]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - drivers/iommu/amd_iommu_init.c
iommu/amd: Prevent binding other PCI drivers to IOMMU PCI devices
[mirror_ubuntu-zesty-kernel.git] / drivers / iommu / amd_iommu_init.c
CommitLineData
f6e2e6b6 1/*
5d0d7156 2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
63ce3ae8 3 * Author: Joerg Roedel <jroedel@suse.de>
f6e2e6b6
JR
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>
f6e2e6b6 22#include <linux/list.h>
5a0e3ad6 23#include <linux/slab.h>
f3c6ea1b 24#include <linux/syscore_ops.h>
a80dc3e0
JR
25#include <linux/interrupt.h>
26#include <linux/msi.h>
403f81d8 27#include <linux/amd-iommu.h>
400a28a0 28#include <linux/export.h>
066f2e98 29#include <linux/iommu.h>
f6e2e6b6 30#include <asm/pci-direct.h>
46a7fa27 31#include <asm/iommu.h>
1d9b16d1 32#include <asm/gart.h>
ea1b0d39 33#include <asm/x86_init.h>
22e6daf4 34#include <asm/iommu_table.h>
eb1eb7ae 35#include <asm/io_apic.h>
6b474b82 36#include <asm/irq_remapping.h>
403f81d8
JR
37
38#include "amd_iommu_proto.h"
39#include "amd_iommu_types.h"
05152a04 40#include "irq_remapping.h"
403f81d8 41
f6e2e6b6
JR
42/*
43 * definitions for the ACPI scanning code
44 */
f6e2e6b6 45#define IVRS_HEADER_LENGTH 48
f6e2e6b6
JR
46
47#define ACPI_IVHD_TYPE 0x10
48#define ACPI_IVMD_TYPE_ALL 0x20
49#define ACPI_IVMD_TYPE 0x21
50#define ACPI_IVMD_TYPE_RANGE 0x22
51
52#define IVHD_DEV_ALL 0x01
53#define IVHD_DEV_SELECT 0x02
54#define IVHD_DEV_SELECT_RANGE_START 0x03
55#define IVHD_DEV_RANGE_END 0x04
56#define IVHD_DEV_ALIAS 0x42
57#define IVHD_DEV_ALIAS_RANGE 0x43
58#define IVHD_DEV_EXT_SELECT 0x46
59#define IVHD_DEV_EXT_SELECT_RANGE 0x47
6efed63b
JR
60#define IVHD_DEV_SPECIAL 0x48
61
62#define IVHD_SPECIAL_IOAPIC 1
63#define IVHD_SPECIAL_HPET 2
f6e2e6b6 64
6da7342f
JR
65#define IVHD_FLAG_HT_TUN_EN_MASK 0x01
66#define IVHD_FLAG_PASSPW_EN_MASK 0x02
67#define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
68#define IVHD_FLAG_ISOC_EN_MASK 0x08
f6e2e6b6
JR
69
70#define IVMD_FLAG_EXCL_RANGE 0x08
71#define IVMD_FLAG_UNITY_MAP 0x01
72
73#define ACPI_DEVFLAG_INITPASS 0x01
74#define ACPI_DEVFLAG_EXTINT 0x02
75#define ACPI_DEVFLAG_NMI 0x04
76#define ACPI_DEVFLAG_SYSMGT1 0x10
77#define ACPI_DEVFLAG_SYSMGT2 0x20
78#define ACPI_DEVFLAG_LINT0 0x40
79#define ACPI_DEVFLAG_LINT1 0x80
80#define ACPI_DEVFLAG_ATSDIS 0x10000000
81
b65233a9
JR
82/*
83 * ACPI table definitions
84 *
85 * These data structures are laid over the table to parse the important values
86 * out of it.
87 */
88
89/*
90 * structure describing one IOMMU in the ACPI table. Typically followed by one
91 * or more ivhd_entrys.
92 */
f6e2e6b6
JR
93struct ivhd_header {
94 u8 type;
95 u8 flags;
96 u16 length;
97 u16 devid;
98 u16 cap_ptr;
99 u64 mmio_phys;
100 u16 pci_seg;
101 u16 info;
30861ddc 102 u32 efr;
f6e2e6b6
JR
103} __attribute__((packed));
104
b65233a9
JR
105/*
106 * A device entry describing which devices a specific IOMMU translates and
107 * which requestor ids they use.
108 */
f6e2e6b6
JR
109struct ivhd_entry {
110 u8 type;
111 u16 devid;
112 u8 flags;
113 u32 ext;
114} __attribute__((packed));
115
b65233a9
JR
116/*
117 * An AMD IOMMU memory definition structure. It defines things like exclusion
118 * ranges for devices and regions that should be unity mapped.
119 */
f6e2e6b6
JR
120struct ivmd_header {
121 u8 type;
122 u8 flags;
123 u16 length;
124 u16 devid;
125 u16 aux;
126 u64 resv;
127 u64 range_start;
128 u64 range_length;
129} __attribute__((packed));
130
fefda117 131bool amd_iommu_dump;
05152a04 132bool amd_iommu_irq_remap __read_mostly;
fefda117 133
02f3b3f5 134static bool amd_iommu_detected;
a5235725 135static bool __initdata amd_iommu_disabled;
c1cbebee 136
b65233a9
JR
137u16 amd_iommu_last_bdf; /* largest PCI device id we have
138 to handle */
2e22847f 139LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
b65233a9 140 we find in ACPI */
3775d481 141u32 amd_iommu_unmap_flush; /* if true, flush on every unmap */
928abd25 142
2e22847f 143LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
b65233a9 144 system */
928abd25 145
bb52777e
JR
146/* Array to assign indices to IOMMUs*/
147struct amd_iommu *amd_iommus[MAX_IOMMUS];
148int amd_iommus_present;
149
318afd41
JR
150/* IOMMUs have a non-present cache? */
151bool amd_iommu_np_cache __read_mostly;
60f723b4 152bool amd_iommu_iotlb_sup __read_mostly = true;
318afd41 153
a919a018 154u32 amd_iommu_max_pasid __read_mostly = ~0;
62f71abb 155
400a28a0 156bool amd_iommu_v2_present __read_mostly;
4160cd9e 157static bool amd_iommu_pc_present __read_mostly;
400a28a0 158
5abcdba4
JR
159bool amd_iommu_force_isolation __read_mostly;
160
aeb26f55
JR
161/*
162 * List of protection domains - used during resume
163 */
164LIST_HEAD(amd_iommu_pd_list);
165spinlock_t amd_iommu_pd_lock;
166
b65233a9
JR
167/*
168 * Pointer to the device table which is shared by all AMD IOMMUs
169 * it is indexed by the PCI device id or the HT unit id and contains
170 * information about the domain the device belongs to as well as the
171 * page table root pointer.
172 */
928abd25 173struct dev_table_entry *amd_iommu_dev_table;
b65233a9
JR
174
175/*
176 * The alias table is a driver specific data structure which contains the
177 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
178 * More than one device can share the same requestor id.
179 */
928abd25 180u16 *amd_iommu_alias_table;
b65233a9
JR
181
182/*
183 * The rlookup table is used to find the IOMMU which is responsible
184 * for a specific device. It is also indexed by the PCI device id.
185 */
928abd25 186struct amd_iommu **amd_iommu_rlookup_table;
b65233a9 187
b65233a9 188/*
0ea2c422
JR
189 * This table is used to find the irq remapping table for a given device id
190 * quickly.
191 */
192struct irq_remap_table **irq_lookup_table;
193
b65233a9 194/*
df805abb 195 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
b65233a9
JR
196 * to know which ones are already in use.
197 */
928abd25
JR
198unsigned long *amd_iommu_pd_alloc_bitmap;
199
b65233a9
JR
200static u32 dev_table_size; /* size of the device table */
201static u32 alias_table_size; /* size of the alias table */
202static u32 rlookup_table_size; /* size if the rlookup table */
3e8064ba 203
2c0ae172
JR
204enum iommu_init_state {
205 IOMMU_START_STATE,
206 IOMMU_IVRS_DETECTED,
207 IOMMU_ACPI_FINISHED,
208 IOMMU_ENABLED,
209 IOMMU_PCI_INIT,
210 IOMMU_INTERRUPTS_EN,
211 IOMMU_DMA_OPS,
212 IOMMU_INITIALIZED,
213 IOMMU_NOT_FOUND,
214 IOMMU_INIT_ERROR,
215};
216
235dacbc
JR
217/* Early ioapic and hpet maps from kernel command line */
218#define EARLY_MAP_SIZE 4
219static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
220static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
221static int __initdata early_ioapic_map_size;
222static int __initdata early_hpet_map_size;
dfbb6d47 223static bool __initdata cmdline_maps;
235dacbc 224
2c0ae172
JR
225static enum iommu_init_state init_state = IOMMU_START_STATE;
226
ae295142 227static int amd_iommu_enable_interrupts(void);
2c0ae172 228static int __init iommu_go_to_state(enum iommu_init_state state);
aafd8ba0 229static void init_device_table_dma(void);
3d9761e7 230
208ec8c9
JR
231static inline void update_last_devid(u16 devid)
232{
233 if (devid > amd_iommu_last_bdf)
234 amd_iommu_last_bdf = devid;
235}
236
c571484e
JR
237static inline unsigned long tbl_size(int entry_size)
238{
239 unsigned shift = PAGE_SHIFT +
421f909c 240 get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
c571484e
JR
241
242 return 1UL << shift;
243}
244
5bcd757f
MG
245/* Access to l1 and l2 indexed register spaces */
246
247static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
248{
249 u32 val;
250
251 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
252 pci_read_config_dword(iommu->dev, 0xfc, &val);
253 return val;
254}
255
256static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
257{
258 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
259 pci_write_config_dword(iommu->dev, 0xfc, val);
260 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
261}
262
263static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
264{
265 u32 val;
266
267 pci_write_config_dword(iommu->dev, 0xf0, address);
268 pci_read_config_dword(iommu->dev, 0xf4, &val);
269 return val;
270}
271
272static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
273{
274 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
275 pci_write_config_dword(iommu->dev, 0xf4, val);
276}
277
b65233a9
JR
278/****************************************************************************
279 *
280 * AMD IOMMU MMIO register space handling functions
281 *
282 * These functions are used to program the IOMMU device registers in
283 * MMIO space required for that driver.
284 *
285 ****************************************************************************/
3e8064ba 286
b65233a9
JR
287/*
288 * This function set the exclusion range in the IOMMU. DMA accesses to the
289 * exclusion range are passed through untranslated
290 */
05f92db9 291static void iommu_set_exclusion_range(struct amd_iommu *iommu)
b2026aa2
JR
292{
293 u64 start = iommu->exclusion_start & PAGE_MASK;
294 u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
295 u64 entry;
296
297 if (!iommu->exclusion_start)
298 return;
299
300 entry = start | MMIO_EXCL_ENABLE_MASK;
301 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
302 &entry, sizeof(entry));
303
304 entry = limit;
305 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
306 &entry, sizeof(entry));
307}
308
b65233a9 309/* Programs the physical address of the device table into the IOMMU hardware */
6b7f000e 310static void iommu_set_device_table(struct amd_iommu *iommu)
b2026aa2 311{
f609891f 312 u64 entry;
b2026aa2
JR
313
314 BUG_ON(iommu->mmio_base == NULL);
315
316 entry = virt_to_phys(amd_iommu_dev_table);
317 entry |= (dev_table_size >> 12) - 1;
318 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
319 &entry, sizeof(entry));
320}
321
b65233a9 322/* Generic functions to enable/disable certain features of the IOMMU. */
05f92db9 323static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
b2026aa2
JR
324{
325 u32 ctrl;
326
327 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
328 ctrl |= (1 << bit);
329 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
330}
331
ca020711 332static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
b2026aa2
JR
333{
334 u32 ctrl;
335
199d0d50 336 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
b2026aa2
JR
337 ctrl &= ~(1 << bit);
338 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
339}
340
1456e9d2
JR
341static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
342{
343 u32 ctrl;
344
345 ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
346 ctrl &= ~CTRL_INV_TO_MASK;
347 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
348 writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
349}
350
b65233a9 351/* Function to enable the hardware */
05f92db9 352static void iommu_enable(struct amd_iommu *iommu)
b2026aa2 353{
b2026aa2 354 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
b2026aa2
JR
355}
356
92ac4320 357static void iommu_disable(struct amd_iommu *iommu)
126c52be 358{
a8c485bb
CW
359 /* Disable command buffer */
360 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
361
362 /* Disable event logging and event interrupts */
363 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
364 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
365
366 /* Disable IOMMU hardware itself */
92ac4320 367 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
126c52be
JR
368}
369
b65233a9
JR
370/*
371 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
372 * the system has one.
373 */
30861ddc 374static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
6c56747b 375{
30861ddc
SK
376 if (!request_mem_region(address, end, "amd_iommu")) {
377 pr_err("AMD-Vi: Can not reserve memory region %llx-%llx for mmio\n",
378 address, end);
e82752d8 379 pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
6c56747b 380 return NULL;
e82752d8 381 }
6c56747b 382
30861ddc 383 return (u8 __iomem *)ioremap_nocache(address, end);
6c56747b
JR
384}
385
386static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
387{
388 if (iommu->mmio_base)
389 iounmap(iommu->mmio_base);
30861ddc 390 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
6c56747b
JR
391}
392
b65233a9
JR
393/****************************************************************************
394 *
395 * The functions below belong to the first pass of AMD IOMMU ACPI table
396 * parsing. In this pass we try to find out the highest device id this
397 * code has to handle. Upon this information the size of the shared data
398 * structures is determined later.
399 *
400 ****************************************************************************/
401
b514e555
JR
402/*
403 * This function calculates the length of a given IVHD entry
404 */
405static inline int ivhd_entry_length(u8 *ivhd)
406{
407 return 0x04 << (*ivhd >> 6);
408}
409
b65233a9
JR
410/*
411 * This function reads the last device id the IOMMU has to handle from the PCI
412 * capability header for this IOMMU
413 */
3e8064ba
JR
414static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
415{
416 u32 cap;
417
418 cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
6f2729ba 419 update_last_devid(PCI_DEVID(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
3e8064ba
JR
420
421 return 0;
422}
423
b65233a9
JR
424/*
425 * After reading the highest device id from the IOMMU PCI capability header
426 * this function looks if there is a higher device id defined in the ACPI table
427 */
3e8064ba
JR
428static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
429{
430 u8 *p = (void *)h, *end = (void *)h;
431 struct ivhd_entry *dev;
432
433 p += sizeof(*h);
434 end += h->length;
435
c5081cd7 436 find_last_devid_on_pci(PCI_BUS_NUM(h->devid),
3e8064ba
JR
437 PCI_SLOT(h->devid),
438 PCI_FUNC(h->devid),
439 h->cap_ptr);
440
441 while (p < end) {
442 dev = (struct ivhd_entry *)p;
443 switch (dev->type) {
444 case IVHD_DEV_SELECT:
445 case IVHD_DEV_RANGE_END:
446 case IVHD_DEV_ALIAS:
447 case IVHD_DEV_EXT_SELECT:
b65233a9 448 /* all the above subfield types refer to device ids */
208ec8c9 449 update_last_devid(dev->devid);
3e8064ba
JR
450 break;
451 default:
452 break;
453 }
b514e555 454 p += ivhd_entry_length(p);
3e8064ba
JR
455 }
456
457 WARN_ON(p != end);
458
459 return 0;
460}
461
b65233a9
JR
462/*
463 * Iterate over all IVHD entries in the ACPI table and find the highest device
464 * id which we need to handle. This is the first of three functions which parse
465 * the ACPI table. So we check the checksum here.
466 */
3e8064ba
JR
467static int __init find_last_devid_acpi(struct acpi_table_header *table)
468{
469 int i;
470 u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
471 struct ivhd_header *h;
472
473 /*
474 * Validate checksum here so we don't need to do it when
475 * we actually parse the table
476 */
477 for (i = 0; i < table->length; ++i)
478 checksum += p[i];
02f3b3f5 479 if (checksum != 0)
3e8064ba 480 /* ACPI table corrupt */
02f3b3f5 481 return -ENODEV;
3e8064ba
JR
482
483 p += IVRS_HEADER_LENGTH;
484
485 end += table->length;
486 while (p < end) {
487 h = (struct ivhd_header *)p;
488 switch (h->type) {
489 case ACPI_IVHD_TYPE:
490 find_last_devid_from_ivhd(h);
491 break;
492 default:
493 break;
494 }
495 p += h->length;
496 }
497 WARN_ON(p != end);
498
499 return 0;
500}
501
b65233a9
JR
502/****************************************************************************
503 *
df805abb 504 * The following functions belong to the code path which parses the ACPI table
b65233a9
JR
505 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
506 * data structures, initialize the device/alias/rlookup table and also
507 * basically initialize the hardware.
508 *
509 ****************************************************************************/
510
511/*
512 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
513 * write commands to that buffer later and the IOMMU will execute them
514 * asynchronously
515 */
b36ca91e
JR
516static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
517{
d0312b21 518 u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
b36ca91e 519 get_order(CMD_BUFFER_SIZE));
b36ca91e
JR
520
521 if (cmd_buf == NULL)
522 return NULL;
523
549c90dc 524 iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
b36ca91e 525
58492e12
JR
526 return cmd_buf;
527}
528
93f1cc67
JR
529/*
530 * This function resets the command buffer if the IOMMU stopped fetching
531 * commands from it.
532 */
533void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
534{
535 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
536
537 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
538 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
539
540 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
541}
542
58492e12
JR
543/*
544 * This function writes the command buffer address to the hardware and
545 * enables it.
546 */
547static void iommu_enable_command_buffer(struct amd_iommu *iommu)
548{
549 u64 entry;
550
551 BUG_ON(iommu->cmd_buf == NULL);
552
553 entry = (u64)virt_to_phys(iommu->cmd_buf);
b36ca91e 554 entry |= MMIO_CMD_SIZE_512;
58492e12 555
b36ca91e 556 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
58492e12 557 &entry, sizeof(entry));
b36ca91e 558
93f1cc67 559 amd_iommu_reset_cmd_buffer(iommu);
549c90dc 560 iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
b36ca91e
JR
561}
562
563static void __init free_command_buffer(struct amd_iommu *iommu)
564{
23c1713f 565 free_pages((unsigned long)iommu->cmd_buf,
549c90dc 566 get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
b36ca91e
JR
567}
568
335503e5
JR
569/* allocates the memory where the IOMMU will log its events to */
570static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
571{
335503e5
JR
572 iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
573 get_order(EVT_BUFFER_SIZE));
574
575 if (iommu->evt_buf == NULL)
576 return NULL;
577
1bc6f838
JR
578 iommu->evt_buf_size = EVT_BUFFER_SIZE;
579
58492e12
JR
580 return iommu->evt_buf;
581}
582
583static void iommu_enable_event_buffer(struct amd_iommu *iommu)
584{
585 u64 entry;
586
587 BUG_ON(iommu->evt_buf == NULL);
588
335503e5 589 entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
58492e12 590
335503e5
JR
591 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
592 &entry, sizeof(entry));
593
09067207
JR
594 /* set head and tail to zero manually */
595 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
596 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
597
58492e12 598 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
335503e5
JR
599}
600
601static void __init free_event_buffer(struct amd_iommu *iommu)
602{
603 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
604}
605
1a29ac01
JR
606/* allocates the memory where the IOMMU will log its events to */
607static u8 * __init alloc_ppr_log(struct amd_iommu *iommu)
608{
609 iommu->ppr_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
610 get_order(PPR_LOG_SIZE));
611
612 if (iommu->ppr_log == NULL)
613 return NULL;
614
615 return iommu->ppr_log;
616}
617
618static void iommu_enable_ppr_log(struct amd_iommu *iommu)
619{
620 u64 entry;
621
622 if (iommu->ppr_log == NULL)
623 return;
624
625 entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
626
627 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
628 &entry, sizeof(entry));
629
630 /* set head and tail to zero manually */
631 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
632 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
633
634 iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
635 iommu_feature_enable(iommu, CONTROL_PPR_EN);
636}
637
638static void __init free_ppr_log(struct amd_iommu *iommu)
639{
640 if (iommu->ppr_log == NULL)
641 return;
642
643 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
644}
645
cbc33a90
JR
646static void iommu_enable_gt(struct amd_iommu *iommu)
647{
648 if (!iommu_feature(iommu, FEATURE_GT))
649 return;
650
651 iommu_feature_enable(iommu, CONTROL_GT_EN);
652}
653
b65233a9 654/* sets a specific bit in the device table entry. */
3566b778
JR
655static void set_dev_entry_bit(u16 devid, u8 bit)
656{
ee6c2868
JR
657 int i = (bit >> 6) & 0x03;
658 int _bit = bit & 0x3f;
3566b778 659
ee6c2868 660 amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
3566b778
JR
661}
662
c5cca146
JR
663static int get_dev_entry_bit(u16 devid, u8 bit)
664{
ee6c2868
JR
665 int i = (bit >> 6) & 0x03;
666 int _bit = bit & 0x3f;
c5cca146 667
ee6c2868 668 return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
c5cca146
JR
669}
670
671
672void amd_iommu_apply_erratum_63(u16 devid)
673{
674 int sysmgt;
675
676 sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
677 (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
678
679 if (sysmgt == 0x01)
680 set_dev_entry_bit(devid, DEV_ENTRY_IW);
681}
682
5ff4789d
JR
683/* Writes the specific IOMMU for a device into the rlookup table */
684static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
685{
686 amd_iommu_rlookup_table[devid] = iommu;
687}
688
b65233a9
JR
689/*
690 * This function takes the device specific flags read from the ACPI
691 * table and sets up the device table entry with that information
692 */
5ff4789d
JR
693static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
694 u16 devid, u32 flags, u32 ext_flags)
3566b778
JR
695{
696 if (flags & ACPI_DEVFLAG_INITPASS)
697 set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
698 if (flags & ACPI_DEVFLAG_EXTINT)
699 set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
700 if (flags & ACPI_DEVFLAG_NMI)
701 set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
702 if (flags & ACPI_DEVFLAG_SYSMGT1)
703 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
704 if (flags & ACPI_DEVFLAG_SYSMGT2)
705 set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
706 if (flags & ACPI_DEVFLAG_LINT0)
707 set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
708 if (flags & ACPI_DEVFLAG_LINT1)
709 set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
3566b778 710
c5cca146
JR
711 amd_iommu_apply_erratum_63(devid);
712
5ff4789d 713 set_iommu_for_device(iommu, devid);
3566b778
JR
714}
715
c50e3247 716static int __init add_special_device(u8 type, u8 id, u16 *devid, bool cmd_line)
6efed63b
JR
717{
718 struct devid_map *entry;
719 struct list_head *list;
720
31cff67f
JR
721 if (type == IVHD_SPECIAL_IOAPIC)
722 list = &ioapic_map;
723 else if (type == IVHD_SPECIAL_HPET)
724 list = &hpet_map;
725 else
6efed63b
JR
726 return -EINVAL;
727
31cff67f
JR
728 list_for_each_entry(entry, list, list) {
729 if (!(entry->id == id && entry->cmd_line))
730 continue;
731
732 pr_info("AMD-Vi: Command-line override present for %s id %d - ignoring\n",
733 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
734
c50e3247
JR
735 *devid = entry->devid;
736
31cff67f
JR
737 return 0;
738 }
739
6efed63b
JR
740 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
741 if (!entry)
742 return -ENOMEM;
743
31cff67f 744 entry->id = id;
c50e3247 745 entry->devid = *devid;
31cff67f 746 entry->cmd_line = cmd_line;
6efed63b
JR
747
748 list_add_tail(&entry->list, list);
749
750 return 0;
751}
752
235dacbc
JR
753static int __init add_early_maps(void)
754{
755 int i, ret;
756
757 for (i = 0; i < early_ioapic_map_size; ++i) {
758 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
759 early_ioapic_map[i].id,
c50e3247 760 &early_ioapic_map[i].devid,
235dacbc
JR
761 early_ioapic_map[i].cmd_line);
762 if (ret)
763 return ret;
764 }
765
766 for (i = 0; i < early_hpet_map_size; ++i) {
767 ret = add_special_device(IVHD_SPECIAL_HPET,
768 early_hpet_map[i].id,
c50e3247 769 &early_hpet_map[i].devid,
235dacbc
JR
770 early_hpet_map[i].cmd_line);
771 if (ret)
772 return ret;
773 }
774
775 return 0;
776}
777
b65233a9 778/*
df805abb 779 * Reads the device exclusion range from ACPI and initializes the IOMMU with
b65233a9
JR
780 * it
781 */
3566b778
JR
782static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
783{
784 struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
785
786 if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
787 return;
788
789 if (iommu) {
b65233a9
JR
790 /*
791 * We only can configure exclusion ranges per IOMMU, not
792 * per device. But we can enable the exclusion range per
793 * device. This is done here
794 */
2c16c9fd 795 set_dev_entry_bit(devid, DEV_ENTRY_EX);
3566b778
JR
796 iommu->exclusion_start = m->range_start;
797 iommu->exclusion_length = m->range_length;
798 }
799}
800
b65233a9
JR
801/*
802 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
803 * initializes the hardware and our data structures with it.
804 */
6efed63b 805static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
5d0c8e49
JR
806 struct ivhd_header *h)
807{
808 u8 *p = (u8 *)h;
809 u8 *end = p, flags = 0;
0de66d5b
JR
810 u16 devid = 0, devid_start = 0, devid_to = 0;
811 u32 dev_i, ext_flags = 0;
58a3bee5 812 bool alias = false;
5d0c8e49 813 struct ivhd_entry *e;
235dacbc
JR
814 int ret;
815
816
817 ret = add_early_maps();
818 if (ret)
819 return ret;
5d0c8e49
JR
820
821 /*
e9bf5197 822 * First save the recommended feature enable bits from ACPI
5d0c8e49 823 */
e9bf5197 824 iommu->acpi_flags = h->flags;
5d0c8e49
JR
825
826 /*
827 * Done. Now parse the device entries
828 */
829 p += sizeof(struct ivhd_header);
830 end += h->length;
831
42a698f4 832
5d0c8e49
JR
833 while (p < end) {
834 e = (struct ivhd_entry *)p;
835 switch (e->type) {
836 case IVHD_DEV_ALL:
42a698f4
JR
837
838 DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
839 " last device %02x:%02x.%x flags: %02x\n",
c5081cd7 840 PCI_BUS_NUM(iommu->first_device),
42a698f4
JR
841 PCI_SLOT(iommu->first_device),
842 PCI_FUNC(iommu->first_device),
c5081cd7 843 PCI_BUS_NUM(iommu->last_device),
42a698f4
JR
844 PCI_SLOT(iommu->last_device),
845 PCI_FUNC(iommu->last_device),
846 e->flags);
847
5d0c8e49
JR
848 for (dev_i = iommu->first_device;
849 dev_i <= iommu->last_device; ++dev_i)
5ff4789d
JR
850 set_dev_entry_from_acpi(iommu, dev_i,
851 e->flags, 0);
5d0c8e49
JR
852 break;
853 case IVHD_DEV_SELECT:
42a698f4
JR
854
855 DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
856 "flags: %02x\n",
c5081cd7 857 PCI_BUS_NUM(e->devid),
42a698f4
JR
858 PCI_SLOT(e->devid),
859 PCI_FUNC(e->devid),
860 e->flags);
861
5d0c8e49 862 devid = e->devid;
5ff4789d 863 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
5d0c8e49
JR
864 break;
865 case IVHD_DEV_SELECT_RANGE_START:
42a698f4
JR
866
867 DUMP_printk(" DEV_SELECT_RANGE_START\t "
868 "devid: %02x:%02x.%x flags: %02x\n",
c5081cd7 869 PCI_BUS_NUM(e->devid),
42a698f4
JR
870 PCI_SLOT(e->devid),
871 PCI_FUNC(e->devid),
872 e->flags);
873
5d0c8e49
JR
874 devid_start = e->devid;
875 flags = e->flags;
876 ext_flags = 0;
58a3bee5 877 alias = false;
5d0c8e49
JR
878 break;
879 case IVHD_DEV_ALIAS:
42a698f4
JR
880
881 DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
882 "flags: %02x devid_to: %02x:%02x.%x\n",
c5081cd7 883 PCI_BUS_NUM(e->devid),
42a698f4
JR
884 PCI_SLOT(e->devid),
885 PCI_FUNC(e->devid),
886 e->flags,
c5081cd7 887 PCI_BUS_NUM(e->ext >> 8),
42a698f4
JR
888 PCI_SLOT(e->ext >> 8),
889 PCI_FUNC(e->ext >> 8));
890
5d0c8e49
JR
891 devid = e->devid;
892 devid_to = e->ext >> 8;
7a6a3a08 893 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
7455aab1 894 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
5d0c8e49
JR
895 amd_iommu_alias_table[devid] = devid_to;
896 break;
897 case IVHD_DEV_ALIAS_RANGE:
42a698f4
JR
898
899 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
900 "devid: %02x:%02x.%x flags: %02x "
901 "devid_to: %02x:%02x.%x\n",
c5081cd7 902 PCI_BUS_NUM(e->devid),
42a698f4
JR
903 PCI_SLOT(e->devid),
904 PCI_FUNC(e->devid),
905 e->flags,
c5081cd7 906 PCI_BUS_NUM(e->ext >> 8),
42a698f4
JR
907 PCI_SLOT(e->ext >> 8),
908 PCI_FUNC(e->ext >> 8));
909
5d0c8e49
JR
910 devid_start = e->devid;
911 flags = e->flags;
912 devid_to = e->ext >> 8;
913 ext_flags = 0;
58a3bee5 914 alias = true;
5d0c8e49
JR
915 break;
916 case IVHD_DEV_EXT_SELECT:
42a698f4
JR
917
918 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
919 "flags: %02x ext: %08x\n",
c5081cd7 920 PCI_BUS_NUM(e->devid),
42a698f4
JR
921 PCI_SLOT(e->devid),
922 PCI_FUNC(e->devid),
923 e->flags, e->ext);
924
5d0c8e49 925 devid = e->devid;
5ff4789d
JR
926 set_dev_entry_from_acpi(iommu, devid, e->flags,
927 e->ext);
5d0c8e49
JR
928 break;
929 case IVHD_DEV_EXT_SELECT_RANGE:
42a698f4
JR
930
931 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
932 "%02x:%02x.%x flags: %02x ext: %08x\n",
c5081cd7 933 PCI_BUS_NUM(e->devid),
42a698f4
JR
934 PCI_SLOT(e->devid),
935 PCI_FUNC(e->devid),
936 e->flags, e->ext);
937
5d0c8e49
JR
938 devid_start = e->devid;
939 flags = e->flags;
940 ext_flags = e->ext;
58a3bee5 941 alias = false;
5d0c8e49
JR
942 break;
943 case IVHD_DEV_RANGE_END:
42a698f4
JR
944
945 DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
c5081cd7 946 PCI_BUS_NUM(e->devid),
42a698f4
JR
947 PCI_SLOT(e->devid),
948 PCI_FUNC(e->devid));
949
5d0c8e49
JR
950 devid = e->devid;
951 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
7a6a3a08 952 if (alias) {
5d0c8e49 953 amd_iommu_alias_table[dev_i] = devid_to;
7a6a3a08
JR
954 set_dev_entry_from_acpi(iommu,
955 devid_to, flags, ext_flags);
956 }
957 set_dev_entry_from_acpi(iommu, dev_i,
958 flags, ext_flags);
5d0c8e49
JR
959 }
960 break;
6efed63b
JR
961 case IVHD_DEV_SPECIAL: {
962 u8 handle, type;
963 const char *var;
964 u16 devid;
965 int ret;
966
967 handle = e->ext & 0xff;
968 devid = (e->ext >> 8) & 0xffff;
969 type = (e->ext >> 24) & 0xff;
970
971 if (type == IVHD_SPECIAL_IOAPIC)
972 var = "IOAPIC";
973 else if (type == IVHD_SPECIAL_HPET)
974 var = "HPET";
975 else
976 var = "UNKNOWN";
977
978 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %02x:%02x.%x\n",
979 var, (int)handle,
c5081cd7 980 PCI_BUS_NUM(devid),
6efed63b
JR
981 PCI_SLOT(devid),
982 PCI_FUNC(devid));
983
c50e3247 984 ret = add_special_device(type, handle, &devid, false);
6efed63b
JR
985 if (ret)
986 return ret;
c50e3247
JR
987
988 /*
989 * add_special_device might update the devid in case a
990 * command-line override is present. So call
991 * set_dev_entry_from_acpi after add_special_device.
992 */
993 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
994
6efed63b
JR
995 break;
996 }
5d0c8e49
JR
997 default:
998 break;
999 }
1000
b514e555 1001 p += ivhd_entry_length(p);
5d0c8e49 1002 }
6efed63b
JR
1003
1004 return 0;
5d0c8e49
JR
1005}
1006
b65233a9 1007/* Initializes the device->iommu mapping for the driver */
5d0c8e49
JR
1008static int __init init_iommu_devices(struct amd_iommu *iommu)
1009{
0de66d5b 1010 u32 i;
5d0c8e49
JR
1011
1012 for (i = iommu->first_device; i <= iommu->last_device; ++i)
1013 set_iommu_for_device(iommu, i);
1014
1015 return 0;
1016}
1017
e47d402d
JR
1018static void __init free_iommu_one(struct amd_iommu *iommu)
1019{
1020 free_command_buffer(iommu);
335503e5 1021 free_event_buffer(iommu);
1a29ac01 1022 free_ppr_log(iommu);
e47d402d
JR
1023 iommu_unmap_mmio_space(iommu);
1024}
1025
1026static void __init free_iommu_all(void)
1027{
1028 struct amd_iommu *iommu, *next;
1029
3bd22172 1030 for_each_iommu_safe(iommu, next) {
e47d402d
JR
1031 list_del(&iommu->list);
1032 free_iommu_one(iommu);
1033 kfree(iommu);
1034 }
1035}
1036
318fe782
SS
1037/*
1038 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1039 * Workaround:
1040 * BIOS should disable L2B micellaneous clock gating by setting
1041 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1042 */
e2f1a3bd 1043static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
318fe782
SS
1044{
1045 u32 value;
1046
1047 if ((boot_cpu_data.x86 != 0x15) ||
1048 (boot_cpu_data.x86_model < 0x10) ||
1049 (boot_cpu_data.x86_model > 0x1f))
1050 return;
1051
1052 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1053 pci_read_config_dword(iommu->dev, 0xf4, &value);
1054
1055 if (value & BIT(2))
1056 return;
1057
1058 /* Select NB indirect register 0x90 and enable writing */
1059 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1060
1061 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1062 pr_info("AMD-Vi: Applying erratum 746 workaround for IOMMU at %s\n",
1063 dev_name(&iommu->dev->dev));
1064
1065 /* Clear the enable writing bit */
1066 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1067}
1068
b65233a9
JR
1069/*
1070 * This function clues the initialization function for one IOMMU
1071 * together and also allocates the command buffer and programs the
1072 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1073 */
e47d402d
JR
1074static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
1075{
6efed63b
JR
1076 int ret;
1077
e47d402d 1078 spin_lock_init(&iommu->lock);
bb52777e
JR
1079
1080 /* Add IOMMU to internal data structures */
e47d402d 1081 list_add_tail(&iommu->list, &amd_iommu_list);
bb52777e
JR
1082 iommu->index = amd_iommus_present++;
1083
1084 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1085 WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
1086 return -ENOSYS;
1087 }
1088
1089 /* Index is fine - add IOMMU to the array */
1090 amd_iommus[iommu->index] = iommu;
e47d402d
JR
1091
1092 /*
1093 * Copy data from ACPI table entry to the iommu struct
1094 */
23c742db 1095 iommu->devid = h->devid;
e47d402d 1096 iommu->cap_ptr = h->cap_ptr;
ee893c24 1097 iommu->pci_seg = h->pci_seg;
e47d402d 1098 iommu->mmio_phys = h->mmio_phys;
30861ddc
SK
1099
1100 /* Check if IVHD EFR contains proper max banks/counters */
1101 if ((h->efr != 0) &&
1102 ((h->efr & (0xF << 13)) != 0) &&
1103 ((h->efr & (0x3F << 17)) != 0)) {
1104 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1105 } else {
1106 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1107 }
1108
1109 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1110 iommu->mmio_phys_end);
e47d402d
JR
1111 if (!iommu->mmio_base)
1112 return -ENOMEM;
1113
e47d402d
JR
1114 iommu->cmd_buf = alloc_command_buffer(iommu);
1115 if (!iommu->cmd_buf)
1116 return -ENOMEM;
1117
335503e5
JR
1118 iommu->evt_buf = alloc_event_buffer(iommu);
1119 if (!iommu->evt_buf)
1120 return -ENOMEM;
1121
a80dc3e0
JR
1122 iommu->int_enabled = false;
1123
6efed63b
JR
1124 ret = init_iommu_from_acpi(iommu, h);
1125 if (ret)
1126 return ret;
f6fec00a 1127
7c71d306
JL
1128 ret = amd_iommu_create_irq_domain(iommu);
1129 if (ret)
1130 return ret;
1131
f6fec00a
JR
1132 /*
1133 * Make sure IOMMU is not considered to translate itself. The IVRS
1134 * table tells us so, but this is a lie!
1135 */
1136 amd_iommu_rlookup_table[iommu->devid] = NULL;
1137
e47d402d
JR
1138 init_iommu_devices(iommu);
1139
23c742db 1140 return 0;
e47d402d
JR
1141}
1142
b65233a9
JR
1143/*
1144 * Iterates over all IOMMU entries in the ACPI table, allocates the
1145 * IOMMU structure and initializes it with init_iommu_one()
1146 */
e47d402d
JR
1147static int __init init_iommu_all(struct acpi_table_header *table)
1148{
1149 u8 *p = (u8 *)table, *end = (u8 *)table;
1150 struct ivhd_header *h;
1151 struct amd_iommu *iommu;
1152 int ret;
1153
e47d402d
JR
1154 end += table->length;
1155 p += IVRS_HEADER_LENGTH;
1156
1157 while (p < end) {
1158 h = (struct ivhd_header *)p;
1159 switch (*p) {
1160 case ACPI_IVHD_TYPE:
9c72041f 1161
ae908c22 1162 DUMP_printk("device: %02x:%02x.%01x cap: %04x "
9c72041f 1163 "seg: %d flags: %01x info %04x\n",
c5081cd7 1164 PCI_BUS_NUM(h->devid), PCI_SLOT(h->devid),
9c72041f
JR
1165 PCI_FUNC(h->devid), h->cap_ptr,
1166 h->pci_seg, h->flags, h->info);
1167 DUMP_printk(" mmio-addr: %016llx\n",
1168 h->mmio_phys);
1169
e47d402d 1170 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
02f3b3f5
JR
1171 if (iommu == NULL)
1172 return -ENOMEM;
3551a708 1173
e47d402d 1174 ret = init_iommu_one(iommu, h);
02f3b3f5
JR
1175 if (ret)
1176 return ret;
e47d402d
JR
1177 break;
1178 default:
1179 break;
1180 }
1181 p += h->length;
1182
1183 }
1184 WARN_ON(p != end);
1185
1186 return 0;
1187}
1188
30861ddc
SK
1189
1190static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1191{
1192 u64 val = 0xabcd, val2 = 0;
1193
1194 if (!iommu_feature(iommu, FEATURE_PC))
1195 return;
1196
1197 amd_iommu_pc_present = true;
1198
1199 /* Check if the performance counters can be written to */
1200 if ((0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val, true)) ||
1201 (0 != amd_iommu_pc_get_set_reg_val(0, 0, 0, 0, &val2, false)) ||
1202 (val != val2)) {
1203 pr_err("AMD-Vi: Unable to write to IOMMU perf counter.\n");
1204 amd_iommu_pc_present = false;
1205 return;
1206 }
1207
1208 pr_info("AMD-Vi: IOMMU performance counters supported\n");
1209
1210 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1211 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1212 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1213}
1214
066f2e98
AW
1215static ssize_t amd_iommu_show_cap(struct device *dev,
1216 struct device_attribute *attr,
1217 char *buf)
1218{
1219 struct amd_iommu *iommu = dev_get_drvdata(dev);
1220 return sprintf(buf, "%x\n", iommu->cap);
1221}
1222static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1223
1224static ssize_t amd_iommu_show_features(struct device *dev,
1225 struct device_attribute *attr,
1226 char *buf)
1227{
1228 struct amd_iommu *iommu = dev_get_drvdata(dev);
1229 return sprintf(buf, "%llx\n", iommu->features);
1230}
1231static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1232
1233static struct attribute *amd_iommu_attrs[] = {
1234 &dev_attr_cap.attr,
1235 &dev_attr_features.attr,
1236 NULL,
1237};
1238
1239static struct attribute_group amd_iommu_group = {
1240 .name = "amd-iommu",
1241 .attrs = amd_iommu_attrs,
1242};
1243
1244static const struct attribute_group *amd_iommu_groups[] = {
1245 &amd_iommu_group,
1246 NULL,
1247};
30861ddc 1248
23c742db
JR
1249static int iommu_init_pci(struct amd_iommu *iommu)
1250{
1251 int cap_ptr = iommu->cap_ptr;
1252 u32 range, misc, low, high;
1253
c5081cd7 1254 iommu->dev = pci_get_bus_and_slot(PCI_BUS_NUM(iommu->devid),
23c742db
JR
1255 iommu->devid & 0xff);
1256 if (!iommu->dev)
1257 return -ENODEV;
1258
cbbc00be
JL
1259 /* Prevent binding other PCI device drivers to IOMMU devices */
1260 iommu->dev->match_driver = false;
1261
23c742db
JR
1262 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1263 &iommu->cap);
1264 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1265 &range);
1266 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1267 &misc);
1268
6f2729ba 1269 iommu->first_device = PCI_DEVID(MMIO_GET_BUS(range),
23c742db 1270 MMIO_GET_FD(range));
6f2729ba 1271 iommu->last_device = PCI_DEVID(MMIO_GET_BUS(range),
23c742db
JR
1272 MMIO_GET_LD(range));
1273
1274 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1275 amd_iommu_iotlb_sup = false;
1276
1277 /* read extended feature bits */
1278 low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1279 high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1280
1281 iommu->features = ((u64)high << 32) | low;
1282
1283 if (iommu_feature(iommu, FEATURE_GT)) {
1284 int glxval;
a919a018
SS
1285 u32 max_pasid;
1286 u64 pasmax;
23c742db 1287
a919a018
SS
1288 pasmax = iommu->features & FEATURE_PASID_MASK;
1289 pasmax >>= FEATURE_PASID_SHIFT;
1290 max_pasid = (1 << (pasmax + 1)) - 1;
23c742db 1291
a919a018
SS
1292 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid);
1293
1294 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK);
23c742db
JR
1295
1296 glxval = iommu->features & FEATURE_GLXVAL_MASK;
1297 glxval >>= FEATURE_GLXVAL_SHIFT;
1298
1299 if (amd_iommu_max_glx_val == -1)
1300 amd_iommu_max_glx_val = glxval;
1301 else
1302 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1303 }
1304
1305 if (iommu_feature(iommu, FEATURE_GT) &&
1306 iommu_feature(iommu, FEATURE_PPR)) {
1307 iommu->is_iommu_v2 = true;
1308 amd_iommu_v2_present = true;
1309 }
1310
1311 if (iommu_feature(iommu, FEATURE_PPR)) {
1312 iommu->ppr_log = alloc_ppr_log(iommu);
1313 if (!iommu->ppr_log)
1314 return -ENOMEM;
1315 }
1316
1317 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1318 amd_iommu_np_cache = true;
1319
30861ddc
SK
1320 init_iommu_perf_ctr(iommu);
1321
23c742db
JR
1322 if (is_rd890_iommu(iommu->dev)) {
1323 int i, j;
1324
1325 iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1326 PCI_DEVFN(0, 0));
1327
1328 /*
1329 * Some rd890 systems may not be fully reconfigured by the
1330 * BIOS, so it's necessary for us to store this information so
1331 * it can be reprogrammed on resume
1332 */
1333 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1334 &iommu->stored_addr_lo);
1335 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1336 &iommu->stored_addr_hi);
1337
1338 /* Low bit locks writes to configuration space */
1339 iommu->stored_addr_lo &= ~1;
1340
1341 for (i = 0; i < 6; i++)
1342 for (j = 0; j < 0x12; j++)
1343 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1344
1345 for (i = 0; i < 0x83; i++)
1346 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1347 }
1348
318fe782
SS
1349 amd_iommu_erratum_746_workaround(iommu);
1350
066f2e98
AW
1351 iommu->iommu_dev = iommu_device_create(&iommu->dev->dev, iommu,
1352 amd_iommu_groups, "ivhd%d",
1353 iommu->index);
1354
23c742db
JR
1355 return pci_enable_device(iommu->dev);
1356}
1357
4d121c32
JR
1358static void print_iommu_info(void)
1359{
1360 static const char * const feat_str[] = {
1361 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1362 "IA", "GA", "HE", "PC"
1363 };
1364 struct amd_iommu *iommu;
1365
1366 for_each_iommu(iommu) {
1367 int i;
1368
1369 pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1370 dev_name(&iommu->dev->dev), iommu->cap_ptr);
1371
1372 if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1373 pr_info("AMD-Vi: Extended features: ");
2bd5ed00 1374 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
4d121c32
JR
1375 if (iommu_feature(iommu, (1ULL << i)))
1376 pr_cont(" %s", feat_str[i]);
1377 }
30861ddc 1378 pr_cont("\n");
500c25ed 1379 }
4d121c32 1380 }
ebe60bbf
JR
1381 if (irq_remapping_enabled)
1382 pr_info("AMD-Vi: Interrupt remapping enabled\n");
4d121c32
JR
1383}
1384
2c0ae172 1385static int __init amd_iommu_init_pci(void)
23c742db
JR
1386{
1387 struct amd_iommu *iommu;
1388 int ret = 0;
1389
1390 for_each_iommu(iommu) {
1391 ret = iommu_init_pci(iommu);
1392 if (ret)
1393 break;
1394 }
1395
aafd8ba0
JR
1396 init_device_table_dma();
1397
1398 for_each_iommu(iommu)
1399 iommu_flush_all_caches(iommu);
1400
3a18404c 1401 ret = amd_iommu_init_api();
23c742db 1402
3a18404c
JR
1403 if (!ret)
1404 print_iommu_info();
4d121c32 1405
23c742db
JR
1406 return ret;
1407}
1408
a80dc3e0
JR
1409/****************************************************************************
1410 *
1411 * The following functions initialize the MSI interrupts for all IOMMUs
df805abb 1412 * in the system. It's a bit challenging because there could be multiple
a80dc3e0
JR
1413 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1414 * pci_dev.
1415 *
1416 ****************************************************************************/
1417
9f800de3 1418static int iommu_setup_msi(struct amd_iommu *iommu)
a80dc3e0
JR
1419{
1420 int r;
a80dc3e0 1421
9ddd592a
JR
1422 r = pci_enable_msi(iommu->dev);
1423 if (r)
1424 return r;
a80dc3e0 1425
72fe00f0
JR
1426 r = request_threaded_irq(iommu->dev->irq,
1427 amd_iommu_int_handler,
1428 amd_iommu_int_thread,
1429 0, "AMD-Vi",
3f398bc7 1430 iommu);
a80dc3e0
JR
1431
1432 if (r) {
1433 pci_disable_msi(iommu->dev);
9ddd592a 1434 return r;
a80dc3e0
JR
1435 }
1436
fab6afa3 1437 iommu->int_enabled = true;
1a29ac01 1438
a80dc3e0
JR
1439 return 0;
1440}
1441
05f92db9 1442static int iommu_init_msi(struct amd_iommu *iommu)
a80dc3e0 1443{
9ddd592a
JR
1444 int ret;
1445
a80dc3e0 1446 if (iommu->int_enabled)
9ddd592a 1447 goto enable_faults;
a80dc3e0 1448
82fcfc67 1449 if (iommu->dev->msi_cap)
9ddd592a
JR
1450 ret = iommu_setup_msi(iommu);
1451 else
1452 ret = -ENODEV;
1453
1454 if (ret)
1455 return ret;
a80dc3e0 1456
9ddd592a
JR
1457enable_faults:
1458 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
a80dc3e0 1459
9ddd592a
JR
1460 if (iommu->ppr_log != NULL)
1461 iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1462
1463 return 0;
a80dc3e0
JR
1464}
1465
b65233a9
JR
1466/****************************************************************************
1467 *
1468 * The next functions belong to the third pass of parsing the ACPI
1469 * table. In this last pass the memory mapping requirements are
df805abb 1470 * gathered (like exclusion and unity mapping ranges).
b65233a9
JR
1471 *
1472 ****************************************************************************/
1473
be2a022c
JR
1474static void __init free_unity_maps(void)
1475{
1476 struct unity_map_entry *entry, *next;
1477
1478 list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1479 list_del(&entry->list);
1480 kfree(entry);
1481 }
1482}
1483
b65233a9 1484/* called when we find an exclusion range definition in ACPI */
be2a022c
JR
1485static int __init init_exclusion_range(struct ivmd_header *m)
1486{
1487 int i;
1488
1489 switch (m->type) {
1490 case ACPI_IVMD_TYPE:
1491 set_device_exclusion_range(m->devid, m);
1492 break;
1493 case ACPI_IVMD_TYPE_ALL:
3a61ec38 1494 for (i = 0; i <= amd_iommu_last_bdf; ++i)
be2a022c
JR
1495 set_device_exclusion_range(i, m);
1496 break;
1497 case ACPI_IVMD_TYPE_RANGE:
1498 for (i = m->devid; i <= m->aux; ++i)
1499 set_device_exclusion_range(i, m);
1500 break;
1501 default:
1502 break;
1503 }
1504
1505 return 0;
1506}
1507
b65233a9 1508/* called for unity map ACPI definition */
be2a022c
JR
1509static int __init init_unity_map_range(struct ivmd_header *m)
1510{
98f1ad25 1511 struct unity_map_entry *e = NULL;
02acc43a 1512 char *s;
be2a022c
JR
1513
1514 e = kzalloc(sizeof(*e), GFP_KERNEL);
1515 if (e == NULL)
1516 return -ENOMEM;
1517
1518 switch (m->type) {
1519 default:
0bc252f4
JR
1520 kfree(e);
1521 return 0;
be2a022c 1522 case ACPI_IVMD_TYPE:
02acc43a 1523 s = "IVMD_TYPEi\t\t\t";
be2a022c
JR
1524 e->devid_start = e->devid_end = m->devid;
1525 break;
1526 case ACPI_IVMD_TYPE_ALL:
02acc43a 1527 s = "IVMD_TYPE_ALL\t\t";
be2a022c
JR
1528 e->devid_start = 0;
1529 e->devid_end = amd_iommu_last_bdf;
1530 break;
1531 case ACPI_IVMD_TYPE_RANGE:
02acc43a 1532 s = "IVMD_TYPE_RANGE\t\t";
be2a022c
JR
1533 e->devid_start = m->devid;
1534 e->devid_end = m->aux;
1535 break;
1536 }
1537 e->address_start = PAGE_ALIGN(m->range_start);
1538 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1539 e->prot = m->flags >> 1;
1540
02acc43a
JR
1541 DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1542 " range_start: %016llx range_end: %016llx flags: %x\n", s,
c5081cd7
SK
1543 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
1544 PCI_FUNC(e->devid_start), PCI_BUS_NUM(e->devid_end),
02acc43a
JR
1545 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1546 e->address_start, e->address_end, m->flags);
1547
be2a022c
JR
1548 list_add_tail(&e->list, &amd_iommu_unity_map);
1549
1550 return 0;
1551}
1552
b65233a9 1553/* iterates over all memory definitions we find in the ACPI table */
be2a022c
JR
1554static int __init init_memory_definitions(struct acpi_table_header *table)
1555{
1556 u8 *p = (u8 *)table, *end = (u8 *)table;
1557 struct ivmd_header *m;
1558
be2a022c
JR
1559 end += table->length;
1560 p += IVRS_HEADER_LENGTH;
1561
1562 while (p < end) {
1563 m = (struct ivmd_header *)p;
1564 if (m->flags & IVMD_FLAG_EXCL_RANGE)
1565 init_exclusion_range(m);
1566 else if (m->flags & IVMD_FLAG_UNITY_MAP)
1567 init_unity_map_range(m);
1568
1569 p += m->length;
1570 }
1571
1572 return 0;
1573}
1574
9f5f5fb3
JR
1575/*
1576 * Init the device table to not allow DMA access for devices and
1577 * suppress all page faults
1578 */
33f28c59 1579static void init_device_table_dma(void)
9f5f5fb3 1580{
0de66d5b 1581 u32 devid;
9f5f5fb3
JR
1582
1583 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1584 set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1585 set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
9f5f5fb3
JR
1586 }
1587}
1588
d04e0ba3
JR
1589static void __init uninit_device_table_dma(void)
1590{
1591 u32 devid;
1592
1593 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1594 amd_iommu_dev_table[devid].data[0] = 0ULL;
1595 amd_iommu_dev_table[devid].data[1] = 0ULL;
1596 }
1597}
1598
33f28c59
JR
1599static void init_device_table(void)
1600{
1601 u32 devid;
1602
1603 if (!amd_iommu_irq_remap)
1604 return;
1605
1606 for (devid = 0; devid <= amd_iommu_last_bdf; ++devid)
1607 set_dev_entry_bit(devid, DEV_ENTRY_IRQ_TBL_EN);
1608}
1609
e9bf5197
JR
1610static void iommu_init_flags(struct amd_iommu *iommu)
1611{
1612 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1613 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1614 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1615
1616 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1617 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1618 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1619
1620 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1621 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1622 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1623
1624 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1625 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1626 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1627
1628 /*
1629 * make IOMMU memory accesses cache coherent
1630 */
1631 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1456e9d2
JR
1632
1633 /* Set IOTLB invalidation timeout to 1s */
1634 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
e9bf5197
JR
1635}
1636
5bcd757f 1637static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
4c894f47 1638{
5bcd757f
MG
1639 int i, j;
1640 u32 ioc_feature_control;
c1bf94ec 1641 struct pci_dev *pdev = iommu->root_pdev;
5bcd757f
MG
1642
1643 /* RD890 BIOSes may not have completely reconfigured the iommu */
c1bf94ec 1644 if (!is_rd890_iommu(iommu->dev) || !pdev)
5bcd757f
MG
1645 return;
1646
1647 /*
1648 * First, we need to ensure that the iommu is enabled. This is
1649 * controlled by a register in the northbridge
1650 */
5bcd757f
MG
1651
1652 /* Select Northbridge indirect register 0x75 and enable writing */
1653 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1654 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1655
1656 /* Enable the iommu */
1657 if (!(ioc_feature_control & 0x1))
1658 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1659
5bcd757f
MG
1660 /* Restore the iommu BAR */
1661 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1662 iommu->stored_addr_lo);
1663 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1664 iommu->stored_addr_hi);
1665
1666 /* Restore the l1 indirect regs for each of the 6 l1s */
1667 for (i = 0; i < 6; i++)
1668 for (j = 0; j < 0x12; j++)
1669 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1670
1671 /* Restore the l2 indirect regs */
1672 for (i = 0; i < 0x83; i++)
1673 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1674
1675 /* Lock PCI setup registers */
1676 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1677 iommu->stored_addr_lo | 1);
4c894f47
JR
1678}
1679
b65233a9
JR
1680/*
1681 * This function finally enables all IOMMUs found in the system after
1682 * they have been initialized
1683 */
11ee5ac4 1684static void early_enable_iommus(void)
8736197b
JR
1685{
1686 struct amd_iommu *iommu;
1687
3bd22172 1688 for_each_iommu(iommu) {
a8c485bb 1689 iommu_disable(iommu);
e9bf5197 1690 iommu_init_flags(iommu);
58492e12
JR
1691 iommu_set_device_table(iommu);
1692 iommu_enable_command_buffer(iommu);
1693 iommu_enable_event_buffer(iommu);
8736197b
JR
1694 iommu_set_exclusion_range(iommu);
1695 iommu_enable(iommu);
7d0c5cc5 1696 iommu_flush_all_caches(iommu);
8736197b
JR
1697 }
1698}
1699
11ee5ac4
JR
1700static void enable_iommus_v2(void)
1701{
1702 struct amd_iommu *iommu;
1703
1704 for_each_iommu(iommu) {
1705 iommu_enable_ppr_log(iommu);
1706 iommu_enable_gt(iommu);
1707 }
1708}
1709
1710static void enable_iommus(void)
1711{
1712 early_enable_iommus();
1713
1714 enable_iommus_v2();
1715}
1716
92ac4320
JR
1717static void disable_iommus(void)
1718{
1719 struct amd_iommu *iommu;
1720
1721 for_each_iommu(iommu)
1722 iommu_disable(iommu);
1723}
1724
7441e9cb
JR
1725/*
1726 * Suspend/Resume support
1727 * disable suspend until real resume implemented
1728 */
1729
f3c6ea1b 1730static void amd_iommu_resume(void)
7441e9cb 1731{
5bcd757f
MG
1732 struct amd_iommu *iommu;
1733
1734 for_each_iommu(iommu)
1735 iommu_apply_resume_quirks(iommu);
1736
736501ee
JR
1737 /* re-load the hardware */
1738 enable_iommus();
3d9761e7
JR
1739
1740 amd_iommu_enable_interrupts();
7441e9cb
JR
1741}
1742
f3c6ea1b 1743static int amd_iommu_suspend(void)
7441e9cb 1744{
736501ee
JR
1745 /* disable IOMMUs to go out of the way for BIOS */
1746 disable_iommus();
1747
1748 return 0;
7441e9cb
JR
1749}
1750
f3c6ea1b 1751static struct syscore_ops amd_iommu_syscore_ops = {
7441e9cb
JR
1752 .suspend = amd_iommu_suspend,
1753 .resume = amd_iommu_resume,
1754};
1755
8704a1ba
JR
1756static void __init free_on_init_error(void)
1757{
0ea2c422
JR
1758 free_pages((unsigned long)irq_lookup_table,
1759 get_order(rlookup_table_size));
8704a1ba 1760
05152a04
JR
1761 if (amd_iommu_irq_cache) {
1762 kmem_cache_destroy(amd_iommu_irq_cache);
1763 amd_iommu_irq_cache = NULL;
0ea2c422 1764
05152a04 1765 }
8704a1ba
JR
1766
1767 free_pages((unsigned long)amd_iommu_rlookup_table,
1768 get_order(rlookup_table_size));
1769
1770 free_pages((unsigned long)amd_iommu_alias_table,
1771 get_order(alias_table_size));
1772
1773 free_pages((unsigned long)amd_iommu_dev_table,
1774 get_order(dev_table_size));
1775
1776 free_iommu_all();
1777
8704a1ba
JR
1778#ifdef CONFIG_GART_IOMMU
1779 /*
1780 * We failed to initialize the AMD IOMMU - try fallback to GART
1781 * if possible.
1782 */
1783 gart_iommu_init();
1784
1785#endif
1786}
1787
c2ff5cf5
JR
1788/* SB IOAPIC is always on this device in AMD systems */
1789#define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
1790
eb1eb7ae
JR
1791static bool __init check_ioapic_information(void)
1792{
dfbb6d47 1793 const char *fw_bug = FW_BUG;
c2ff5cf5 1794 bool ret, has_sb_ioapic;
eb1eb7ae
JR
1795 int idx;
1796
c2ff5cf5
JR
1797 has_sb_ioapic = false;
1798 ret = false;
eb1eb7ae 1799
dfbb6d47
JR
1800 /*
1801 * If we have map overrides on the kernel command line the
1802 * messages in this function might not describe firmware bugs
1803 * anymore - so be careful
1804 */
1805 if (cmdline_maps)
1806 fw_bug = "";
1807
c2ff5cf5
JR
1808 for (idx = 0; idx < nr_ioapics; idx++) {
1809 int devid, id = mpc_ioapic_id(idx);
1810
1811 devid = get_ioapic_devid(id);
1812 if (devid < 0) {
dfbb6d47
JR
1813 pr_err("%sAMD-Vi: IOAPIC[%d] not in IVRS table\n",
1814 fw_bug, id);
c2ff5cf5
JR
1815 ret = false;
1816 } else if (devid == IOAPIC_SB_DEVID) {
1817 has_sb_ioapic = true;
1818 ret = true;
eb1eb7ae
JR
1819 }
1820 }
1821
c2ff5cf5
JR
1822 if (!has_sb_ioapic) {
1823 /*
1824 * We expect the SB IOAPIC to be listed in the IVRS
1825 * table. The system timer is connected to the SB IOAPIC
1826 * and if we don't have it in the list the system will
1827 * panic at boot time. This situation usually happens
1828 * when the BIOS is buggy and provides us the wrong
1829 * device id for the IOAPIC in the system.
1830 */
dfbb6d47 1831 pr_err("%sAMD-Vi: No southbridge IOAPIC found\n", fw_bug);
c2ff5cf5
JR
1832 }
1833
1834 if (!ret)
dfbb6d47 1835 pr_err("AMD-Vi: Disabling interrupt remapping\n");
c2ff5cf5
JR
1836
1837 return ret;
eb1eb7ae
JR
1838}
1839
d04e0ba3
JR
1840static void __init free_dma_resources(void)
1841{
d04e0ba3
JR
1842 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1843 get_order(MAX_DOMAIN_ID/8));
1844
1845 free_unity_maps();
1846}
1847
b65233a9 1848/*
8704a1ba
JR
1849 * This is the hardware init function for AMD IOMMU in the system.
1850 * This function is called either from amd_iommu_init or from the interrupt
1851 * remapping setup code.
b65233a9
JR
1852 *
1853 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1854 * three times:
1855 *
1856 * 1 pass) Find the highest PCI device id the driver has to handle.
1857 * Upon this information the size of the data structures is
1858 * determined that needs to be allocated.
1859 *
1860 * 2 pass) Initialize the data structures just allocated with the
1861 * information in the ACPI table about available AMD IOMMUs
1862 * in the system. It also maps the PCI devices in the
1863 * system to specific IOMMUs
1864 *
1865 * 3 pass) After the basic data structures are allocated and
1866 * initialized we update them with information about memory
1867 * remapping requirements parsed out of the ACPI table in
1868 * this last pass.
1869 *
8704a1ba
JR
1870 * After everything is set up the IOMMUs are enabled and the necessary
1871 * hotplug and suspend notifiers are registered.
b65233a9 1872 */
643511b3 1873static int __init early_amd_iommu_init(void)
fe74c9cf 1874{
02f3b3f5
JR
1875 struct acpi_table_header *ivrs_base;
1876 acpi_size ivrs_size;
1877 acpi_status status;
fe74c9cf
JR
1878 int i, ret = 0;
1879
643511b3 1880 if (!amd_iommu_detected)
8704a1ba
JR
1881 return -ENODEV;
1882
02f3b3f5
JR
1883 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1884 if (status == AE_NOT_FOUND)
1885 return -ENODEV;
1886 else if (ACPI_FAILURE(status)) {
1887 const char *err = acpi_format_exception(status);
1888 pr_err("AMD-Vi: IVRS table error: %s\n", err);
1889 return -EINVAL;
1890 }
1891
fe74c9cf
JR
1892 /*
1893 * First parse ACPI tables to find the largest Bus/Dev/Func
1894 * we need to handle. Upon this information the shared data
1895 * structures for the IOMMUs in the system will be allocated
1896 */
2c0ae172
JR
1897 ret = find_last_devid_acpi(ivrs_base);
1898 if (ret)
3551a708
JR
1899 goto out;
1900
c571484e
JR
1901 dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1902 alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1903 rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
fe74c9cf 1904
fe74c9cf 1905 /* Device table - directly used by all IOMMUs */
8704a1ba 1906 ret = -ENOMEM;
5dc8bff0 1907 amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
fe74c9cf
JR
1908 get_order(dev_table_size));
1909 if (amd_iommu_dev_table == NULL)
1910 goto out;
1911
1912 /*
1913 * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1914 * IOMMU see for that device
1915 */
1916 amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1917 get_order(alias_table_size));
1918 if (amd_iommu_alias_table == NULL)
2c0ae172 1919 goto out;
fe74c9cf
JR
1920
1921 /* IOMMU rlookup table - find the IOMMU for a specific device */
83fd5cc6
JR
1922 amd_iommu_rlookup_table = (void *)__get_free_pages(
1923 GFP_KERNEL | __GFP_ZERO,
fe74c9cf
JR
1924 get_order(rlookup_table_size));
1925 if (amd_iommu_rlookup_table == NULL)
2c0ae172 1926 goto out;
fe74c9cf 1927
5dc8bff0
JR
1928 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1929 GFP_KERNEL | __GFP_ZERO,
fe74c9cf
JR
1930 get_order(MAX_DOMAIN_ID/8));
1931 if (amd_iommu_pd_alloc_bitmap == NULL)
2c0ae172 1932 goto out;
fe74c9cf
JR
1933
1934 /*
5dc8bff0 1935 * let all alias entries point to itself
fe74c9cf 1936 */
3a61ec38 1937 for (i = 0; i <= amd_iommu_last_bdf; ++i)
fe74c9cf
JR
1938 amd_iommu_alias_table[i] = i;
1939
fe74c9cf
JR
1940 /*
1941 * never allocate domain 0 because its used as the non-allocated and
1942 * error value placeholder
1943 */
1944 amd_iommu_pd_alloc_bitmap[0] = 1;
1945
aeb26f55
JR
1946 spin_lock_init(&amd_iommu_pd_lock);
1947
fe74c9cf
JR
1948 /*
1949 * now the data structures are allocated and basically initialized
1950 * start the real acpi table scan
1951 */
02f3b3f5
JR
1952 ret = init_iommu_all(ivrs_base);
1953 if (ret)
2c0ae172 1954 goto out;
fe74c9cf 1955
eb1eb7ae
JR
1956 if (amd_iommu_irq_remap)
1957 amd_iommu_irq_remap = check_ioapic_information();
1958
05152a04
JR
1959 if (amd_iommu_irq_remap) {
1960 /*
1961 * Interrupt remapping enabled, create kmem_cache for the
1962 * remapping tables.
1963 */
83ed9c13 1964 ret = -ENOMEM;
05152a04
JR
1965 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
1966 MAX_IRQS_PER_TABLE * sizeof(u32),
1967 IRQ_TABLE_ALIGNMENT,
1968 0, NULL);
1969 if (!amd_iommu_irq_cache)
1970 goto out;
0ea2c422
JR
1971
1972 irq_lookup_table = (void *)__get_free_pages(
1973 GFP_KERNEL | __GFP_ZERO,
1974 get_order(rlookup_table_size));
1975 if (!irq_lookup_table)
1976 goto out;
05152a04
JR
1977 }
1978
02f3b3f5
JR
1979 ret = init_memory_definitions(ivrs_base);
1980 if (ret)
2c0ae172 1981 goto out;
3551a708 1982
eb1eb7ae
JR
1983 /* init the device table */
1984 init_device_table();
1985
8704a1ba 1986out:
02f3b3f5
JR
1987 /* Don't leak any ACPI memory */
1988 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1989 ivrs_base = NULL;
1990
643511b3
JR
1991 return ret;
1992}
1993
ae295142 1994static int amd_iommu_enable_interrupts(void)
3d9761e7
JR
1995{
1996 struct amd_iommu *iommu;
1997 int ret = 0;
1998
1999 for_each_iommu(iommu) {
2000 ret = iommu_init_msi(iommu);
2001 if (ret)
2002 goto out;
2003 }
2004
2005out:
2006 return ret;
2007}
2008
02f3b3f5
JR
2009static bool detect_ivrs(void)
2010{
2011 struct acpi_table_header *ivrs_base;
2012 acpi_size ivrs_size;
2013 acpi_status status;
2014
2015 status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
2016 if (status == AE_NOT_FOUND)
2017 return false;
2018 else if (ACPI_FAILURE(status)) {
2019 const char *err = acpi_format_exception(status);
2020 pr_err("AMD-Vi: IVRS table error: %s\n", err);
2021 return false;
2022 }
2023
2024 early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
2025
1adb7d31
JR
2026 /* Make sure ACS will be enabled during PCI probe */
2027 pci_request_acs();
2028
02f3b3f5
JR
2029 return true;
2030}
2031
2c0ae172 2032/****************************************************************************
8704a1ba 2033 *
2c0ae172
JR
2034 * AMD IOMMU Initialization State Machine
2035 *
2036 ****************************************************************************/
2037
2038static int __init state_next(void)
8704a1ba
JR
2039{
2040 int ret = 0;
2041
2c0ae172
JR
2042 switch (init_state) {
2043 case IOMMU_START_STATE:
2044 if (!detect_ivrs()) {
2045 init_state = IOMMU_NOT_FOUND;
2046 ret = -ENODEV;
2047 } else {
2048 init_state = IOMMU_IVRS_DETECTED;
2049 }
2050 break;
2051 case IOMMU_IVRS_DETECTED:
2052 ret = early_amd_iommu_init();
2053 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
2054 break;
2055 case IOMMU_ACPI_FINISHED:
2056 early_enable_iommus();
2057 register_syscore_ops(&amd_iommu_syscore_ops);
2058 x86_platform.iommu_shutdown = disable_iommus;
2059 init_state = IOMMU_ENABLED;
2060 break;
2061 case IOMMU_ENABLED:
2062 ret = amd_iommu_init_pci();
2063 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
2064 enable_iommus_v2();
2065 break;
2066 case IOMMU_PCI_INIT:
2067 ret = amd_iommu_enable_interrupts();
2068 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
2069 break;
2070 case IOMMU_INTERRUPTS_EN:
1e6a7b04 2071 ret = amd_iommu_init_dma_ops();
2c0ae172
JR
2072 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
2073 break;
2074 case IOMMU_DMA_OPS:
2075 init_state = IOMMU_INITIALIZED;
2076 break;
2077 case IOMMU_INITIALIZED:
2078 /* Nothing to do */
2079 break;
2080 case IOMMU_NOT_FOUND:
2081 case IOMMU_INIT_ERROR:
2082 /* Error states => do nothing */
2083 ret = -EINVAL;
2084 break;
2085 default:
2086 /* Unknown state */
2087 BUG();
2088 }
3d9761e7 2089
2c0ae172
JR
2090 return ret;
2091}
7441e9cb 2092
2c0ae172
JR
2093static int __init iommu_go_to_state(enum iommu_init_state state)
2094{
2095 int ret = 0;
f5325094 2096
2c0ae172
JR
2097 while (init_state != state) {
2098 ret = state_next();
2099 if (init_state == IOMMU_NOT_FOUND ||
2100 init_state == IOMMU_INIT_ERROR)
2101 break;
2102 }
f2f12b6f 2103
fe74c9cf 2104 return ret;
2c0ae172 2105}
fe74c9cf 2106
6b474b82
JR
2107#ifdef CONFIG_IRQ_REMAP
2108int __init amd_iommu_prepare(void)
2109{
3f4cb7c0
TG
2110 int ret;
2111
7fa1c842 2112 amd_iommu_irq_remap = true;
84d07793 2113
3f4cb7c0
TG
2114 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
2115 if (ret)
2116 return ret;
2117 return amd_iommu_irq_remap ? 0 : -ENODEV;
6b474b82 2118}
d7f07769 2119
6b474b82
JR
2120int __init amd_iommu_enable(void)
2121{
2122 int ret;
2123
2124 ret = iommu_go_to_state(IOMMU_ENABLED);
2125 if (ret)
2126 return ret;
d7f07769 2127
6b474b82 2128 irq_remapping_enabled = 1;
d7f07769 2129
6b474b82
JR
2130 return 0;
2131}
2132
2133void amd_iommu_disable(void)
2134{
2135 amd_iommu_suspend();
2136}
2137
2138int amd_iommu_reenable(int mode)
2139{
2140 amd_iommu_resume();
2141
2142 return 0;
2143}
d7f07769 2144
6b474b82
JR
2145int __init amd_iommu_enable_faulting(void)
2146{
2147 /* We enable MSI later when PCI is initialized */
2148 return 0;
2149}
2150#endif
d7f07769 2151
2c0ae172
JR
2152/*
2153 * This is the core init function for AMD IOMMU hardware in the system.
2154 * This function is called from the generic x86 DMA layer initialization
2155 * code.
2156 */
2157static int __init amd_iommu_init(void)
2158{
2159 int ret;
2160
2161 ret = iommu_go_to_state(IOMMU_INITIALIZED);
2162 if (ret) {
d04e0ba3
JR
2163 free_dma_resources();
2164 if (!irq_remapping_enabled) {
2165 disable_iommus();
2166 free_on_init_error();
2167 } else {
2168 struct amd_iommu *iommu;
2169
2170 uninit_device_table_dma();
2171 for_each_iommu(iommu)
2172 iommu_flush_all_caches(iommu);
2173 }
2c0ae172
JR
2174 }
2175
2176 return ret;
fe74c9cf
JR
2177}
2178
b65233a9
JR
2179/****************************************************************************
2180 *
2181 * Early detect code. This code runs at IOMMU detection time in the DMA
2182 * layer. It just looks if there is an IVRS ACPI table to detect AMD
2183 * IOMMUs
2184 *
2185 ****************************************************************************/
480125ba 2186int __init amd_iommu_detect(void)
ae7877de 2187{
2c0ae172 2188 int ret;
02f3b3f5 2189
75f1cdf1 2190 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
480125ba 2191 return -ENODEV;
ae7877de 2192
a5235725 2193 if (amd_iommu_disabled)
480125ba 2194 return -ENODEV;
a5235725 2195
2c0ae172
JR
2196 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
2197 if (ret)
2198 return ret;
11bd04f6 2199
02f3b3f5
JR
2200 amd_iommu_detected = true;
2201 iommu_detected = 1;
2202 x86_init.iommu.iommu_init = amd_iommu_init;
2203
02f3b3f5 2204 return 0;
ae7877de
JR
2205}
2206
b65233a9
JR
2207/****************************************************************************
2208 *
2209 * Parsing functions for the AMD IOMMU specific kernel command line
2210 * options.
2211 *
2212 ****************************************************************************/
2213
fefda117
JR
2214static int __init parse_amd_iommu_dump(char *str)
2215{
2216 amd_iommu_dump = true;
2217
2218 return 1;
2219}
2220
918ad6c5
JR
2221static int __init parse_amd_iommu_options(char *str)
2222{
2223 for (; *str; ++str) {
695b5676 2224 if (strncmp(str, "fullflush", 9) == 0)
afa9fdc2 2225 amd_iommu_unmap_flush = true;
a5235725
JR
2226 if (strncmp(str, "off", 3) == 0)
2227 amd_iommu_disabled = true;
5abcdba4
JR
2228 if (strncmp(str, "force_isolation", 15) == 0)
2229 amd_iommu_force_isolation = true;
918ad6c5
JR
2230 }
2231
2232 return 1;
2233}
2234
440e8998
JR
2235static int __init parse_ivrs_ioapic(char *str)
2236{
2237 unsigned int bus, dev, fn;
2238 int ret, id, i;
2239 u16 devid;
2240
2241 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2242
2243 if (ret != 4) {
2244 pr_err("AMD-Vi: Invalid command line: ivrs_ioapic%s\n", str);
2245 return 1;
2246 }
2247
2248 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
2249 pr_err("AMD-Vi: Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
2250 str);
2251 return 1;
2252 }
2253
2254 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2255
dfbb6d47 2256 cmdline_maps = true;
440e8998
JR
2257 i = early_ioapic_map_size++;
2258 early_ioapic_map[i].id = id;
2259 early_ioapic_map[i].devid = devid;
2260 early_ioapic_map[i].cmd_line = true;
2261
2262 return 1;
2263}
2264
2265static int __init parse_ivrs_hpet(char *str)
2266{
2267 unsigned int bus, dev, fn;
2268 int ret, id, i;
2269 u16 devid;
2270
2271 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn);
2272
2273 if (ret != 4) {
2274 pr_err("AMD-Vi: Invalid command line: ivrs_hpet%s\n", str);
2275 return 1;
2276 }
2277
2278 if (early_hpet_map_size == EARLY_MAP_SIZE) {
2279 pr_err("AMD-Vi: Early HPET map overflow - ignoring ivrs_hpet%s\n",
2280 str);
2281 return 1;
2282 }
2283
2284 devid = ((bus & 0xff) << 8) | ((dev & 0x1f) << 3) | (fn & 0x7);
2285
dfbb6d47 2286 cmdline_maps = true;
440e8998
JR
2287 i = early_hpet_map_size++;
2288 early_hpet_map[i].id = id;
2289 early_hpet_map[i].devid = devid;
2290 early_hpet_map[i].cmd_line = true;
2291
2292 return 1;
2293}
2294
2295__setup("amd_iommu_dump", parse_amd_iommu_dump);
2296__setup("amd_iommu=", parse_amd_iommu_options);
2297__setup("ivrs_ioapic", parse_ivrs_ioapic);
2298__setup("ivrs_hpet", parse_ivrs_hpet);
22e6daf4
KRW
2299
2300IOMMU_INIT_FINISH(amd_iommu_detect,
2301 gart_iommu_hole_init,
98f1ad25
JR
2302 NULL,
2303 NULL);
400a28a0
JR
2304
2305bool amd_iommu_v2_supported(void)
2306{
2307 return amd_iommu_v2_present;
2308}
2309EXPORT_SYMBOL(amd_iommu_v2_supported);
30861ddc
SK
2310
2311/****************************************************************************
2312 *
2313 * IOMMU EFR Performance Counter support functionality. This code allows
2314 * access to the IOMMU PC functionality.
2315 *
2316 ****************************************************************************/
2317
2318u8 amd_iommu_pc_get_max_banks(u16 devid)
2319{
2320 struct amd_iommu *iommu;
2321 u8 ret = 0;
2322
2323 /* locate the iommu governing the devid */
2324 iommu = amd_iommu_rlookup_table[devid];
2325 if (iommu)
2326 ret = iommu->max_banks;
2327
2328 return ret;
2329}
2330EXPORT_SYMBOL(amd_iommu_pc_get_max_banks);
2331
2332bool amd_iommu_pc_supported(void)
2333{
2334 return amd_iommu_pc_present;
2335}
2336EXPORT_SYMBOL(amd_iommu_pc_supported);
2337
2338u8 amd_iommu_pc_get_max_counters(u16 devid)
2339{
2340 struct amd_iommu *iommu;
2341 u8 ret = 0;
2342
2343 /* locate the iommu governing the devid */
2344 iommu = amd_iommu_rlookup_table[devid];
2345 if (iommu)
2346 ret = iommu->max_counters;
2347
2348 return ret;
2349}
2350EXPORT_SYMBOL(amd_iommu_pc_get_max_counters);
2351
2352int amd_iommu_pc_get_set_reg_val(u16 devid, u8 bank, u8 cntr, u8 fxn,
2353 u64 *value, bool is_write)
2354{
2355 struct amd_iommu *iommu;
2356 u32 offset;
2357 u32 max_offset_lim;
2358
2359 /* Make sure the IOMMU PC resource is available */
2360 if (!amd_iommu_pc_present)
2361 return -ENODEV;
2362
2363 /* Locate the iommu associated with the device ID */
2364 iommu = amd_iommu_rlookup_table[devid];
2365
2366 /* Check for valid iommu and pc register indexing */
2367 if (WARN_ON((iommu == NULL) || (fxn > 0x28) || (fxn & 7)))
2368 return -ENODEV;
2369
2370 offset = (u32)(((0x40|bank) << 12) | (cntr << 8) | fxn);
2371
2372 /* Limit the offset to the hw defined mmio region aperture */
2373 max_offset_lim = (u32)(((0x40|iommu->max_banks) << 12) |
2374 (iommu->max_counters << 8) | 0x28);
2375 if ((offset < MMIO_CNTR_REG_OFFSET) ||
2376 (offset > max_offset_lim))
2377 return -EINVAL;
2378
2379 if (is_write) {
2380 writel((u32)*value, iommu->mmio_base + offset);
2381 writel((*value >> 32), iommu->mmio_base + offset + 4);
2382 } else {
2383 *value = readl(iommu->mmio_base + offset + 4);
2384 *value <<= 32;
2385 *value = readl(iommu->mmio_base + offset);
2386 }
2387
2388 return 0;
2389}
2390EXPORT_SYMBOL(amd_iommu_pc_get_set_reg_val);