Root/drivers/iommu/amd_iommu_init.c

1/*
2 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3 * Author: Joerg Roedel <joerg.roedel@amd.com>
4 * Leo Duran <leo.duran@amd.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19
20#include <linux/pci.h>
21#include <linux/acpi.h>
22#include <linux/list.h>
23#include <linux/slab.h>
24#include <linux/syscore_ops.h>
25#include <linux/interrupt.h>
26#include <linux/msi.h>
27#include <linux/amd-iommu.h>
28#include <linux/export.h>
29#include <linux/acpi.h>
30#include <acpi/acpi.h>
31#include <asm/pci-direct.h>
32#include <asm/iommu.h>
33#include <asm/gart.h>
34#include <asm/x86_init.h>
35#include <asm/iommu_table.h>
36
37#include "amd_iommu_proto.h"
38#include "amd_iommu_types.h"
39
40/*
41 * definitions for the ACPI scanning code
42 */
43#define IVRS_HEADER_LENGTH 48
44
45#define ACPI_IVHD_TYPE 0x10
46#define ACPI_IVMD_TYPE_ALL 0x20
47#define ACPI_IVMD_TYPE 0x21
48#define ACPI_IVMD_TYPE_RANGE 0x22
49
50#define IVHD_DEV_ALL 0x01
51#define IVHD_DEV_SELECT 0x02
52#define IVHD_DEV_SELECT_RANGE_START 0x03
53#define IVHD_DEV_RANGE_END 0x04
54#define IVHD_DEV_ALIAS 0x42
55#define IVHD_DEV_ALIAS_RANGE 0x43
56#define IVHD_DEV_EXT_SELECT 0x46
57#define IVHD_DEV_EXT_SELECT_RANGE 0x47
58
59#define IVHD_FLAG_HT_TUN_EN_MASK 0x01
60#define IVHD_FLAG_PASSPW_EN_MASK 0x02
61#define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
62#define IVHD_FLAG_ISOC_EN_MASK 0x08
63
64#define IVMD_FLAG_EXCL_RANGE 0x08
65#define IVMD_FLAG_UNITY_MAP 0x01
66
67#define ACPI_DEVFLAG_INITPASS 0x01
68#define ACPI_DEVFLAG_EXTINT 0x02
69#define ACPI_DEVFLAG_NMI 0x04
70#define ACPI_DEVFLAG_SYSMGT1 0x10
71#define ACPI_DEVFLAG_SYSMGT2 0x20
72#define ACPI_DEVFLAG_LINT0 0x40
73#define ACPI_DEVFLAG_LINT1 0x80
74#define ACPI_DEVFLAG_ATSDIS 0x10000000
75
76/*
77 * ACPI table definitions
78 *
79 * These data structures are laid over the table to parse the important values
80 * out of it.
81 */
82
83/*
84 * structure describing one IOMMU in the ACPI table. Typically followed by one
85 * or more ivhd_entrys.
86 */
87struct ivhd_header {
88    u8 type;
89    u8 flags;
90    u16 length;
91    u16 devid;
92    u16 cap_ptr;
93    u64 mmio_phys;
94    u16 pci_seg;
95    u16 info;
96    u32 reserved;
97} __attribute__((packed));
98
99/*
100 * A device entry describing which devices a specific IOMMU translates and
101 * which requestor ids they use.
102 */
103struct ivhd_entry {
104    u8 type;
105    u16 devid;
106    u8 flags;
107    u32 ext;
108} __attribute__((packed));
109
110/*
111 * An AMD IOMMU memory definition structure. It defines things like exclusion
112 * ranges for devices and regions that should be unity mapped.
113 */
114struct ivmd_header {
115    u8 type;
116    u8 flags;
117    u16 length;
118    u16 devid;
119    u16 aux;
120    u64 resv;
121    u64 range_start;
122    u64 range_length;
123} __attribute__((packed));
124
125bool amd_iommu_dump;
126
127static bool amd_iommu_detected;
128static bool __initdata amd_iommu_disabled;
129
130u16 amd_iommu_last_bdf; /* largest PCI device id we have
131                       to handle */
132LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
133                       we find in ACPI */
134u32 amd_iommu_unmap_flush; /* if true, flush on every unmap */
135
136LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
137                       system */
138
139/* Array to assign indices to IOMMUs*/
140struct amd_iommu *amd_iommus[MAX_IOMMUS];
141int amd_iommus_present;
142
143/* IOMMUs have a non-present cache? */
144bool amd_iommu_np_cache __read_mostly;
145bool amd_iommu_iotlb_sup __read_mostly = true;
146
147u32 amd_iommu_max_pasids __read_mostly = ~0;
148
149bool amd_iommu_v2_present __read_mostly;
150
151bool amd_iommu_force_isolation __read_mostly;
152
153/*
154 * List of protection domains - used during resume
155 */
156LIST_HEAD(amd_iommu_pd_list);
157spinlock_t amd_iommu_pd_lock;
158
159/*
160 * Pointer to the device table which is shared by all AMD IOMMUs
161 * it is indexed by the PCI device id or the HT unit id and contains
162 * information about the domain the device belongs to as well as the
163 * page table root pointer.
164 */
165struct dev_table_entry *amd_iommu_dev_table;
166
167/*
168 * The alias table is a driver specific data structure which contains the
169 * mappings of the PCI device ids to the actual requestor ids on the IOMMU.
170 * More than one device can share the same requestor id.
171 */
172u16 *amd_iommu_alias_table;
173
174/*
175 * The rlookup table is used to find the IOMMU which is responsible
176 * for a specific device. It is also indexed by the PCI device id.
177 */
178struct amd_iommu **amd_iommu_rlookup_table;
179
180/*
181 * AMD IOMMU allows up to 2^16 differend protection domains. This is a bitmap
182 * to know which ones are already in use.
183 */
184unsigned long *amd_iommu_pd_alloc_bitmap;
185
186static u32 dev_table_size; /* size of the device table */
187static u32 alias_table_size; /* size of the alias table */
188static u32 rlookup_table_size; /* size if the rlookup table */
189
190enum iommu_init_state {
191    IOMMU_START_STATE,
192    IOMMU_IVRS_DETECTED,
193    IOMMU_ACPI_FINISHED,
194    IOMMU_ENABLED,
195    IOMMU_PCI_INIT,
196    IOMMU_INTERRUPTS_EN,
197    IOMMU_DMA_OPS,
198    IOMMU_INITIALIZED,
199    IOMMU_NOT_FOUND,
200    IOMMU_INIT_ERROR,
201};
202
203static enum iommu_init_state init_state = IOMMU_START_STATE;
204
205static int amd_iommu_enable_interrupts(void);
206static int __init iommu_go_to_state(enum iommu_init_state state);
207
208static inline void update_last_devid(u16 devid)
209{
210    if (devid > amd_iommu_last_bdf)
211        amd_iommu_last_bdf = devid;
212}
213
214static inline unsigned long tbl_size(int entry_size)
215{
216    unsigned shift = PAGE_SHIFT +
217             get_order(((int)amd_iommu_last_bdf + 1) * entry_size);
218
219    return 1UL << shift;
220}
221
222/* Access to l1 and l2 indexed register spaces */
223
224static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
225{
226    u32 val;
227
228    pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
229    pci_read_config_dword(iommu->dev, 0xfc, &val);
230    return val;
231}
232
233static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
234{
235    pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
236    pci_write_config_dword(iommu->dev, 0xfc, val);
237    pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
238}
239
240static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
241{
242    u32 val;
243
244    pci_write_config_dword(iommu->dev, 0xf0, address);
245    pci_read_config_dword(iommu->dev, 0xf4, &val);
246    return val;
247}
248
249static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
250{
251    pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
252    pci_write_config_dword(iommu->dev, 0xf4, val);
253}
254
255/****************************************************************************
256 *
257 * AMD IOMMU MMIO register space handling functions
258 *
259 * These functions are used to program the IOMMU device registers in
260 * MMIO space required for that driver.
261 *
262 ****************************************************************************/
263
264/*
265 * This function set the exclusion range in the IOMMU. DMA accesses to the
266 * exclusion range are passed through untranslated
267 */
268static void iommu_set_exclusion_range(struct amd_iommu *iommu)
269{
270    u64 start = iommu->exclusion_start & PAGE_MASK;
271    u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
272    u64 entry;
273
274    if (!iommu->exclusion_start)
275        return;
276
277    entry = start | MMIO_EXCL_ENABLE_MASK;
278    memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
279            &entry, sizeof(entry));
280
281    entry = limit;
282    memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
283            &entry, sizeof(entry));
284}
285
286/* Programs the physical address of the device table into the IOMMU hardware */
287static void iommu_set_device_table(struct amd_iommu *iommu)
288{
289    u64 entry;
290
291    BUG_ON(iommu->mmio_base == NULL);
292
293    entry = virt_to_phys(amd_iommu_dev_table);
294    entry |= (dev_table_size >> 12) - 1;
295    memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
296            &entry, sizeof(entry));
297}
298
299/* Generic functions to enable/disable certain features of the IOMMU. */
300static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
301{
302    u32 ctrl;
303
304    ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
305    ctrl |= (1 << bit);
306    writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
307}
308
309static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
310{
311    u32 ctrl;
312
313    ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
314    ctrl &= ~(1 << bit);
315    writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
316}
317
318static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
319{
320    u32 ctrl;
321
322    ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
323    ctrl &= ~CTRL_INV_TO_MASK;
324    ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
325    writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
326}
327
328/* Function to enable the hardware */
329static void iommu_enable(struct amd_iommu *iommu)
330{
331    iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
332}
333
334static void iommu_disable(struct amd_iommu *iommu)
335{
336    /* Disable command buffer */
337    iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
338
339    /* Disable event logging and event interrupts */
340    iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
341    iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
342
343    /* Disable IOMMU hardware itself */
344    iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
345}
346
347/*
348 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
349 * the system has one.
350 */
351static u8 __iomem * __init iommu_map_mmio_space(u64 address)
352{
353    if (!request_mem_region(address, MMIO_REGION_LENGTH, "amd_iommu")) {
354        pr_err("AMD-Vi: Can not reserve memory region %llx for mmio\n",
355            address);
356        pr_err("AMD-Vi: This is a BIOS bug. Please contact your hardware vendor\n");
357        return NULL;
358    }
359
360    return (u8 __iomem *)ioremap_nocache(address, MMIO_REGION_LENGTH);
361}
362
363static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
364{
365    if (iommu->mmio_base)
366        iounmap(iommu->mmio_base);
367    release_mem_region(iommu->mmio_phys, MMIO_REGION_LENGTH);
368}
369
370/****************************************************************************
371 *
372 * The functions below belong to the first pass of AMD IOMMU ACPI table
373 * parsing. In this pass we try to find out the highest device id this
374 * code has to handle. Upon this information the size of the shared data
375 * structures is determined later.
376 *
377 ****************************************************************************/
378
379/*
380 * This function calculates the length of a given IVHD entry
381 */
382static inline int ivhd_entry_length(u8 *ivhd)
383{
384    return 0x04 << (*ivhd >> 6);
385}
386
387/*
388 * This function reads the last device id the IOMMU has to handle from the PCI
389 * capability header for this IOMMU
390 */
391static int __init find_last_devid_on_pci(int bus, int dev, int fn, int cap_ptr)
392{
393    u32 cap;
394
395    cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
396    update_last_devid(calc_devid(MMIO_GET_BUS(cap), MMIO_GET_LD(cap)));
397
398    return 0;
399}
400
401/*
402 * After reading the highest device id from the IOMMU PCI capability header
403 * this function looks if there is a higher device id defined in the ACPI table
404 */
405static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
406{
407    u8 *p = (void *)h, *end = (void *)h;
408    struct ivhd_entry *dev;
409
410    p += sizeof(*h);
411    end += h->length;
412
413    find_last_devid_on_pci(PCI_BUS(h->devid),
414            PCI_SLOT(h->devid),
415            PCI_FUNC(h->devid),
416            h->cap_ptr);
417
418    while (p < end) {
419        dev = (struct ivhd_entry *)p;
420        switch (dev->type) {
421        case IVHD_DEV_SELECT:
422        case IVHD_DEV_RANGE_END:
423        case IVHD_DEV_ALIAS:
424        case IVHD_DEV_EXT_SELECT:
425            /* all the above subfield types refer to device ids */
426            update_last_devid(dev->devid);
427            break;
428        default:
429            break;
430        }
431        p += ivhd_entry_length(p);
432    }
433
434    WARN_ON(p != end);
435
436    return 0;
437}
438
439/*
440 * Iterate over all IVHD entries in the ACPI table and find the highest device
441 * id which we need to handle. This is the first of three functions which parse
442 * the ACPI table. So we check the checksum here.
443 */
444static int __init find_last_devid_acpi(struct acpi_table_header *table)
445{
446    int i;
447    u8 checksum = 0, *p = (u8 *)table, *end = (u8 *)table;
448    struct ivhd_header *h;
449
450    /*
451     * Validate checksum here so we don't need to do it when
452     * we actually parse the table
453     */
454    for (i = 0; i < table->length; ++i)
455        checksum += p[i];
456    if (checksum != 0)
457        /* ACPI table corrupt */
458        return -ENODEV;
459
460    p += IVRS_HEADER_LENGTH;
461
462    end += table->length;
463    while (p < end) {
464        h = (struct ivhd_header *)p;
465        switch (h->type) {
466        case ACPI_IVHD_TYPE:
467            find_last_devid_from_ivhd(h);
468            break;
469        default:
470            break;
471        }
472        p += h->length;
473    }
474    WARN_ON(p != end);
475
476    return 0;
477}
478
479/****************************************************************************
480 *
481 * The following functions belong the the code path which parses the ACPI table
482 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
483 * data structures, initialize the device/alias/rlookup table and also
484 * basically initialize the hardware.
485 *
486 ****************************************************************************/
487
488/*
489 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
490 * write commands to that buffer later and the IOMMU will execute them
491 * asynchronously
492 */
493static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
494{
495    u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
496            get_order(CMD_BUFFER_SIZE));
497
498    if (cmd_buf == NULL)
499        return NULL;
500
501    iommu->cmd_buf_size = CMD_BUFFER_SIZE | CMD_BUFFER_UNINITIALIZED;
502
503    return cmd_buf;
504}
505
506/*
507 * This function resets the command buffer if the IOMMU stopped fetching
508 * commands from it.
509 */
510void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
511{
512    iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
513
514    writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
515    writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
516
517    iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
518}
519
520/*
521 * This function writes the command buffer address to the hardware and
522 * enables it.
523 */
524static void iommu_enable_command_buffer(struct amd_iommu *iommu)
525{
526    u64 entry;
527
528    BUG_ON(iommu->cmd_buf == NULL);
529
530    entry = (u64)virt_to_phys(iommu->cmd_buf);
531    entry |= MMIO_CMD_SIZE_512;
532
533    memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
534            &entry, sizeof(entry));
535
536    amd_iommu_reset_cmd_buffer(iommu);
537    iommu->cmd_buf_size &= ~(CMD_BUFFER_UNINITIALIZED);
538}
539
540static void __init free_command_buffer(struct amd_iommu *iommu)
541{
542    free_pages((unsigned long)iommu->cmd_buf,
543           get_order(iommu->cmd_buf_size & ~(CMD_BUFFER_UNINITIALIZED)));
544}
545
546/* allocates the memory where the IOMMU will log its events to */
547static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
548{
549    iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
550                        get_order(EVT_BUFFER_SIZE));
551
552    if (iommu->evt_buf == NULL)
553        return NULL;
554
555    iommu->evt_buf_size = EVT_BUFFER_SIZE;
556
557    return iommu->evt_buf;
558}
559
560static void iommu_enable_event_buffer(struct amd_iommu *iommu)
561{
562    u64 entry;
563
564    BUG_ON(iommu->evt_buf == NULL);
565
566    entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
567
568    memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
569            &entry, sizeof(entry));
570
571    /* set head and tail to zero manually */
572    writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
573    writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
574
575    iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
576}
577
578static void __init free_event_buffer(struct amd_iommu *iommu)
579{
580    free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
581}
582
583/* allocates the memory where the IOMMU will log its events to */
584static u8 * __init alloc_ppr_log(struct amd_iommu *iommu)
585{
586    iommu->ppr_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
587                        get_order(PPR_LOG_SIZE));
588
589    if (iommu->ppr_log == NULL)
590        return NULL;
591
592    return iommu->ppr_log;
593}
594
595static void iommu_enable_ppr_log(struct amd_iommu *iommu)
596{
597    u64 entry;
598
599    if (iommu->ppr_log == NULL)
600        return;
601
602    entry = (u64)virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
603
604    memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
605            &entry, sizeof(entry));
606
607    /* set head and tail to zero manually */
608    writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
609    writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
610
611    iommu_feature_enable(iommu, CONTROL_PPFLOG_EN);
612    iommu_feature_enable(iommu, CONTROL_PPR_EN);
613}
614
615static void __init free_ppr_log(struct amd_iommu *iommu)
616{
617    if (iommu->ppr_log == NULL)
618        return;
619
620    free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
621}
622
623static void iommu_enable_gt(struct amd_iommu *iommu)
624{
625    if (!iommu_feature(iommu, FEATURE_GT))
626        return;
627
628    iommu_feature_enable(iommu, CONTROL_GT_EN);
629}
630
631/* sets a specific bit in the device table entry. */
632static void set_dev_entry_bit(u16 devid, u8 bit)
633{
634    int i = (bit >> 6) & 0x03;
635    int _bit = bit & 0x3f;
636
637    amd_iommu_dev_table[devid].data[i] |= (1UL << _bit);
638}
639
640static int get_dev_entry_bit(u16 devid, u8 bit)
641{
642    int i = (bit >> 6) & 0x03;
643    int _bit = bit & 0x3f;
644
645    return (amd_iommu_dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
646}
647
648
649void amd_iommu_apply_erratum_63(u16 devid)
650{
651    int sysmgt;
652
653    sysmgt = get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1) |
654         (get_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2) << 1);
655
656    if (sysmgt == 0x01)
657        set_dev_entry_bit(devid, DEV_ENTRY_IW);
658}
659
660/* Writes the specific IOMMU for a device into the rlookup table */
661static void __init set_iommu_for_device(struct amd_iommu *iommu, u16 devid)
662{
663    amd_iommu_rlookup_table[devid] = iommu;
664}
665
666/*
667 * This function takes the device specific flags read from the ACPI
668 * table and sets up the device table entry with that information
669 */
670static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
671                       u16 devid, u32 flags, u32 ext_flags)
672{
673    if (flags & ACPI_DEVFLAG_INITPASS)
674        set_dev_entry_bit(devid, DEV_ENTRY_INIT_PASS);
675    if (flags & ACPI_DEVFLAG_EXTINT)
676        set_dev_entry_bit(devid, DEV_ENTRY_EINT_PASS);
677    if (flags & ACPI_DEVFLAG_NMI)
678        set_dev_entry_bit(devid, DEV_ENTRY_NMI_PASS);
679    if (flags & ACPI_DEVFLAG_SYSMGT1)
680        set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT1);
681    if (flags & ACPI_DEVFLAG_SYSMGT2)
682        set_dev_entry_bit(devid, DEV_ENTRY_SYSMGT2);
683    if (flags & ACPI_DEVFLAG_LINT0)
684        set_dev_entry_bit(devid, DEV_ENTRY_LINT0_PASS);
685    if (flags & ACPI_DEVFLAG_LINT1)
686        set_dev_entry_bit(devid, DEV_ENTRY_LINT1_PASS);
687
688    amd_iommu_apply_erratum_63(devid);
689
690    set_iommu_for_device(iommu, devid);
691}
692
693/*
694 * Reads the device exclusion range from ACPI and initialize IOMMU with
695 * it
696 */
697static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
698{
699    struct amd_iommu *iommu = amd_iommu_rlookup_table[devid];
700
701    if (!(m->flags & IVMD_FLAG_EXCL_RANGE))
702        return;
703
704    if (iommu) {
705        /*
706         * We only can configure exclusion ranges per IOMMU, not
707         * per device. But we can enable the exclusion range per
708         * device. This is done here
709         */
710        set_dev_entry_bit(m->devid, DEV_ENTRY_EX);
711        iommu->exclusion_start = m->range_start;
712        iommu->exclusion_length = m->range_length;
713    }
714}
715
716/*
717 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
718 * initializes the hardware and our data structures with it.
719 */
720static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
721                    struct ivhd_header *h)
722{
723    u8 *p = (u8 *)h;
724    u8 *end = p, flags = 0;
725    u16 devid = 0, devid_start = 0, devid_to = 0;
726    u32 dev_i, ext_flags = 0;
727    bool alias = false;
728    struct ivhd_entry *e;
729
730    /*
731     * First save the recommended feature enable bits from ACPI
732     */
733    iommu->acpi_flags = h->flags;
734
735    /*
736     * Done. Now parse the device entries
737     */
738    p += sizeof(struct ivhd_header);
739    end += h->length;
740
741
742    while (p < end) {
743        e = (struct ivhd_entry *)p;
744        switch (e->type) {
745        case IVHD_DEV_ALL:
746
747            DUMP_printk(" DEV_ALL\t\t\t first devid: %02x:%02x.%x"
748                    " last device %02x:%02x.%x flags: %02x\n",
749                    PCI_BUS(iommu->first_device),
750                    PCI_SLOT(iommu->first_device),
751                    PCI_FUNC(iommu->first_device),
752                    PCI_BUS(iommu->last_device),
753                    PCI_SLOT(iommu->last_device),
754                    PCI_FUNC(iommu->last_device),
755                    e->flags);
756
757            for (dev_i = iommu->first_device;
758                    dev_i <= iommu->last_device; ++dev_i)
759                set_dev_entry_from_acpi(iommu, dev_i,
760                            e->flags, 0);
761            break;
762        case IVHD_DEV_SELECT:
763
764            DUMP_printk(" DEV_SELECT\t\t\t devid: %02x:%02x.%x "
765                    "flags: %02x\n",
766                    PCI_BUS(e->devid),
767                    PCI_SLOT(e->devid),
768                    PCI_FUNC(e->devid),
769                    e->flags);
770
771            devid = e->devid;
772            set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
773            break;
774        case IVHD_DEV_SELECT_RANGE_START:
775
776            DUMP_printk(" DEV_SELECT_RANGE_START\t "
777                    "devid: %02x:%02x.%x flags: %02x\n",
778                    PCI_BUS(e->devid),
779                    PCI_SLOT(e->devid),
780                    PCI_FUNC(e->devid),
781                    e->flags);
782
783            devid_start = e->devid;
784            flags = e->flags;
785            ext_flags = 0;
786            alias = false;
787            break;
788        case IVHD_DEV_ALIAS:
789
790            DUMP_printk(" DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
791                    "flags: %02x devid_to: %02x:%02x.%x\n",
792                    PCI_BUS(e->devid),
793                    PCI_SLOT(e->devid),
794                    PCI_FUNC(e->devid),
795                    e->flags,
796                    PCI_BUS(e->ext >> 8),
797                    PCI_SLOT(e->ext >> 8),
798                    PCI_FUNC(e->ext >> 8));
799
800            devid = e->devid;
801            devid_to = e->ext >> 8;
802            set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
803            set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
804            amd_iommu_alias_table[devid] = devid_to;
805            break;
806        case IVHD_DEV_ALIAS_RANGE:
807
808            DUMP_printk(" DEV_ALIAS_RANGE\t\t "
809                    "devid: %02x:%02x.%x flags: %02x "
810                    "devid_to: %02x:%02x.%x\n",
811                    PCI_BUS(e->devid),
812                    PCI_SLOT(e->devid),
813                    PCI_FUNC(e->devid),
814                    e->flags,
815                    PCI_BUS(e->ext >> 8),
816                    PCI_SLOT(e->ext >> 8),
817                    PCI_FUNC(e->ext >> 8));
818
819            devid_start = e->devid;
820            flags = e->flags;
821            devid_to = e->ext >> 8;
822            ext_flags = 0;
823            alias = true;
824            break;
825        case IVHD_DEV_EXT_SELECT:
826
827            DUMP_printk(" DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
828                    "flags: %02x ext: %08x\n",
829                    PCI_BUS(e->devid),
830                    PCI_SLOT(e->devid),
831                    PCI_FUNC(e->devid),
832                    e->flags, e->ext);
833
834            devid = e->devid;
835            set_dev_entry_from_acpi(iommu, devid, e->flags,
836                        e->ext);
837            break;
838        case IVHD_DEV_EXT_SELECT_RANGE:
839
840            DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
841                    "%02x:%02x.%x flags: %02x ext: %08x\n",
842                    PCI_BUS(e->devid),
843                    PCI_SLOT(e->devid),
844                    PCI_FUNC(e->devid),
845                    e->flags, e->ext);
846
847            devid_start = e->devid;
848            flags = e->flags;
849            ext_flags = e->ext;
850            alias = false;
851            break;
852        case IVHD_DEV_RANGE_END:
853
854            DUMP_printk(" DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
855                    PCI_BUS(e->devid),
856                    PCI_SLOT(e->devid),
857                    PCI_FUNC(e->devid));
858
859            devid = e->devid;
860            for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
861                if (alias) {
862                    amd_iommu_alias_table[dev_i] = devid_to;
863                    set_dev_entry_from_acpi(iommu,
864                        devid_to, flags, ext_flags);
865                }
866                set_dev_entry_from_acpi(iommu, dev_i,
867                            flags, ext_flags);
868            }
869            break;
870        default:
871            break;
872        }
873
874        p += ivhd_entry_length(p);
875    }
876}
877
878/* Initializes the device->iommu mapping for the driver */
879static int __init init_iommu_devices(struct amd_iommu *iommu)
880{
881    u32 i;
882
883    for (i = iommu->first_device; i <= iommu->last_device; ++i)
884        set_iommu_for_device(iommu, i);
885
886    return 0;
887}
888
889static void __init free_iommu_one(struct amd_iommu *iommu)
890{
891    free_command_buffer(iommu);
892    free_event_buffer(iommu);
893    free_ppr_log(iommu);
894    iommu_unmap_mmio_space(iommu);
895}
896
897static void __init free_iommu_all(void)
898{
899    struct amd_iommu *iommu, *next;
900
901    for_each_iommu_safe(iommu, next) {
902        list_del(&iommu->list);
903        free_iommu_one(iommu);
904        kfree(iommu);
905    }
906}
907
908/*
909 * This function clues the initialization function for one IOMMU
910 * together and also allocates the command buffer and programs the
911 * hardware. It does NOT enable the IOMMU. This is done afterwards.
912 */
913static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
914{
915    spin_lock_init(&iommu->lock);
916
917    /* Add IOMMU to internal data structures */
918    list_add_tail(&iommu->list, &amd_iommu_list);
919    iommu->index = amd_iommus_present++;
920
921    if (unlikely(iommu->index >= MAX_IOMMUS)) {
922        WARN(1, "AMD-Vi: System has more IOMMUs than supported by this driver\n");
923        return -ENOSYS;
924    }
925
926    /* Index is fine - add IOMMU to the array */
927    amd_iommus[iommu->index] = iommu;
928
929    /*
930     * Copy data from ACPI table entry to the iommu struct
931     */
932    iommu->devid = h->devid;
933    iommu->cap_ptr = h->cap_ptr;
934    iommu->pci_seg = h->pci_seg;
935    iommu->mmio_phys = h->mmio_phys;
936    iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
937    if (!iommu->mmio_base)
938        return -ENOMEM;
939
940    iommu->cmd_buf = alloc_command_buffer(iommu);
941    if (!iommu->cmd_buf)
942        return -ENOMEM;
943
944    iommu->evt_buf = alloc_event_buffer(iommu);
945    if (!iommu->evt_buf)
946        return -ENOMEM;
947
948    iommu->int_enabled = false;
949
950    init_iommu_from_acpi(iommu, h);
951    init_iommu_devices(iommu);
952
953    return 0;
954}
955
956/*
957 * Iterates over all IOMMU entries in the ACPI table, allocates the
958 * IOMMU structure and initializes it with init_iommu_one()
959 */
960static int __init init_iommu_all(struct acpi_table_header *table)
961{
962    u8 *p = (u8 *)table, *end = (u8 *)table;
963    struct ivhd_header *h;
964    struct amd_iommu *iommu;
965    int ret;
966
967    end += table->length;
968    p += IVRS_HEADER_LENGTH;
969
970    while (p < end) {
971        h = (struct ivhd_header *)p;
972        switch (*p) {
973        case ACPI_IVHD_TYPE:
974
975            DUMP_printk("device: %02x:%02x.%01x cap: %04x "
976                    "seg: %d flags: %01x info %04x\n",
977                    PCI_BUS(h->devid), PCI_SLOT(h->devid),
978                    PCI_FUNC(h->devid), h->cap_ptr,
979                    h->pci_seg, h->flags, h->info);
980            DUMP_printk(" mmio-addr: %016llx\n",
981                    h->mmio_phys);
982
983            iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
984            if (iommu == NULL)
985                return -ENOMEM;
986
987            ret = init_iommu_one(iommu, h);
988            if (ret)
989                return ret;
990            break;
991        default:
992            break;
993        }
994        p += h->length;
995
996    }
997    WARN_ON(p != end);
998
999    return 0;
1000}
1001
1002static int iommu_init_pci(struct amd_iommu *iommu)
1003{
1004    int cap_ptr = iommu->cap_ptr;
1005    u32 range, misc, low, high;
1006
1007    iommu->dev = pci_get_bus_and_slot(PCI_BUS(iommu->devid),
1008                      iommu->devid & 0xff);
1009    if (!iommu->dev)
1010        return -ENODEV;
1011
1012    pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
1013                  &iommu->cap);
1014    pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
1015                  &range);
1016    pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
1017                  &misc);
1018
1019    iommu->first_device = calc_devid(MMIO_GET_BUS(range),
1020                     MMIO_GET_FD(range));
1021    iommu->last_device = calc_devid(MMIO_GET_BUS(range),
1022                    MMIO_GET_LD(range));
1023
1024    if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
1025        amd_iommu_iotlb_sup = false;
1026
1027    /* read extended feature bits */
1028    low = readl(iommu->mmio_base + MMIO_EXT_FEATURES);
1029    high = readl(iommu->mmio_base + MMIO_EXT_FEATURES + 4);
1030
1031    iommu->features = ((u64)high << 32) | low;
1032
1033    if (iommu_feature(iommu, FEATURE_GT)) {
1034        int glxval;
1035        u32 pasids;
1036        u64 shift;
1037
1038        shift = iommu->features & FEATURE_PASID_MASK;
1039        shift >>= FEATURE_PASID_SHIFT;
1040        pasids = (1 << shift);
1041
1042        amd_iommu_max_pasids = min(amd_iommu_max_pasids, pasids);
1043
1044        glxval = iommu->features & FEATURE_GLXVAL_MASK;
1045        glxval >>= FEATURE_GLXVAL_SHIFT;
1046
1047        if (amd_iommu_max_glx_val == -1)
1048            amd_iommu_max_glx_val = glxval;
1049        else
1050            amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
1051    }
1052
1053    if (iommu_feature(iommu, FEATURE_GT) &&
1054        iommu_feature(iommu, FEATURE_PPR)) {
1055        iommu->is_iommu_v2 = true;
1056        amd_iommu_v2_present = true;
1057    }
1058
1059    if (iommu_feature(iommu, FEATURE_PPR)) {
1060        iommu->ppr_log = alloc_ppr_log(iommu);
1061        if (!iommu->ppr_log)
1062            return -ENOMEM;
1063    }
1064
1065    if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE))
1066        amd_iommu_np_cache = true;
1067
1068    if (is_rd890_iommu(iommu->dev)) {
1069        int i, j;
1070
1071        iommu->root_pdev = pci_get_bus_and_slot(iommu->dev->bus->number,
1072                PCI_DEVFN(0, 0));
1073
1074        /*
1075         * Some rd890 systems may not be fully reconfigured by the
1076         * BIOS, so it's necessary for us to store this information so
1077         * it can be reprogrammed on resume
1078         */
1079        pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
1080                &iommu->stored_addr_lo);
1081        pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
1082                &iommu->stored_addr_hi);
1083
1084        /* Low bit locks writes to configuration space */
1085        iommu->stored_addr_lo &= ~1;
1086
1087        for (i = 0; i < 6; i++)
1088            for (j = 0; j < 0x12; j++)
1089                iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
1090
1091        for (i = 0; i < 0x83; i++)
1092            iommu->stored_l2[i] = iommu_read_l2(iommu, i);
1093    }
1094
1095    return pci_enable_device(iommu->dev);
1096}
1097
1098static void print_iommu_info(void)
1099{
1100    static const char * const feat_str[] = {
1101        "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
1102        "IA", "GA", "HE", "PC"
1103    };
1104    struct amd_iommu *iommu;
1105
1106    for_each_iommu(iommu) {
1107        int i;
1108
1109        pr_info("AMD-Vi: Found IOMMU at %s cap 0x%hx\n",
1110            dev_name(&iommu->dev->dev), iommu->cap_ptr);
1111
1112        if (iommu->cap & (1 << IOMMU_CAP_EFR)) {
1113            pr_info("AMD-Vi: Extended features: ");
1114            for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
1115                if (iommu_feature(iommu, (1ULL << i)))
1116                    pr_cont(" %s", feat_str[i]);
1117            }
1118        }
1119        pr_cont("\n");
1120    }
1121}
1122
1123static int __init amd_iommu_init_pci(void)
1124{
1125    struct amd_iommu *iommu;
1126    int ret = 0;
1127
1128    for_each_iommu(iommu) {
1129        ret = iommu_init_pci(iommu);
1130        if (ret)
1131            break;
1132    }
1133
1134    ret = amd_iommu_init_devices();
1135
1136    print_iommu_info();
1137
1138    return ret;
1139}
1140
1141/****************************************************************************
1142 *
1143 * The following functions initialize the MSI interrupts for all IOMMUs
1144 * in the system. Its a bit challenging because there could be multiple
1145 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
1146 * pci_dev.
1147 *
1148 ****************************************************************************/
1149
1150static int iommu_setup_msi(struct amd_iommu *iommu)
1151{
1152    int r;
1153
1154    r = pci_enable_msi(iommu->dev);
1155    if (r)
1156        return r;
1157
1158    r = request_threaded_irq(iommu->dev->irq,
1159                 amd_iommu_int_handler,
1160                 amd_iommu_int_thread,
1161                 0, "AMD-Vi",
1162                 iommu->dev);
1163
1164    if (r) {
1165        pci_disable_msi(iommu->dev);
1166        return r;
1167    }
1168
1169    iommu->int_enabled = true;
1170
1171    return 0;
1172}
1173
1174static int iommu_init_msi(struct amd_iommu *iommu)
1175{
1176    int ret;
1177
1178    if (iommu->int_enabled)
1179        goto enable_faults;
1180
1181    if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
1182        ret = iommu_setup_msi(iommu);
1183    else
1184        ret = -ENODEV;
1185
1186    if (ret)
1187        return ret;
1188
1189enable_faults:
1190    iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
1191
1192    if (iommu->ppr_log != NULL)
1193        iommu_feature_enable(iommu, CONTROL_PPFINT_EN);
1194
1195    return 0;
1196}
1197
1198/****************************************************************************
1199 *
1200 * The next functions belong to the third pass of parsing the ACPI
1201 * table. In this last pass the memory mapping requirements are
1202 * gathered (like exclusion and unity mapping reanges).
1203 *
1204 ****************************************************************************/
1205
1206static void __init free_unity_maps(void)
1207{
1208    struct unity_map_entry *entry, *next;
1209
1210    list_for_each_entry_safe(entry, next, &amd_iommu_unity_map, list) {
1211        list_del(&entry->list);
1212        kfree(entry);
1213    }
1214}
1215
1216/* called when we find an exclusion range definition in ACPI */
1217static int __init init_exclusion_range(struct ivmd_header *m)
1218{
1219    int i;
1220
1221    switch (m->type) {
1222    case ACPI_IVMD_TYPE:
1223        set_device_exclusion_range(m->devid, m);
1224        break;
1225    case ACPI_IVMD_TYPE_ALL:
1226        for (i = 0; i <= amd_iommu_last_bdf; ++i)
1227            set_device_exclusion_range(i, m);
1228        break;
1229    case ACPI_IVMD_TYPE_RANGE:
1230        for (i = m->devid; i <= m->aux; ++i)
1231            set_device_exclusion_range(i, m);
1232        break;
1233    default:
1234        break;
1235    }
1236
1237    return 0;
1238}
1239
1240/* called for unity map ACPI definition */
1241static int __init init_unity_map_range(struct ivmd_header *m)
1242{
1243    struct unity_map_entry *e = NULL;
1244    char *s;
1245
1246    e = kzalloc(sizeof(*e), GFP_KERNEL);
1247    if (e == NULL)
1248        return -ENOMEM;
1249
1250    switch (m->type) {
1251    default:
1252        kfree(e);
1253        return 0;
1254    case ACPI_IVMD_TYPE:
1255        s = "IVMD_TYPEi\t\t\t";
1256        e->devid_start = e->devid_end = m->devid;
1257        break;
1258    case ACPI_IVMD_TYPE_ALL:
1259        s = "IVMD_TYPE_ALL\t\t";
1260        e->devid_start = 0;
1261        e->devid_end = amd_iommu_last_bdf;
1262        break;
1263    case ACPI_IVMD_TYPE_RANGE:
1264        s = "IVMD_TYPE_RANGE\t\t";
1265        e->devid_start = m->devid;
1266        e->devid_end = m->aux;
1267        break;
1268    }
1269    e->address_start = PAGE_ALIGN(m->range_start);
1270    e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
1271    e->prot = m->flags >> 1;
1272
1273    DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
1274            " range_start: %016llx range_end: %016llx flags: %x\n", s,
1275            PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
1276            PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
1277            PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
1278            e->address_start, e->address_end, m->flags);
1279
1280    list_add_tail(&e->list, &amd_iommu_unity_map);
1281
1282    return 0;
1283}
1284
1285/* iterates over all memory definitions we find in the ACPI table */
1286static int __init init_memory_definitions(struct acpi_table_header *table)
1287{
1288    u8 *p = (u8 *)table, *end = (u8 *)table;
1289    struct ivmd_header *m;
1290
1291    end += table->length;
1292    p += IVRS_HEADER_LENGTH;
1293
1294    while (p < end) {
1295        m = (struct ivmd_header *)p;
1296        if (m->flags & IVMD_FLAG_EXCL_RANGE)
1297            init_exclusion_range(m);
1298        else if (m->flags & IVMD_FLAG_UNITY_MAP)
1299            init_unity_map_range(m);
1300
1301        p += m->length;
1302    }
1303
1304    return 0;
1305}
1306
1307/*
1308 * Init the device table to not allow DMA access for devices and
1309 * suppress all page faults
1310 */
1311static void init_device_table(void)
1312{
1313    u32 devid;
1314
1315    for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
1316        set_dev_entry_bit(devid, DEV_ENTRY_VALID);
1317        set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
1318    }
1319}
1320
1321static void iommu_init_flags(struct amd_iommu *iommu)
1322{
1323    iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
1324        iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
1325        iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
1326
1327    iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
1328        iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
1329        iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
1330
1331    iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
1332        iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
1333        iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
1334
1335    iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
1336        iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
1337        iommu_feature_disable(iommu, CONTROL_ISOC_EN);
1338
1339    /*
1340     * make IOMMU memory accesses cache coherent
1341     */
1342    iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
1343
1344    /* Set IOTLB invalidation timeout to 1s */
1345    iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
1346}
1347
1348static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
1349{
1350    int i, j;
1351    u32 ioc_feature_control;
1352    struct pci_dev *pdev = iommu->root_pdev;
1353
1354    /* RD890 BIOSes may not have completely reconfigured the iommu */
1355    if (!is_rd890_iommu(iommu->dev) || !pdev)
1356        return;
1357
1358    /*
1359     * First, we need to ensure that the iommu is enabled. This is
1360     * controlled by a register in the northbridge
1361     */
1362
1363    /* Select Northbridge indirect register 0x75 and enable writing */
1364    pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
1365    pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
1366
1367    /* Enable the iommu */
1368    if (!(ioc_feature_control & 0x1))
1369        pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
1370
1371    /* Restore the iommu BAR */
1372    pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1373                   iommu->stored_addr_lo);
1374    pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
1375                   iommu->stored_addr_hi);
1376
1377    /* Restore the l1 indirect regs for each of the 6 l1s */
1378    for (i = 0; i < 6; i++)
1379        for (j = 0; j < 0x12; j++)
1380            iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
1381
1382    /* Restore the l2 indirect regs */
1383    for (i = 0; i < 0x83; i++)
1384        iommu_write_l2(iommu, i, iommu->stored_l2[i]);
1385
1386    /* Lock PCI setup registers */
1387    pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
1388                   iommu->stored_addr_lo | 1);
1389}
1390
1391/*
1392 * This function finally enables all IOMMUs found in the system after
1393 * they have been initialized
1394 */
1395static void early_enable_iommus(void)
1396{
1397    struct amd_iommu *iommu;
1398
1399    for_each_iommu(iommu) {
1400        iommu_disable(iommu);
1401        iommu_init_flags(iommu);
1402        iommu_set_device_table(iommu);
1403        iommu_enable_command_buffer(iommu);
1404        iommu_enable_event_buffer(iommu);
1405        iommu_set_exclusion_range(iommu);
1406        iommu_enable(iommu);
1407        iommu_flush_all_caches(iommu);
1408    }
1409}
1410
1411static void enable_iommus_v2(void)
1412{
1413    struct amd_iommu *iommu;
1414
1415    for_each_iommu(iommu) {
1416        iommu_enable_ppr_log(iommu);
1417        iommu_enable_gt(iommu);
1418    }
1419}
1420
1421static void enable_iommus(void)
1422{
1423    early_enable_iommus();
1424
1425    enable_iommus_v2();
1426}
1427
1428static void disable_iommus(void)
1429{
1430    struct amd_iommu *iommu;
1431
1432    for_each_iommu(iommu)
1433        iommu_disable(iommu);
1434}
1435
1436/*
1437 * Suspend/Resume support
1438 * disable suspend until real resume implemented
1439 */
1440
1441static void amd_iommu_resume(void)
1442{
1443    struct amd_iommu *iommu;
1444
1445    for_each_iommu(iommu)
1446        iommu_apply_resume_quirks(iommu);
1447
1448    /* re-load the hardware */
1449    enable_iommus();
1450
1451    amd_iommu_enable_interrupts();
1452}
1453
1454static int amd_iommu_suspend(void)
1455{
1456    /* disable IOMMUs to go out of the way for BIOS */
1457    disable_iommus();
1458
1459    return 0;
1460}
1461
1462static struct syscore_ops amd_iommu_syscore_ops = {
1463    .suspend = amd_iommu_suspend,
1464    .resume = amd_iommu_resume,
1465};
1466
1467static void __init free_on_init_error(void)
1468{
1469    amd_iommu_uninit_devices();
1470
1471    free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
1472           get_order(MAX_DOMAIN_ID/8));
1473
1474    free_pages((unsigned long)amd_iommu_rlookup_table,
1475           get_order(rlookup_table_size));
1476
1477    free_pages((unsigned long)amd_iommu_alias_table,
1478           get_order(alias_table_size));
1479
1480    free_pages((unsigned long)amd_iommu_dev_table,
1481           get_order(dev_table_size));
1482
1483    free_iommu_all();
1484
1485    free_unity_maps();
1486
1487#ifdef CONFIG_GART_IOMMU
1488    /*
1489     * We failed to initialize the AMD IOMMU - try fallback to GART
1490     * if possible.
1491     */
1492    gart_iommu_init();
1493
1494#endif
1495}
1496
1497/*
1498 * This is the hardware init function for AMD IOMMU in the system.
1499 * This function is called either from amd_iommu_init or from the interrupt
1500 * remapping setup code.
1501 *
1502 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
1503 * three times:
1504 *
1505 * 1 pass) Find the highest PCI device id the driver has to handle.
1506 * Upon this information the size of the data structures is
1507 * determined that needs to be allocated.
1508 *
1509 * 2 pass) Initialize the data structures just allocated with the
1510 * information in the ACPI table about available AMD IOMMUs
1511 * in the system. It also maps the PCI devices in the
1512 * system to specific IOMMUs
1513 *
1514 * 3 pass) After the basic data structures are allocated and
1515 * initialized we update them with information about memory
1516 * remapping requirements parsed out of the ACPI table in
1517 * this last pass.
1518 *
1519 * After everything is set up the IOMMUs are enabled and the necessary
1520 * hotplug and suspend notifiers are registered.
1521 */
1522static int __init early_amd_iommu_init(void)
1523{
1524    struct acpi_table_header *ivrs_base;
1525    acpi_size ivrs_size;
1526    acpi_status status;
1527    int i, ret = 0;
1528
1529    if (!amd_iommu_detected)
1530        return -ENODEV;
1531
1532    status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1533    if (status == AE_NOT_FOUND)
1534        return -ENODEV;
1535    else if (ACPI_FAILURE(status)) {
1536        const char *err = acpi_format_exception(status);
1537        pr_err("AMD-Vi: IVRS table error: %s\n", err);
1538        return -EINVAL;
1539    }
1540
1541    /*
1542     * First parse ACPI tables to find the largest Bus/Dev/Func
1543     * we need to handle. Upon this information the shared data
1544     * structures for the IOMMUs in the system will be allocated
1545     */
1546    ret = find_last_devid_acpi(ivrs_base);
1547    if (ret)
1548        goto out;
1549
1550    dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE);
1551    alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE);
1552    rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE);
1553
1554    /* Device table - directly used by all IOMMUs */
1555    ret = -ENOMEM;
1556    amd_iommu_dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1557                      get_order(dev_table_size));
1558    if (amd_iommu_dev_table == NULL)
1559        goto out;
1560
1561    /*
1562     * Alias table - map PCI Bus/Dev/Func to Bus/Dev/Func the
1563     * IOMMU see for that device
1564     */
1565    amd_iommu_alias_table = (void *)__get_free_pages(GFP_KERNEL,
1566            get_order(alias_table_size));
1567    if (amd_iommu_alias_table == NULL)
1568        goto out;
1569
1570    /* IOMMU rlookup table - find the IOMMU for a specific device */
1571    amd_iommu_rlookup_table = (void *)__get_free_pages(
1572            GFP_KERNEL | __GFP_ZERO,
1573            get_order(rlookup_table_size));
1574    if (amd_iommu_rlookup_table == NULL)
1575        goto out;
1576
1577    amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages(
1578                        GFP_KERNEL | __GFP_ZERO,
1579                        get_order(MAX_DOMAIN_ID/8));
1580    if (amd_iommu_pd_alloc_bitmap == NULL)
1581        goto out;
1582
1583    /* init the device table */
1584    init_device_table();
1585
1586    /*
1587     * let all alias entries point to itself
1588     */
1589    for (i = 0; i <= amd_iommu_last_bdf; ++i)
1590        amd_iommu_alias_table[i] = i;
1591
1592    /*
1593     * never allocate domain 0 because its used as the non-allocated and
1594     * error value placeholder
1595     */
1596    amd_iommu_pd_alloc_bitmap[0] = 1;
1597
1598    spin_lock_init(&amd_iommu_pd_lock);
1599
1600    /*
1601     * now the data structures are allocated and basically initialized
1602     * start the real acpi table scan
1603     */
1604    ret = init_iommu_all(ivrs_base);
1605    if (ret)
1606        goto out;
1607
1608    ret = init_memory_definitions(ivrs_base);
1609    if (ret)
1610        goto out;
1611
1612out:
1613    /* Don't leak any ACPI memory */
1614    early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1615    ivrs_base = NULL;
1616
1617    return ret;
1618}
1619
1620static int amd_iommu_enable_interrupts(void)
1621{
1622    struct amd_iommu *iommu;
1623    int ret = 0;
1624
1625    for_each_iommu(iommu) {
1626        ret = iommu_init_msi(iommu);
1627        if (ret)
1628            goto out;
1629    }
1630
1631out:
1632    return ret;
1633}
1634
1635static bool detect_ivrs(void)
1636{
1637    struct acpi_table_header *ivrs_base;
1638    acpi_size ivrs_size;
1639    acpi_status status;
1640
1641    status = acpi_get_table_with_size("IVRS", 0, &ivrs_base, &ivrs_size);
1642    if (status == AE_NOT_FOUND)
1643        return false;
1644    else if (ACPI_FAILURE(status)) {
1645        const char *err = acpi_format_exception(status);
1646        pr_err("AMD-Vi: IVRS table error: %s\n", err);
1647        return false;
1648    }
1649
1650    early_acpi_os_unmap_memory((char __iomem *)ivrs_base, ivrs_size);
1651
1652    /* Make sure ACS will be enabled during PCI probe */
1653    pci_request_acs();
1654
1655    return true;
1656}
1657
1658static int amd_iommu_init_dma(void)
1659{
1660    int ret;
1661
1662    if (iommu_pass_through)
1663        ret = amd_iommu_init_passthrough();
1664    else
1665        ret = amd_iommu_init_dma_ops();
1666
1667    if (ret)
1668        return ret;
1669
1670    amd_iommu_init_api();
1671
1672    amd_iommu_init_notifier();
1673
1674    return 0;
1675}
1676
1677/****************************************************************************
1678 *
1679 * AMD IOMMU Initialization State Machine
1680 *
1681 ****************************************************************************/
1682
1683static int __init state_next(void)
1684{
1685    int ret = 0;
1686
1687    switch (init_state) {
1688    case IOMMU_START_STATE:
1689        if (!detect_ivrs()) {
1690            init_state = IOMMU_NOT_FOUND;
1691            ret = -ENODEV;
1692        } else {
1693            init_state = IOMMU_IVRS_DETECTED;
1694        }
1695        break;
1696    case IOMMU_IVRS_DETECTED:
1697        ret = early_amd_iommu_init();
1698        init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
1699        break;
1700    case IOMMU_ACPI_FINISHED:
1701        early_enable_iommus();
1702        register_syscore_ops(&amd_iommu_syscore_ops);
1703        x86_platform.iommu_shutdown = disable_iommus;
1704        init_state = IOMMU_ENABLED;
1705        break;
1706    case IOMMU_ENABLED:
1707        ret = amd_iommu_init_pci();
1708        init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
1709        enable_iommus_v2();
1710        break;
1711    case IOMMU_PCI_INIT:
1712        ret = amd_iommu_enable_interrupts();
1713        init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
1714        break;
1715    case IOMMU_INTERRUPTS_EN:
1716        ret = amd_iommu_init_dma();
1717        init_state = ret ? IOMMU_INIT_ERROR : IOMMU_DMA_OPS;
1718        break;
1719    case IOMMU_DMA_OPS:
1720        init_state = IOMMU_INITIALIZED;
1721        break;
1722    case IOMMU_INITIALIZED:
1723        /* Nothing to do */
1724        break;
1725    case IOMMU_NOT_FOUND:
1726    case IOMMU_INIT_ERROR:
1727        /* Error states => do nothing */
1728        ret = -EINVAL;
1729        break;
1730    default:
1731        /* Unknown state */
1732        BUG();
1733    }
1734
1735    return ret;
1736}
1737
1738static int __init iommu_go_to_state(enum iommu_init_state state)
1739{
1740    int ret = 0;
1741
1742    while (init_state != state) {
1743        ret = state_next();
1744        if (init_state == IOMMU_NOT_FOUND ||
1745            init_state == IOMMU_INIT_ERROR)
1746            break;
1747    }
1748
1749    return ret;
1750}
1751
1752
1753
1754/*
1755 * This is the core init function for AMD IOMMU hardware in the system.
1756 * This function is called from the generic x86 DMA layer initialization
1757 * code.
1758 */
1759static int __init amd_iommu_init(void)
1760{
1761    int ret;
1762
1763    ret = iommu_go_to_state(IOMMU_INITIALIZED);
1764    if (ret) {
1765        disable_iommus();
1766        free_on_init_error();
1767    }
1768
1769    return ret;
1770}
1771
1772/****************************************************************************
1773 *
1774 * Early detect code. This code runs at IOMMU detection time in the DMA
1775 * layer. It just looks if there is an IVRS ACPI table to detect AMD
1776 * IOMMUs
1777 *
1778 ****************************************************************************/
1779int __init amd_iommu_detect(void)
1780{
1781    int ret;
1782
1783    if (no_iommu || (iommu_detected && !gart_iommu_aperture))
1784        return -ENODEV;
1785
1786    if (amd_iommu_disabled)
1787        return -ENODEV;
1788
1789    ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
1790    if (ret)
1791        return ret;
1792
1793    amd_iommu_detected = true;
1794    iommu_detected = 1;
1795    x86_init.iommu.iommu_init = amd_iommu_init;
1796
1797    return 0;
1798}
1799
1800/****************************************************************************
1801 *
1802 * Parsing functions for the AMD IOMMU specific kernel command line
1803 * options.
1804 *
1805 ****************************************************************************/
1806
1807static int __init parse_amd_iommu_dump(char *str)
1808{
1809    amd_iommu_dump = true;
1810
1811    return 1;
1812}
1813
1814static int __init parse_amd_iommu_options(char *str)
1815{
1816    for (; *str; ++str) {
1817        if (strncmp(str, "fullflush", 9) == 0)
1818            amd_iommu_unmap_flush = true;
1819        if (strncmp(str, "off", 3) == 0)
1820            amd_iommu_disabled = true;
1821        if (strncmp(str, "force_isolation", 15) == 0)
1822            amd_iommu_force_isolation = true;
1823    }
1824
1825    return 1;
1826}
1827
1828__setup("amd_iommu_dump", parse_amd_iommu_dump);
1829__setup("amd_iommu=", parse_amd_iommu_options);
1830
1831IOMMU_INIT_FINISH(amd_iommu_detect,
1832          gart_iommu_hole_init,
1833          NULL,
1834          NULL);
1835
1836bool amd_iommu_v2_supported(void)
1837{
1838    return amd_iommu_v2_present;
1839}
1840EXPORT_SYMBOL(amd_iommu_v2_supported);
1841

Archive Download this file



interactive