8),
1062 PCI_FUNC(e->ext >> 8));
1063
1064 devid = e->devid;
1065 devid_to = e->ext >> 8;
1066 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1067 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1068 amd_iommu_alias_table[devid] = devid_to;
1069 break;
1070 case IVHD_DEV_ALIAS_RANGE:
1071
1072 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1073 "devid: %02x:%02x.%x flags: %02x "
1074 "devid_to: %02x:%02x.%x\n",
1075 PCI_BUS_NUM(e->devid),
1076 PCI_SLOT(e->devid),
1077 PCI_FUNC(e->devid),
1078 e->flags,
1079 PCI_BUS_NUM(e->ext >> 8),
1080 PCI_SLOT(e->ext >> 8),
1081 PCI_FUNC(e->ext >> 8));
1082
1083 devid_start = e->devid;
1084 flags = e->flags;
1085 devid_to = e->ext >> 8;
1086 ext_flags = 0;
1087 alias = true;
1088 break;
1089 case IVHD_DEV_EXT_SELECT:
1090
1091 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
1092 "flags: %02x ext: %08x\n",
1093 PCI_BUS_NUM(e->devid),
1094 PCI_SLOT(e->devid),
1095 PCI_FUNC(e->devid),
1096 e->flags, e->ext);
1097
1098 devid = e->devid;
1099 set_dev_entry_from_acpi(iommu, devid, e->flags,
1100 e->ext);
1101 break;
1102 case IVHD_DEV_EXT_SELECT_RANGE:
1103
1104 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1105 "%02x:%02x.%x flags: %02x ext: %08x\n",
1106 PCI_BUS_NUM(e->devid),
1107 PCI_SLOT(e->devid),
1108 PCI_FUNC(e->devid),
1109 e->flags, e->ext);
1110
1111 devid_start = e->devid;
1112 flags = e->flags;
1113 ext_flags = e->ext;
1114 alias = false;
1115 break;
1116 case IVHD_DEV_RANGE_END:
1117
1118 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
1119 PCI_BUS_NUM(e->devid),
1120 PCI_SLOT(e->devid),
1121 PCI_FUNC(e->devid));
1122
1123 devid = e->devid;
1124 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1125 if (alias) {
1126 amd_iommu_alias_table[dev_i] = devid_to;
1127 set_dev_entry_from_acpi(iommu,
1128 devid_to, flags, ext_flags);
1129 }
1130 set_dev_entry_from_acpi(iommu, dev_i,
1131 flags, ext_flags);
1132 }
1133 break;
1134 case IVHD_DEV_SPECIAL: {
1135 u8 handle, type;
1136 const char *var;
1137 u16 devid;
1138 int ret;
1139
1140 handle = e->ext & 0xff;
1141 devid = (e->ext >> 8) & 0xffff;
1142 type = (e->ext >> 24) & 0xff;
1143
1144 if (type == IVHD_SPECIAL_IOAPIC)
1145 var = "IOAPIC";
1146 else if (type == IVHD_SPECIAL_HPET)
1147 var = "HPET";
1148 else
1149 var = "UNKNOWN";
1150
1151 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
1152 var, (int)handle,
1153 PCI_BUS_NUM(devid),
1154 PCI_SLOT(devid),
1155 PCI_FUNC(devid));
1156
1157 ret = add_special_device(type, handle, &devid, false);
1158 if (ret)
1159 return ret;
1160
1161 /*
1162 * add_special_device might update the devid in case a
1163 * command-line override is present. So call
1164 * set_dev_entry_from_acpi after add_special_device.
1165 */
1166 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1167
1168 break;
1169 }
1170 case IVHD_DEV_ACPI_HID: {
1171 u16 devid;
1172 u8 hid[ACPIHID_HID_LEN] = {0};
1173 u8 uid[ACPIHID_UID_LEN] = {0};
1174 int ret;
1175
1176 if (h->type != 0x40) {
1177 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1178 e->type);
1179 break;
1180 }
1181
1182 memcpy(hid, (u8 *)(&e->ext), ACPIHID_HID_LEN - 1);
1183 hid[ACPIHID_HID_LEN - 1] = '\0';
1184
1185 if (!(*hid)) {
1186 pr_err(FW_BUG "Invalid HID.\n");
1187 break;
1188 }
1189
1190 switch (e->uidf) {
1191 case UID_NOT_PRESENT:
1192
1193 if (e->uidl != 0)
1194 pr_warn(FW_BUG "Invalid UID length.\n");
1195
1196 break;
1197 case UID_IS_INTEGER:
1198
1199 sprintf(uid, "%d", e->uid);
1200
1201 break;
1202 case UID_IS_CHARACTER:
1203
1204 memcpy(uid, (u8 *)(&e->uid), ACPIHID_UID_LEN - 1);
1205 uid[ACPIHID_UID_LEN - 1] = '\0';
1206
1207 break;
1208 default:
1209 break;
1210 }
1211
1212 devid = e->devid;
1213 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %02x:%02x.%x\n",
1214 hid, uid,
1215 PCI_BUS_NUM(devid),
1216 PCI_SLOT(devid),
1217 PCI_FUNC(devid));
1218
1219 flags = e->flags;
1220
1221 ret = add_acpi_hid_device(hid, uid, &devid, false);
1222 if (ret)
1223 return ret;
1224
1225 /*
1226 * add_special_device might update the devid in case a
1227 * command-line override is present. So call
1228 * set_dev_entry_from_acpi after add_special_device.
1229 */
1230 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1231
1232 break;
1233 }
1234 default:
1235 break;
1236 }
1237
1238 p += ivhd_entry_length(p);
1239 }
1240
1241 return 0;
1242}
1243
1244static void __init free_iommu_one(struct amd_iommu *iommu)
1245{
1246 free_command_buffer(iommu);
1247 free_event_buffer(iommu);
1248 free_ppr_log(iommu);
1249 free_ga_log(iommu);
1250 iommu_unmap_mmio_space(iommu);
1251}
1252
1253static void __init free_iommu_all(void)
1254{
1255 struct amd_iommu *iommu, *next;
1256
1257 for_each_iommu_safe(iommu, next) {
1258 list_del(&iommu->list);
1259 free_iommu_one(iommu);
1260 kfree(iommu);
1261 }
1262}
1263
1264/*
1265 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1266 * Workaround:
1267 * BIOS should disable L2B micellaneous clock gating by setting
1268 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1269 */
1270static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1271{
1272 u32 value;
1273
1274 if ((boot_cpu_data.x86 != 0x15) ||
1275 (boot_cpu_data.x86_model < 0x10) ||
1276 (boot_cpu_data.x86_model > 0x1f))
1277 return;
1278
1279 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1280 pci_read_config_dword(iommu->dev, 0xf4, &value);
1281
1282 if (value & BIT(2))
1283 return;
1284
1285 /* Select NB indirect register 0x90 and enable writing */
1286 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1287
1288 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1289 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1290 dev_name(&iommu->dev->dev));
1291
1292 /* Clear the enable writing bit */
1293 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1294}
1295
1296/*
1297 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1298 * Workaround:
1299 * BIOS should enable ATS write permission check by setting
1300 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1301 */
1302static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1303{
1304 u32 value;
1305
1306 if ((boot_cpu_data.x86 != 0x15) ||
1307 (boot_cpu_data.x86_model < 0x30) ||
1308 (boot_cpu_data.x86_model > 0x3f))
1309 return;
1310
1311 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1312 value = iommu_read_l2(iommu, 0x47);
1313
1314 if (value & BIT(0))
1315 return;
1316
1317 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1318 iommu_write_l2(iommu, 0x47, value | BIT(0));
1319
1320 pr_info("AMD-Vi: Applying ATS write check workaround for IOMMU at %s\n",
1321 dev_name(&iommu->dev->dev));
1322}
1323
1324/*
1325 * This function clues the initialization function for one IOMMU
1326 * together and also allocates the command buffer and programs the
1327 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1328 */
1329static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1330{
1331 int ret;
1332
1333 spin_lock_init(&iommu->lock);
1334
1335 /* Add IOMMU to internal data structures */
1336 list_add_tail(&iommu->list, &amd_iommu_list);
1337 iommu->index = amd_iommus_present++;
1338
1339 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1340 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1341 return -ENOSYS;
1342 }
1343
1344 /* Index is fine - add IOMMU to the array */
1345 amd_iommus[iommu->index] = iommu;
1346
1347 /*
1348 * Copy data from ACPI table entry to the iommu struct
1349 */
1350 iommu->devid = h->devid;
1351 iommu->cap_ptr = h->cap_ptr;
1352 iommu->pci_seg = h->pci_seg;
1353 iommu->mmio_phys = h->mmio_phys;
1354
1355 switch (h->type) {
1356 case 0x10:
1357 /* Check if IVHD EFR contains proper max banks/counters */
1358 if ((h->efr_attr != 0) &&
1359 ((h->efr_attr & (0xF << 13)) != 0) &&
1360 ((h->efr_attr & (0x3F << 17)) != 0))
1361 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1362 else
1363 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1364 if (((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1365 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1366 break;
1367 case 0x11:
1368 case 0x40:
1369 if (h->efr_reg & (1 << 9))
1370 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1371 else
1372 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1373 if (((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0))
1374 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1375 break;
1376 default:
1377 return -EINVAL;
1378 }
1379
1380 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1381 iommu->mmio_phys_end);
1382 if (!iommu->mmio_base)
1383 return -ENOMEM;
1384
1385 if (alloc_command_buffer(iommu))
1386 return -ENOMEM;
1387
1388 if (alloc_event_buffer(iommu))
1389 return -ENOMEM;
1390
1391 iommu->int_enabled = false;
1392
1393 ret = init_iommu_from_acpi(iommu, h);
1394 if (ret)
1395 return ret;
1396
1397 ret = amd_iommu_create_irq_domain(iommu);
1398 if (ret)
1399 return ret;
1400
1401 /*
1402 * Make sure IOMMU is not considered to translate itself. The IVRS
1403 * table tells us so, but this is a lie!
1404 */
1405 amd_iommu_rlookup_table[iommu->devid] = NULL;
1406
1407 return 0;
1408}
1409
1410/**
1411 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1412 * @ivrs Pointer to the IVRS header
1413 *
1414 * This function search through all IVDB of the maximum supported IVHD
1415 */
1416static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1417{
1418 u8 *base = (u8 *)ivrs;
1419 struct ivhd_header *ivhd = (struct ivhd_header *)
1420 (base + IVRS_HEADER_LENGTH);
1421 u8 last_type = ivhd->type;
1422 u16 devid = ivhd->devid;
1423
1424 while (((u8 *)ivhd - base < ivrs->length) &&
1425 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1426 u8 *p = (u8 *) ivhd;
1427
1428 if (ivhd->devid == devid)
1429 last_type = ivhd->type;
1430 ivhd = (struct ivhd_header *)(p + ivhd->length);
1431 }
1432
1433 return last_type;
1434}
1435
1436/*
1437 * Iterates over all IOMMU entries in the ACPI table, allocates the
1438 * IOMMU structure and initializes it with init_iommu_one()
1439 */
1440static int __init init_iommu_all(struct acpi_table_header *table)
1441{
1442 u8 *p = (u8 *)table, *end = (u8 *)table;
1443 struct ivhd_header *h;
1444 struct amd_iommu *iommu;
1445 int ret;
1446
1447 end += table->length;
1448 p += IVRS_HEADER_LENGTH;
1449
1450 while (p < end) {
1451 h = (struct ivhd_header *)p;
1452 if (*p == amd_iommu_target_ivhd_type) {
1453
1454 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
1455 "seg: %d flags: %01x info %04x\n",
1456 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
1457 PCI_FUNC(h->devid), h->cap_ptr,
1458 h->pci_seg, h->flags, h->info);
1459 DUMP_printk(" mmio-addr: %016llx\n",
1460 h->mmio_phys);
1461
1462 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1463 if (iommu == NULL)
1464 return -ENOMEM;
1465
1466 ret = init_iommu_one(iommu, h);
1467 if (ret)
1468 return ret;
1469 }
1470 p += h->length;
1471
1472 }
1473 WARN_ON(p != end);
1474
1475 return 0;
1476}
1477
1478
1479static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1480{
1481 u64 val = 0xabcd, val2 = 0;
1482
1483 if (!iommu_feature(iommu, FEATURE_PC))
1484 return;
1485
1486 amd_iommu_pc_present = true;
1487
1488 /* Check if the performance counters can be written to */
1489 if ((0 != iommu_pc_get_set_reg_val(iommu, 0, 0, 0, &val, true)) ||
1490 (0 != iommu_pc_get_set_reg_val(iommu, 0, 0, 0, &val2, false)) ||
1491 (val != val2)) {
1492 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1493 amd_iommu_pc_present = false;
1494 return;
1495 }
1496
1497 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1498
1499 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1500 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1501 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1502}
1503
1504static ssize_t amd_iommu_show_cap(struct device *dev,
1505 struct device_attribute *attr,
1506 char *buf)
1507{
1508 struct amd_iommu *iommu = dev_get_drvdata(dev);
1509 return sprintf(buf, "%x\n", iommu->cap);
1510}
1511static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1512
1513static ssize_t amd_iommu_show_features(struct device *dev,
1514 struct device_attribute *attr,
1515 char *buf)
1516{
1517 struct amd_iommu *iommu = dev_get_drvdata(dev);
1518 return sprintf(buf, "%llx\n", iommu->features);
1519}
1520static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1521
1522static struct attribute *amd_iommu_attrs[] = {
1523 &dev_attr_cap.attr,
1524 &dev_attr_features.attr,
1525 NULL,
1526};
1527
1528static struct attribute_group amd_iommu_group = {
1529 .name = "amd-iommu",
1530 .attrs = amd_iommu_attrs,
1531};
1532
1533static const struct attribute_group *amd_iommu_groups[] = {
1534 &amd_iommu_group,
1535 NULL,
1536};
1537
1538static int iommu_init_pci(struct amd_iommu *iommu)
1539{
1540 int cap_ptr = iommu->cap_ptr;
1541 u32 range, misc, low, high;
1542 int ret;
1543
1544 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
1545 iommu->devid & 0xff);
1546 if (!iommu->dev)
1547 return -ENODEV;
1548
1549 /* Prevent binding other PCI device drivers to IOMMU devices */
1550 iommu->dev->match_driver = false;
1551
1552 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1553 &iommu->cap);
1554 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1555 &range);
1556 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1557 &misc);
1558
1559 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1560 amd_iommu_iotlb_sup = false;
1561
1562 /* read extended feature bits */
1563 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1564 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1565
1566 iommu->features = ((u64)high << 32) | low;
1567
1568 if (iommu_feature(iommu, FEATURE_GT)) {
1569 int glxval;
1570 u32 max_pasid;
1571 u64 pasmax;
1572
1573 pasmax = iommu->features & FEATURE_PASID_MASK;
1574 pasmax >>= FEATURE_PASID_SHIFT;
1575 max_pasid = (1 << (pasmax + 1)) - 1;
1576
1577 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1578
1579 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
1580
1581 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1582 glxval >>= FEATURE_GLXVAL_SHIFT;
1583
1584 if (amd_iommu_max_glx_val == -1)
1585 amd_iommu_max_glx_val = glxval;
1586 else
1587 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1588 }
1589
1590 if (iommu_feature(iommu, FEATURE_GT) &&
1591 iommu_feature(iommu, FEATURE_PPR)) {
1592 iommu->is_iommu_v2 = true;
1593 amd_iommu_v2_present = true;
1594 }
1595
1596 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu))
1597 return -ENOMEM;
1598
1599 ret = iommu_init_ga(iommu);
1600 if (ret)
1601 return ret;
1602
1603 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1604 amd_iommu_np_cache = true;
1605
1606 init_iommu_perf_ctr(iommu);
1607
1608 if (is_rd890_iommu(iommu->dev)) {
1609 int i, j;
1610
1611 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1612 PCI_DEVFN(0, 0));
1613
1614 /*
1615 * Some rd890 systems may not be fully reconfigured by the
1616 * BIOS, so it's necessary for us to store this information so
1617 * it can be reprogrammed on resume
1618 */
1619 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1620 &iommu->stored_addr_lo);
1621 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1622 &iommu->stored_addr_hi);
1623
1624 /* Low bit locks writes to configuration space */
1625 iommu->stored_addr_lo &= ~1;
1626
1627 for (i = 0; i < 6; i++)
1628 for (j = 0; j < 0x12; j++)
1629 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1630
1631 for (i = 0; i < 0x83; i++)
1632 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1633 }
1634
1635 amd_iommu_erratum_746_workaround(iommu);
1636 amd_iommu_ats_write_check_workaround(iommu);
1637
1638 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1639 amd_iommu_groups, "ivhd%d",
1640 iommu->index);
1641
1642 return pci_enable_device(iommu->dev);
1643}
1644
1645static void print_iommu_info(void)
1646{
1647 static const char * const feat_str[] = {
1648 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1649 "IA", "GA", "HE", "PC"
1650 };
1651 struct amd_iommu *iommu;
1652
1653 for_each_iommu(iommu) {
1654 int i;
1655
1656 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1657 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1658
1659 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1660 pr_info("AMD-Vi: Extended features (%#llx):\n",
1661 iommu->features);
1662 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1663 if (iommu_feature(iommu, (1ULL << i)))
1664 pr_cont(" %s", feat_str[i]);
1665 }
1666
1667 if (iommu->features & FEATURE_GAM_VAPIC)
1668 pr_cont(" GA_vAPIC");
1669
1670 pr_cont("\n");
1671 }
1672 }
1673 if (irq_remapping_enabled) {
1674 pr_info("AMD-Vi: Interrupt remapping enabled\n");
1675 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
1676 pr_info("AMD-Vi: virtual APIC enabled\n");
1677 }
1678}
1679
1680static int __init amd_iommu_init_pci(void)
1681{
1682 struct amd_iommu *iommu;
1683 int ret = 0;
1684
1685 for_each_iommu(iommu) {
1686 ret = iommu_init_pci(iommu);
1687 if (ret)
1688 break;
1689 }
1690
1691 /*
1692 * Order is important here to make sure any unity map requirements are
1693 * fulfilled. The unity mappings are created and written to the device
1694 * table during the amd_iommu_init_api() call.
1695 *
1696 * After that we call init_device_table_dma() to make sure any
1697 * uninitialized DTE will block DMA, and in the end we flush the caches
1698 * of all IOMMUs to make sure the changes to the device table are
1699 * active.
1700 */
1701 ret = amd_iommu_init_api();
1702
1703 init_device_table_dma();
1704
1705 for_each_iommu(iommu)
1706 iommu_flush_all_caches(iommu);
1707
1708 if (!ret)
1709 print_iommu_info();
1710
1711 return ret;
1712}
1713
1714/****************************************************************************
1715 *
1716 * The following functions initialize the MSI interrupts for all IOMMUs
1717 * in the system. It's a bit challenging because there could be multiple
1718 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1719 * pci_dev.
1720 *
1721 ****************************************************************************/
1722
1723static int iommu_setup_msi(struct amd_iommu *iommu)
1724{
1725 int r;
1726
1727 r = pci_enable_msi(iommu->dev);
1728 if (r)
1729 return r;
1730
1731 r = request_threaded_irq(iommu->dev->irq,
1732 amd_iommu_int_handler,
1733 amd_iommu_int_thread,
1734 0, "AMD-Vi",
1735 iommu);
1736
1737 if (r) {
1738 pci_disable_msi(iommu->dev);
1739 return r;
1740 }
1741
1742 iommu->int_enabled = true;
1743
1744 return 0;
1745}
1746
1747static int iommu_init_msi(struct amd_iommu *iommu)
1748{
1749 int ret;
1750
1751 if (iommu->int_enabled)
1752 goto enable_faults;
1753
1754 if (iommu->dev->msi_cap)
1755 ret = iommu_setup_msi(iommu);
1756 else
1757 ret = -ENODEV;
1758
1759 if (ret)
1760 return ret;
1761
1762enable_faults:
1763 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1764
1765 if (iommu->ppr_log != NULL)
1766 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1767
1768 iommu_ga_log_enable(iommu);
1769
1770 return 0;
1771}
1772
1773/****************************************************************************
1774 *
1775 * The next functions belong to the third pass of parsing the ACPI
1776 * table. In this last pass the memory mapping requirements are
1777 * gathered (like exclusion and unity mapping ranges).
1778 *
1779 ****************************************************************************/
1780
1781static void __init free_unity_maps(void)
1782{
1783 struct unity_map_entry *entry, *next;
1784
1785 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1786 list_del(&entry->list);
1787 kfree(entry);
1788 }
1789}
1790
1791/* called when we find an exclusion range definition in ACPI */
1792static int __init init_exclusion_range(struct ivmd_header *m)
1793{
1794 int i;
1795
1796 switch (m->type) {
1797 case ACPI_IVMD_TYPE:
1798 set_device_exclusion_range(m->devid, m);
1799 break;
1800 case ACPI_IVMD_TYPE_ALL:
1801 for (i = 0; i <= amd_iommu_last_bdf; ++i)
1802 set_device_exclusion_range(i, m);
1803 break;
1804 case ACPI_IVMD_TYPE_RANGE:
1805 for (i = m->devid; i <= m->aux; ++i)
1806 set_device_exclusion_range(i, m);
1807 break;
1808 default:
1809 break;
1810 }
1811
1812 return 0;
1813}
1814
1815/* called for unity map ACPI definition */
1816static int __init init_unity_map_range(struct ivmd_header *m)
1817{
1818 struct unity_map_entry *e = NULL;
1819 char *s;
1820
1821 e = kzalloc(sizeof(*e), GFP_KERNEL);
1822 if (e == NULL)
1823 return -ENOMEM;
1824
1825 switch (m->type) {
1826 default:
1827 kfree(e);
1828 return 0;
1829 case ACPI_IVMD_TYPE:
1830 s = "IVMD_TYPEi\t\t\t";
1831 e->devid_start = e->devid_end = m->devid;
1832 break;
1833 case ACPI_IVMD_TYPE_ALL:
1834 s = "IVMD_TYPE_ALL\t\t";
1835 e->devid_start = 0;
1836 e->devid_end = amd_iommu_last_bdf;
1837 break;
1838 case ACPI_IVMD_TYPE_RANGE:
1839 s = "IVMD_TYPE_RANGE\t\t";
1840 e->devid_start = m->devid;
1841 e->devid_end = m->aux;
1842 break;
1843 }
1844 e->address_start = PAGE_ALIGN(m->range_start);
1845 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1846 e->prot = m->flags >> 1;
1847
1848 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1849 " range_start: %016llx range_end: %016llx flags: %x\n", s,
1850 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1851 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
1852 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1853 e->address_start, e->address_end, m->flags);
1854
1855 list_add_tail(&e->list, &amd_iommu_unity_map);
1856
1857 return 0;
1858}
1859
1860/* iterates over all memory definitions we find in the ACPI table */
1861static int __init init_memory_definitions(struct acpi_table_header *table)
1862{
1863 u8 *p = (u8 *)table, *end = (u8 *)table;
1864 struct ivmd_header *m;
1865
1866 end += table->length;
1867 p += IVRS_HEADER_LENGTH;
1868
1869 while (p < end) {
1870 m = (struct ivmd_header *)p;
1871 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1872 init_exclusion_range(m);
1873 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1874 init_unity_map_range(m);
1875
1876 p += m->length;
1877 }
1878
1879 return 0;
1880}
1881
1882/*
1883 * Init the device table to not allow DMA access for devices and
1884 * suppress all page faults
1885 */
1886static void init_device_table_dma(void)
1887{
1888 u32 devid;
1889
1890 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1891 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1892 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1893 }
1894}
1895
1896static void __init uninit_device_table_dma(void)
1897{
1898 u32 devid;
1899
1900 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1901 amd_iommu_dev_table[devid].data[0] = 0ULL;
1902 amd_iommu_dev_table[devid].data[1] = 0ULL;
1903 }
1904}
1905
1906static void init_device_table(void)
1907{
1908 u32 devid;
1909
1910 if (!amd_iommu_irq_remap)
1911 return;
1912
1913 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1914 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1915}
1916
1917static void iommu_init_flags(struct amd_iommu *iommu)
1918{
1919 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1920 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1921 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1922
1923 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1924 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1925 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1926
1927 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1928 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1929 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1930
1931 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1932 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1933 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1934
1935 /*
1936 * make IOMMU memory accesses cache coherent
1937 */
1938 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1939
1940 /* Set IOTLB invalidation timeout to 1s */
1941 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1942}
1943
1944static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1945{
1946 int i, j;
1947 u32 ioc_feature_control;
1948 struct pci_dev *pdev = iommu->root_pdev;
1949
1950 /* RD890 BIOSes may not have completely reconfigured the iommu */
1951 if (!is_rd890_iommu(iommu->dev) || !pdev)
1952 return;
1953
1954 /*
1955 * First, we need to ensure that the iommu is enabled. This is
1956 * controlled by a register in the northbridge
1957 */
1958
1959 /* Select Northbridge indirect register 0x75 and enable writing */
1960 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1961 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1962
1963 /* Enable the iommu */
1964 if (!(ioc_feature_control & 0x1))
1965 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1966
1967 /* Restore the iommu BAR */
1968 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1969 iommu->stored_addr_lo);
1970 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1971 iommu->stored_addr_hi);
1972
1973 /* Restore the l1 indirect regs for each of the 6 l1s */
1974 for (i = 0; i < 6; i++)
1975 for (j = 0; j < 0x12; j++)
1976 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1977
1978 /* Restore the l2 indirect regs */
1979 for (i = 0; i < 0x83; i++)
1980 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1981
1982 /* Lock PCI setup registers */
1983 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1984 iommu->stored_addr_lo | 1);
1985}
1986
1987static void iommu_enable_ga(struct amd_iommu *iommu)
1988{
1989#ifdef CONFIG_IRQ_REMAP
1990 switch (amd_iommu_guest_ir) {
1991 case AMD_IOMMU_GUEST_IR_VAPIC:
1992 iommu_feature_enable(iommu, CONTROL_GAM_EN);
1993 /* Fall through */
1994 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
1995 iommu_feature_enable(iommu, CONTROL_GA_EN);
1996 iommu->irte_ops = &irte_128_ops;
1997 break;
1998 default:
1999 iommu->irte_ops = &irte_32_ops;
2000 break;
2001 }
2002#endif
2003}
2004
2005/*
2006 * This function finally enables all IOMMUs found in the system after
2007 * they have been initialized
2008 */
2009static void early_enable_iommus(void)
2010{
2011 struct amd_iommu *iommu;
2012
2013 for_each_iommu(iommu) {
2014 iommu_disable(iommu);
2015 iommu_init_flags(iommu);
2016 iommu_set_device_table(iommu);
2017 iommu_enable_command_buffer(iommu);
2018 iommu_enable_event_buffer(iommu);
2019 iommu_set_exclusion_range(iommu);
2020 iommu_enable_ga(iommu);
2021 iommu_enable(iommu);
2022 iommu_flush_all_caches(iommu);
2023 }
2024
2025#ifdef CONFIG_IRQ_REMAP
2026 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2027 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2028#endif
2029}
2030
2031static void enable_iommus_v2(void)
2032{
2033 struct amd_iommu *iommu;
2034
2035 for_each_iommu(iommu) {
2036 iommu_enable_ppr_log(iommu);
2037 iommu_enable_gt(iommu);
2038 }
2039}
2040
2041static void enable_iommus(void)
2042{
2043 early_enable_iommus();
2044
2045 enable_iommus_v2();
2046}
2047
2048static void disable_iommus(void)
2049{
2050 struct amd_iommu *iommu;
2051
2052 for_each_iommu(iommu)
2053 iommu_disable(iommu);
2054
2055#ifdef CONFIG_IRQ_REMAP
2056 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2057 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2058#endif
2059}
2060
2061/*
2062 * Suspend/Resume support
2063 * disable suspend until real resume implemented
2064 */
2065
2066static void amd_iommu_resume(void)
2067{
2068 struct amd_iommu *iommu;
2069
2070 for_each_iommu(iommu)
2071 iommu_apply_resume_quirks(iommu);
2072
2073 /* re-load the hardware */
2074 enable_iommus();
2075
2076 amd_iommu_enable_interrupts();
2077}
2078
2079static int amd_iommu_suspend(void)
2080{
2081 /* disable IOMMUs to go out of the way for BIOS */
2082 disable_iommus();
2083
2084 return 0;
2085}
2086
2087static struct syscore_ops amd_iommu_syscore_ops = {
2088 .suspend = amd_iommu_suspend,
2089 .resume = amd_iommu_resume,
2090};
2091
2092static void __init free_on_init_error(void)
2093{
2094 kmemleak_free(irq_lookup_table);
2095 free_pages((unsigned long)irq_lookup_table,
2096 get_order(rlookup_table_size));
2097
2098 kmem_cache_destroy(amd_iommu_irq_cache);
2099 amd_iommu_irq_cache = NULL;
2100
2101 free_pages((unsigned long)amd_iommu_rlookup_table,
2102 get_order(rlookup_table_size));
2103
2104 free_pages((unsigned long)amd_iommu_alias_table,
2105 get_order(alias_table_size));
2106
2107 free_pages((unsigned long)amd_iommu_dev_table,
2108 get_order(dev_table_size));
2109
2110 free_iommu_all();
2111
2112#ifdef CONFIG_GART_IOMMU
2113 /*
2114 * We failed to initialize the AMD IOMMU - try fallback to GART
2115 * if possible.
2116 */
2117 gart_iommu_init();
2118
2119#endif
2120}
2121
2122/* SB IOAPIC is always on this device in AMD systems */
2123#define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2124
2125static bool __init check_ioapic_information(void)
2126{
2127 const char *fw_bug = FW_BUG;
2128 bool ret, has_sb_ioapic;
2129 int idx;
2130
2131 has_sb_ioapic = false;
2132 ret = false;
2133
2134 /*
2135 * If we have map overrides on the kernel command line the
2136 * messages in this function might not describe firmware bugs
2137 * anymore - so be careful
2138 */
2139 if (cmdline_maps)
2140 fw_bug = "";
2141
2142 for (idx = 0; idx < nr_ioapics; idx++) {
2143 int devid, id = mpc_ioapic_id(idx);
2144
2145 devid = get_ioapic_devid(id);
2146 if (devid < 0) {
2147 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
2148 fw_bug, id);
2149 ret = false;
2150 } else if (devid == IOAPIC_SB_DEVID) {
2151 has_sb_ioapic = true;
2152 ret = true;
2153 }
2154 }
2155
2156 if (!has_sb_ioapic) {
2157 /*
2158 * We expect the SB IOAPIC to be listed in the IVRS
2159 * table. The system timer is connected to the SB IOAPIC
2160 * and if we don't have it in the list the system will
2161 * panic at boot time. This situation usually happens
2162 * when the BIOS is buggy and provides us the wrong
2163 * device id for the IOAPIC in the system.
2164 */
2165 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
2166 }
2167
2168 if (!ret)
2169 pr_err("AMD-Vi: Disabling interrupt remapping\n");
2170
2171 return ret;
2172}
2173
2174static void __init free_dma_resources(void)
2175{
2176 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
2177 get_order(MAX_DOMAIN_ID/8));
2178
2179 free_unity_maps();
2180}
2181
2182/*
2183 * This is the hardware init function for AMD IOMMU in the system.
2184 * This function is called either from amd_iommu_init or from the interrupt
2185 * remapping setup code.
2186 *
2187 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
2188 * four times:
2189 *
2190 * 1 pass) Discover the most comprehensive IVHD type to use.
2191 *
2192 * 2 pass) Find the highest PCI device id the driver has to handle.
2193 * Upon this information the size of the data structures is
2194 * determined that needs to be allocated.
2195 *
2196 * 3 pass) Initialize the data structures just allocated with the
2197 * information in the ACPI table about available AMD IOMMUs
2198 * in the system. It also maps the PCI devices in the
2199 * system to specific IOMMUs
2200 *
2201 * 4 pass) After the basic data structures are allocated and
2202 * initialized we update them with information about memory
2203 * remapping requirements parsed out of the ACPI table in
2204 * this last pass.
2205 *
2206 * After everything is set up the IOMMUs are enabled and the necessary
2207 * hotplug and suspend notifiers are registered.
2208 */
2209static int __init early_amd_iommu_init(void)
2210{
2211 struct acpi_table_header *ivrs_base;
2212 acpi_status status;
2213 int i, remap_cache_sz, ret = 0;
2214
2215 if (!amd_iommu_detected)
2216 return -ENODEV;
2217
2218 status = acpi_get_table("IVRS", 0, &ivrs_base);
2219 if (status == AE_NOT_FOUND)
2220 return -ENODEV;
2221 else if (ACPI_FAILURE(status)) {
2222 const char *err = acpi_format_exception(status);
2223 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2224 return -EINVAL;
2225 }
2226
2227 /*
2228 * Validate checksum here so we don't need to do it when
2229 * we actually parse the table
2230 */
2231 ret = check_ivrs_checksum(ivrs_base);
2232 if (ret)
2233 return ret;
2234
2235 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
2236 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
2237
2238 /*
2239 * First parse ACPI tables to find the largest Bus/Dev/Func
2240 * we need to handle. Upon this information the shared data
2241 * structures for the IOMMUs in the system will be allocated
2242 */
2243 ret = find_last_devid_acpi(ivrs_base);
2244 if (ret)
2245 goto out;
2246
2247 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
2248 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
2249 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
2250
2251 /* Device table - directly used by all IOMMUs */
2252 ret = -ENOMEM;
2253 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
2254 get_order(dev_table_size));
2255 if (amd_iommu_dev_table == NULL)
2256 goto out;
2257
2258 /*
2259 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
2260 * IOMMU see for that device
2261 */
2262 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
2263 get_order(alias_table_size));
2264 if (amd_iommu_alias_table == NULL)
2265 goto out;
2266
2267 /* IOMMU rlookup table - find the IOMMU for a specific device */
2268 amd_iommu_rlookup_table = (void *)__get_free_pages(
2269 GFP_KERNEL | __GFP_ZERO,
2270 get_order(rlookup_table_size));
2271 if (amd_iommu_rlookup_table == NULL)
2272 goto out;
2273
2274 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
2275 GFP_KERNEL | __GFP_ZERO,
2276 get_order(MAX_DOMAIN_ID/8));
2277 if (amd_iommu_pd_alloc_bitmap == NULL)
2278 goto out;
2279
2280 /*
2281 * let all alias entries point to itself
2282 */
2283 for (i = 0; i <= amd_iommu_last_bdf; ++i)
2284 amd_iommu_alias_table[i] = i;
2285
2286 /*
2287 * never allocate domain 0 because its used as the non-allocated and
2288 * error value placeholder
2289 */
2290 __set_bit(0, amd_iommu_pd_alloc_bitmap);
2291
2292 spin_lock_init(&amd_iommu_pd_lock);
2293
2294 /*
2295 * now the data structures are allocated and basically initialized
2296 * start the real acpi table scan
2297 */
2298 ret = init_iommu_all(ivrs_base);
2299 if (ret)
2300 goto out;
2301
2302 if (amd_iommu_irq_remap)
2303 amd_iommu_irq_remap = check_ioapic_information();
2304
2305 if (amd_iommu_irq_remap) {
2306 /*
2307 * Interrupt remapping enabled, create kmem_cache for the
2308 * remapping tables.
2309 */
2310 ret = -ENOMEM;
2311 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
2312 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
2313 else
2314 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
2315 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
2316 remap_cache_sz,
2317 IRQ_TABLE_ALIGNMENT,
2318 0, NULL);
2319 if (!amd_iommu_irq_cache)
2320 goto out;
2321
2322 irq_lookup_table = (void *)__get_free_pages(
2323 GFP_KERNEL | __GFP_ZERO,
2324 get_order(rlookup_table_size));
2325 kmemleak_alloc(irq_lookup_table, rlookup_table_size,
2326 1, GFP_KERNEL);
2327 if (!irq_lookup_table)
2328 goto out;
2329 }
2330
2331 ret = init_memory_definitions(ivrs_base);
2332 if (ret)
2333 goto out;
2334
2335 /* init the device table */
2336 init_device_table();
2337
2338out:
2339 /* Don't leak any ACPI memory */
2340 acpi_put_table(ivrs_base);
2341 ivrs_base = NULL;
2342
2343 return ret;
2344}
2345
2346static int amd_iommu_enable_interrupts(void)
2347{
2348 struct amd_iommu *iommu;
2349 int ret = 0;
2350
2351 for_each_iommu(iommu) {
2352 ret = iommu_init_msi(iommu);
2353 if (ret)
2354 goto out;
2355 }
2356
2357out:
2358 return ret;
2359}
2360
2361static bool detect_ivrs(void)
2362{
2363 struct acpi_table_header *ivrs_base;
2364 acpi_status status;
2365
2366 status = acpi_get_table("IVRS", 0, &ivrs_base);
2367 if (status == AE_NOT_FOUND)
2368 return false;
2369 else if (ACPI_FAILURE(status)) {
2370 const char *err = acpi_format_exception(status);
2371 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2372 return false;
2373 }
2374
2375 acpi_put_table(ivrs_base);
2376
2377 /* Make sure ACS will be enabled during PCI probe */
2378 pci_request_acs();
2379
2380 return true;
2381}
2382
2383/****************************************************************************
2384 *
2385 * AMD IOMMU Initialization State Machine
2386 *
2387 ****************************************************************************/
2388
2389static int __init state_next(void)
2390{
2391 int ret = 0;
2392
2393 switch (init_state) {
2394 case IOMMU_START_STATE:
2395 if (!detect_ivrs()) {
2396 init_state = IOMMU_NOT_FOUND;
2397 ret = -ENODEV;
2398 } else {
2399 init_state = IOMMU_IVRS_DETECTED;
2400 }
2401 break;
2402 case IOMMU_IVRS_DETECTED:
2403 ret = early_amd_iommu_init();
2404 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2405 break;
2406 case IOMMU_ACPI_FINISHED:
2407 early_enable_iommus();
2408 register_syscore_ops(&amd_iommu_syscore_ops);
2409 x86_platform.iommu_shutdown = disable_iommus;
2410 init_state = IOMMU_ENABLED;
2411 break;
2412 case IOMMU_ENABLED:
2413 ret = amd_iommu_init_pci();
2414 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2415 enable_iommus_v2();
2416 break;
2417 case IOMMU_PCI_INIT:
2418 ret = amd_iommu_enable_interrupts();
2419 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2420 break;
2421 case IOMMU_INTERRUPTS_EN:
2422 ret = amd_iommu_init_dma_ops();
2423 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2424 break;
2425 case IOMMU_DMA_OPS:
2426 init_state = IOMMU_INITIALIZED;
2427 break;
2428 case IOMMU_INITIALIZED:
2429 /* Nothing to do */
2430 break;
2431 case IOMMU_NOT_FOUND:
2432 case IOMMU_INIT_ERROR:
2433 /* Error states => do nothing */
2434 ret = -EINVAL;
2435 break;
2436 default:
2437 /* Unknown state */
2438 BUG();
2439 }
2440
2441 return ret;
2442}
2443
2444static int __init iommu_go_to_state(enum iommu_init_state state)
2445{
2446 int ret = 0;
2447
2448 while (init_state != state) {
2449 ret = state_next();
2450 if (init_state == IOMMU_NOT_FOUND ||
2451 init_state == IOMMU_INIT_ERROR)
2452 break;
2453 }
2454
2455 return ret;
2456}
2457
2458#ifdef CONFIG_IRQ_REMAP
2459int __init amd_iommu_prepare(void)
2460{
2461 int ret;
2462
2463 amd_iommu_irq_remap = true;
2464
2465 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2466 if (ret)
2467 return ret;
2468 return amd_iommu_irq_remap ? 0 : -ENODEV;
2469}
2470
2471int __init amd_iommu_enable(void)
2472{
2473 int ret;
2474
2475 ret = iommu_go_to_state(IOMMU_ENABLED);
2476 if (ret)
2477 return ret;
2478
2479 irq_remapping_enabled = 1;
2480
2481 return 0;
2482}
2483
2484void amd_iommu_disable(void)
2485{
2486 amd_iommu_suspend();
2487}
2488
2489int amd_iommu_reenable(int mode)
2490{
2491 amd_iommu_resume();
2492
2493 return 0;
2494}
2495
2496int __init amd_iommu_enable_faulting(void)
2497{
2498 /* We enable MSI later when PCI is initialized */
2499 return 0;
2500}
2501#endif
2502
2503/*
2504 * This is the core init function for AMD IOMMU hardware in the system.
2505 * This function is called from the generic x86 DMA layer initialization
2506 * code.
2507 */
2508static int __init amd_iommu_init(void)
2509{
2510 int ret;
2511
2512 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2513 if (ret) {
2514 free_dma_resources();
2515 if (!irq_remapping_enabled) {
2516 disable_iommus();
2517 free_on_init_error();
2518 } else {
2519 struct amd_iommu *iommu;
2520
2521 uninit_device_table_dma();
2522 for_each_iommu(iommu)
2523 iommu_flush_all_caches(iommu);
2524 }
2525 }
2526
2527 return ret;
2528}
2529
2530/****************************************************************************
2531 *
2532 * Early detect code. This code runs at IOMMU detection time in the DMA
2533 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2534 * IOMMUs
2535 *
2536 ****************************************************************************/
2537int __init amd_iommu_detect(void)
2538{
2539 int ret;
2540
2541 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
2542 return -ENODEV;
2543
2544 if (amd_iommu_disabled)
2545 return -ENODEV;
2546
2547 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2548 if (ret)
2549 return ret;
2550
2551 amd_iommu_detected = true;
2552 iommu_detected = 1;
2553 x86_init.iommu.iommu_init = amd_iommu_init;
2554
2555 return 1;
2556}
2557
2558/****************************************************************************
2559 *
2560 * Parsing functions for the AMD IOMMU specific kernel command line
2561 * options.
2562 *
2563 ****************************************************************************/
2564
2565static int __init parse_amd_iommu_dump(char *str)
2566{
2567 amd_iommu_dump = true;
2568
2569 return 1;
2570}
2571
2572static int __init parse_amd_iommu_intr(char *str)
2573{
2574 for (; *str; ++str) {
2575 if (strncmp(str, "legacy", 6) == 0) {
2576 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
2577 break;
2578 }
2579 if (strncmp(str, "vapic", 5) == 0) {
2580 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
2581 break;
2582 }
2583 }
2584 return 1;
2585}
2586
2587static int __init parse_amd_iommu_options(char *str)
2588{
2589 for (; *str; ++str) {
2590 if (strncmp(str, "fullflush", 9) == 0)
2591 amd_iommu_unmap_flush = true;
2592 if (strncmp(str, "off", 3) == 0)
2593 amd_iommu_disabled = true;
2594 if (strncmp(str, "force_isolation", 15) == 0)
2595 amd_iommu_force_isolation = true;
2596 }
2597
2598 return 1;
2599}
2600
2601static int __init parse_ivrs_ioapic(char *str)
2602{
2603 unsigned int bus, dev, fn;
2604 int ret, id, i;
2605 u16 devid;
2606
2607 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2608
2609 if (ret != 4) {
2610 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2611 return 1;
2612 }
2613
2614 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2615 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2616 str);
2617 return 1;
2618 }
2619
2620 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2621
2622 cmdline_maps = true;
2623 i = early_ioapic_map_size++;
2624 early_ioapic_map[i].id = id;
2625 early_ioapic_map[i].devid = devid;
2626 early_ioapic_map[i].cmd_line = true;
2627
2628 return 1;
2629}
2630
2631static int __init parse_ivrs_hpet(char *str)
2632{
2633 unsigned int bus, dev, fn;
2634 int ret, id, i;
2635 u16 devid;
2636
2637 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2638
2639 if (ret != 4) {
2640 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2641 return 1;
2642 }
2643
2644 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2645 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2646 str);
2647 return 1;
2648 }
2649
2650 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2651
2652 cmdline_maps = true;
2653 i = early_hpet_map_size++;
2654 early_hpet_map[i].id = id;
2655 early_hpet_map[i].devid = devid;
2656 early_hpet_map[i].cmd_line = true;
2657
2658 return 1;
2659}
2660
2661static int __init parse_ivrs_acpihid(char *str)
2662{
2663 u32 bus, dev, fn;
2664 char *hid, *uid, *p;
2665 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0};
2666 int ret, i;
2667
2668 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid);
2669 if (ret != 4) {
2670 pr_err("AMD-Vi: Invalid command line: ivrs_acpihid(%s)\n", str);
2671 return 1;
2672 }
2673
2674 p = acpiid;
2675 hid = strsep(&p, ":");
2676 uid = p;
2677
2678 if (!hid || !(*hid) || !uid) {
2679 pr_err("AMD-Vi: Invalid command line: hid or uid\n");
2680 return 1;
2681 }
2682
2683 i = early_acpihid_map_size++;
2684 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
2685 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
2686 early_acpihid_map[i].devid =
2687 ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2688 early_acpihid_map[i].cmd_line = true;
2689
2690 return 1;
2691}
2692
2693__setup("amd_iommu_dump", parse_amd_iommu_dump);
2694__setup("amd_iommu=", parse_amd_iommu_options);
2695__setup("amd_iommu_intr=", parse_amd_iommu_intr);
2696__setup("ivrs_ioapic", parse_ivrs_ioapic);
2697__setup("ivrs_hpet", parse_ivrs_hpet);
2698__setup("ivrs_acpihid", parse_ivrs_acpihid);
2699
2700IOMMU_INIT_FINISH(amd_iommu_detect,
2701 gart_iommu_hole_init,
2702 NULL,
2703 NULL);
2704
2705bool amd_iommu_v2_supported(void)
2706{
2707 return amd_iommu_v2_present;
2708}
2709EXPORT_SYMBOL(amd_iommu_v2_supported);
2710
2711/****************************************************************************
2712 *
2713 * IOMMU EFR Performance Counter support functionality. This code allows
2714 * access to the IOMMU PC functionality.
2715 *
2716 ****************************************************************************/
2717
2718u8 amd_iommu_pc_get_max_banks(u16 devid)
2719{
2720 struct amd_iommu *iommu;
2721 u8 ret = 0;
2722
2723 /* locate the iommu governing the devid */
2724 iommu = amd_iommu_rlookup_table[devid];
2725 if (iommu)
2726 ret = iommu->max_banks;
2727
2728 return ret;
2729}
2730EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2731
2732bool amd_iommu_pc_supported(void)
2733{
2734 return amd_iommu_pc_present;
2735}
2736EXPORT_SYMBOL(amd_iommu_pc_supported);
2737
2738u8 amd_iommu_pc_get_max_counters(u16 devid)
2739{
2740 struct amd_iommu *iommu;
2741 u8 ret = 0;
2742
2743 /* locate the iommu governing the devid */
2744 iommu = amd_iommu_rlookup_table[devid];
2745 if (iommu)
2746 ret = iommu->max_counters;
2747
2748 return ret;
2749}
2750EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2751
2752static int iommu_pc_get_set_reg_val(struct amd_iommu *iommu,
2753 u8 bank, u8 cntr, u8 fxn,
2754 u64 *value, bool is_write)
2755{
2756 u32 offset;
2757 u32 max_offset_lim;
2758
2759 /* Check for valid iommu and pc register indexing */
2760 if (WARN_ON((fxn > 0x28) || (fxn & 7)))
2761 return -ENODEV;
2762
2763 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2764
2765 /* Limit the offset to the hw defined mmio region aperture */
2766 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2767 (iommu->max_counters << 8) | 0x28);
2768 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2769 (offset > max_offset_lim))
2770 return -EINVAL;
2771
2772 if (is_write) {
2773 writel((u32)*value, iommu->mmio_base + offset);
2774 writel((*value >> 32), iommu->mmio_base + offset + 4);
2775 } else {
2776 *value = readl(iommu->mmio_base + offset + 4);
2777 *value <<= 32;
2778 *value = readl(iommu->mmio_base + offset);
2779 }
2780
2781 return 0;
2782}
2783EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);
2784
2785int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2786 u64 *value, bool is_write)
2787{
2788 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
2789
2790 /* Make sure the IOMMU PC resource is available */
2791 if (!amd_iommu_pc_present || iommu == NULL)
2792 return -ENODEV;
2793
2794 return iommu_pc_get_set_reg_val(iommu, bank, cntr, fxn,
2795 value, is_write);
2796}