Commit | Line | Data |
---|---|---|
b6c02715 | 1 | /* |
5d0d7156 | 2 | * Copyright (C) 2007-2010 Advanced Micro Devices, Inc. |
b6c02715 JR |
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 | ||
72e1dcc4 | 20 | #include <linux/ratelimit.h> |
b6c02715 | 21 | #include <linux/pci.h> |
cb41ed85 | 22 | #include <linux/pci-ats.h> |
a66022c4 | 23 | #include <linux/bitmap.h> |
5a0e3ad6 | 24 | #include <linux/slab.h> |
7f26508b | 25 | #include <linux/debugfs.h> |
b6c02715 | 26 | #include <linux/scatterlist.h> |
51491367 | 27 | #include <linux/dma-mapping.h> |
b6c02715 | 28 | #include <linux/iommu-helper.h> |
c156e347 | 29 | #include <linux/iommu.h> |
815b33fd | 30 | #include <linux/delay.h> |
403f81d8 | 31 | #include <linux/amd-iommu.h> |
72e1dcc4 JR |
32 | #include <linux/notifier.h> |
33 | #include <linux/export.h> | |
2b324506 JR |
34 | #include <linux/irq.h> |
35 | #include <linux/msi.h> | |
36 | #include <asm/irq_remapping.h> | |
37 | #include <asm/io_apic.h> | |
38 | #include <asm/apic.h> | |
39 | #include <asm/hw_irq.h> | |
17f5b569 | 40 | #include <asm/msidef.h> |
b6c02715 | 41 | #include <asm/proto.h> |
46a7fa27 | 42 | #include <asm/iommu.h> |
1d9b16d1 | 43 | #include <asm/gart.h> |
27c2127a | 44 | #include <asm/dma.h> |
403f81d8 JR |
45 | |
46 | #include "amd_iommu_proto.h" | |
47 | #include "amd_iommu_types.h" | |
6b474b82 | 48 | #include "irq_remapping.h" |
b6c02715 JR |
49 | |
50 | #define CMD_SET_TYPE(cmd, t) ((cmd)->data[1] |= ((t) << 28)) | |
51 | ||
815b33fd | 52 | #define LOOP_TIMEOUT 100000 |
136f78a1 | 53 | |
aa3de9c0 OBC |
54 | /* |
55 | * This bitmap is used to advertise the page sizes our hardware support | |
56 | * to the IOMMU core, which will then use this information to split | |
57 | * physically contiguous memory regions it is mapping into page sizes | |
58 | * that we support. | |
59 | * | |
954e3dd8 | 60 | * 512GB Pages are not supported due to a hardware bug |
aa3de9c0 | 61 | */ |
954e3dd8 | 62 | #define AMD_IOMMU_PGSIZES ((~0xFFFUL) & ~(2ULL << 38)) |
aa3de9c0 | 63 | |
b6c02715 JR |
64 | static DEFINE_RWLOCK(amd_iommu_devtable_lock); |
65 | ||
bd60b735 JR |
66 | /* A list of preallocated protection domains */ |
67 | static LIST_HEAD(iommu_pd_list); | |
68 | static DEFINE_SPINLOCK(iommu_pd_list_lock); | |
69 | ||
8fa5f802 JR |
70 | /* List of all available dev_data structures */ |
71 | static LIST_HEAD(dev_data_list); | |
72 | static DEFINE_SPINLOCK(dev_data_list_lock); | |
73 | ||
6efed63b JR |
74 | LIST_HEAD(ioapic_map); |
75 | LIST_HEAD(hpet_map); | |
76 | ||
0feae533 JR |
77 | /* |
78 | * Domain for untranslated devices - only allocated | |
79 | * if iommu=pt passed on kernel cmd line. | |
80 | */ | |
81 | static struct protection_domain *pt_domain; | |
82 | ||
b22f6434 | 83 | static const struct iommu_ops amd_iommu_ops; |
26961efe | 84 | |
72e1dcc4 | 85 | static ATOMIC_NOTIFIER_HEAD(ppr_notifier); |
52815b75 | 86 | int amd_iommu_max_glx_val = -1; |
72e1dcc4 | 87 | |
ac1534a5 JR |
88 | static struct dma_map_ops amd_iommu_dma_ops; |
89 | ||
431b2a20 JR |
90 | /* |
91 | * general struct to manage commands send to an IOMMU | |
92 | */ | |
d6449536 | 93 | struct iommu_cmd { |
b6c02715 JR |
94 | u32 data[4]; |
95 | }; | |
96 | ||
05152a04 JR |
97 | struct kmem_cache *amd_iommu_irq_cache; |
98 | ||
04bfdd84 | 99 | static void update_domain(struct protection_domain *domain); |
5abcdba4 | 100 | static int __init alloc_passthrough_domain(void); |
c1eee67b | 101 | |
15898bbc JR |
102 | /**************************************************************************** |
103 | * | |
104 | * Helper functions | |
105 | * | |
106 | ****************************************************************************/ | |
107 | ||
f62dda66 | 108 | static struct iommu_dev_data *alloc_dev_data(u16 devid) |
8fa5f802 JR |
109 | { |
110 | struct iommu_dev_data *dev_data; | |
111 | unsigned long flags; | |
112 | ||
113 | dev_data = kzalloc(sizeof(*dev_data), GFP_KERNEL); | |
114 | if (!dev_data) | |
115 | return NULL; | |
116 | ||
f62dda66 | 117 | dev_data->devid = devid; |
8fa5f802 JR |
118 | atomic_set(&dev_data->bind, 0); |
119 | ||
120 | spin_lock_irqsave(&dev_data_list_lock, flags); | |
121 | list_add_tail(&dev_data->dev_data_list, &dev_data_list); | |
122 | spin_unlock_irqrestore(&dev_data_list_lock, flags); | |
123 | ||
124 | return dev_data; | |
125 | } | |
126 | ||
127 | static void free_dev_data(struct iommu_dev_data *dev_data) | |
128 | { | |
129 | unsigned long flags; | |
130 | ||
131 | spin_lock_irqsave(&dev_data_list_lock, flags); | |
132 | list_del(&dev_data->dev_data_list); | |
133 | spin_unlock_irqrestore(&dev_data_list_lock, flags); | |
134 | ||
135 | kfree(dev_data); | |
136 | } | |
137 | ||
3b03bb74 JR |
138 | static struct iommu_dev_data *search_dev_data(u16 devid) |
139 | { | |
140 | struct iommu_dev_data *dev_data; | |
141 | unsigned long flags; | |
142 | ||
143 | spin_lock_irqsave(&dev_data_list_lock, flags); | |
144 | list_for_each_entry(dev_data, &dev_data_list, dev_data_list) { | |
145 | if (dev_data->devid == devid) | |
146 | goto out_unlock; | |
147 | } | |
148 | ||
149 | dev_data = NULL; | |
150 | ||
151 | out_unlock: | |
152 | spin_unlock_irqrestore(&dev_data_list_lock, flags); | |
153 | ||
154 | return dev_data; | |
155 | } | |
156 | ||
157 | static struct iommu_dev_data *find_dev_data(u16 devid) | |
158 | { | |
159 | struct iommu_dev_data *dev_data; | |
160 | ||
161 | dev_data = search_dev_data(devid); | |
162 | ||
163 | if (dev_data == NULL) | |
164 | dev_data = alloc_dev_data(devid); | |
165 | ||
166 | return dev_data; | |
167 | } | |
168 | ||
15898bbc JR |
169 | static inline u16 get_device_id(struct device *dev) |
170 | { | |
171 | struct pci_dev *pdev = to_pci_dev(dev); | |
172 | ||
6f2729ba | 173 | return PCI_DEVID(pdev->bus->number, pdev->devfn); |
15898bbc JR |
174 | } |
175 | ||
657cbb6b JR |
176 | static struct iommu_dev_data *get_dev_data(struct device *dev) |
177 | { | |
178 | return dev->archdata.iommu; | |
179 | } | |
180 | ||
5abcdba4 JR |
181 | static bool pci_iommuv2_capable(struct pci_dev *pdev) |
182 | { | |
183 | static const int caps[] = { | |
184 | PCI_EXT_CAP_ID_ATS, | |
46277b75 JR |
185 | PCI_EXT_CAP_ID_PRI, |
186 | PCI_EXT_CAP_ID_PASID, | |
5abcdba4 JR |
187 | }; |
188 | int i, pos; | |
189 | ||
190 | for (i = 0; i < 3; ++i) { | |
191 | pos = pci_find_ext_capability(pdev, caps[i]); | |
192 | if (pos == 0) | |
193 | return false; | |
194 | } | |
195 | ||
196 | return true; | |
197 | } | |
198 | ||
6a113ddc JR |
199 | static bool pdev_pri_erratum(struct pci_dev *pdev, u32 erratum) |
200 | { | |
201 | struct iommu_dev_data *dev_data; | |
202 | ||
203 | dev_data = get_dev_data(&pdev->dev); | |
204 | ||
205 | return dev_data->errata & (1 << erratum) ? true : false; | |
206 | } | |
207 | ||
71c70984 JR |
208 | /* |
209 | * In this function the list of preallocated protection domains is traversed to | |
210 | * find the domain for a specific device | |
211 | */ | |
212 | static struct dma_ops_domain *find_protection_domain(u16 devid) | |
213 | { | |
214 | struct dma_ops_domain *entry, *ret = NULL; | |
215 | unsigned long flags; | |
216 | u16 alias = amd_iommu_alias_table[devid]; | |
217 | ||
218 | if (list_empty(&iommu_pd_list)) | |
219 | return NULL; | |
220 | ||
221 | spin_lock_irqsave(&iommu_pd_list_lock, flags); | |
222 | ||
223 | list_for_each_entry(entry, &iommu_pd_list, list) { | |
224 | if (entry->target_dev == devid || | |
225 | entry->target_dev == alias) { | |
226 | ret = entry; | |
227 | break; | |
228 | } | |
229 | } | |
230 | ||
231 | spin_unlock_irqrestore(&iommu_pd_list_lock, flags); | |
232 | ||
233 | return ret; | |
234 | } | |
235 | ||
98fc5a69 JR |
236 | /* |
237 | * This function checks if the driver got a valid device from the caller to | |
238 | * avoid dereferencing invalid pointers. | |
239 | */ | |
240 | static bool check_device(struct device *dev) | |
241 | { | |
242 | u16 devid; | |
243 | ||
244 | if (!dev || !dev->dma_mask) | |
245 | return false; | |
246 | ||
b82a2272 YW |
247 | /* No PCI device */ |
248 | if (!dev_is_pci(dev)) | |
98fc5a69 JR |
249 | return false; |
250 | ||
251 | devid = get_device_id(dev); | |
252 | ||
253 | /* Out of our scope? */ | |
254 | if (devid > amd_iommu_last_bdf) | |
255 | return false; | |
256 | ||
257 | if (amd_iommu_rlookup_table[devid] == NULL) | |
258 | return false; | |
259 | ||
260 | return true; | |
261 | } | |
262 | ||
2851db21 AW |
263 | static int init_iommu_group(struct device *dev) |
264 | { | |
2851db21 | 265 | struct iommu_group *group; |
2851db21 | 266 | |
65d5352f | 267 | group = iommu_group_get_for_dev(dev); |
2851db21 | 268 | |
65d5352f AW |
269 | if (IS_ERR(group)) |
270 | return PTR_ERR(group); | |
2851db21 | 271 | |
65d5352f AW |
272 | iommu_group_put(group); |
273 | return 0; | |
eb9c9527 AW |
274 | } |
275 | ||
c1931090 AW |
276 | static int __last_alias(struct pci_dev *pdev, u16 alias, void *data) |
277 | { | |
278 | *(u16 *)data = alias; | |
279 | return 0; | |
280 | } | |
281 | ||
282 | static u16 get_alias(struct device *dev) | |
283 | { | |
284 | struct pci_dev *pdev = to_pci_dev(dev); | |
285 | u16 devid, ivrs_alias, pci_alias; | |
286 | ||
287 | devid = get_device_id(dev); | |
288 | ivrs_alias = amd_iommu_alias_table[devid]; | |
289 | pci_for_each_dma_alias(pdev, __last_alias, &pci_alias); | |
290 | ||
291 | if (ivrs_alias == pci_alias) | |
292 | return ivrs_alias; | |
293 | ||
294 | /* | |
295 | * DMA alias showdown | |
296 | * | |
297 | * The IVRS is fairly reliable in telling us about aliases, but it | |
298 | * can't know about every screwy device. If we don't have an IVRS | |
299 | * reported alias, use the PCI reported alias. In that case we may | |
300 | * still need to initialize the rlookup and dev_table entries if the | |
301 | * alias is to a non-existent device. | |
302 | */ | |
303 | if (ivrs_alias == devid) { | |
304 | if (!amd_iommu_rlookup_table[pci_alias]) { | |
305 | amd_iommu_rlookup_table[pci_alias] = | |
306 | amd_iommu_rlookup_table[devid]; | |
307 | memcpy(amd_iommu_dev_table[pci_alias].data, | |
308 | amd_iommu_dev_table[devid].data, | |
309 | sizeof(amd_iommu_dev_table[pci_alias].data)); | |
310 | } | |
311 | ||
312 | return pci_alias; | |
313 | } | |
314 | ||
315 | pr_info("AMD-Vi: Using IVRS reported alias %02x:%02x.%d " | |
316 | "for device %s[%04x:%04x], kernel reported alias " | |
317 | "%02x:%02x.%d\n", PCI_BUS_NUM(ivrs_alias), PCI_SLOT(ivrs_alias), | |
318 | PCI_FUNC(ivrs_alias), dev_name(dev), pdev->vendor, pdev->device, | |
319 | PCI_BUS_NUM(pci_alias), PCI_SLOT(pci_alias), | |
320 | PCI_FUNC(pci_alias)); | |
321 | ||
322 | /* | |
323 | * If we don't have a PCI DMA alias and the IVRS alias is on the same | |
324 | * bus, then the IVRS table may know about a quirk that we don't. | |
325 | */ | |
326 | if (pci_alias == devid && | |
327 | PCI_BUS_NUM(ivrs_alias) == pdev->bus->number) { | |
328 | pdev->dev_flags |= PCI_DEV_FLAGS_DMA_ALIAS_DEVFN; | |
329 | pdev->dma_alias_devfn = ivrs_alias & 0xff; | |
330 | pr_info("AMD-Vi: Added PCI DMA alias %02x.%d for %s\n", | |
331 | PCI_SLOT(ivrs_alias), PCI_FUNC(ivrs_alias), | |
332 | dev_name(dev)); | |
333 | } | |
334 | ||
335 | return ivrs_alias; | |
336 | } | |
337 | ||
eb9c9527 AW |
338 | static int iommu_init_device(struct device *dev) |
339 | { | |
340 | struct pci_dev *pdev = to_pci_dev(dev); | |
341 | struct iommu_dev_data *dev_data; | |
342 | u16 alias; | |
343 | int ret; | |
344 | ||
345 | if (dev->archdata.iommu) | |
346 | return 0; | |
347 | ||
348 | dev_data = find_dev_data(get_device_id(dev)); | |
349 | if (!dev_data) | |
350 | return -ENOMEM; | |
351 | ||
c1931090 AW |
352 | alias = get_alias(dev); |
353 | ||
eb9c9527 AW |
354 | if (alias != dev_data->devid) { |
355 | struct iommu_dev_data *alias_data; | |
356 | ||
357 | alias_data = find_dev_data(alias); | |
358 | if (alias_data == NULL) { | |
359 | pr_err("AMD-Vi: Warning: Unhandled device %s\n", | |
360 | dev_name(dev)); | |
361 | free_dev_data(dev_data); | |
362 | return -ENOTSUPP; | |
363 | } | |
364 | dev_data->alias_data = alias_data; | |
365 | } | |
366 | ||
367 | ret = init_iommu_group(dev); | |
e644a013 RK |
368 | if (ret) { |
369 | free_dev_data(dev_data); | |
9dcd6130 | 370 | return ret; |
e644a013 | 371 | } |
9dcd6130 | 372 | |
5abcdba4 JR |
373 | if (pci_iommuv2_capable(pdev)) { |
374 | struct amd_iommu *iommu; | |
375 | ||
376 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
377 | dev_data->iommu_v2 = iommu->is_iommu_v2; | |
378 | } | |
379 | ||
657cbb6b JR |
380 | dev->archdata.iommu = dev_data; |
381 | ||
066f2e98 AW |
382 | iommu_device_link(amd_iommu_rlookup_table[dev_data->devid]->iommu_dev, |
383 | dev); | |
384 | ||
657cbb6b JR |
385 | return 0; |
386 | } | |
387 | ||
26018874 JR |
388 | static void iommu_ignore_device(struct device *dev) |
389 | { | |
390 | u16 devid, alias; | |
391 | ||
392 | devid = get_device_id(dev); | |
393 | alias = amd_iommu_alias_table[devid]; | |
394 | ||
395 | memset(&amd_iommu_dev_table[devid], 0, sizeof(struct dev_table_entry)); | |
396 | memset(&amd_iommu_dev_table[alias], 0, sizeof(struct dev_table_entry)); | |
397 | ||
398 | amd_iommu_rlookup_table[devid] = NULL; | |
399 | amd_iommu_rlookup_table[alias] = NULL; | |
400 | } | |
401 | ||
657cbb6b JR |
402 | static void iommu_uninit_device(struct device *dev) |
403 | { | |
c1931090 AW |
404 | struct iommu_dev_data *dev_data = search_dev_data(get_device_id(dev)); |
405 | ||
406 | if (!dev_data) | |
407 | return; | |
408 | ||
066f2e98 AW |
409 | iommu_device_unlink(amd_iommu_rlookup_table[dev_data->devid]->iommu_dev, |
410 | dev); | |
411 | ||
9dcd6130 AW |
412 | iommu_group_remove_device(dev); |
413 | ||
c1931090 AW |
414 | /* Unlink from alias, it may change if another device is re-plugged */ |
415 | dev_data->alias_data = NULL; | |
416 | ||
8fa5f802 | 417 | /* |
c1931090 AW |
418 | * We keep dev_data around for unplugged devices and reuse it when the |
419 | * device is re-plugged - not doing so would introduce a ton of races. | |
8fa5f802 | 420 | */ |
657cbb6b | 421 | } |
b7cc9554 JR |
422 | |
423 | void __init amd_iommu_uninit_devices(void) | |
424 | { | |
8fa5f802 | 425 | struct iommu_dev_data *dev_data, *n; |
b7cc9554 JR |
426 | struct pci_dev *pdev = NULL; |
427 | ||
428 | for_each_pci_dev(pdev) { | |
429 | ||
430 | if (!check_device(&pdev->dev)) | |
431 | continue; | |
432 | ||
433 | iommu_uninit_device(&pdev->dev); | |
434 | } | |
8fa5f802 JR |
435 | |
436 | /* Free all of our dev_data structures */ | |
437 | list_for_each_entry_safe(dev_data, n, &dev_data_list, dev_data_list) | |
438 | free_dev_data(dev_data); | |
b7cc9554 JR |
439 | } |
440 | ||
441 | int __init amd_iommu_init_devices(void) | |
442 | { | |
443 | struct pci_dev *pdev = NULL; | |
444 | int ret = 0; | |
445 | ||
446 | for_each_pci_dev(pdev) { | |
447 | ||
448 | if (!check_device(&pdev->dev)) | |
449 | continue; | |
450 | ||
451 | ret = iommu_init_device(&pdev->dev); | |
26018874 JR |
452 | if (ret == -ENOTSUPP) |
453 | iommu_ignore_device(&pdev->dev); | |
454 | else if (ret) | |
b7cc9554 JR |
455 | goto out_free; |
456 | } | |
457 | ||
458 | return 0; | |
459 | ||
460 | out_free: | |
461 | ||
462 | amd_iommu_uninit_devices(); | |
463 | ||
464 | return ret; | |
465 | } | |
7f26508b JR |
466 | #ifdef CONFIG_AMD_IOMMU_STATS |
467 | ||
468 | /* | |
469 | * Initialization code for statistics collection | |
470 | */ | |
471 | ||
da49f6df | 472 | DECLARE_STATS_COUNTER(compl_wait); |
0f2a86f2 | 473 | DECLARE_STATS_COUNTER(cnt_map_single); |
146a6917 | 474 | DECLARE_STATS_COUNTER(cnt_unmap_single); |
d03f067a | 475 | DECLARE_STATS_COUNTER(cnt_map_sg); |
55877a6b | 476 | DECLARE_STATS_COUNTER(cnt_unmap_sg); |
c8f0fb36 | 477 | DECLARE_STATS_COUNTER(cnt_alloc_coherent); |
5d31ee7e | 478 | DECLARE_STATS_COUNTER(cnt_free_coherent); |
c1858976 | 479 | DECLARE_STATS_COUNTER(cross_page); |
f57d98ae | 480 | DECLARE_STATS_COUNTER(domain_flush_single); |
18811f55 | 481 | DECLARE_STATS_COUNTER(domain_flush_all); |
5774f7c5 | 482 | DECLARE_STATS_COUNTER(alloced_io_mem); |
8ecaf8f1 | 483 | DECLARE_STATS_COUNTER(total_map_requests); |
399be2f5 JR |
484 | DECLARE_STATS_COUNTER(complete_ppr); |
485 | DECLARE_STATS_COUNTER(invalidate_iotlb); | |
486 | DECLARE_STATS_COUNTER(invalidate_iotlb_all); | |
487 | DECLARE_STATS_COUNTER(pri_requests); | |
488 | ||
7f26508b | 489 | static struct dentry *stats_dir; |
7f26508b JR |
490 | static struct dentry *de_fflush; |
491 | ||
492 | static void amd_iommu_stats_add(struct __iommu_counter *cnt) | |
493 | { | |
494 | if (stats_dir == NULL) | |
495 | return; | |
496 | ||
497 | cnt->dent = debugfs_create_u64(cnt->name, 0444, stats_dir, | |
498 | &cnt->value); | |
499 | } | |
500 | ||
501 | static void amd_iommu_stats_init(void) | |
502 | { | |
503 | stats_dir = debugfs_create_dir("amd-iommu", NULL); | |
504 | if (stats_dir == NULL) | |
505 | return; | |
506 | ||
7f26508b | 507 | de_fflush = debugfs_create_bool("fullflush", 0444, stats_dir, |
3775d481 | 508 | &amd_iommu_unmap_flush); |
da49f6df JR |
509 | |
510 | amd_iommu_stats_add(&compl_wait); | |
0f2a86f2 | 511 | amd_iommu_stats_add(&cnt_map_single); |
146a6917 | 512 | amd_iommu_stats_add(&cnt_unmap_single); |
d03f067a | 513 | amd_iommu_stats_add(&cnt_map_sg); |
55877a6b | 514 | amd_iommu_stats_add(&cnt_unmap_sg); |
c8f0fb36 | 515 | amd_iommu_stats_add(&cnt_alloc_coherent); |
5d31ee7e | 516 | amd_iommu_stats_add(&cnt_free_coherent); |
c1858976 | 517 | amd_iommu_stats_add(&cross_page); |
f57d98ae | 518 | amd_iommu_stats_add(&domain_flush_single); |
18811f55 | 519 | amd_iommu_stats_add(&domain_flush_all); |
5774f7c5 | 520 | amd_iommu_stats_add(&alloced_io_mem); |
8ecaf8f1 | 521 | amd_iommu_stats_add(&total_map_requests); |
399be2f5 JR |
522 | amd_iommu_stats_add(&complete_ppr); |
523 | amd_iommu_stats_add(&invalidate_iotlb); | |
524 | amd_iommu_stats_add(&invalidate_iotlb_all); | |
525 | amd_iommu_stats_add(&pri_requests); | |
7f26508b JR |
526 | } |
527 | ||
528 | #endif | |
529 | ||
a80dc3e0 JR |
530 | /**************************************************************************** |
531 | * | |
532 | * Interrupt handling functions | |
533 | * | |
534 | ****************************************************************************/ | |
535 | ||
e3e59876 JR |
536 | static void dump_dte_entry(u16 devid) |
537 | { | |
538 | int i; | |
539 | ||
ee6c2868 JR |
540 | for (i = 0; i < 4; ++i) |
541 | pr_err("AMD-Vi: DTE[%d]: %016llx\n", i, | |
e3e59876 JR |
542 | amd_iommu_dev_table[devid].data[i]); |
543 | } | |
544 | ||
945b4ac4 JR |
545 | static void dump_command(unsigned long phys_addr) |
546 | { | |
547 | struct iommu_cmd *cmd = phys_to_virt(phys_addr); | |
548 | int i; | |
549 | ||
550 | for (i = 0; i < 4; ++i) | |
551 | pr_err("AMD-Vi: CMD[%d]: %08x\n", i, cmd->data[i]); | |
552 | } | |
553 | ||
a345b23b | 554 | static void iommu_print_event(struct amd_iommu *iommu, void *__evt) |
90008ee4 | 555 | { |
3d06fca8 JR |
556 | int type, devid, domid, flags; |
557 | volatile u32 *event = __evt; | |
558 | int count = 0; | |
559 | u64 address; | |
560 | ||
561 | retry: | |
562 | type = (event[1] >> EVENT_TYPE_SHIFT) & EVENT_TYPE_MASK; | |
563 | devid = (event[0] >> EVENT_DEVID_SHIFT) & EVENT_DEVID_MASK; | |
564 | domid = (event[1] >> EVENT_DOMID_SHIFT) & EVENT_DOMID_MASK; | |
565 | flags = (event[1] >> EVENT_FLAGS_SHIFT) & EVENT_FLAGS_MASK; | |
566 | address = (u64)(((u64)event[3]) << 32) | event[2]; | |
567 | ||
568 | if (type == 0) { | |
569 | /* Did we hit the erratum? */ | |
570 | if (++count == LOOP_TIMEOUT) { | |
571 | pr_err("AMD-Vi: No event written to event log\n"); | |
572 | return; | |
573 | } | |
574 | udelay(1); | |
575 | goto retry; | |
576 | } | |
90008ee4 | 577 | |
4c6f40d4 | 578 | printk(KERN_ERR "AMD-Vi: Event logged ["); |
90008ee4 JR |
579 | |
580 | switch (type) { | |
581 | case EVENT_TYPE_ILL_DEV: | |
582 | printk("ILLEGAL_DEV_TABLE_ENTRY device=%02x:%02x.%x " | |
583 | "address=0x%016llx flags=0x%04x]\n", | |
c5081cd7 | 584 | PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
90008ee4 | 585 | address, flags); |
e3e59876 | 586 | dump_dte_entry(devid); |
90008ee4 JR |
587 | break; |
588 | case EVENT_TYPE_IO_FAULT: | |
589 | printk("IO_PAGE_FAULT device=%02x:%02x.%x " | |
590 | "domain=0x%04x address=0x%016llx flags=0x%04x]\n", | |
c5081cd7 | 591 | PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
90008ee4 JR |
592 | domid, address, flags); |
593 | break; | |
594 | case EVENT_TYPE_DEV_TAB_ERR: | |
595 | printk("DEV_TAB_HARDWARE_ERROR device=%02x:%02x.%x " | |
596 | "address=0x%016llx flags=0x%04x]\n", | |
c5081cd7 | 597 | PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
90008ee4 JR |
598 | address, flags); |
599 | break; | |
600 | case EVENT_TYPE_PAGE_TAB_ERR: | |
601 | printk("PAGE_TAB_HARDWARE_ERROR device=%02x:%02x.%x " | |
602 | "domain=0x%04x address=0x%016llx flags=0x%04x]\n", | |
c5081cd7 | 603 | PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
90008ee4 JR |
604 | domid, address, flags); |
605 | break; | |
606 | case EVENT_TYPE_ILL_CMD: | |
607 | printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address); | |
945b4ac4 | 608 | dump_command(address); |
90008ee4 JR |
609 | break; |
610 | case EVENT_TYPE_CMD_HARD_ERR: | |
611 | printk("COMMAND_HARDWARE_ERROR address=0x%016llx " | |
612 | "flags=0x%04x]\n", address, flags); | |
613 | break; | |
614 | case EVENT_TYPE_IOTLB_INV_TO: | |
615 | printk("IOTLB_INV_TIMEOUT device=%02x:%02x.%x " | |
616 | "address=0x%016llx]\n", | |
c5081cd7 | 617 | PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
90008ee4 JR |
618 | address); |
619 | break; | |
620 | case EVENT_TYPE_INV_DEV_REQ: | |
621 | printk("INVALID_DEVICE_REQUEST device=%02x:%02x.%x " | |
622 | "address=0x%016llx flags=0x%04x]\n", | |
c5081cd7 | 623 | PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
90008ee4 JR |
624 | address, flags); |
625 | break; | |
626 | default: | |
627 | printk(KERN_ERR "UNKNOWN type=0x%02x]\n", type); | |
628 | } | |
3d06fca8 JR |
629 | |
630 | memset(__evt, 0, 4 * sizeof(u32)); | |
90008ee4 JR |
631 | } |
632 | ||
633 | static void iommu_poll_events(struct amd_iommu *iommu) | |
634 | { | |
635 | u32 head, tail; | |
90008ee4 JR |
636 | |
637 | head = readl(iommu->mmio_base + MMIO_EVT_HEAD_OFFSET); | |
638 | tail = readl(iommu->mmio_base + MMIO_EVT_TAIL_OFFSET); | |
639 | ||
640 | while (head != tail) { | |
a345b23b | 641 | iommu_print_event(iommu, iommu->evt_buf + head); |
90008ee4 JR |
642 | head = (head + EVENT_ENTRY_SIZE) % iommu->evt_buf_size; |
643 | } | |
644 | ||
645 | writel(head, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET); | |
90008ee4 JR |
646 | } |
647 | ||
eee53537 | 648 | static void iommu_handle_ppr_entry(struct amd_iommu *iommu, u64 *raw) |
72e1dcc4 JR |
649 | { |
650 | struct amd_iommu_fault fault; | |
72e1dcc4 | 651 | |
399be2f5 JR |
652 | INC_STATS_COUNTER(pri_requests); |
653 | ||
72e1dcc4 JR |
654 | if (PPR_REQ_TYPE(raw[0]) != PPR_REQ_FAULT) { |
655 | pr_err_ratelimited("AMD-Vi: Unknown PPR request received\n"); | |
656 | return; | |
657 | } | |
658 | ||
659 | fault.address = raw[1]; | |
660 | fault.pasid = PPR_PASID(raw[0]); | |
661 | fault.device_id = PPR_DEVID(raw[0]); | |
662 | fault.tag = PPR_TAG(raw[0]); | |
663 | fault.flags = PPR_FLAGS(raw[0]); | |
664 | ||
72e1dcc4 JR |
665 | atomic_notifier_call_chain(&ppr_notifier, 0, &fault); |
666 | } | |
667 | ||
668 | static void iommu_poll_ppr_log(struct amd_iommu *iommu) | |
669 | { | |
72e1dcc4 JR |
670 | u32 head, tail; |
671 | ||
672 | if (iommu->ppr_log == NULL) | |
673 | return; | |
674 | ||
72e1dcc4 JR |
675 | head = readl(iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); |
676 | tail = readl(iommu->mmio_base + MMIO_PPR_TAIL_OFFSET); | |
677 | ||
678 | while (head != tail) { | |
eee53537 JR |
679 | volatile u64 *raw; |
680 | u64 entry[2]; | |
681 | int i; | |
682 | ||
683 | raw = (u64 *)(iommu->ppr_log + head); | |
684 | ||
685 | /* | |
686 | * Hardware bug: Interrupt may arrive before the entry is | |
687 | * written to memory. If this happens we need to wait for the | |
688 | * entry to arrive. | |
689 | */ | |
690 | for (i = 0; i < LOOP_TIMEOUT; ++i) { | |
691 | if (PPR_REQ_TYPE(raw[0]) != 0) | |
692 | break; | |
693 | udelay(1); | |
694 | } | |
72e1dcc4 | 695 | |
eee53537 JR |
696 | /* Avoid memcpy function-call overhead */ |
697 | entry[0] = raw[0]; | |
698 | entry[1] = raw[1]; | |
72e1dcc4 | 699 | |
eee53537 JR |
700 | /* |
701 | * To detect the hardware bug we need to clear the entry | |
702 | * back to zero. | |
703 | */ | |
704 | raw[0] = raw[1] = 0UL; | |
705 | ||
706 | /* Update head pointer of hardware ring-buffer */ | |
72e1dcc4 JR |
707 | head = (head + PPR_ENTRY_SIZE) % PPR_LOG_SIZE; |
708 | writel(head, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); | |
eee53537 | 709 | |
eee53537 JR |
710 | /* Handle PPR entry */ |
711 | iommu_handle_ppr_entry(iommu, entry); | |
712 | ||
eee53537 JR |
713 | /* Refresh ring-buffer information */ |
714 | head = readl(iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); | |
72e1dcc4 JR |
715 | tail = readl(iommu->mmio_base + MMIO_PPR_TAIL_OFFSET); |
716 | } | |
72e1dcc4 JR |
717 | } |
718 | ||
72fe00f0 | 719 | irqreturn_t amd_iommu_int_thread(int irq, void *data) |
a80dc3e0 | 720 | { |
3f398bc7 SS |
721 | struct amd_iommu *iommu = (struct amd_iommu *) data; |
722 | u32 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); | |
90008ee4 | 723 | |
3f398bc7 SS |
724 | while (status & (MMIO_STATUS_EVT_INT_MASK | MMIO_STATUS_PPR_INT_MASK)) { |
725 | /* Enable EVT and PPR interrupts again */ | |
726 | writel((MMIO_STATUS_EVT_INT_MASK | MMIO_STATUS_PPR_INT_MASK), | |
727 | iommu->mmio_base + MMIO_STATUS_OFFSET); | |
90008ee4 | 728 | |
3f398bc7 SS |
729 | if (status & MMIO_STATUS_EVT_INT_MASK) { |
730 | pr_devel("AMD-Vi: Processing IOMMU Event Log\n"); | |
731 | iommu_poll_events(iommu); | |
732 | } | |
90008ee4 | 733 | |
3f398bc7 SS |
734 | if (status & MMIO_STATUS_PPR_INT_MASK) { |
735 | pr_devel("AMD-Vi: Processing IOMMU PPR Log\n"); | |
736 | iommu_poll_ppr_log(iommu); | |
737 | } | |
90008ee4 | 738 | |
3f398bc7 SS |
739 | /* |
740 | * Hardware bug: ERBT1312 | |
741 | * When re-enabling interrupt (by writing 1 | |
742 | * to clear the bit), the hardware might also try to set | |
743 | * the interrupt bit in the event status register. | |
744 | * In this scenario, the bit will be set, and disable | |
745 | * subsequent interrupts. | |
746 | * | |
747 | * Workaround: The IOMMU driver should read back the | |
748 | * status register and check if the interrupt bits are cleared. | |
749 | * If not, driver will need to go through the interrupt handler | |
750 | * again and re-clear the bits | |
751 | */ | |
752 | status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); | |
753 | } | |
90008ee4 | 754 | return IRQ_HANDLED; |
a80dc3e0 JR |
755 | } |
756 | ||
72fe00f0 JR |
757 | irqreturn_t amd_iommu_int_handler(int irq, void *data) |
758 | { | |
759 | return IRQ_WAKE_THREAD; | |
760 | } | |
761 | ||
431b2a20 JR |
762 | /**************************************************************************** |
763 | * | |
764 | * IOMMU command queuing functions | |
765 | * | |
766 | ****************************************************************************/ | |
767 | ||
ac0ea6e9 JR |
768 | static int wait_on_sem(volatile u64 *sem) |
769 | { | |
770 | int i = 0; | |
771 | ||
772 | while (*sem == 0 && i < LOOP_TIMEOUT) { | |
773 | udelay(1); | |
774 | i += 1; | |
775 | } | |
776 | ||
777 | if (i == LOOP_TIMEOUT) { | |
778 | pr_alert("AMD-Vi: Completion-Wait loop timed out\n"); | |
779 | return -EIO; | |
780 | } | |
781 | ||
782 | return 0; | |
783 | } | |
784 | ||
785 | static void copy_cmd_to_buffer(struct amd_iommu *iommu, | |
786 | struct iommu_cmd *cmd, | |
787 | u32 tail) | |
a19ae1ec | 788 | { |
a19ae1ec JR |
789 | u8 *target; |
790 | ||
8a7c5ef3 | 791 | target = iommu->cmd_buf + tail; |
ac0ea6e9 JR |
792 | tail = (tail + sizeof(*cmd)) % iommu->cmd_buf_size; |
793 | ||
794 | /* Copy command to buffer */ | |
795 | memcpy(target, cmd, sizeof(*cmd)); | |
796 | ||
797 | /* Tell the IOMMU about it */ | |
a19ae1ec | 798 | writel(tail, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); |
ac0ea6e9 | 799 | } |
a19ae1ec | 800 | |
815b33fd | 801 | static void build_completion_wait(struct iommu_cmd *cmd, u64 address) |
ded46737 | 802 | { |
815b33fd JR |
803 | WARN_ON(address & 0x7ULL); |
804 | ||
ded46737 | 805 | memset(cmd, 0, sizeof(*cmd)); |
815b33fd JR |
806 | cmd->data[0] = lower_32_bits(__pa(address)) | CMD_COMPL_WAIT_STORE_MASK; |
807 | cmd->data[1] = upper_32_bits(__pa(address)); | |
808 | cmd->data[2] = 1; | |
ded46737 JR |
809 | CMD_SET_TYPE(cmd, CMD_COMPL_WAIT); |
810 | } | |
811 | ||
94fe79e2 JR |
812 | static void build_inv_dte(struct iommu_cmd *cmd, u16 devid) |
813 | { | |
814 | memset(cmd, 0, sizeof(*cmd)); | |
815 | cmd->data[0] = devid; | |
816 | CMD_SET_TYPE(cmd, CMD_INV_DEV_ENTRY); | |
817 | } | |
818 | ||
11b6402c JR |
819 | static void build_inv_iommu_pages(struct iommu_cmd *cmd, u64 address, |
820 | size_t size, u16 domid, int pde) | |
821 | { | |
822 | u64 pages; | |
823 | int s; | |
824 | ||
825 | pages = iommu_num_pages(address, size, PAGE_SIZE); | |
826 | s = 0; | |
827 | ||
828 | if (pages > 1) { | |
829 | /* | |
830 | * If we have to flush more than one page, flush all | |
831 | * TLB entries for this domain | |
832 | */ | |
833 | address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS; | |
834 | s = 1; | |
835 | } | |
836 | ||
837 | address &= PAGE_MASK; | |
838 | ||
839 | memset(cmd, 0, sizeof(*cmd)); | |
840 | cmd->data[1] |= domid; | |
841 | cmd->data[2] = lower_32_bits(address); | |
842 | cmd->data[3] = upper_32_bits(address); | |
843 | CMD_SET_TYPE(cmd, CMD_INV_IOMMU_PAGES); | |
844 | if (s) /* size bit - we flush more than one 4kb page */ | |
845 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
df805abb | 846 | if (pde) /* PDE bit - we want to flush everything, not only the PTEs */ |
11b6402c JR |
847 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK; |
848 | } | |
849 | ||
cb41ed85 JR |
850 | static void build_inv_iotlb_pages(struct iommu_cmd *cmd, u16 devid, int qdep, |
851 | u64 address, size_t size) | |
852 | { | |
853 | u64 pages; | |
854 | int s; | |
855 | ||
856 | pages = iommu_num_pages(address, size, PAGE_SIZE); | |
857 | s = 0; | |
858 | ||
859 | if (pages > 1) { | |
860 | /* | |
861 | * If we have to flush more than one page, flush all | |
862 | * TLB entries for this domain | |
863 | */ | |
864 | address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS; | |
865 | s = 1; | |
866 | } | |
867 | ||
868 | address &= PAGE_MASK; | |
869 | ||
870 | memset(cmd, 0, sizeof(*cmd)); | |
871 | cmd->data[0] = devid; | |
872 | cmd->data[0] |= (qdep & 0xff) << 24; | |
873 | cmd->data[1] = devid; | |
874 | cmd->data[2] = lower_32_bits(address); | |
875 | cmd->data[3] = upper_32_bits(address); | |
876 | CMD_SET_TYPE(cmd, CMD_INV_IOTLB_PAGES); | |
877 | if (s) | |
878 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
879 | } | |
880 | ||
22e266c7 JR |
881 | static void build_inv_iommu_pasid(struct iommu_cmd *cmd, u16 domid, int pasid, |
882 | u64 address, bool size) | |
883 | { | |
884 | memset(cmd, 0, sizeof(*cmd)); | |
885 | ||
886 | address &= ~(0xfffULL); | |
887 | ||
a919a018 | 888 | cmd->data[0] = pasid; |
22e266c7 JR |
889 | cmd->data[1] = domid; |
890 | cmd->data[2] = lower_32_bits(address); | |
891 | cmd->data[3] = upper_32_bits(address); | |
892 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK; | |
893 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_GN_MASK; | |
894 | if (size) | |
895 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
896 | CMD_SET_TYPE(cmd, CMD_INV_IOMMU_PAGES); | |
897 | } | |
898 | ||
899 | static void build_inv_iotlb_pasid(struct iommu_cmd *cmd, u16 devid, int pasid, | |
900 | int qdep, u64 address, bool size) | |
901 | { | |
902 | memset(cmd, 0, sizeof(*cmd)); | |
903 | ||
904 | address &= ~(0xfffULL); | |
905 | ||
906 | cmd->data[0] = devid; | |
e8d2d82d | 907 | cmd->data[0] |= ((pasid >> 8) & 0xff) << 16; |
22e266c7 JR |
908 | cmd->data[0] |= (qdep & 0xff) << 24; |
909 | cmd->data[1] = devid; | |
e8d2d82d | 910 | cmd->data[1] |= (pasid & 0xff) << 16; |
22e266c7 JR |
911 | cmd->data[2] = lower_32_bits(address); |
912 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_GN_MASK; | |
913 | cmd->data[3] = upper_32_bits(address); | |
914 | if (size) | |
915 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
916 | CMD_SET_TYPE(cmd, CMD_INV_IOTLB_PAGES); | |
917 | } | |
918 | ||
c99afa25 JR |
919 | static void build_complete_ppr(struct iommu_cmd *cmd, u16 devid, int pasid, |
920 | int status, int tag, bool gn) | |
921 | { | |
922 | memset(cmd, 0, sizeof(*cmd)); | |
923 | ||
924 | cmd->data[0] = devid; | |
925 | if (gn) { | |
a919a018 | 926 | cmd->data[1] = pasid; |
c99afa25 JR |
927 | cmd->data[2] = CMD_INV_IOMMU_PAGES_GN_MASK; |
928 | } | |
929 | cmd->data[3] = tag & 0x1ff; | |
930 | cmd->data[3] |= (status & PPR_STATUS_MASK) << PPR_STATUS_SHIFT; | |
931 | ||
932 | CMD_SET_TYPE(cmd, CMD_COMPLETE_PPR); | |
933 | } | |
934 | ||
58fc7f14 JR |
935 | static void build_inv_all(struct iommu_cmd *cmd) |
936 | { | |
937 | memset(cmd, 0, sizeof(*cmd)); | |
938 | CMD_SET_TYPE(cmd, CMD_INV_ALL); | |
a19ae1ec JR |
939 | } |
940 | ||
7ef2798d JR |
941 | static void build_inv_irt(struct iommu_cmd *cmd, u16 devid) |
942 | { | |
943 | memset(cmd, 0, sizeof(*cmd)); | |
944 | cmd->data[0] = devid; | |
945 | CMD_SET_TYPE(cmd, CMD_INV_IRT); | |
946 | } | |
947 | ||
431b2a20 | 948 | /* |
431b2a20 | 949 | * Writes the command to the IOMMUs command buffer and informs the |
ac0ea6e9 | 950 | * hardware about the new command. |
431b2a20 | 951 | */ |
f1ca1512 JR |
952 | static int iommu_queue_command_sync(struct amd_iommu *iommu, |
953 | struct iommu_cmd *cmd, | |
954 | bool sync) | |
a19ae1ec | 955 | { |
ac0ea6e9 | 956 | u32 left, tail, head, next_tail; |
a19ae1ec | 957 | unsigned long flags; |
a19ae1ec | 958 | |
549c90dc | 959 | WARN_ON(iommu->cmd_buf_size & CMD_BUFFER_UNINITIALIZED); |
ac0ea6e9 JR |
960 | |
961 | again: | |
a19ae1ec | 962 | spin_lock_irqsave(&iommu->lock, flags); |
a19ae1ec | 963 | |
ac0ea6e9 JR |
964 | head = readl(iommu->mmio_base + MMIO_CMD_HEAD_OFFSET); |
965 | tail = readl(iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); | |
966 | next_tail = (tail + sizeof(*cmd)) % iommu->cmd_buf_size; | |
967 | left = (head - next_tail) % iommu->cmd_buf_size; | |
a19ae1ec | 968 | |
ac0ea6e9 JR |
969 | if (left <= 2) { |
970 | struct iommu_cmd sync_cmd; | |
971 | volatile u64 sem = 0; | |
972 | int ret; | |
8d201968 | 973 | |
ac0ea6e9 JR |
974 | build_completion_wait(&sync_cmd, (u64)&sem); |
975 | copy_cmd_to_buffer(iommu, &sync_cmd, tail); | |
da49f6df | 976 | |
ac0ea6e9 JR |
977 | spin_unlock_irqrestore(&iommu->lock, flags); |
978 | ||
979 | if ((ret = wait_on_sem(&sem)) != 0) | |
980 | return ret; | |
981 | ||
982 | goto again; | |
8d201968 JR |
983 | } |
984 | ||
ac0ea6e9 JR |
985 | copy_cmd_to_buffer(iommu, cmd, tail); |
986 | ||
987 | /* We need to sync now to make sure all commands are processed */ | |
f1ca1512 | 988 | iommu->need_sync = sync; |
ac0ea6e9 | 989 | |
a19ae1ec | 990 | spin_unlock_irqrestore(&iommu->lock, flags); |
8d201968 | 991 | |
815b33fd | 992 | return 0; |
8d201968 JR |
993 | } |
994 | ||
f1ca1512 JR |
995 | static int iommu_queue_command(struct amd_iommu *iommu, struct iommu_cmd *cmd) |
996 | { | |
997 | return iommu_queue_command_sync(iommu, cmd, true); | |
998 | } | |
999 | ||
8d201968 JR |
1000 | /* |
1001 | * This function queues a completion wait command into the command | |
1002 | * buffer of an IOMMU | |
1003 | */ | |
a19ae1ec | 1004 | static int iommu_completion_wait(struct amd_iommu *iommu) |
8d201968 JR |
1005 | { |
1006 | struct iommu_cmd cmd; | |
815b33fd | 1007 | volatile u64 sem = 0; |
ac0ea6e9 | 1008 | int ret; |
8d201968 | 1009 | |
09ee17eb | 1010 | if (!iommu->need_sync) |
815b33fd | 1011 | return 0; |
09ee17eb | 1012 | |
815b33fd | 1013 | build_completion_wait(&cmd, (u64)&sem); |
a19ae1ec | 1014 | |
f1ca1512 | 1015 | ret = iommu_queue_command_sync(iommu, &cmd, false); |
a19ae1ec | 1016 | if (ret) |
815b33fd | 1017 | return ret; |
8d201968 | 1018 | |
ac0ea6e9 | 1019 | return wait_on_sem(&sem); |
8d201968 JR |
1020 | } |
1021 | ||
d8c13085 | 1022 | static int iommu_flush_dte(struct amd_iommu *iommu, u16 devid) |
a19ae1ec | 1023 | { |
d8c13085 | 1024 | struct iommu_cmd cmd; |
a19ae1ec | 1025 | |
d8c13085 | 1026 | build_inv_dte(&cmd, devid); |
7e4f88da | 1027 | |
d8c13085 JR |
1028 | return iommu_queue_command(iommu, &cmd); |
1029 | } | |
09ee17eb | 1030 | |
7d0c5cc5 JR |
1031 | static void iommu_flush_dte_all(struct amd_iommu *iommu) |
1032 | { | |
1033 | u32 devid; | |
09ee17eb | 1034 | |
7d0c5cc5 JR |
1035 | for (devid = 0; devid <= 0xffff; ++devid) |
1036 | iommu_flush_dte(iommu, devid); | |
a19ae1ec | 1037 | |
7d0c5cc5 JR |
1038 | iommu_completion_wait(iommu); |
1039 | } | |
84df8175 | 1040 | |
7d0c5cc5 JR |
1041 | /* |
1042 | * This function uses heavy locking and may disable irqs for some time. But | |
1043 | * this is no issue because it is only called during resume. | |
1044 | */ | |
1045 | static void iommu_flush_tlb_all(struct amd_iommu *iommu) | |
1046 | { | |
1047 | u32 dom_id; | |
a19ae1ec | 1048 | |
7d0c5cc5 JR |
1049 | for (dom_id = 0; dom_id <= 0xffff; ++dom_id) { |
1050 | struct iommu_cmd cmd; | |
1051 | build_inv_iommu_pages(&cmd, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, | |
1052 | dom_id, 1); | |
1053 | iommu_queue_command(iommu, &cmd); | |
1054 | } | |
8eed9833 | 1055 | |
7d0c5cc5 | 1056 | iommu_completion_wait(iommu); |
a19ae1ec JR |
1057 | } |
1058 | ||
58fc7f14 | 1059 | static void iommu_flush_all(struct amd_iommu *iommu) |
0518a3a4 | 1060 | { |
58fc7f14 | 1061 | struct iommu_cmd cmd; |
0518a3a4 | 1062 | |
58fc7f14 | 1063 | build_inv_all(&cmd); |
0518a3a4 | 1064 | |
58fc7f14 JR |
1065 | iommu_queue_command(iommu, &cmd); |
1066 | iommu_completion_wait(iommu); | |
1067 | } | |
1068 | ||
7ef2798d JR |
1069 | static void iommu_flush_irt(struct amd_iommu *iommu, u16 devid) |
1070 | { | |
1071 | struct iommu_cmd cmd; | |
1072 | ||
1073 | build_inv_irt(&cmd, devid); | |
1074 | ||
1075 | iommu_queue_command(iommu, &cmd); | |
1076 | } | |
1077 | ||
1078 | static void iommu_flush_irt_all(struct amd_iommu *iommu) | |
1079 | { | |
1080 | u32 devid; | |
1081 | ||
1082 | for (devid = 0; devid <= MAX_DEV_TABLE_ENTRIES; devid++) | |
1083 | iommu_flush_irt(iommu, devid); | |
1084 | ||
1085 | iommu_completion_wait(iommu); | |
1086 | } | |
1087 | ||
7d0c5cc5 JR |
1088 | void iommu_flush_all_caches(struct amd_iommu *iommu) |
1089 | { | |
58fc7f14 JR |
1090 | if (iommu_feature(iommu, FEATURE_IA)) { |
1091 | iommu_flush_all(iommu); | |
1092 | } else { | |
1093 | iommu_flush_dte_all(iommu); | |
7ef2798d | 1094 | iommu_flush_irt_all(iommu); |
58fc7f14 | 1095 | iommu_flush_tlb_all(iommu); |
0518a3a4 JR |
1096 | } |
1097 | } | |
1098 | ||
431b2a20 | 1099 | /* |
cb41ed85 | 1100 | * Command send function for flushing on-device TLB |
431b2a20 | 1101 | */ |
6c542047 JR |
1102 | static int device_flush_iotlb(struct iommu_dev_data *dev_data, |
1103 | u64 address, size_t size) | |
3fa43655 JR |
1104 | { |
1105 | struct amd_iommu *iommu; | |
b00d3bcf | 1106 | struct iommu_cmd cmd; |
cb41ed85 | 1107 | int qdep; |
3fa43655 | 1108 | |
ea61cddb JR |
1109 | qdep = dev_data->ats.qdep; |
1110 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
3fa43655 | 1111 | |
ea61cddb | 1112 | build_inv_iotlb_pages(&cmd, dev_data->devid, qdep, address, size); |
b00d3bcf JR |
1113 | |
1114 | return iommu_queue_command(iommu, &cmd); | |
3fa43655 JR |
1115 | } |
1116 | ||
431b2a20 | 1117 | /* |
431b2a20 | 1118 | * Command send function for invalidating a device table entry |
431b2a20 | 1119 | */ |
6c542047 | 1120 | static int device_flush_dte(struct iommu_dev_data *dev_data) |
a19ae1ec | 1121 | { |
3fa43655 | 1122 | struct amd_iommu *iommu; |
ee2fa743 | 1123 | int ret; |
a19ae1ec | 1124 | |
6c542047 | 1125 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
a19ae1ec | 1126 | |
f62dda66 | 1127 | ret = iommu_flush_dte(iommu, dev_data->devid); |
cb41ed85 JR |
1128 | if (ret) |
1129 | return ret; | |
1130 | ||
ea61cddb | 1131 | if (dev_data->ats.enabled) |
6c542047 | 1132 | ret = device_flush_iotlb(dev_data, 0, ~0UL); |
ee2fa743 | 1133 | |
ee2fa743 | 1134 | return ret; |
a19ae1ec JR |
1135 | } |
1136 | ||
431b2a20 JR |
1137 | /* |
1138 | * TLB invalidation function which is called from the mapping functions. | |
1139 | * It invalidates a single PTE if the range to flush is within a single | |
1140 | * page. Otherwise it flushes the whole TLB of the IOMMU. | |
1141 | */ | |
17b124bf JR |
1142 | static void __domain_flush_pages(struct protection_domain *domain, |
1143 | u64 address, size_t size, int pde) | |
a19ae1ec | 1144 | { |
cb41ed85 | 1145 | struct iommu_dev_data *dev_data; |
11b6402c JR |
1146 | struct iommu_cmd cmd; |
1147 | int ret = 0, i; | |
a19ae1ec | 1148 | |
11b6402c | 1149 | build_inv_iommu_pages(&cmd, address, size, domain->id, pde); |
999ba417 | 1150 | |
6de8ad9b JR |
1151 | for (i = 0; i < amd_iommus_present; ++i) { |
1152 | if (!domain->dev_iommu[i]) | |
1153 | continue; | |
1154 | ||
1155 | /* | |
1156 | * Devices of this domain are behind this IOMMU | |
1157 | * We need a TLB flush | |
1158 | */ | |
11b6402c | 1159 | ret |= iommu_queue_command(amd_iommus[i], &cmd); |
6de8ad9b JR |
1160 | } |
1161 | ||
cb41ed85 | 1162 | list_for_each_entry(dev_data, &domain->dev_list, list) { |
cb41ed85 | 1163 | |
ea61cddb | 1164 | if (!dev_data->ats.enabled) |
cb41ed85 JR |
1165 | continue; |
1166 | ||
6c542047 | 1167 | ret |= device_flush_iotlb(dev_data, address, size); |
cb41ed85 JR |
1168 | } |
1169 | ||
11b6402c | 1170 | WARN_ON(ret); |
6de8ad9b JR |
1171 | } |
1172 | ||
17b124bf JR |
1173 | static void domain_flush_pages(struct protection_domain *domain, |
1174 | u64 address, size_t size) | |
6de8ad9b | 1175 | { |
17b124bf | 1176 | __domain_flush_pages(domain, address, size, 0); |
a19ae1ec | 1177 | } |
b6c02715 | 1178 | |
1c655773 | 1179 | /* Flush the whole IO/TLB for a given protection domain */ |
17b124bf | 1180 | static void domain_flush_tlb(struct protection_domain *domain) |
1c655773 | 1181 | { |
17b124bf | 1182 | __domain_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 0); |
1c655773 JR |
1183 | } |
1184 | ||
42a49f96 | 1185 | /* Flush the whole IO/TLB for a given protection domain - including PDE */ |
17b124bf | 1186 | static void domain_flush_tlb_pde(struct protection_domain *domain) |
42a49f96 | 1187 | { |
17b124bf | 1188 | __domain_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 1); |
42a49f96 CW |
1189 | } |
1190 | ||
17b124bf | 1191 | static void domain_flush_complete(struct protection_domain *domain) |
b00d3bcf | 1192 | { |
17b124bf | 1193 | int i; |
18811f55 | 1194 | |
17b124bf JR |
1195 | for (i = 0; i < amd_iommus_present; ++i) { |
1196 | if (!domain->dev_iommu[i]) | |
1197 | continue; | |
bfd1be18 | 1198 | |
17b124bf JR |
1199 | /* |
1200 | * Devices of this domain are behind this IOMMU | |
1201 | * We need to wait for completion of all commands. | |
1202 | */ | |
1203 | iommu_completion_wait(amd_iommus[i]); | |
bfd1be18 | 1204 | } |
e394d72a JR |
1205 | } |
1206 | ||
b00d3bcf | 1207 | |
09b42804 | 1208 | /* |
b00d3bcf | 1209 | * This function flushes the DTEs for all devices in domain |
09b42804 | 1210 | */ |
17b124bf | 1211 | static void domain_flush_devices(struct protection_domain *domain) |
e394d72a | 1212 | { |
b00d3bcf | 1213 | struct iommu_dev_data *dev_data; |
b26e81b8 | 1214 | |
b00d3bcf | 1215 | list_for_each_entry(dev_data, &domain->dev_list, list) |
6c542047 | 1216 | device_flush_dte(dev_data); |
a345b23b JR |
1217 | } |
1218 | ||
431b2a20 JR |
1219 | /**************************************************************************** |
1220 | * | |
1221 | * The functions below are used the create the page table mappings for | |
1222 | * unity mapped regions. | |
1223 | * | |
1224 | ****************************************************************************/ | |
1225 | ||
308973d3 JR |
1226 | /* |
1227 | * This function is used to add another level to an IO page table. Adding | |
1228 | * another level increases the size of the address space by 9 bits to a size up | |
1229 | * to 64 bits. | |
1230 | */ | |
1231 | static bool increase_address_space(struct protection_domain *domain, | |
1232 | gfp_t gfp) | |
1233 | { | |
1234 | u64 *pte; | |
1235 | ||
1236 | if (domain->mode == PAGE_MODE_6_LEVEL) | |
1237 | /* address space already 64 bit large */ | |
1238 | return false; | |
1239 | ||
1240 | pte = (void *)get_zeroed_page(gfp); | |
1241 | if (!pte) | |
1242 | return false; | |
1243 | ||
1244 | *pte = PM_LEVEL_PDE(domain->mode, | |
1245 | virt_to_phys(domain->pt_root)); | |
1246 | domain->pt_root = pte; | |
1247 | domain->mode += 1; | |
1248 | domain->updated = true; | |
1249 | ||
1250 | return true; | |
1251 | } | |
1252 | ||
1253 | static u64 *alloc_pte(struct protection_domain *domain, | |
1254 | unsigned long address, | |
cbb9d729 | 1255 | unsigned long page_size, |
308973d3 JR |
1256 | u64 **pte_page, |
1257 | gfp_t gfp) | |
1258 | { | |
cbb9d729 | 1259 | int level, end_lvl; |
308973d3 | 1260 | u64 *pte, *page; |
cbb9d729 JR |
1261 | |
1262 | BUG_ON(!is_power_of_2(page_size)); | |
308973d3 JR |
1263 | |
1264 | while (address > PM_LEVEL_SIZE(domain->mode)) | |
1265 | increase_address_space(domain, gfp); | |
1266 | ||
cbb9d729 JR |
1267 | level = domain->mode - 1; |
1268 | pte = &domain->pt_root[PM_LEVEL_INDEX(level, address)]; | |
1269 | address = PAGE_SIZE_ALIGN(address, page_size); | |
1270 | end_lvl = PAGE_SIZE_LEVEL(page_size); | |
308973d3 JR |
1271 | |
1272 | while (level > end_lvl) { | |
1273 | if (!IOMMU_PTE_PRESENT(*pte)) { | |
1274 | page = (u64 *)get_zeroed_page(gfp); | |
1275 | if (!page) | |
1276 | return NULL; | |
1277 | *pte = PM_LEVEL_PDE(level, virt_to_phys(page)); | |
1278 | } | |
1279 | ||
cbb9d729 JR |
1280 | /* No level skipping support yet */ |
1281 | if (PM_PTE_LEVEL(*pte) != level) | |
1282 | return NULL; | |
1283 | ||
308973d3 JR |
1284 | level -= 1; |
1285 | ||
1286 | pte = IOMMU_PTE_PAGE(*pte); | |
1287 | ||
1288 | if (pte_page && level == end_lvl) | |
1289 | *pte_page = pte; | |
1290 | ||
1291 | pte = &pte[PM_LEVEL_INDEX(level, address)]; | |
1292 | } | |
1293 | ||
1294 | return pte; | |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * This function checks if there is a PTE for a given dma address. If | |
1299 | * there is one, it returns the pointer to it. | |
1300 | */ | |
24cd7723 | 1301 | static u64 *fetch_pte(struct protection_domain *domain, unsigned long address) |
308973d3 JR |
1302 | { |
1303 | int level; | |
1304 | u64 *pte; | |
1305 | ||
24cd7723 JR |
1306 | if (address > PM_LEVEL_SIZE(domain->mode)) |
1307 | return NULL; | |
1308 | ||
1309 | level = domain->mode - 1; | |
1310 | pte = &domain->pt_root[PM_LEVEL_INDEX(level, address)]; | |
308973d3 | 1311 | |
24cd7723 JR |
1312 | while (level > 0) { |
1313 | ||
1314 | /* Not Present */ | |
308973d3 JR |
1315 | if (!IOMMU_PTE_PRESENT(*pte)) |
1316 | return NULL; | |
1317 | ||
24cd7723 JR |
1318 | /* Large PTE */ |
1319 | if (PM_PTE_LEVEL(*pte) == 0x07) { | |
1320 | unsigned long pte_mask, __pte; | |
1321 | ||
1322 | /* | |
1323 | * If we have a series of large PTEs, make | |
1324 | * sure to return a pointer to the first one. | |
1325 | */ | |
1326 | pte_mask = PTE_PAGE_SIZE(*pte); | |
1327 | pte_mask = ~((PAGE_SIZE_PTE_COUNT(pte_mask) << 3) - 1); | |
1328 | __pte = ((unsigned long)pte) & pte_mask; | |
1329 | ||
1330 | return (u64 *)__pte; | |
1331 | } | |
1332 | ||
1333 | /* No level skipping support yet */ | |
1334 | if (PM_PTE_LEVEL(*pte) != level) | |
1335 | return NULL; | |
1336 | ||
308973d3 JR |
1337 | level -= 1; |
1338 | ||
24cd7723 | 1339 | /* Walk to the next level */ |
308973d3 JR |
1340 | pte = IOMMU_PTE_PAGE(*pte); |
1341 | pte = &pte[PM_LEVEL_INDEX(level, address)]; | |
308973d3 JR |
1342 | } |
1343 | ||
1344 | return pte; | |
1345 | } | |
1346 | ||
431b2a20 JR |
1347 | /* |
1348 | * Generic mapping functions. It maps a physical address into a DMA | |
1349 | * address space. It allocates the page table pages if necessary. | |
1350 | * In the future it can be extended to a generic mapping function | |
1351 | * supporting all features of AMD IOMMU page tables like level skipping | |
1352 | * and full 64 bit address spaces. | |
1353 | */ | |
38e817fe JR |
1354 | static int iommu_map_page(struct protection_domain *dom, |
1355 | unsigned long bus_addr, | |
1356 | unsigned long phys_addr, | |
abdc5eb3 | 1357 | int prot, |
cbb9d729 | 1358 | unsigned long page_size) |
bd0e5211 | 1359 | { |
8bda3092 | 1360 | u64 __pte, *pte; |
cbb9d729 | 1361 | int i, count; |
abdc5eb3 | 1362 | |
bad1cac2 | 1363 | if (!(prot & IOMMU_PROT_MASK)) |
bd0e5211 JR |
1364 | return -EINVAL; |
1365 | ||
cbb9d729 JR |
1366 | bus_addr = PAGE_ALIGN(bus_addr); |
1367 | phys_addr = PAGE_ALIGN(phys_addr); | |
1368 | count = PAGE_SIZE_PTE_COUNT(page_size); | |
1369 | pte = alloc_pte(dom, bus_addr, page_size, NULL, GFP_KERNEL); | |
1370 | ||
1371 | for (i = 0; i < count; ++i) | |
1372 | if (IOMMU_PTE_PRESENT(pte[i])) | |
1373 | return -EBUSY; | |
bd0e5211 | 1374 | |
cbb9d729 JR |
1375 | if (page_size > PAGE_SIZE) { |
1376 | __pte = PAGE_SIZE_PTE(phys_addr, page_size); | |
1377 | __pte |= PM_LEVEL_ENC(7) | IOMMU_PTE_P | IOMMU_PTE_FC; | |
1378 | } else | |
1379 | __pte = phys_addr | IOMMU_PTE_P | IOMMU_PTE_FC; | |
bd0e5211 | 1380 | |
bd0e5211 JR |
1381 | if (prot & IOMMU_PROT_IR) |
1382 | __pte |= IOMMU_PTE_IR; | |
1383 | if (prot & IOMMU_PROT_IW) | |
1384 | __pte |= IOMMU_PTE_IW; | |
1385 | ||
cbb9d729 JR |
1386 | for (i = 0; i < count; ++i) |
1387 | pte[i] = __pte; | |
bd0e5211 | 1388 | |
04bfdd84 JR |
1389 | update_domain(dom); |
1390 | ||
bd0e5211 JR |
1391 | return 0; |
1392 | } | |
1393 | ||
24cd7723 JR |
1394 | static unsigned long iommu_unmap_page(struct protection_domain *dom, |
1395 | unsigned long bus_addr, | |
1396 | unsigned long page_size) | |
eb74ff6c | 1397 | { |
24cd7723 JR |
1398 | unsigned long long unmap_size, unmapped; |
1399 | u64 *pte; | |
1400 | ||
1401 | BUG_ON(!is_power_of_2(page_size)); | |
1402 | ||
1403 | unmapped = 0; | |
eb74ff6c | 1404 | |
24cd7723 JR |
1405 | while (unmapped < page_size) { |
1406 | ||
1407 | pte = fetch_pte(dom, bus_addr); | |
1408 | ||
1409 | if (!pte) { | |
1410 | /* | |
1411 | * No PTE for this address | |
1412 | * move forward in 4kb steps | |
1413 | */ | |
1414 | unmap_size = PAGE_SIZE; | |
1415 | } else if (PM_PTE_LEVEL(*pte) == 0) { | |
1416 | /* 4kb PTE found for this address */ | |
1417 | unmap_size = PAGE_SIZE; | |
1418 | *pte = 0ULL; | |
1419 | } else { | |
1420 | int count, i; | |
1421 | ||
1422 | /* Large PTE found which maps this address */ | |
1423 | unmap_size = PTE_PAGE_SIZE(*pte); | |
60d0ca3c AW |
1424 | |
1425 | /* Only unmap from the first pte in the page */ | |
1426 | if ((unmap_size - 1) & bus_addr) | |
1427 | break; | |
24cd7723 JR |
1428 | count = PAGE_SIZE_PTE_COUNT(unmap_size); |
1429 | for (i = 0; i < count; i++) | |
1430 | pte[i] = 0ULL; | |
1431 | } | |
1432 | ||
1433 | bus_addr = (bus_addr & ~(unmap_size - 1)) + unmap_size; | |
1434 | unmapped += unmap_size; | |
1435 | } | |
1436 | ||
60d0ca3c | 1437 | BUG_ON(unmapped && !is_power_of_2(unmapped)); |
eb74ff6c | 1438 | |
24cd7723 | 1439 | return unmapped; |
eb74ff6c | 1440 | } |
eb74ff6c | 1441 | |
431b2a20 JR |
1442 | /* |
1443 | * This function checks if a specific unity mapping entry is needed for | |
1444 | * this specific IOMMU. | |
1445 | */ | |
bd0e5211 JR |
1446 | static int iommu_for_unity_map(struct amd_iommu *iommu, |
1447 | struct unity_map_entry *entry) | |
1448 | { | |
1449 | u16 bdf, i; | |
1450 | ||
1451 | for (i = entry->devid_start; i <= entry->devid_end; ++i) { | |
1452 | bdf = amd_iommu_alias_table[i]; | |
1453 | if (amd_iommu_rlookup_table[bdf] == iommu) | |
1454 | return 1; | |
1455 | } | |
1456 | ||
1457 | return 0; | |
1458 | } | |
1459 | ||
431b2a20 JR |
1460 | /* |
1461 | * This function actually applies the mapping to the page table of the | |
1462 | * dma_ops domain. | |
1463 | */ | |
bd0e5211 JR |
1464 | static int dma_ops_unity_map(struct dma_ops_domain *dma_dom, |
1465 | struct unity_map_entry *e) | |
1466 | { | |
1467 | u64 addr; | |
1468 | int ret; | |
1469 | ||
1470 | for (addr = e->address_start; addr < e->address_end; | |
1471 | addr += PAGE_SIZE) { | |
abdc5eb3 | 1472 | ret = iommu_map_page(&dma_dom->domain, addr, addr, e->prot, |
cbb9d729 | 1473 | PAGE_SIZE); |
bd0e5211 JR |
1474 | if (ret) |
1475 | return ret; | |
1476 | /* | |
1477 | * if unity mapping is in aperture range mark the page | |
1478 | * as allocated in the aperture | |
1479 | */ | |
1480 | if (addr < dma_dom->aperture_size) | |
c3239567 | 1481 | __set_bit(addr >> PAGE_SHIFT, |
384de729 | 1482 | dma_dom->aperture[0]->bitmap); |
bd0e5211 JR |
1483 | } |
1484 | ||
1485 | return 0; | |
1486 | } | |
1487 | ||
171e7b37 JR |
1488 | /* |
1489 | * Init the unity mappings for a specific IOMMU in the system | |
1490 | * | |
1491 | * Basically iterates over all unity mapping entries and applies them to | |
1492 | * the default domain DMA of that IOMMU if necessary. | |
1493 | */ | |
1494 | static int iommu_init_unity_mappings(struct amd_iommu *iommu) | |
1495 | { | |
1496 | struct unity_map_entry *entry; | |
1497 | int ret; | |
1498 | ||
1499 | list_for_each_entry(entry, &amd_iommu_unity_map, list) { | |
1500 | if (!iommu_for_unity_map(iommu, entry)) | |
1501 | continue; | |
1502 | ret = dma_ops_unity_map(iommu->default_dom, entry); | |
1503 | if (ret) | |
1504 | return ret; | |
1505 | } | |
1506 | ||
1507 | return 0; | |
1508 | } | |
1509 | ||
431b2a20 JR |
1510 | /* |
1511 | * Inits the unity mappings required for a specific device | |
1512 | */ | |
bd0e5211 JR |
1513 | static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom, |
1514 | u16 devid) | |
1515 | { | |
1516 | struct unity_map_entry *e; | |
1517 | int ret; | |
1518 | ||
1519 | list_for_each_entry(e, &amd_iommu_unity_map, list) { | |
1520 | if (!(devid >= e->devid_start && devid <= e->devid_end)) | |
1521 | continue; | |
1522 | ret = dma_ops_unity_map(dma_dom, e); | |
1523 | if (ret) | |
1524 | return ret; | |
1525 | } | |
1526 | ||
1527 | return 0; | |
1528 | } | |
1529 | ||
431b2a20 JR |
1530 | /**************************************************************************** |
1531 | * | |
1532 | * The next functions belong to the address allocator for the dma_ops | |
1533 | * interface functions. They work like the allocators in the other IOMMU | |
1534 | * drivers. Its basically a bitmap which marks the allocated pages in | |
1535 | * the aperture. Maybe it could be enhanced in the future to a more | |
1536 | * efficient allocator. | |
1537 | * | |
1538 | ****************************************************************************/ | |
d3086444 | 1539 | |
431b2a20 | 1540 | /* |
384de729 | 1541 | * The address allocator core functions. |
431b2a20 JR |
1542 | * |
1543 | * called with domain->lock held | |
1544 | */ | |
384de729 | 1545 | |
171e7b37 JR |
1546 | /* |
1547 | * Used to reserve address ranges in the aperture (e.g. for exclusion | |
1548 | * ranges. | |
1549 | */ | |
1550 | static void dma_ops_reserve_addresses(struct dma_ops_domain *dom, | |
1551 | unsigned long start_page, | |
1552 | unsigned int pages) | |
1553 | { | |
1554 | unsigned int i, last_page = dom->aperture_size >> PAGE_SHIFT; | |
1555 | ||
1556 | if (start_page + pages > last_page) | |
1557 | pages = last_page - start_page; | |
1558 | ||
1559 | for (i = start_page; i < start_page + pages; ++i) { | |
1560 | int index = i / APERTURE_RANGE_PAGES; | |
1561 | int page = i % APERTURE_RANGE_PAGES; | |
1562 | __set_bit(page, dom->aperture[index]->bitmap); | |
1563 | } | |
1564 | } | |
1565 | ||
9cabe89b JR |
1566 | /* |
1567 | * This function is used to add a new aperture range to an existing | |
1568 | * aperture in case of dma_ops domain allocation or address allocation | |
1569 | * failure. | |
1570 | */ | |
576175c2 | 1571 | static int alloc_new_range(struct dma_ops_domain *dma_dom, |
9cabe89b JR |
1572 | bool populate, gfp_t gfp) |
1573 | { | |
1574 | int index = dma_dom->aperture_size >> APERTURE_RANGE_SHIFT; | |
576175c2 | 1575 | struct amd_iommu *iommu; |
17f5b569 | 1576 | unsigned long i, old_size; |
9cabe89b | 1577 | |
f5e9705c JR |
1578 | #ifdef CONFIG_IOMMU_STRESS |
1579 | populate = false; | |
1580 | #endif | |
1581 | ||
9cabe89b JR |
1582 | if (index >= APERTURE_MAX_RANGES) |
1583 | return -ENOMEM; | |
1584 | ||
1585 | dma_dom->aperture[index] = kzalloc(sizeof(struct aperture_range), gfp); | |
1586 | if (!dma_dom->aperture[index]) | |
1587 | return -ENOMEM; | |
1588 | ||
1589 | dma_dom->aperture[index]->bitmap = (void *)get_zeroed_page(gfp); | |
1590 | if (!dma_dom->aperture[index]->bitmap) | |
1591 | goto out_free; | |
1592 | ||
1593 | dma_dom->aperture[index]->offset = dma_dom->aperture_size; | |
1594 | ||
1595 | if (populate) { | |
1596 | unsigned long address = dma_dom->aperture_size; | |
1597 | int i, num_ptes = APERTURE_RANGE_PAGES / 512; | |
1598 | u64 *pte, *pte_page; | |
1599 | ||
1600 | for (i = 0; i < num_ptes; ++i) { | |
cbb9d729 | 1601 | pte = alloc_pte(&dma_dom->domain, address, PAGE_SIZE, |
9cabe89b JR |
1602 | &pte_page, gfp); |
1603 | if (!pte) | |
1604 | goto out_free; | |
1605 | ||
1606 | dma_dom->aperture[index]->pte_pages[i] = pte_page; | |
1607 | ||
1608 | address += APERTURE_RANGE_SIZE / 64; | |
1609 | } | |
1610 | } | |
1611 | ||
17f5b569 | 1612 | old_size = dma_dom->aperture_size; |
9cabe89b JR |
1613 | dma_dom->aperture_size += APERTURE_RANGE_SIZE; |
1614 | ||
17f5b569 JR |
1615 | /* Reserve address range used for MSI messages */ |
1616 | if (old_size < MSI_ADDR_BASE_LO && | |
1617 | dma_dom->aperture_size > MSI_ADDR_BASE_LO) { | |
1618 | unsigned long spage; | |
1619 | int pages; | |
1620 | ||
1621 | pages = iommu_num_pages(MSI_ADDR_BASE_LO, 0x10000, PAGE_SIZE); | |
1622 | spage = MSI_ADDR_BASE_LO >> PAGE_SHIFT; | |
1623 | ||
1624 | dma_ops_reserve_addresses(dma_dom, spage, pages); | |
1625 | } | |
1626 | ||
b595076a | 1627 | /* Initialize the exclusion range if necessary */ |
576175c2 JR |
1628 | for_each_iommu(iommu) { |
1629 | if (iommu->exclusion_start && | |
1630 | iommu->exclusion_start >= dma_dom->aperture[index]->offset | |
1631 | && iommu->exclusion_start < dma_dom->aperture_size) { | |
1632 | unsigned long startpage; | |
1633 | int pages = iommu_num_pages(iommu->exclusion_start, | |
1634 | iommu->exclusion_length, | |
1635 | PAGE_SIZE); | |
1636 | startpage = iommu->exclusion_start >> PAGE_SHIFT; | |
1637 | dma_ops_reserve_addresses(dma_dom, startpage, pages); | |
1638 | } | |
00cd122a JR |
1639 | } |
1640 | ||
1641 | /* | |
1642 | * Check for areas already mapped as present in the new aperture | |
1643 | * range and mark those pages as reserved in the allocator. Such | |
1644 | * mappings may already exist as a result of requested unity | |
1645 | * mappings for devices. | |
1646 | */ | |
1647 | for (i = dma_dom->aperture[index]->offset; | |
1648 | i < dma_dom->aperture_size; | |
1649 | i += PAGE_SIZE) { | |
24cd7723 | 1650 | u64 *pte = fetch_pte(&dma_dom->domain, i); |
00cd122a JR |
1651 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) |
1652 | continue; | |
1653 | ||
fcd0861d | 1654 | dma_ops_reserve_addresses(dma_dom, i >> PAGE_SHIFT, 1); |
00cd122a JR |
1655 | } |
1656 | ||
04bfdd84 JR |
1657 | update_domain(&dma_dom->domain); |
1658 | ||
9cabe89b JR |
1659 | return 0; |
1660 | ||
1661 | out_free: | |
04bfdd84 JR |
1662 | update_domain(&dma_dom->domain); |
1663 | ||
9cabe89b JR |
1664 | free_page((unsigned long)dma_dom->aperture[index]->bitmap); |
1665 | ||
1666 | kfree(dma_dom->aperture[index]); | |
1667 | dma_dom->aperture[index] = NULL; | |
1668 | ||
1669 | return -ENOMEM; | |
1670 | } | |
1671 | ||
384de729 JR |
1672 | static unsigned long dma_ops_area_alloc(struct device *dev, |
1673 | struct dma_ops_domain *dom, | |
1674 | unsigned int pages, | |
1675 | unsigned long align_mask, | |
1676 | u64 dma_mask, | |
1677 | unsigned long start) | |
1678 | { | |
803b8cb4 | 1679 | unsigned long next_bit = dom->next_address % APERTURE_RANGE_SIZE; |
384de729 JR |
1680 | int max_index = dom->aperture_size >> APERTURE_RANGE_SHIFT; |
1681 | int i = start >> APERTURE_RANGE_SHIFT; | |
1682 | unsigned long boundary_size; | |
1683 | unsigned long address = -1; | |
1684 | unsigned long limit; | |
1685 | ||
803b8cb4 JR |
1686 | next_bit >>= PAGE_SHIFT; |
1687 | ||
384de729 JR |
1688 | boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1, |
1689 | PAGE_SIZE) >> PAGE_SHIFT; | |
1690 | ||
1691 | for (;i < max_index; ++i) { | |
1692 | unsigned long offset = dom->aperture[i]->offset >> PAGE_SHIFT; | |
1693 | ||
1694 | if (dom->aperture[i]->offset >= dma_mask) | |
1695 | break; | |
1696 | ||
1697 | limit = iommu_device_max_index(APERTURE_RANGE_PAGES, offset, | |
1698 | dma_mask >> PAGE_SHIFT); | |
1699 | ||
1700 | address = iommu_area_alloc(dom->aperture[i]->bitmap, | |
1701 | limit, next_bit, pages, 0, | |
1702 | boundary_size, align_mask); | |
1703 | if (address != -1) { | |
1704 | address = dom->aperture[i]->offset + | |
1705 | (address << PAGE_SHIFT); | |
803b8cb4 | 1706 | dom->next_address = address + (pages << PAGE_SHIFT); |
384de729 JR |
1707 | break; |
1708 | } | |
1709 | ||
1710 | next_bit = 0; | |
1711 | } | |
1712 | ||
1713 | return address; | |
1714 | } | |
1715 | ||
d3086444 JR |
1716 | static unsigned long dma_ops_alloc_addresses(struct device *dev, |
1717 | struct dma_ops_domain *dom, | |
6d4f343f | 1718 | unsigned int pages, |
832a90c3 JR |
1719 | unsigned long align_mask, |
1720 | u64 dma_mask) | |
d3086444 | 1721 | { |
d3086444 | 1722 | unsigned long address; |
d3086444 | 1723 | |
fe16f088 JR |
1724 | #ifdef CONFIG_IOMMU_STRESS |
1725 | dom->next_address = 0; | |
1726 | dom->need_flush = true; | |
1727 | #endif | |
d3086444 | 1728 | |
384de729 | 1729 | address = dma_ops_area_alloc(dev, dom, pages, align_mask, |
803b8cb4 | 1730 | dma_mask, dom->next_address); |
d3086444 | 1731 | |
1c655773 | 1732 | if (address == -1) { |
803b8cb4 | 1733 | dom->next_address = 0; |
384de729 JR |
1734 | address = dma_ops_area_alloc(dev, dom, pages, align_mask, |
1735 | dma_mask, 0); | |
1c655773 JR |
1736 | dom->need_flush = true; |
1737 | } | |
d3086444 | 1738 | |
384de729 | 1739 | if (unlikely(address == -1)) |
8fd524b3 | 1740 | address = DMA_ERROR_CODE; |
d3086444 JR |
1741 | |
1742 | WARN_ON((address + (PAGE_SIZE*pages)) > dom->aperture_size); | |
1743 | ||
1744 | return address; | |
1745 | } | |
1746 | ||
431b2a20 JR |
1747 | /* |
1748 | * The address free function. | |
1749 | * | |
1750 | * called with domain->lock held | |
1751 | */ | |
d3086444 JR |
1752 | static void dma_ops_free_addresses(struct dma_ops_domain *dom, |
1753 | unsigned long address, | |
1754 | unsigned int pages) | |
1755 | { | |
384de729 JR |
1756 | unsigned i = address >> APERTURE_RANGE_SHIFT; |
1757 | struct aperture_range *range = dom->aperture[i]; | |
80be308d | 1758 | |
384de729 JR |
1759 | BUG_ON(i >= APERTURE_MAX_RANGES || range == NULL); |
1760 | ||
47bccd6b JR |
1761 | #ifdef CONFIG_IOMMU_STRESS |
1762 | if (i < 4) | |
1763 | return; | |
1764 | #endif | |
80be308d | 1765 | |
803b8cb4 | 1766 | if (address >= dom->next_address) |
80be308d | 1767 | dom->need_flush = true; |
384de729 JR |
1768 | |
1769 | address = (address % APERTURE_RANGE_SIZE) >> PAGE_SHIFT; | |
803b8cb4 | 1770 | |
a66022c4 | 1771 | bitmap_clear(range->bitmap, address, pages); |
384de729 | 1772 | |
d3086444 JR |
1773 | } |
1774 | ||
431b2a20 JR |
1775 | /**************************************************************************** |
1776 | * | |
1777 | * The next functions belong to the domain allocation. A domain is | |
1778 | * allocated for every IOMMU as the default domain. If device isolation | |
1779 | * is enabled, every device get its own domain. The most important thing | |
1780 | * about domains is the page table mapping the DMA address space they | |
1781 | * contain. | |
1782 | * | |
1783 | ****************************************************************************/ | |
1784 | ||
aeb26f55 JR |
1785 | /* |
1786 | * This function adds a protection domain to the global protection domain list | |
1787 | */ | |
1788 | static void add_domain_to_list(struct protection_domain *domain) | |
1789 | { | |
1790 | unsigned long flags; | |
1791 | ||
1792 | spin_lock_irqsave(&amd_iommu_pd_lock, flags); | |
1793 | list_add(&domain->list, &amd_iommu_pd_list); | |
1794 | spin_unlock_irqrestore(&amd_iommu_pd_lock, flags); | |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * This function removes a protection domain to the global | |
1799 | * protection domain list | |
1800 | */ | |
1801 | static void del_domain_from_list(struct protection_domain *domain) | |
1802 | { | |
1803 | unsigned long flags; | |
1804 | ||
1805 | spin_lock_irqsave(&amd_iommu_pd_lock, flags); | |
1806 | list_del(&domain->list); | |
1807 | spin_unlock_irqrestore(&amd_iommu_pd_lock, flags); | |
1808 | } | |
1809 | ||
ec487d1a JR |
1810 | static u16 domain_id_alloc(void) |
1811 | { | |
1812 | unsigned long flags; | |
1813 | int id; | |
1814 | ||
1815 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
1816 | id = find_first_zero_bit(amd_iommu_pd_alloc_bitmap, MAX_DOMAIN_ID); | |
1817 | BUG_ON(id == 0); | |
1818 | if (id > 0 && id < MAX_DOMAIN_ID) | |
1819 | __set_bit(id, amd_iommu_pd_alloc_bitmap); | |
1820 | else | |
1821 | id = 0; | |
1822 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
1823 | ||
1824 | return id; | |
1825 | } | |
1826 | ||
a2acfb75 JR |
1827 | static void domain_id_free(int id) |
1828 | { | |
1829 | unsigned long flags; | |
1830 | ||
1831 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
1832 | if (id > 0 && id < MAX_DOMAIN_ID) | |
1833 | __clear_bit(id, amd_iommu_pd_alloc_bitmap); | |
1834 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
1835 | } | |
a2acfb75 | 1836 | |
5c34c403 JR |
1837 | #define DEFINE_FREE_PT_FN(LVL, FN) \ |
1838 | static void free_pt_##LVL (unsigned long __pt) \ | |
1839 | { \ | |
1840 | unsigned long p; \ | |
1841 | u64 *pt; \ | |
1842 | int i; \ | |
1843 | \ | |
1844 | pt = (u64 *)__pt; \ | |
1845 | \ | |
1846 | for (i = 0; i < 512; ++i) { \ | |
1847 | if (!IOMMU_PTE_PRESENT(pt[i])) \ | |
1848 | continue; \ | |
1849 | \ | |
1850 | p = (unsigned long)IOMMU_PTE_PAGE(pt[i]); \ | |
1851 | FN(p); \ | |
1852 | } \ | |
1853 | free_page((unsigned long)pt); \ | |
1854 | } | |
1855 | ||
1856 | DEFINE_FREE_PT_FN(l2, free_page) | |
1857 | DEFINE_FREE_PT_FN(l3, free_pt_l2) | |
1858 | DEFINE_FREE_PT_FN(l4, free_pt_l3) | |
1859 | DEFINE_FREE_PT_FN(l5, free_pt_l4) | |
1860 | DEFINE_FREE_PT_FN(l6, free_pt_l5) | |
1861 | ||
86db2e5d | 1862 | static void free_pagetable(struct protection_domain *domain) |
ec487d1a | 1863 | { |
5c34c403 | 1864 | unsigned long root = (unsigned long)domain->pt_root; |
ec487d1a | 1865 | |
5c34c403 JR |
1866 | switch (domain->mode) { |
1867 | case PAGE_MODE_NONE: | |
1868 | break; | |
1869 | case PAGE_MODE_1_LEVEL: | |
1870 | free_page(root); | |
1871 | break; | |
1872 | case PAGE_MODE_2_LEVEL: | |
1873 | free_pt_l2(root); | |
1874 | break; | |
1875 | case PAGE_MODE_3_LEVEL: | |
1876 | free_pt_l3(root); | |
1877 | break; | |
1878 | case PAGE_MODE_4_LEVEL: | |
1879 | free_pt_l4(root); | |
1880 | break; | |
1881 | case PAGE_MODE_5_LEVEL: | |
1882 | free_pt_l5(root); | |
1883 | break; | |
1884 | case PAGE_MODE_6_LEVEL: | |
1885 | free_pt_l6(root); | |
1886 | break; | |
1887 | default: | |
1888 | BUG(); | |
ec487d1a | 1889 | } |
ec487d1a JR |
1890 | } |
1891 | ||
b16137b1 JR |
1892 | static void free_gcr3_tbl_level1(u64 *tbl) |
1893 | { | |
1894 | u64 *ptr; | |
1895 | int i; | |
1896 | ||
1897 | for (i = 0; i < 512; ++i) { | |
1898 | if (!(tbl[i] & GCR3_VALID)) | |
1899 | continue; | |
1900 | ||
1901 | ptr = __va(tbl[i] & PAGE_MASK); | |
1902 | ||
1903 | free_page((unsigned long)ptr); | |
1904 | } | |
1905 | } | |
1906 | ||
1907 | static void free_gcr3_tbl_level2(u64 *tbl) | |
1908 | { | |
1909 | u64 *ptr; | |
1910 | int i; | |
1911 | ||
1912 | for (i = 0; i < 512; ++i) { | |
1913 | if (!(tbl[i] & GCR3_VALID)) | |
1914 | continue; | |
1915 | ||
1916 | ptr = __va(tbl[i] & PAGE_MASK); | |
1917 | ||
1918 | free_gcr3_tbl_level1(ptr); | |
1919 | } | |
1920 | } | |
1921 | ||
52815b75 JR |
1922 | static void free_gcr3_table(struct protection_domain *domain) |
1923 | { | |
b16137b1 JR |
1924 | if (domain->glx == 2) |
1925 | free_gcr3_tbl_level2(domain->gcr3_tbl); | |
1926 | else if (domain->glx == 1) | |
1927 | free_gcr3_tbl_level1(domain->gcr3_tbl); | |
1928 | else if (domain->glx != 0) | |
1929 | BUG(); | |
1930 | ||
52815b75 JR |
1931 | free_page((unsigned long)domain->gcr3_tbl); |
1932 | } | |
1933 | ||
431b2a20 JR |
1934 | /* |
1935 | * Free a domain, only used if something went wrong in the | |
1936 | * allocation path and we need to free an already allocated page table | |
1937 | */ | |
ec487d1a JR |
1938 | static void dma_ops_domain_free(struct dma_ops_domain *dom) |
1939 | { | |
384de729 JR |
1940 | int i; |
1941 | ||
ec487d1a JR |
1942 | if (!dom) |
1943 | return; | |
1944 | ||
aeb26f55 JR |
1945 | del_domain_from_list(&dom->domain); |
1946 | ||
86db2e5d | 1947 | free_pagetable(&dom->domain); |
ec487d1a | 1948 | |
384de729 JR |
1949 | for (i = 0; i < APERTURE_MAX_RANGES; ++i) { |
1950 | if (!dom->aperture[i]) | |
1951 | continue; | |
1952 | free_page((unsigned long)dom->aperture[i]->bitmap); | |
1953 | kfree(dom->aperture[i]); | |
1954 | } | |
ec487d1a JR |
1955 | |
1956 | kfree(dom); | |
1957 | } | |
1958 | ||
431b2a20 JR |
1959 | /* |
1960 | * Allocates a new protection domain usable for the dma_ops functions. | |
b595076a | 1961 | * It also initializes the page table and the address allocator data |
431b2a20 JR |
1962 | * structures required for the dma_ops interface |
1963 | */ | |
87a64d52 | 1964 | static struct dma_ops_domain *dma_ops_domain_alloc(void) |
ec487d1a JR |
1965 | { |
1966 | struct dma_ops_domain *dma_dom; | |
ec487d1a JR |
1967 | |
1968 | dma_dom = kzalloc(sizeof(struct dma_ops_domain), GFP_KERNEL); | |
1969 | if (!dma_dom) | |
1970 | return NULL; | |
1971 | ||
1972 | spin_lock_init(&dma_dom->domain.lock); | |
1973 | ||
1974 | dma_dom->domain.id = domain_id_alloc(); | |
1975 | if (dma_dom->domain.id == 0) | |
1976 | goto free_dma_dom; | |
7c392cbe | 1977 | INIT_LIST_HEAD(&dma_dom->domain.dev_list); |
8f7a017c | 1978 | dma_dom->domain.mode = PAGE_MODE_2_LEVEL; |
ec487d1a | 1979 | dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL); |
9fdb19d6 | 1980 | dma_dom->domain.flags = PD_DMA_OPS_MASK; |
ec487d1a JR |
1981 | dma_dom->domain.priv = dma_dom; |
1982 | if (!dma_dom->domain.pt_root) | |
1983 | goto free_dma_dom; | |
ec487d1a | 1984 | |
1c655773 | 1985 | dma_dom->need_flush = false; |
bd60b735 | 1986 | dma_dom->target_dev = 0xffff; |
1c655773 | 1987 | |
aeb26f55 JR |
1988 | add_domain_to_list(&dma_dom->domain); |
1989 | ||
576175c2 | 1990 | if (alloc_new_range(dma_dom, true, GFP_KERNEL)) |
ec487d1a | 1991 | goto free_dma_dom; |
ec487d1a | 1992 | |
431b2a20 | 1993 | /* |
ec487d1a JR |
1994 | * mark the first page as allocated so we never return 0 as |
1995 | * a valid dma-address. So we can use 0 as error value | |
431b2a20 | 1996 | */ |
384de729 | 1997 | dma_dom->aperture[0]->bitmap[0] = 1; |
803b8cb4 | 1998 | dma_dom->next_address = 0; |
ec487d1a | 1999 | |
ec487d1a JR |
2000 | |
2001 | return dma_dom; | |
2002 | ||
2003 | free_dma_dom: | |
2004 | dma_ops_domain_free(dma_dom); | |
2005 | ||
2006 | return NULL; | |
2007 | } | |
2008 | ||
5b28df6f JR |
2009 | /* |
2010 | * little helper function to check whether a given protection domain is a | |
2011 | * dma_ops domain | |
2012 | */ | |
2013 | static bool dma_ops_domain(struct protection_domain *domain) | |
2014 | { | |
2015 | return domain->flags & PD_DMA_OPS_MASK; | |
2016 | } | |
2017 | ||
fd7b5535 | 2018 | static void set_dte_entry(u16 devid, struct protection_domain *domain, bool ats) |
b20ac0d4 | 2019 | { |
132bd68f | 2020 | u64 pte_root = 0; |
ee6c2868 | 2021 | u64 flags = 0; |
863c74eb | 2022 | |
132bd68f JR |
2023 | if (domain->mode != PAGE_MODE_NONE) |
2024 | pte_root = virt_to_phys(domain->pt_root); | |
2025 | ||
38ddf41b JR |
2026 | pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK) |
2027 | << DEV_ENTRY_MODE_SHIFT; | |
2028 | pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV; | |
b20ac0d4 | 2029 | |
ee6c2868 JR |
2030 | flags = amd_iommu_dev_table[devid].data[1]; |
2031 | ||
fd7b5535 JR |
2032 | if (ats) |
2033 | flags |= DTE_FLAG_IOTLB; | |
2034 | ||
52815b75 JR |
2035 | if (domain->flags & PD_IOMMUV2_MASK) { |
2036 | u64 gcr3 = __pa(domain->gcr3_tbl); | |
2037 | u64 glx = domain->glx; | |
2038 | u64 tmp; | |
2039 | ||
2040 | pte_root |= DTE_FLAG_GV; | |
2041 | pte_root |= (glx & DTE_GLX_MASK) << DTE_GLX_SHIFT; | |
2042 | ||
2043 | /* First mask out possible old values for GCR3 table */ | |
2044 | tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B; | |
2045 | flags &= ~tmp; | |
2046 | ||
2047 | tmp = DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C; | |
2048 | flags &= ~tmp; | |
2049 | ||
2050 | /* Encode GCR3 table into DTE */ | |
2051 | tmp = DTE_GCR3_VAL_A(gcr3) << DTE_GCR3_SHIFT_A; | |
2052 | pte_root |= tmp; | |
2053 | ||
2054 | tmp = DTE_GCR3_VAL_B(gcr3) << DTE_GCR3_SHIFT_B; | |
2055 | flags |= tmp; | |
2056 | ||
2057 | tmp = DTE_GCR3_VAL_C(gcr3) << DTE_GCR3_SHIFT_C; | |
2058 | flags |= tmp; | |
2059 | } | |
2060 | ||
ee6c2868 JR |
2061 | flags &= ~(0xffffUL); |
2062 | flags |= domain->id; | |
2063 | ||
2064 | amd_iommu_dev_table[devid].data[1] = flags; | |
2065 | amd_iommu_dev_table[devid].data[0] = pte_root; | |
15898bbc JR |
2066 | } |
2067 | ||
2068 | static void clear_dte_entry(u16 devid) | |
2069 | { | |
15898bbc JR |
2070 | /* remove entry from the device table seen by the hardware */ |
2071 | amd_iommu_dev_table[devid].data[0] = IOMMU_PTE_P | IOMMU_PTE_TV; | |
2072 | amd_iommu_dev_table[devid].data[1] = 0; | |
15898bbc JR |
2073 | |
2074 | amd_iommu_apply_erratum_63(devid); | |
7f760ddd JR |
2075 | } |
2076 | ||
ec9e79ef JR |
2077 | static void do_attach(struct iommu_dev_data *dev_data, |
2078 | struct protection_domain *domain) | |
7f760ddd | 2079 | { |
7f760ddd | 2080 | struct amd_iommu *iommu; |
ec9e79ef | 2081 | bool ats; |
fd7b5535 | 2082 | |
ec9e79ef JR |
2083 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
2084 | ats = dev_data->ats.enabled; | |
7f760ddd JR |
2085 | |
2086 | /* Update data structures */ | |
2087 | dev_data->domain = domain; | |
2088 | list_add(&dev_data->list, &domain->dev_list); | |
f62dda66 | 2089 | set_dte_entry(dev_data->devid, domain, ats); |
7f760ddd JR |
2090 | |
2091 | /* Do reference counting */ | |
2092 | domain->dev_iommu[iommu->index] += 1; | |
2093 | domain->dev_cnt += 1; | |
2094 | ||
2095 | /* Flush the DTE entry */ | |
6c542047 | 2096 | device_flush_dte(dev_data); |
7f760ddd JR |
2097 | } |
2098 | ||
ec9e79ef | 2099 | static void do_detach(struct iommu_dev_data *dev_data) |
7f760ddd | 2100 | { |
7f760ddd | 2101 | struct amd_iommu *iommu; |
7f760ddd | 2102 | |
ec9e79ef | 2103 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
15898bbc JR |
2104 | |
2105 | /* decrease reference counters */ | |
7f760ddd JR |
2106 | dev_data->domain->dev_iommu[iommu->index] -= 1; |
2107 | dev_data->domain->dev_cnt -= 1; | |
2108 | ||
2109 | /* Update data structures */ | |
2110 | dev_data->domain = NULL; | |
2111 | list_del(&dev_data->list); | |
f62dda66 | 2112 | clear_dte_entry(dev_data->devid); |
15898bbc | 2113 | |
7f760ddd | 2114 | /* Flush the DTE entry */ |
6c542047 | 2115 | device_flush_dte(dev_data); |
2b681faf JR |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * If a device is not yet associated with a domain, this function does | |
2120 | * assigns it visible for the hardware | |
2121 | */ | |
ec9e79ef | 2122 | static int __attach_device(struct iommu_dev_data *dev_data, |
15898bbc | 2123 | struct protection_domain *domain) |
2b681faf | 2124 | { |
84fe6c19 | 2125 | int ret; |
657cbb6b | 2126 | |
2b681faf JR |
2127 | /* lock domain */ |
2128 | spin_lock(&domain->lock); | |
2129 | ||
71f77580 JR |
2130 | if (dev_data->alias_data != NULL) { |
2131 | struct iommu_dev_data *alias_data = dev_data->alias_data; | |
15898bbc | 2132 | |
2b02b091 JR |
2133 | /* Some sanity checks */ |
2134 | ret = -EBUSY; | |
2135 | if (alias_data->domain != NULL && | |
2136 | alias_data->domain != domain) | |
2137 | goto out_unlock; | |
eba6ac60 | 2138 | |
2b02b091 JR |
2139 | if (dev_data->domain != NULL && |
2140 | dev_data->domain != domain) | |
2141 | goto out_unlock; | |
15898bbc | 2142 | |
2b02b091 | 2143 | /* Do real assignment */ |
7f760ddd | 2144 | if (alias_data->domain == NULL) |
ec9e79ef | 2145 | do_attach(alias_data, domain); |
24100055 JR |
2146 | |
2147 | atomic_inc(&alias_data->bind); | |
657cbb6b | 2148 | } |
15898bbc | 2149 | |
7f760ddd | 2150 | if (dev_data->domain == NULL) |
ec9e79ef | 2151 | do_attach(dev_data, domain); |
eba6ac60 | 2152 | |
24100055 JR |
2153 | atomic_inc(&dev_data->bind); |
2154 | ||
84fe6c19 JL |
2155 | ret = 0; |
2156 | ||
2157 | out_unlock: | |
2158 | ||
eba6ac60 JR |
2159 | /* ready */ |
2160 | spin_unlock(&domain->lock); | |
15898bbc | 2161 | |
84fe6c19 | 2162 | return ret; |
0feae533 | 2163 | } |
b20ac0d4 | 2164 | |
52815b75 JR |
2165 | |
2166 | static void pdev_iommuv2_disable(struct pci_dev *pdev) | |
2167 | { | |
2168 | pci_disable_ats(pdev); | |
2169 | pci_disable_pri(pdev); | |
2170 | pci_disable_pasid(pdev); | |
2171 | } | |
2172 | ||
6a113ddc JR |
2173 | /* FIXME: Change generic reset-function to do the same */ |
2174 | static int pri_reset_while_enabled(struct pci_dev *pdev) | |
2175 | { | |
2176 | u16 control; | |
2177 | int pos; | |
2178 | ||
46277b75 | 2179 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PRI); |
6a113ddc JR |
2180 | if (!pos) |
2181 | return -EINVAL; | |
2182 | ||
46277b75 JR |
2183 | pci_read_config_word(pdev, pos + PCI_PRI_CTRL, &control); |
2184 | control |= PCI_PRI_CTRL_RESET; | |
2185 | pci_write_config_word(pdev, pos + PCI_PRI_CTRL, control); | |
6a113ddc JR |
2186 | |
2187 | return 0; | |
2188 | } | |
2189 | ||
52815b75 JR |
2190 | static int pdev_iommuv2_enable(struct pci_dev *pdev) |
2191 | { | |
6a113ddc JR |
2192 | bool reset_enable; |
2193 | int reqs, ret; | |
2194 | ||
2195 | /* FIXME: Hardcode number of outstanding requests for now */ | |
2196 | reqs = 32; | |
2197 | if (pdev_pri_erratum(pdev, AMD_PRI_DEV_ERRATUM_LIMIT_REQ_ONE)) | |
2198 | reqs = 1; | |
2199 | reset_enable = pdev_pri_erratum(pdev, AMD_PRI_DEV_ERRATUM_ENABLE_RESET); | |
52815b75 JR |
2200 | |
2201 | /* Only allow access to user-accessible pages */ | |
2202 | ret = pci_enable_pasid(pdev, 0); | |
2203 | if (ret) | |
2204 | goto out_err; | |
2205 | ||
2206 | /* First reset the PRI state of the device */ | |
2207 | ret = pci_reset_pri(pdev); | |
2208 | if (ret) | |
2209 | goto out_err; | |
2210 | ||
6a113ddc JR |
2211 | /* Enable PRI */ |
2212 | ret = pci_enable_pri(pdev, reqs); | |
52815b75 JR |
2213 | if (ret) |
2214 | goto out_err; | |
2215 | ||
6a113ddc JR |
2216 | if (reset_enable) { |
2217 | ret = pri_reset_while_enabled(pdev); | |
2218 | if (ret) | |
2219 | goto out_err; | |
2220 | } | |
2221 | ||
52815b75 JR |
2222 | ret = pci_enable_ats(pdev, PAGE_SHIFT); |
2223 | if (ret) | |
2224 | goto out_err; | |
2225 | ||
2226 | return 0; | |
2227 | ||
2228 | out_err: | |
2229 | pci_disable_pri(pdev); | |
2230 | pci_disable_pasid(pdev); | |
2231 | ||
2232 | return ret; | |
2233 | } | |
2234 | ||
c99afa25 | 2235 | /* FIXME: Move this to PCI code */ |
a3b93121 | 2236 | #define PCI_PRI_TLP_OFF (1 << 15) |
c99afa25 | 2237 | |
98f1ad25 | 2238 | static bool pci_pri_tlp_required(struct pci_dev *pdev) |
c99afa25 | 2239 | { |
a3b93121 | 2240 | u16 status; |
c99afa25 JR |
2241 | int pos; |
2242 | ||
46277b75 | 2243 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PRI); |
c99afa25 JR |
2244 | if (!pos) |
2245 | return false; | |
2246 | ||
a3b93121 | 2247 | pci_read_config_word(pdev, pos + PCI_PRI_STATUS, &status); |
c99afa25 | 2248 | |
a3b93121 | 2249 | return (status & PCI_PRI_TLP_OFF) ? true : false; |
c99afa25 JR |
2250 | } |
2251 | ||
407d733e | 2252 | /* |
df805abb | 2253 | * If a device is not yet associated with a domain, this function |
407d733e JR |
2254 | * assigns it visible for the hardware |
2255 | */ | |
15898bbc JR |
2256 | static int attach_device(struct device *dev, |
2257 | struct protection_domain *domain) | |
0feae533 | 2258 | { |
fd7b5535 | 2259 | struct pci_dev *pdev = to_pci_dev(dev); |
ea61cddb | 2260 | struct iommu_dev_data *dev_data; |
eba6ac60 | 2261 | unsigned long flags; |
15898bbc | 2262 | int ret; |
eba6ac60 | 2263 | |
ea61cddb JR |
2264 | dev_data = get_dev_data(dev); |
2265 | ||
52815b75 JR |
2266 | if (domain->flags & PD_IOMMUV2_MASK) { |
2267 | if (!dev_data->iommu_v2 || !dev_data->passthrough) | |
2268 | return -EINVAL; | |
2269 | ||
2270 | if (pdev_iommuv2_enable(pdev) != 0) | |
2271 | return -EINVAL; | |
2272 | ||
2273 | dev_data->ats.enabled = true; | |
2274 | dev_data->ats.qdep = pci_ats_queue_depth(pdev); | |
c99afa25 | 2275 | dev_data->pri_tlp = pci_pri_tlp_required(pdev); |
52815b75 JR |
2276 | } else if (amd_iommu_iotlb_sup && |
2277 | pci_enable_ats(pdev, PAGE_SHIFT) == 0) { | |
ea61cddb JR |
2278 | dev_data->ats.enabled = true; |
2279 | dev_data->ats.qdep = pci_ats_queue_depth(pdev); | |
2280 | } | |
fd7b5535 | 2281 | |
eba6ac60 | 2282 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); |
ec9e79ef | 2283 | ret = __attach_device(dev_data, domain); |
b20ac0d4 JR |
2284 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); |
2285 | ||
0feae533 JR |
2286 | /* |
2287 | * We might boot into a crash-kernel here. The crashed kernel | |
2288 | * left the caches in the IOMMU dirty. So we have to flush | |
2289 | * here to evict all dirty stuff. | |
2290 | */ | |
17b124bf | 2291 | domain_flush_tlb_pde(domain); |
15898bbc JR |
2292 | |
2293 | return ret; | |
b20ac0d4 JR |
2294 | } |
2295 | ||
355bf553 JR |
2296 | /* |
2297 | * Removes a device from a protection domain (unlocked) | |
2298 | */ | |
ec9e79ef | 2299 | static void __detach_device(struct iommu_dev_data *dev_data) |
355bf553 | 2300 | { |
2ca76279 | 2301 | struct protection_domain *domain; |
7c392cbe | 2302 | unsigned long flags; |
c4596114 | 2303 | |
7f760ddd | 2304 | BUG_ON(!dev_data->domain); |
355bf553 | 2305 | |
2ca76279 JR |
2306 | domain = dev_data->domain; |
2307 | ||
2308 | spin_lock_irqsave(&domain->lock, flags); | |
24100055 | 2309 | |
71f77580 JR |
2310 | if (dev_data->alias_data != NULL) { |
2311 | struct iommu_dev_data *alias_data = dev_data->alias_data; | |
2312 | ||
7f760ddd | 2313 | if (atomic_dec_and_test(&alias_data->bind)) |
ec9e79ef | 2314 | do_detach(alias_data); |
24100055 JR |
2315 | } |
2316 | ||
7f760ddd | 2317 | if (atomic_dec_and_test(&dev_data->bind)) |
ec9e79ef | 2318 | do_detach(dev_data); |
7f760ddd | 2319 | |
2ca76279 | 2320 | spin_unlock_irqrestore(&domain->lock, flags); |
21129f78 JR |
2321 | |
2322 | /* | |
2323 | * If we run in passthrough mode the device must be assigned to the | |
d3ad9373 JR |
2324 | * passthrough domain if it is detached from any other domain. |
2325 | * Make sure we can deassign from the pt_domain itself. | |
21129f78 | 2326 | */ |
5abcdba4 | 2327 | if (dev_data->passthrough && |
d3ad9373 | 2328 | (dev_data->domain == NULL && domain != pt_domain)) |
ec9e79ef | 2329 | __attach_device(dev_data, pt_domain); |
355bf553 JR |
2330 | } |
2331 | ||
2332 | /* | |
2333 | * Removes a device from a protection domain (with devtable_lock held) | |
2334 | */ | |
15898bbc | 2335 | static void detach_device(struct device *dev) |
355bf553 | 2336 | { |
52815b75 | 2337 | struct protection_domain *domain; |
ea61cddb | 2338 | struct iommu_dev_data *dev_data; |
355bf553 JR |
2339 | unsigned long flags; |
2340 | ||
ec9e79ef | 2341 | dev_data = get_dev_data(dev); |
52815b75 | 2342 | domain = dev_data->domain; |
ec9e79ef | 2343 | |
355bf553 JR |
2344 | /* lock device table */ |
2345 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
ec9e79ef | 2346 | __detach_device(dev_data); |
355bf553 | 2347 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); |
fd7b5535 | 2348 | |
52815b75 JR |
2349 | if (domain->flags & PD_IOMMUV2_MASK) |
2350 | pdev_iommuv2_disable(to_pci_dev(dev)); | |
2351 | else if (dev_data->ats.enabled) | |
ea61cddb | 2352 | pci_disable_ats(to_pci_dev(dev)); |
52815b75 JR |
2353 | |
2354 | dev_data->ats.enabled = false; | |
355bf553 | 2355 | } |
e275a2a0 | 2356 | |
15898bbc JR |
2357 | /* |
2358 | * Find out the protection domain structure for a given PCI device. This | |
2359 | * will give us the pointer to the page table root for example. | |
2360 | */ | |
2361 | static struct protection_domain *domain_for_device(struct device *dev) | |
2362 | { | |
71f77580 | 2363 | struct iommu_dev_data *dev_data; |
2b02b091 | 2364 | struct protection_domain *dom = NULL; |
15898bbc | 2365 | unsigned long flags; |
15898bbc | 2366 | |
657cbb6b | 2367 | dev_data = get_dev_data(dev); |
15898bbc | 2368 | |
2b02b091 JR |
2369 | if (dev_data->domain) |
2370 | return dev_data->domain; | |
15898bbc | 2371 | |
71f77580 JR |
2372 | if (dev_data->alias_data != NULL) { |
2373 | struct iommu_dev_data *alias_data = dev_data->alias_data; | |
2b02b091 JR |
2374 | |
2375 | read_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
2376 | if (alias_data->domain != NULL) { | |
2377 | __attach_device(dev_data, alias_data->domain); | |
2378 | dom = alias_data->domain; | |
2379 | } | |
2380 | read_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
2381 | } | |
15898bbc JR |
2382 | |
2383 | return dom; | |
2384 | } | |
2385 | ||
e275a2a0 JR |
2386 | static int device_change_notifier(struct notifier_block *nb, |
2387 | unsigned long action, void *data) | |
2388 | { | |
e275a2a0 | 2389 | struct dma_ops_domain *dma_domain; |
5abcdba4 JR |
2390 | struct protection_domain *domain; |
2391 | struct iommu_dev_data *dev_data; | |
2392 | struct device *dev = data; | |
e275a2a0 | 2393 | struct amd_iommu *iommu; |
1ac4cbbc | 2394 | unsigned long flags; |
5abcdba4 | 2395 | u16 devid; |
e275a2a0 | 2396 | |
98fc5a69 JR |
2397 | if (!check_device(dev)) |
2398 | return 0; | |
e275a2a0 | 2399 | |
5abcdba4 JR |
2400 | devid = get_device_id(dev); |
2401 | iommu = amd_iommu_rlookup_table[devid]; | |
2402 | dev_data = get_dev_data(dev); | |
e275a2a0 JR |
2403 | |
2404 | switch (action) { | |
c1eee67b | 2405 | case BUS_NOTIFY_UNBOUND_DRIVER: |
657cbb6b JR |
2406 | |
2407 | domain = domain_for_device(dev); | |
2408 | ||
e275a2a0 JR |
2409 | if (!domain) |
2410 | goto out; | |
5abcdba4 | 2411 | if (dev_data->passthrough) |
a1ca331c | 2412 | break; |
15898bbc | 2413 | detach_device(dev); |
1ac4cbbc JR |
2414 | break; |
2415 | case BUS_NOTIFY_ADD_DEVICE: | |
657cbb6b JR |
2416 | |
2417 | iommu_init_device(dev); | |
2418 | ||
2c9195e9 JR |
2419 | /* |
2420 | * dev_data is still NULL and | |
2421 | * got initialized in iommu_init_device | |
2422 | */ | |
2423 | dev_data = get_dev_data(dev); | |
2424 | ||
2425 | if (iommu_pass_through || dev_data->iommu_v2) { | |
2426 | dev_data->passthrough = true; | |
2427 | attach_device(dev, pt_domain); | |
2428 | break; | |
2429 | } | |
2430 | ||
657cbb6b JR |
2431 | domain = domain_for_device(dev); |
2432 | ||
1ac4cbbc JR |
2433 | /* allocate a protection domain if a device is added */ |
2434 | dma_domain = find_protection_domain(devid); | |
c2a2876e JR |
2435 | if (!dma_domain) { |
2436 | dma_domain = dma_ops_domain_alloc(); | |
2437 | if (!dma_domain) | |
2438 | goto out; | |
2439 | dma_domain->target_dev = devid; | |
2440 | ||
2441 | spin_lock_irqsave(&iommu_pd_list_lock, flags); | |
2442 | list_add_tail(&dma_domain->list, &iommu_pd_list); | |
2443 | spin_unlock_irqrestore(&iommu_pd_list_lock, flags); | |
2444 | } | |
ac1534a5 | 2445 | |
2c9195e9 | 2446 | dev->archdata.dma_ops = &amd_iommu_dma_ops; |
ac1534a5 | 2447 | |
e275a2a0 | 2448 | break; |
657cbb6b JR |
2449 | case BUS_NOTIFY_DEL_DEVICE: |
2450 | ||
2451 | iommu_uninit_device(dev); | |
2452 | ||
e275a2a0 JR |
2453 | default: |
2454 | goto out; | |
2455 | } | |
2456 | ||
e275a2a0 JR |
2457 | iommu_completion_wait(iommu); |
2458 | ||
2459 | out: | |
2460 | return 0; | |
2461 | } | |
2462 | ||
b25ae679 | 2463 | static struct notifier_block device_nb = { |
e275a2a0 JR |
2464 | .notifier_call = device_change_notifier, |
2465 | }; | |
355bf553 | 2466 | |
8638c491 JR |
2467 | void amd_iommu_init_notifier(void) |
2468 | { | |
2469 | bus_register_notifier(&pci_bus_type, &device_nb); | |
2470 | } | |
2471 | ||
431b2a20 JR |
2472 | /***************************************************************************** |
2473 | * | |
2474 | * The next functions belong to the dma_ops mapping/unmapping code. | |
2475 | * | |
2476 | *****************************************************************************/ | |
2477 | ||
2478 | /* | |
2479 | * In the dma_ops path we only have the struct device. This function | |
2480 | * finds the corresponding IOMMU, the protection domain and the | |
2481 | * requestor id for a given device. | |
2482 | * If the device is not yet associated with a domain this is also done | |
2483 | * in this function. | |
2484 | */ | |
94f6d190 | 2485 | static struct protection_domain *get_domain(struct device *dev) |
b20ac0d4 | 2486 | { |
94f6d190 | 2487 | struct protection_domain *domain; |
b20ac0d4 | 2488 | struct dma_ops_domain *dma_dom; |
94f6d190 | 2489 | u16 devid = get_device_id(dev); |
b20ac0d4 | 2490 | |
f99c0f1c | 2491 | if (!check_device(dev)) |
94f6d190 | 2492 | return ERR_PTR(-EINVAL); |
b20ac0d4 | 2493 | |
94f6d190 JR |
2494 | domain = domain_for_device(dev); |
2495 | if (domain != NULL && !dma_ops_domain(domain)) | |
2496 | return ERR_PTR(-EBUSY); | |
f99c0f1c | 2497 | |
94f6d190 JR |
2498 | if (domain != NULL) |
2499 | return domain; | |
b20ac0d4 | 2500 | |
df805abb | 2501 | /* Device not bound yet - bind it */ |
94f6d190 | 2502 | dma_dom = find_protection_domain(devid); |
15898bbc | 2503 | if (!dma_dom) |
94f6d190 JR |
2504 | dma_dom = amd_iommu_rlookup_table[devid]->default_dom; |
2505 | attach_device(dev, &dma_dom->domain); | |
15898bbc | 2506 | DUMP_printk("Using protection domain %d for device %s\n", |
94f6d190 | 2507 | dma_dom->domain.id, dev_name(dev)); |
f91ba190 | 2508 | |
94f6d190 | 2509 | return &dma_dom->domain; |
b20ac0d4 JR |
2510 | } |
2511 | ||
04bfdd84 JR |
2512 | static void update_device_table(struct protection_domain *domain) |
2513 | { | |
492667da | 2514 | struct iommu_dev_data *dev_data; |
04bfdd84 | 2515 | |
ea61cddb JR |
2516 | list_for_each_entry(dev_data, &domain->dev_list, list) |
2517 | set_dte_entry(dev_data->devid, domain, dev_data->ats.enabled); | |
04bfdd84 JR |
2518 | } |
2519 | ||
2520 | static void update_domain(struct protection_domain *domain) | |
2521 | { | |
2522 | if (!domain->updated) | |
2523 | return; | |
2524 | ||
2525 | update_device_table(domain); | |
17b124bf JR |
2526 | |
2527 | domain_flush_devices(domain); | |
2528 | domain_flush_tlb_pde(domain); | |
04bfdd84 JR |
2529 | |
2530 | domain->updated = false; | |
2531 | } | |
2532 | ||
8bda3092 JR |
2533 | /* |
2534 | * This function fetches the PTE for a given address in the aperture | |
2535 | */ | |
2536 | static u64* dma_ops_get_pte(struct dma_ops_domain *dom, | |
2537 | unsigned long address) | |
2538 | { | |
384de729 | 2539 | struct aperture_range *aperture; |
8bda3092 JR |
2540 | u64 *pte, *pte_page; |
2541 | ||
384de729 JR |
2542 | aperture = dom->aperture[APERTURE_RANGE_INDEX(address)]; |
2543 | if (!aperture) | |
2544 | return NULL; | |
2545 | ||
2546 | pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)]; | |
8bda3092 | 2547 | if (!pte) { |
cbb9d729 | 2548 | pte = alloc_pte(&dom->domain, address, PAGE_SIZE, &pte_page, |
abdc5eb3 | 2549 | GFP_ATOMIC); |
384de729 JR |
2550 | aperture->pte_pages[APERTURE_PAGE_INDEX(address)] = pte_page; |
2551 | } else | |
8c8c143c | 2552 | pte += PM_LEVEL_INDEX(0, address); |
8bda3092 | 2553 | |
04bfdd84 | 2554 | update_domain(&dom->domain); |
8bda3092 JR |
2555 | |
2556 | return pte; | |
2557 | } | |
2558 | ||
431b2a20 JR |
2559 | /* |
2560 | * This is the generic map function. It maps one 4kb page at paddr to | |
2561 | * the given address in the DMA address space for the domain. | |
2562 | */ | |
680525e0 | 2563 | static dma_addr_t dma_ops_domain_map(struct dma_ops_domain *dom, |
cb76c322 JR |
2564 | unsigned long address, |
2565 | phys_addr_t paddr, | |
2566 | int direction) | |
2567 | { | |
2568 | u64 *pte, __pte; | |
2569 | ||
2570 | WARN_ON(address > dom->aperture_size); | |
2571 | ||
2572 | paddr &= PAGE_MASK; | |
2573 | ||
8bda3092 | 2574 | pte = dma_ops_get_pte(dom, address); |
53812c11 | 2575 | if (!pte) |
8fd524b3 | 2576 | return DMA_ERROR_CODE; |
cb76c322 JR |
2577 | |
2578 | __pte = paddr | IOMMU_PTE_P | IOMMU_PTE_FC; | |
2579 | ||
2580 | if (direction == DMA_TO_DEVICE) | |
2581 | __pte |= IOMMU_PTE_IR; | |
2582 | else if (direction == DMA_FROM_DEVICE) | |
2583 | __pte |= IOMMU_PTE_IW; | |
2584 | else if (direction == DMA_BIDIRECTIONAL) | |
2585 | __pte |= IOMMU_PTE_IR | IOMMU_PTE_IW; | |
2586 | ||
2587 | WARN_ON(*pte); | |
2588 | ||
2589 | *pte = __pte; | |
2590 | ||
2591 | return (dma_addr_t)address; | |
2592 | } | |
2593 | ||
431b2a20 JR |
2594 | /* |
2595 | * The generic unmapping function for on page in the DMA address space. | |
2596 | */ | |
680525e0 | 2597 | static void dma_ops_domain_unmap(struct dma_ops_domain *dom, |
cb76c322 JR |
2598 | unsigned long address) |
2599 | { | |
384de729 | 2600 | struct aperture_range *aperture; |
cb76c322 JR |
2601 | u64 *pte; |
2602 | ||
2603 | if (address >= dom->aperture_size) | |
2604 | return; | |
2605 | ||
384de729 JR |
2606 | aperture = dom->aperture[APERTURE_RANGE_INDEX(address)]; |
2607 | if (!aperture) | |
2608 | return; | |
2609 | ||
2610 | pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)]; | |
2611 | if (!pte) | |
2612 | return; | |
cb76c322 | 2613 | |
8c8c143c | 2614 | pte += PM_LEVEL_INDEX(0, address); |
cb76c322 JR |
2615 | |
2616 | WARN_ON(!*pte); | |
2617 | ||
2618 | *pte = 0ULL; | |
2619 | } | |
2620 | ||
431b2a20 JR |
2621 | /* |
2622 | * This function contains common code for mapping of a physically | |
24f81160 JR |
2623 | * contiguous memory region into DMA address space. It is used by all |
2624 | * mapping functions provided with this IOMMU driver. | |
431b2a20 JR |
2625 | * Must be called with the domain lock held. |
2626 | */ | |
cb76c322 | 2627 | static dma_addr_t __map_single(struct device *dev, |
cb76c322 JR |
2628 | struct dma_ops_domain *dma_dom, |
2629 | phys_addr_t paddr, | |
2630 | size_t size, | |
6d4f343f | 2631 | int dir, |
832a90c3 JR |
2632 | bool align, |
2633 | u64 dma_mask) | |
cb76c322 JR |
2634 | { |
2635 | dma_addr_t offset = paddr & ~PAGE_MASK; | |
53812c11 | 2636 | dma_addr_t address, start, ret; |
cb76c322 | 2637 | unsigned int pages; |
6d4f343f | 2638 | unsigned long align_mask = 0; |
cb76c322 JR |
2639 | int i; |
2640 | ||
e3c449f5 | 2641 | pages = iommu_num_pages(paddr, size, PAGE_SIZE); |
cb76c322 JR |
2642 | paddr &= PAGE_MASK; |
2643 | ||
8ecaf8f1 JR |
2644 | INC_STATS_COUNTER(total_map_requests); |
2645 | ||
c1858976 JR |
2646 | if (pages > 1) |
2647 | INC_STATS_COUNTER(cross_page); | |
2648 | ||
6d4f343f JR |
2649 | if (align) |
2650 | align_mask = (1UL << get_order(size)) - 1; | |
2651 | ||
11b83888 | 2652 | retry: |
832a90c3 JR |
2653 | address = dma_ops_alloc_addresses(dev, dma_dom, pages, align_mask, |
2654 | dma_mask); | |
8fd524b3 | 2655 | if (unlikely(address == DMA_ERROR_CODE)) { |
11b83888 JR |
2656 | /* |
2657 | * setting next_address here will let the address | |
2658 | * allocator only scan the new allocated range in the | |
2659 | * first run. This is a small optimization. | |
2660 | */ | |
2661 | dma_dom->next_address = dma_dom->aperture_size; | |
2662 | ||
576175c2 | 2663 | if (alloc_new_range(dma_dom, false, GFP_ATOMIC)) |
11b83888 JR |
2664 | goto out; |
2665 | ||
2666 | /* | |
af901ca1 | 2667 | * aperture was successfully enlarged by 128 MB, try |
11b83888 JR |
2668 | * allocation again |
2669 | */ | |
2670 | goto retry; | |
2671 | } | |
cb76c322 JR |
2672 | |
2673 | start = address; | |
2674 | for (i = 0; i < pages; ++i) { | |
680525e0 | 2675 | ret = dma_ops_domain_map(dma_dom, start, paddr, dir); |
8fd524b3 | 2676 | if (ret == DMA_ERROR_CODE) |
53812c11 JR |
2677 | goto out_unmap; |
2678 | ||
cb76c322 JR |
2679 | paddr += PAGE_SIZE; |
2680 | start += PAGE_SIZE; | |
2681 | } | |
2682 | address += offset; | |
2683 | ||
5774f7c5 JR |
2684 | ADD_STATS_COUNTER(alloced_io_mem, size); |
2685 | ||
afa9fdc2 | 2686 | if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) { |
17b124bf | 2687 | domain_flush_tlb(&dma_dom->domain); |
1c655773 | 2688 | dma_dom->need_flush = false; |
318afd41 | 2689 | } else if (unlikely(amd_iommu_np_cache)) |
17b124bf | 2690 | domain_flush_pages(&dma_dom->domain, address, size); |
270cab24 | 2691 | |
cb76c322 JR |
2692 | out: |
2693 | return address; | |
53812c11 JR |
2694 | |
2695 | out_unmap: | |
2696 | ||
2697 | for (--i; i >= 0; --i) { | |
2698 | start -= PAGE_SIZE; | |
680525e0 | 2699 | dma_ops_domain_unmap(dma_dom, start); |
53812c11 JR |
2700 | } |
2701 | ||
2702 | dma_ops_free_addresses(dma_dom, address, pages); | |
2703 | ||
8fd524b3 | 2704 | return DMA_ERROR_CODE; |
cb76c322 JR |
2705 | } |
2706 | ||
431b2a20 JR |
2707 | /* |
2708 | * Does the reverse of the __map_single function. Must be called with | |
2709 | * the domain lock held too | |
2710 | */ | |
cd8c82e8 | 2711 | static void __unmap_single(struct dma_ops_domain *dma_dom, |
cb76c322 JR |
2712 | dma_addr_t dma_addr, |
2713 | size_t size, | |
2714 | int dir) | |
2715 | { | |
04e0463e | 2716 | dma_addr_t flush_addr; |
cb76c322 JR |
2717 | dma_addr_t i, start; |
2718 | unsigned int pages; | |
2719 | ||
8fd524b3 | 2720 | if ((dma_addr == DMA_ERROR_CODE) || |
b8d9905d | 2721 | (dma_addr + size > dma_dom->aperture_size)) |
cb76c322 JR |
2722 | return; |
2723 | ||
04e0463e | 2724 | flush_addr = dma_addr; |
e3c449f5 | 2725 | pages = iommu_num_pages(dma_addr, size, PAGE_SIZE); |
cb76c322 JR |
2726 | dma_addr &= PAGE_MASK; |
2727 | start = dma_addr; | |
2728 | ||
2729 | for (i = 0; i < pages; ++i) { | |
680525e0 | 2730 | dma_ops_domain_unmap(dma_dom, start); |
cb76c322 JR |
2731 | start += PAGE_SIZE; |
2732 | } | |
2733 | ||
5774f7c5 JR |
2734 | SUB_STATS_COUNTER(alloced_io_mem, size); |
2735 | ||
cb76c322 | 2736 | dma_ops_free_addresses(dma_dom, dma_addr, pages); |
270cab24 | 2737 | |
80be308d | 2738 | if (amd_iommu_unmap_flush || dma_dom->need_flush) { |
17b124bf | 2739 | domain_flush_pages(&dma_dom->domain, flush_addr, size); |
80be308d JR |
2740 | dma_dom->need_flush = false; |
2741 | } | |
cb76c322 JR |
2742 | } |
2743 | ||
431b2a20 JR |
2744 | /* |
2745 | * The exported map_single function for dma_ops. | |
2746 | */ | |
51491367 FT |
2747 | static dma_addr_t map_page(struct device *dev, struct page *page, |
2748 | unsigned long offset, size_t size, | |
2749 | enum dma_data_direction dir, | |
2750 | struct dma_attrs *attrs) | |
4da70b9e JR |
2751 | { |
2752 | unsigned long flags; | |
4da70b9e | 2753 | struct protection_domain *domain; |
4da70b9e | 2754 | dma_addr_t addr; |
832a90c3 | 2755 | u64 dma_mask; |
51491367 | 2756 | phys_addr_t paddr = page_to_phys(page) + offset; |
4da70b9e | 2757 | |
0f2a86f2 JR |
2758 | INC_STATS_COUNTER(cnt_map_single); |
2759 | ||
94f6d190 JR |
2760 | domain = get_domain(dev); |
2761 | if (PTR_ERR(domain) == -EINVAL) | |
4da70b9e | 2762 | return (dma_addr_t)paddr; |
94f6d190 JR |
2763 | else if (IS_ERR(domain)) |
2764 | return DMA_ERROR_CODE; | |
4da70b9e | 2765 | |
f99c0f1c JR |
2766 | dma_mask = *dev->dma_mask; |
2767 | ||
4da70b9e | 2768 | spin_lock_irqsave(&domain->lock, flags); |
94f6d190 | 2769 | |
cd8c82e8 | 2770 | addr = __map_single(dev, domain->priv, paddr, size, dir, false, |
832a90c3 | 2771 | dma_mask); |
8fd524b3 | 2772 | if (addr == DMA_ERROR_CODE) |
4da70b9e JR |
2773 | goto out; |
2774 | ||
17b124bf | 2775 | domain_flush_complete(domain); |
4da70b9e JR |
2776 | |
2777 | out: | |
2778 | spin_unlock_irqrestore(&domain->lock, flags); | |
2779 | ||
2780 | return addr; | |
2781 | } | |
2782 | ||
431b2a20 JR |
2783 | /* |
2784 | * The exported unmap_single function for dma_ops. | |
2785 | */ | |
51491367 FT |
2786 | static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size, |
2787 | enum dma_data_direction dir, struct dma_attrs *attrs) | |
4da70b9e JR |
2788 | { |
2789 | unsigned long flags; | |
4da70b9e | 2790 | struct protection_domain *domain; |
4da70b9e | 2791 | |
146a6917 JR |
2792 | INC_STATS_COUNTER(cnt_unmap_single); |
2793 | ||
94f6d190 JR |
2794 | domain = get_domain(dev); |
2795 | if (IS_ERR(domain)) | |
5b28df6f JR |
2796 | return; |
2797 | ||
4da70b9e JR |
2798 | spin_lock_irqsave(&domain->lock, flags); |
2799 | ||
cd8c82e8 | 2800 | __unmap_single(domain->priv, dma_addr, size, dir); |
4da70b9e | 2801 | |
17b124bf | 2802 | domain_flush_complete(domain); |
4da70b9e JR |
2803 | |
2804 | spin_unlock_irqrestore(&domain->lock, flags); | |
2805 | } | |
2806 | ||
431b2a20 JR |
2807 | /* |
2808 | * The exported map_sg function for dma_ops (handles scatter-gather | |
2809 | * lists). | |
2810 | */ | |
65b050ad | 2811 | static int map_sg(struct device *dev, struct scatterlist *sglist, |
160c1d8e FT |
2812 | int nelems, enum dma_data_direction dir, |
2813 | struct dma_attrs *attrs) | |
65b050ad JR |
2814 | { |
2815 | unsigned long flags; | |
65b050ad | 2816 | struct protection_domain *domain; |
65b050ad JR |
2817 | int i; |
2818 | struct scatterlist *s; | |
2819 | phys_addr_t paddr; | |
2820 | int mapped_elems = 0; | |
832a90c3 | 2821 | u64 dma_mask; |
65b050ad | 2822 | |
d03f067a JR |
2823 | INC_STATS_COUNTER(cnt_map_sg); |
2824 | ||
94f6d190 | 2825 | domain = get_domain(dev); |
a0e191b2 | 2826 | if (IS_ERR(domain)) |
94f6d190 | 2827 | return 0; |
dbcc112e | 2828 | |
832a90c3 | 2829 | dma_mask = *dev->dma_mask; |
65b050ad | 2830 | |
65b050ad JR |
2831 | spin_lock_irqsave(&domain->lock, flags); |
2832 | ||
2833 | for_each_sg(sglist, s, nelems, i) { | |
2834 | paddr = sg_phys(s); | |
2835 | ||
cd8c82e8 | 2836 | s->dma_address = __map_single(dev, domain->priv, |
832a90c3 JR |
2837 | paddr, s->length, dir, false, |
2838 | dma_mask); | |
65b050ad JR |
2839 | |
2840 | if (s->dma_address) { | |
2841 | s->dma_length = s->length; | |
2842 | mapped_elems++; | |
2843 | } else | |
2844 | goto unmap; | |
65b050ad JR |
2845 | } |
2846 | ||
17b124bf | 2847 | domain_flush_complete(domain); |
65b050ad JR |
2848 | |
2849 | out: | |
2850 | spin_unlock_irqrestore(&domain->lock, flags); | |
2851 | ||
2852 | return mapped_elems; | |
2853 | unmap: | |
2854 | for_each_sg(sglist, s, mapped_elems, i) { | |
2855 | if (s->dma_address) | |
cd8c82e8 | 2856 | __unmap_single(domain->priv, s->dma_address, |
65b050ad JR |
2857 | s->dma_length, dir); |
2858 | s->dma_address = s->dma_length = 0; | |
2859 | } | |
2860 | ||
2861 | mapped_elems = 0; | |
2862 | ||
2863 | goto out; | |
2864 | } | |
2865 | ||
431b2a20 JR |
2866 | /* |
2867 | * The exported map_sg function for dma_ops (handles scatter-gather | |
2868 | * lists). | |
2869 | */ | |
65b050ad | 2870 | static void unmap_sg(struct device *dev, struct scatterlist *sglist, |
160c1d8e FT |
2871 | int nelems, enum dma_data_direction dir, |
2872 | struct dma_attrs *attrs) | |
65b050ad JR |
2873 | { |
2874 | unsigned long flags; | |
65b050ad JR |
2875 | struct protection_domain *domain; |
2876 | struct scatterlist *s; | |
65b050ad JR |
2877 | int i; |
2878 | ||
55877a6b JR |
2879 | INC_STATS_COUNTER(cnt_unmap_sg); |
2880 | ||
94f6d190 JR |
2881 | domain = get_domain(dev); |
2882 | if (IS_ERR(domain)) | |
5b28df6f JR |
2883 | return; |
2884 | ||
65b050ad JR |
2885 | spin_lock_irqsave(&domain->lock, flags); |
2886 | ||
2887 | for_each_sg(sglist, s, nelems, i) { | |
cd8c82e8 | 2888 | __unmap_single(domain->priv, s->dma_address, |
65b050ad | 2889 | s->dma_length, dir); |
65b050ad JR |
2890 | s->dma_address = s->dma_length = 0; |
2891 | } | |
2892 | ||
17b124bf | 2893 | domain_flush_complete(domain); |
65b050ad JR |
2894 | |
2895 | spin_unlock_irqrestore(&domain->lock, flags); | |
2896 | } | |
2897 | ||
431b2a20 JR |
2898 | /* |
2899 | * The exported alloc_coherent function for dma_ops. | |
2900 | */ | |
5d8b53cf | 2901 | static void *alloc_coherent(struct device *dev, size_t size, |
baa676fc AP |
2902 | dma_addr_t *dma_addr, gfp_t flag, |
2903 | struct dma_attrs *attrs) | |
5d8b53cf JR |
2904 | { |
2905 | unsigned long flags; | |
2906 | void *virt_addr; | |
5d8b53cf | 2907 | struct protection_domain *domain; |
5d8b53cf | 2908 | phys_addr_t paddr; |
832a90c3 | 2909 | u64 dma_mask = dev->coherent_dma_mask; |
5d8b53cf | 2910 | |
c8f0fb36 JR |
2911 | INC_STATS_COUNTER(cnt_alloc_coherent); |
2912 | ||
94f6d190 JR |
2913 | domain = get_domain(dev); |
2914 | if (PTR_ERR(domain) == -EINVAL) { | |
f99c0f1c JR |
2915 | virt_addr = (void *)__get_free_pages(flag, get_order(size)); |
2916 | *dma_addr = __pa(virt_addr); | |
2917 | return virt_addr; | |
94f6d190 JR |
2918 | } else if (IS_ERR(domain)) |
2919 | return NULL; | |
5d8b53cf | 2920 | |
f99c0f1c JR |
2921 | dma_mask = dev->coherent_dma_mask; |
2922 | flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32); | |
2923 | flag |= __GFP_ZERO; | |
5d8b53cf JR |
2924 | |
2925 | virt_addr = (void *)__get_free_pages(flag, get_order(size)); | |
2926 | if (!virt_addr) | |
b25ae679 | 2927 | return NULL; |
5d8b53cf | 2928 | |
5d8b53cf JR |
2929 | paddr = virt_to_phys(virt_addr); |
2930 | ||
832a90c3 JR |
2931 | if (!dma_mask) |
2932 | dma_mask = *dev->dma_mask; | |
2933 | ||
5d8b53cf JR |
2934 | spin_lock_irqsave(&domain->lock, flags); |
2935 | ||
cd8c82e8 | 2936 | *dma_addr = __map_single(dev, domain->priv, paddr, |
832a90c3 | 2937 | size, DMA_BIDIRECTIONAL, true, dma_mask); |
5d8b53cf | 2938 | |
8fd524b3 | 2939 | if (*dma_addr == DMA_ERROR_CODE) { |
367d04c4 | 2940 | spin_unlock_irqrestore(&domain->lock, flags); |
5b28df6f | 2941 | goto out_free; |
367d04c4 | 2942 | } |
5d8b53cf | 2943 | |
17b124bf | 2944 | domain_flush_complete(domain); |
5d8b53cf | 2945 | |
5d8b53cf JR |
2946 | spin_unlock_irqrestore(&domain->lock, flags); |
2947 | ||
2948 | return virt_addr; | |
5b28df6f JR |
2949 | |
2950 | out_free: | |
2951 | ||
2952 | free_pages((unsigned long)virt_addr, get_order(size)); | |
2953 | ||
2954 | return NULL; | |
5d8b53cf JR |
2955 | } |
2956 | ||
431b2a20 JR |
2957 | /* |
2958 | * The exported free_coherent function for dma_ops. | |
431b2a20 | 2959 | */ |
5d8b53cf | 2960 | static void free_coherent(struct device *dev, size_t size, |
baa676fc AP |
2961 | void *virt_addr, dma_addr_t dma_addr, |
2962 | struct dma_attrs *attrs) | |
5d8b53cf JR |
2963 | { |
2964 | unsigned long flags; | |
5d8b53cf | 2965 | struct protection_domain *domain; |
5d8b53cf | 2966 | |
5d31ee7e JR |
2967 | INC_STATS_COUNTER(cnt_free_coherent); |
2968 | ||
94f6d190 JR |
2969 | domain = get_domain(dev); |
2970 | if (IS_ERR(domain)) | |
5b28df6f JR |
2971 | goto free_mem; |
2972 | ||
5d8b53cf JR |
2973 | spin_lock_irqsave(&domain->lock, flags); |
2974 | ||
cd8c82e8 | 2975 | __unmap_single(domain->priv, dma_addr, size, DMA_BIDIRECTIONAL); |
5d8b53cf | 2976 | |
17b124bf | 2977 | domain_flush_complete(domain); |
5d8b53cf JR |
2978 | |
2979 | spin_unlock_irqrestore(&domain->lock, flags); | |
2980 | ||
2981 | free_mem: | |
2982 | free_pages((unsigned long)virt_addr, get_order(size)); | |
2983 | } | |
2984 | ||
b39ba6ad JR |
2985 | /* |
2986 | * This function is called by the DMA layer to find out if we can handle a | |
2987 | * particular device. It is part of the dma_ops. | |
2988 | */ | |
2989 | static int amd_iommu_dma_supported(struct device *dev, u64 mask) | |
2990 | { | |
420aef8a | 2991 | return check_device(dev); |
b39ba6ad JR |
2992 | } |
2993 | ||
c432f3df | 2994 | /* |
431b2a20 JR |
2995 | * The function for pre-allocating protection domains. |
2996 | * | |
c432f3df JR |
2997 | * If the driver core informs the DMA layer if a driver grabs a device |
2998 | * we don't need to preallocate the protection domains anymore. | |
2999 | * For now we have to. | |
3000 | */ | |
943bc7e1 | 3001 | static void __init prealloc_protection_domains(void) |
c432f3df | 3002 | { |
5abcdba4 | 3003 | struct iommu_dev_data *dev_data; |
c432f3df | 3004 | struct dma_ops_domain *dma_dom; |
5abcdba4 | 3005 | struct pci_dev *dev = NULL; |
98fc5a69 | 3006 | u16 devid; |
c432f3df | 3007 | |
d18c69d3 | 3008 | for_each_pci_dev(dev) { |
98fc5a69 JR |
3009 | |
3010 | /* Do we handle this device? */ | |
3011 | if (!check_device(&dev->dev)) | |
c432f3df | 3012 | continue; |
98fc5a69 | 3013 | |
5abcdba4 JR |
3014 | dev_data = get_dev_data(&dev->dev); |
3015 | if (!amd_iommu_force_isolation && dev_data->iommu_v2) { | |
3016 | /* Make sure passthrough domain is allocated */ | |
3017 | alloc_passthrough_domain(); | |
3018 | dev_data->passthrough = true; | |
3019 | attach_device(&dev->dev, pt_domain); | |
df805abb | 3020 | pr_info("AMD-Vi: Using passthrough domain for device %s\n", |
5abcdba4 JR |
3021 | dev_name(&dev->dev)); |
3022 | } | |
3023 | ||
98fc5a69 | 3024 | /* Is there already any domain for it? */ |
15898bbc | 3025 | if (domain_for_device(&dev->dev)) |
c432f3df | 3026 | continue; |
98fc5a69 JR |
3027 | |
3028 | devid = get_device_id(&dev->dev); | |
3029 | ||
87a64d52 | 3030 | dma_dom = dma_ops_domain_alloc(); |
c432f3df JR |
3031 | if (!dma_dom) |
3032 | continue; | |
3033 | init_unity_mappings_for_device(dma_dom, devid); | |
bd60b735 JR |
3034 | dma_dom->target_dev = devid; |
3035 | ||
15898bbc | 3036 | attach_device(&dev->dev, &dma_dom->domain); |
be831297 | 3037 | |
bd60b735 | 3038 | list_add_tail(&dma_dom->list, &iommu_pd_list); |
c432f3df JR |
3039 | } |
3040 | } | |
3041 | ||
160c1d8e | 3042 | static struct dma_map_ops amd_iommu_dma_ops = { |
baa676fc AP |
3043 | .alloc = alloc_coherent, |
3044 | .free = free_coherent, | |
51491367 FT |
3045 | .map_page = map_page, |
3046 | .unmap_page = unmap_page, | |
6631ee9d JR |
3047 | .map_sg = map_sg, |
3048 | .unmap_sg = unmap_sg, | |
b39ba6ad | 3049 | .dma_supported = amd_iommu_dma_supported, |
6631ee9d JR |
3050 | }; |
3051 | ||
27c2127a JR |
3052 | static unsigned device_dma_ops_init(void) |
3053 | { | |
5abcdba4 | 3054 | struct iommu_dev_data *dev_data; |
27c2127a JR |
3055 | struct pci_dev *pdev = NULL; |
3056 | unsigned unhandled = 0; | |
3057 | ||
3058 | for_each_pci_dev(pdev) { | |
3059 | if (!check_device(&pdev->dev)) { | |
af1be049 JR |
3060 | |
3061 | iommu_ignore_device(&pdev->dev); | |
3062 | ||
27c2127a JR |
3063 | unhandled += 1; |
3064 | continue; | |
3065 | } | |
3066 | ||
5abcdba4 JR |
3067 | dev_data = get_dev_data(&pdev->dev); |
3068 | ||
3069 | if (!dev_data->passthrough) | |
3070 | pdev->dev.archdata.dma_ops = &amd_iommu_dma_ops; | |
3071 | else | |
3072 | pdev->dev.archdata.dma_ops = &nommu_dma_ops; | |
27c2127a JR |
3073 | } |
3074 | ||
3075 | return unhandled; | |
3076 | } | |
3077 | ||
431b2a20 JR |
3078 | /* |
3079 | * The function which clues the AMD IOMMU driver into dma_ops. | |
3080 | */ | |
f5325094 JR |
3081 | |
3082 | void __init amd_iommu_init_api(void) | |
3083 | { | |
2cc21c42 | 3084 | bus_set_iommu(&pci_bus_type, &amd_iommu_ops); |
f5325094 JR |
3085 | } |
3086 | ||
6631ee9d JR |
3087 | int __init amd_iommu_init_dma_ops(void) |
3088 | { | |
3089 | struct amd_iommu *iommu; | |
27c2127a | 3090 | int ret, unhandled; |
6631ee9d | 3091 | |
431b2a20 JR |
3092 | /* |
3093 | * first allocate a default protection domain for every IOMMU we | |
3094 | * found in the system. Devices not assigned to any other | |
3095 | * protection domain will be assigned to the default one. | |
3096 | */ | |
3bd22172 | 3097 | for_each_iommu(iommu) { |
87a64d52 | 3098 | iommu->default_dom = dma_ops_domain_alloc(); |
6631ee9d JR |
3099 | if (iommu->default_dom == NULL) |
3100 | return -ENOMEM; | |
e2dc14a2 | 3101 | iommu->default_dom->domain.flags |= PD_DEFAULT_MASK; |
6631ee9d JR |
3102 | ret = iommu_init_unity_mappings(iommu); |
3103 | if (ret) | |
3104 | goto free_domains; | |
3105 | } | |
3106 | ||
431b2a20 | 3107 | /* |
8793abeb | 3108 | * Pre-allocate the protection domains for each device. |
431b2a20 | 3109 | */ |
8793abeb | 3110 | prealloc_protection_domains(); |
6631ee9d JR |
3111 | |
3112 | iommu_detected = 1; | |
75f1cdf1 | 3113 | swiotlb = 0; |
6631ee9d | 3114 | |
431b2a20 | 3115 | /* Make the driver finally visible to the drivers */ |
27c2127a JR |
3116 | unhandled = device_dma_ops_init(); |
3117 | if (unhandled && max_pfn > MAX_DMA32_PFN) { | |
3118 | /* There are unhandled devices - initialize swiotlb for them */ | |
3119 | swiotlb = 1; | |
3120 | } | |
6631ee9d | 3121 | |
7f26508b JR |
3122 | amd_iommu_stats_init(); |
3123 | ||
62410eeb JR |
3124 | if (amd_iommu_unmap_flush) |
3125 | pr_info("AMD-Vi: IO/TLB flush on unmap enabled\n"); | |
3126 | else | |
3127 | pr_info("AMD-Vi: Lazy IO/TLB flushing enabled\n"); | |
3128 | ||
6631ee9d JR |
3129 | return 0; |
3130 | ||
3131 | free_domains: | |
3132 | ||
3bd22172 | 3133 | for_each_iommu(iommu) { |
91457df7 | 3134 | dma_ops_domain_free(iommu->default_dom); |
6631ee9d JR |
3135 | } |
3136 | ||
3137 | return ret; | |
3138 | } | |
6d98cd80 JR |
3139 | |
3140 | /***************************************************************************** | |
3141 | * | |
3142 | * The following functions belong to the exported interface of AMD IOMMU | |
3143 | * | |
3144 | * This interface allows access to lower level functions of the IOMMU | |
3145 | * like protection domain handling and assignement of devices to domains | |
3146 | * which is not possible with the dma_ops interface. | |
3147 | * | |
3148 | *****************************************************************************/ | |
3149 | ||
6d98cd80 JR |
3150 | static void cleanup_domain(struct protection_domain *domain) |
3151 | { | |
492667da | 3152 | struct iommu_dev_data *dev_data, *next; |
6d98cd80 | 3153 | unsigned long flags; |
6d98cd80 JR |
3154 | |
3155 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
3156 | ||
492667da | 3157 | list_for_each_entry_safe(dev_data, next, &domain->dev_list, list) { |
ec9e79ef | 3158 | __detach_device(dev_data); |
492667da JR |
3159 | atomic_set(&dev_data->bind, 0); |
3160 | } | |
6d98cd80 JR |
3161 | |
3162 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
3163 | } | |
3164 | ||
2650815f JR |
3165 | static void protection_domain_free(struct protection_domain *domain) |
3166 | { | |
3167 | if (!domain) | |
3168 | return; | |
3169 | ||
aeb26f55 JR |
3170 | del_domain_from_list(domain); |
3171 | ||
2650815f JR |
3172 | if (domain->id) |
3173 | domain_id_free(domain->id); | |
3174 | ||
3175 | kfree(domain); | |
3176 | } | |
3177 | ||
3178 | static struct protection_domain *protection_domain_alloc(void) | |
c156e347 JR |
3179 | { |
3180 | struct protection_domain *domain; | |
3181 | ||
3182 | domain = kzalloc(sizeof(*domain), GFP_KERNEL); | |
3183 | if (!domain) | |
2650815f | 3184 | return NULL; |
c156e347 JR |
3185 | |
3186 | spin_lock_init(&domain->lock); | |
5d214fe6 | 3187 | mutex_init(&domain->api_lock); |
c156e347 JR |
3188 | domain->id = domain_id_alloc(); |
3189 | if (!domain->id) | |
2650815f | 3190 | goto out_err; |
7c392cbe | 3191 | INIT_LIST_HEAD(&domain->dev_list); |
2650815f | 3192 | |
aeb26f55 JR |
3193 | add_domain_to_list(domain); |
3194 | ||
2650815f JR |
3195 | return domain; |
3196 | ||
3197 | out_err: | |
3198 | kfree(domain); | |
3199 | ||
3200 | return NULL; | |
3201 | } | |
3202 | ||
5abcdba4 JR |
3203 | static int __init alloc_passthrough_domain(void) |
3204 | { | |
3205 | if (pt_domain != NULL) | |
3206 | return 0; | |
3207 | ||
3208 | /* allocate passthrough domain */ | |
3209 | pt_domain = protection_domain_alloc(); | |
3210 | if (!pt_domain) | |
3211 | return -ENOMEM; | |
3212 | ||
3213 | pt_domain->mode = PAGE_MODE_NONE; | |
3214 | ||
3215 | return 0; | |
3216 | } | |
2650815f JR |
3217 | static int amd_iommu_domain_init(struct iommu_domain *dom) |
3218 | { | |
3219 | struct protection_domain *domain; | |
3220 | ||
3221 | domain = protection_domain_alloc(); | |
3222 | if (!domain) | |
c156e347 | 3223 | goto out_free; |
2650815f JR |
3224 | |
3225 | domain->mode = PAGE_MODE_3_LEVEL; | |
c156e347 JR |
3226 | domain->pt_root = (void *)get_zeroed_page(GFP_KERNEL); |
3227 | if (!domain->pt_root) | |
3228 | goto out_free; | |
3229 | ||
f3572db8 JR |
3230 | domain->iommu_domain = dom; |
3231 | ||
c156e347 JR |
3232 | dom->priv = domain; |
3233 | ||
0ff64f80 JR |
3234 | dom->geometry.aperture_start = 0; |
3235 | dom->geometry.aperture_end = ~0ULL; | |
3236 | dom->geometry.force_aperture = true; | |
3237 | ||
c156e347 JR |
3238 | return 0; |
3239 | ||
3240 | out_free: | |
2650815f | 3241 | protection_domain_free(domain); |
c156e347 JR |
3242 | |
3243 | return -ENOMEM; | |
3244 | } | |
3245 | ||
98383fc3 JR |
3246 | static void amd_iommu_domain_destroy(struct iommu_domain *dom) |
3247 | { | |
3248 | struct protection_domain *domain = dom->priv; | |
3249 | ||
3250 | if (!domain) | |
3251 | return; | |
3252 | ||
3253 | if (domain->dev_cnt > 0) | |
3254 | cleanup_domain(domain); | |
3255 | ||
3256 | BUG_ON(domain->dev_cnt != 0); | |
3257 | ||
132bd68f JR |
3258 | if (domain->mode != PAGE_MODE_NONE) |
3259 | free_pagetable(domain); | |
98383fc3 | 3260 | |
52815b75 JR |
3261 | if (domain->flags & PD_IOMMUV2_MASK) |
3262 | free_gcr3_table(domain); | |
3263 | ||
8b408fe4 | 3264 | protection_domain_free(domain); |
98383fc3 JR |
3265 | |
3266 | dom->priv = NULL; | |
3267 | } | |
3268 | ||
684f2888 JR |
3269 | static void amd_iommu_detach_device(struct iommu_domain *dom, |
3270 | struct device *dev) | |
3271 | { | |
657cbb6b | 3272 | struct iommu_dev_data *dev_data = dev->archdata.iommu; |
684f2888 | 3273 | struct amd_iommu *iommu; |
684f2888 JR |
3274 | u16 devid; |
3275 | ||
98fc5a69 | 3276 | if (!check_device(dev)) |
684f2888 JR |
3277 | return; |
3278 | ||
98fc5a69 | 3279 | devid = get_device_id(dev); |
684f2888 | 3280 | |
657cbb6b | 3281 | if (dev_data->domain != NULL) |
15898bbc | 3282 | detach_device(dev); |
684f2888 JR |
3283 | |
3284 | iommu = amd_iommu_rlookup_table[devid]; | |
3285 | if (!iommu) | |
3286 | return; | |
3287 | ||
684f2888 JR |
3288 | iommu_completion_wait(iommu); |
3289 | } | |
3290 | ||
01106066 JR |
3291 | static int amd_iommu_attach_device(struct iommu_domain *dom, |
3292 | struct device *dev) | |
3293 | { | |
3294 | struct protection_domain *domain = dom->priv; | |
657cbb6b | 3295 | struct iommu_dev_data *dev_data; |
01106066 | 3296 | struct amd_iommu *iommu; |
15898bbc | 3297 | int ret; |
01106066 | 3298 | |
98fc5a69 | 3299 | if (!check_device(dev)) |
01106066 JR |
3300 | return -EINVAL; |
3301 | ||
657cbb6b JR |
3302 | dev_data = dev->archdata.iommu; |
3303 | ||
f62dda66 | 3304 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
01106066 JR |
3305 | if (!iommu) |
3306 | return -EINVAL; | |
3307 | ||
657cbb6b | 3308 | if (dev_data->domain) |
15898bbc | 3309 | detach_device(dev); |
01106066 | 3310 | |
15898bbc | 3311 | ret = attach_device(dev, domain); |
01106066 JR |
3312 | |
3313 | iommu_completion_wait(iommu); | |
3314 | ||
15898bbc | 3315 | return ret; |
01106066 JR |
3316 | } |
3317 | ||
468e2366 | 3318 | static int amd_iommu_map(struct iommu_domain *dom, unsigned long iova, |
5009065d | 3319 | phys_addr_t paddr, size_t page_size, int iommu_prot) |
c6229ca6 JR |
3320 | { |
3321 | struct protection_domain *domain = dom->priv; | |
c6229ca6 JR |
3322 | int prot = 0; |
3323 | int ret; | |
3324 | ||
132bd68f JR |
3325 | if (domain->mode == PAGE_MODE_NONE) |
3326 | return -EINVAL; | |
3327 | ||
c6229ca6 JR |
3328 | if (iommu_prot & IOMMU_READ) |
3329 | prot |= IOMMU_PROT_IR; | |
3330 | if (iommu_prot & IOMMU_WRITE) | |
3331 | prot |= IOMMU_PROT_IW; | |
3332 | ||
5d214fe6 | 3333 | mutex_lock(&domain->api_lock); |
795e74f7 | 3334 | ret = iommu_map_page(domain, iova, paddr, prot, page_size); |
5d214fe6 JR |
3335 | mutex_unlock(&domain->api_lock); |
3336 | ||
795e74f7 | 3337 | return ret; |
c6229ca6 JR |
3338 | } |
3339 | ||
5009065d OBC |
3340 | static size_t amd_iommu_unmap(struct iommu_domain *dom, unsigned long iova, |
3341 | size_t page_size) | |
eb74ff6c | 3342 | { |
eb74ff6c | 3343 | struct protection_domain *domain = dom->priv; |
5009065d | 3344 | size_t unmap_size; |
eb74ff6c | 3345 | |
132bd68f JR |
3346 | if (domain->mode == PAGE_MODE_NONE) |
3347 | return -EINVAL; | |
3348 | ||
5d214fe6 | 3349 | mutex_lock(&domain->api_lock); |
468e2366 | 3350 | unmap_size = iommu_unmap_page(domain, iova, page_size); |
795e74f7 | 3351 | mutex_unlock(&domain->api_lock); |
eb74ff6c | 3352 | |
17b124bf | 3353 | domain_flush_tlb_pde(domain); |
5d214fe6 | 3354 | |
5009065d | 3355 | return unmap_size; |
eb74ff6c JR |
3356 | } |
3357 | ||
645c4c8d | 3358 | static phys_addr_t amd_iommu_iova_to_phys(struct iommu_domain *dom, |
bb5547ac | 3359 | dma_addr_t iova) |
645c4c8d JR |
3360 | { |
3361 | struct protection_domain *domain = dom->priv; | |
f03152bb | 3362 | unsigned long offset_mask; |
645c4c8d | 3363 | phys_addr_t paddr; |
f03152bb | 3364 | u64 *pte, __pte; |
645c4c8d | 3365 | |
132bd68f JR |
3366 | if (domain->mode == PAGE_MODE_NONE) |
3367 | return iova; | |
3368 | ||
24cd7723 | 3369 | pte = fetch_pte(domain, iova); |
645c4c8d | 3370 | |
a6d41a40 | 3371 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) |
645c4c8d JR |
3372 | return 0; |
3373 | ||
f03152bb JR |
3374 | if (PM_PTE_LEVEL(*pte) == 0) |
3375 | offset_mask = PAGE_SIZE - 1; | |
3376 | else | |
3377 | offset_mask = PTE_PAGE_SIZE(*pte) - 1; | |
3378 | ||
3379 | __pte = *pte & PM_ADDR_MASK; | |
3380 | paddr = (__pte & ~offset_mask) | (iova & offset_mask); | |
645c4c8d JR |
3381 | |
3382 | return paddr; | |
3383 | } | |
3384 | ||
dbb9fd86 SY |
3385 | static int amd_iommu_domain_has_cap(struct iommu_domain *domain, |
3386 | unsigned long cap) | |
3387 | { | |
80a506b8 JR |
3388 | switch (cap) { |
3389 | case IOMMU_CAP_CACHE_COHERENCY: | |
3390 | return 1; | |
bdddadcb JR |
3391 | case IOMMU_CAP_INTR_REMAP: |
3392 | return irq_remapping_enabled; | |
80a506b8 JR |
3393 | } |
3394 | ||
dbb9fd86 SY |
3395 | return 0; |
3396 | } | |
3397 | ||
b22f6434 | 3398 | static const struct iommu_ops amd_iommu_ops = { |
26961efe JR |
3399 | .domain_init = amd_iommu_domain_init, |
3400 | .domain_destroy = amd_iommu_domain_destroy, | |
3401 | .attach_dev = amd_iommu_attach_device, | |
3402 | .detach_dev = amd_iommu_detach_device, | |
468e2366 JR |
3403 | .map = amd_iommu_map, |
3404 | .unmap = amd_iommu_unmap, | |
26961efe | 3405 | .iova_to_phys = amd_iommu_iova_to_phys, |
dbb9fd86 | 3406 | .domain_has_cap = amd_iommu_domain_has_cap, |
aa3de9c0 | 3407 | .pgsize_bitmap = AMD_IOMMU_PGSIZES, |
26961efe JR |
3408 | }; |
3409 | ||
0feae533 JR |
3410 | /***************************************************************************** |
3411 | * | |
3412 | * The next functions do a basic initialization of IOMMU for pass through | |
3413 | * mode | |
3414 | * | |
3415 | * In passthrough mode the IOMMU is initialized and enabled but not used for | |
3416 | * DMA-API translation. | |
3417 | * | |
3418 | *****************************************************************************/ | |
3419 | ||
3420 | int __init amd_iommu_init_passthrough(void) | |
3421 | { | |
5abcdba4 | 3422 | struct iommu_dev_data *dev_data; |
0feae533 | 3423 | struct pci_dev *dev = NULL; |
5abcdba4 | 3424 | int ret; |
0feae533 | 3425 | |
5abcdba4 JR |
3426 | ret = alloc_passthrough_domain(); |
3427 | if (ret) | |
3428 | return ret; | |
0feae533 | 3429 | |
6c54aabd | 3430 | for_each_pci_dev(dev) { |
98fc5a69 | 3431 | if (!check_device(&dev->dev)) |
0feae533 JR |
3432 | continue; |
3433 | ||
5abcdba4 JR |
3434 | dev_data = get_dev_data(&dev->dev); |
3435 | dev_data->passthrough = true; | |
3436 | ||
15898bbc | 3437 | attach_device(&dev->dev, pt_domain); |
0feae533 JR |
3438 | } |
3439 | ||
2655d7a2 JR |
3440 | amd_iommu_stats_init(); |
3441 | ||
0feae533 JR |
3442 | pr_info("AMD-Vi: Initialized for Passthrough Mode\n"); |
3443 | ||
3444 | return 0; | |
3445 | } | |
72e1dcc4 JR |
3446 | |
3447 | /* IOMMUv2 specific functions */ | |
3448 | int amd_iommu_register_ppr_notifier(struct notifier_block *nb) | |
3449 | { | |
3450 | return atomic_notifier_chain_register(&ppr_notifier, nb); | |
3451 | } | |
3452 | EXPORT_SYMBOL(amd_iommu_register_ppr_notifier); | |
3453 | ||
3454 | int amd_iommu_unregister_ppr_notifier(struct notifier_block *nb) | |
3455 | { | |
3456 | return atomic_notifier_chain_unregister(&ppr_notifier, nb); | |
3457 | } | |
3458 | EXPORT_SYMBOL(amd_iommu_unregister_ppr_notifier); | |
132bd68f JR |
3459 | |
3460 | void amd_iommu_domain_direct_map(struct iommu_domain *dom) | |
3461 | { | |
3462 | struct protection_domain *domain = dom->priv; | |
3463 | unsigned long flags; | |
3464 | ||
3465 | spin_lock_irqsave(&domain->lock, flags); | |
3466 | ||
3467 | /* Update data structure */ | |
3468 | domain->mode = PAGE_MODE_NONE; | |
3469 | domain->updated = true; | |
3470 | ||
3471 | /* Make changes visible to IOMMUs */ | |
3472 | update_domain(domain); | |
3473 | ||
3474 | /* Page-table is not visible to IOMMU anymore, so free it */ | |
3475 | free_pagetable(domain); | |
3476 | ||
3477 | spin_unlock_irqrestore(&domain->lock, flags); | |
3478 | } | |
3479 | EXPORT_SYMBOL(amd_iommu_domain_direct_map); | |
52815b75 JR |
3480 | |
3481 | int amd_iommu_domain_enable_v2(struct iommu_domain *dom, int pasids) | |
3482 | { | |
3483 | struct protection_domain *domain = dom->priv; | |
3484 | unsigned long flags; | |
3485 | int levels, ret; | |
3486 | ||
3487 | if (pasids <= 0 || pasids > (PASID_MASK + 1)) | |
3488 | return -EINVAL; | |
3489 | ||
3490 | /* Number of GCR3 table levels required */ | |
3491 | for (levels = 0; (pasids - 1) & ~0x1ff; pasids >>= 9) | |
3492 | levels += 1; | |
3493 | ||
3494 | if (levels > amd_iommu_max_glx_val) | |
3495 | return -EINVAL; | |
3496 | ||
3497 | spin_lock_irqsave(&domain->lock, flags); | |
3498 | ||
3499 | /* | |
3500 | * Save us all sanity checks whether devices already in the | |
3501 | * domain support IOMMUv2. Just force that the domain has no | |
3502 | * devices attached when it is switched into IOMMUv2 mode. | |
3503 | */ | |
3504 | ret = -EBUSY; | |
3505 | if (domain->dev_cnt > 0 || domain->flags & PD_IOMMUV2_MASK) | |
3506 | goto out; | |
3507 | ||
3508 | ret = -ENOMEM; | |
3509 | domain->gcr3_tbl = (void *)get_zeroed_page(GFP_ATOMIC); | |
3510 | if (domain->gcr3_tbl == NULL) | |
3511 | goto out; | |
3512 | ||
3513 | domain->glx = levels; | |
3514 | domain->flags |= PD_IOMMUV2_MASK; | |
3515 | domain->updated = true; | |
3516 | ||
3517 | update_domain(domain); | |
3518 | ||
3519 | ret = 0; | |
3520 | ||
3521 | out: | |
3522 | spin_unlock_irqrestore(&domain->lock, flags); | |
3523 | ||
3524 | return ret; | |
3525 | } | |
3526 | EXPORT_SYMBOL(amd_iommu_domain_enable_v2); | |
22e266c7 JR |
3527 | |
3528 | static int __flush_pasid(struct protection_domain *domain, int pasid, | |
3529 | u64 address, bool size) | |
3530 | { | |
3531 | struct iommu_dev_data *dev_data; | |
3532 | struct iommu_cmd cmd; | |
3533 | int i, ret; | |
3534 | ||
3535 | if (!(domain->flags & PD_IOMMUV2_MASK)) | |
3536 | return -EINVAL; | |
3537 | ||
3538 | build_inv_iommu_pasid(&cmd, domain->id, pasid, address, size); | |
3539 | ||
3540 | /* | |
3541 | * IOMMU TLB needs to be flushed before Device TLB to | |
3542 | * prevent device TLB refill from IOMMU TLB | |
3543 | */ | |
3544 | for (i = 0; i < amd_iommus_present; ++i) { | |
3545 | if (domain->dev_iommu[i] == 0) | |
3546 | continue; | |
3547 | ||
3548 | ret = iommu_queue_command(amd_iommus[i], &cmd); | |
3549 | if (ret != 0) | |
3550 | goto out; | |
3551 | } | |
3552 | ||
3553 | /* Wait until IOMMU TLB flushes are complete */ | |
3554 | domain_flush_complete(domain); | |
3555 | ||
3556 | /* Now flush device TLBs */ | |
3557 | list_for_each_entry(dev_data, &domain->dev_list, list) { | |
3558 | struct amd_iommu *iommu; | |
3559 | int qdep; | |
3560 | ||
3561 | BUG_ON(!dev_data->ats.enabled); | |
3562 | ||
3563 | qdep = dev_data->ats.qdep; | |
3564 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
3565 | ||
3566 | build_inv_iotlb_pasid(&cmd, dev_data->devid, pasid, | |
3567 | qdep, address, size); | |
3568 | ||
3569 | ret = iommu_queue_command(iommu, &cmd); | |
3570 | if (ret != 0) | |
3571 | goto out; | |
3572 | } | |
3573 | ||
3574 | /* Wait until all device TLBs are flushed */ | |
3575 | domain_flush_complete(domain); | |
3576 | ||
3577 | ret = 0; | |
3578 | ||
3579 | out: | |
3580 | ||
3581 | return ret; | |
3582 | } | |
3583 | ||
3584 | static int __amd_iommu_flush_page(struct protection_domain *domain, int pasid, | |
3585 | u64 address) | |
3586 | { | |
399be2f5 JR |
3587 | INC_STATS_COUNTER(invalidate_iotlb); |
3588 | ||
22e266c7 JR |
3589 | return __flush_pasid(domain, pasid, address, false); |
3590 | } | |
3591 | ||
3592 | int amd_iommu_flush_page(struct iommu_domain *dom, int pasid, | |
3593 | u64 address) | |
3594 | { | |
3595 | struct protection_domain *domain = dom->priv; | |
3596 | unsigned long flags; | |
3597 | int ret; | |
3598 | ||
3599 | spin_lock_irqsave(&domain->lock, flags); | |
3600 | ret = __amd_iommu_flush_page(domain, pasid, address); | |
3601 | spin_unlock_irqrestore(&domain->lock, flags); | |
3602 | ||
3603 | return ret; | |
3604 | } | |
3605 | EXPORT_SYMBOL(amd_iommu_flush_page); | |
3606 | ||
3607 | static int __amd_iommu_flush_tlb(struct protection_domain *domain, int pasid) | |
3608 | { | |
399be2f5 JR |
3609 | INC_STATS_COUNTER(invalidate_iotlb_all); |
3610 | ||
22e266c7 JR |
3611 | return __flush_pasid(domain, pasid, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, |
3612 | true); | |
3613 | } | |
3614 | ||
3615 | int amd_iommu_flush_tlb(struct iommu_domain *dom, int pasid) | |
3616 | { | |
3617 | struct protection_domain *domain = dom->priv; | |
3618 | unsigned long flags; | |
3619 | int ret; | |
3620 | ||
3621 | spin_lock_irqsave(&domain->lock, flags); | |
3622 | ret = __amd_iommu_flush_tlb(domain, pasid); | |
3623 | spin_unlock_irqrestore(&domain->lock, flags); | |
3624 | ||
3625 | return ret; | |
3626 | } | |
3627 | EXPORT_SYMBOL(amd_iommu_flush_tlb); | |
3628 | ||
b16137b1 JR |
3629 | static u64 *__get_gcr3_pte(u64 *root, int level, int pasid, bool alloc) |
3630 | { | |
3631 | int index; | |
3632 | u64 *pte; | |
3633 | ||
3634 | while (true) { | |
3635 | ||
3636 | index = (pasid >> (9 * level)) & 0x1ff; | |
3637 | pte = &root[index]; | |
3638 | ||
3639 | if (level == 0) | |
3640 | break; | |
3641 | ||
3642 | if (!(*pte & GCR3_VALID)) { | |
3643 | if (!alloc) | |
3644 | return NULL; | |
3645 | ||
3646 | root = (void *)get_zeroed_page(GFP_ATOMIC); | |
3647 | if (root == NULL) | |
3648 | return NULL; | |
3649 | ||
3650 | *pte = __pa(root) | GCR3_VALID; | |
3651 | } | |
3652 | ||
3653 | root = __va(*pte & PAGE_MASK); | |
3654 | ||
3655 | level -= 1; | |
3656 | } | |
3657 | ||
3658 | return pte; | |
3659 | } | |
3660 | ||
3661 | static int __set_gcr3(struct protection_domain *domain, int pasid, | |
3662 | unsigned long cr3) | |
3663 | { | |
3664 | u64 *pte; | |
3665 | ||
3666 | if (domain->mode != PAGE_MODE_NONE) | |
3667 | return -EINVAL; | |
3668 | ||
3669 | pte = __get_gcr3_pte(domain->gcr3_tbl, domain->glx, pasid, true); | |
3670 | if (pte == NULL) | |
3671 | return -ENOMEM; | |
3672 | ||
3673 | *pte = (cr3 & PAGE_MASK) | GCR3_VALID; | |
3674 | ||
3675 | return __amd_iommu_flush_tlb(domain, pasid); | |
3676 | } | |
3677 | ||
3678 | static int __clear_gcr3(struct protection_domain *domain, int pasid) | |
3679 | { | |
3680 | u64 *pte; | |
3681 | ||
3682 | if (domain->mode != PAGE_MODE_NONE) | |
3683 | return -EINVAL; | |
3684 | ||
3685 | pte = __get_gcr3_pte(domain->gcr3_tbl, domain->glx, pasid, false); | |
3686 | if (pte == NULL) | |
3687 | return 0; | |
3688 | ||
3689 | *pte = 0; | |
3690 | ||
3691 | return __amd_iommu_flush_tlb(domain, pasid); | |
3692 | } | |
3693 | ||
3694 | int amd_iommu_domain_set_gcr3(struct iommu_domain *dom, int pasid, | |
3695 | unsigned long cr3) | |
3696 | { | |
3697 | struct protection_domain *domain = dom->priv; | |
3698 | unsigned long flags; | |
3699 | int ret; | |
3700 | ||
3701 | spin_lock_irqsave(&domain->lock, flags); | |
3702 | ret = __set_gcr3(domain, pasid, cr3); | |
3703 | spin_unlock_irqrestore(&domain->lock, flags); | |
3704 | ||
3705 | return ret; | |
3706 | } | |
3707 | EXPORT_SYMBOL(amd_iommu_domain_set_gcr3); | |
3708 | ||
3709 | int amd_iommu_domain_clear_gcr3(struct iommu_domain *dom, int pasid) | |
3710 | { | |
3711 | struct protection_domain *domain = dom->priv; | |
3712 | unsigned long flags; | |
3713 | int ret; | |
3714 | ||
3715 | spin_lock_irqsave(&domain->lock, flags); | |
3716 | ret = __clear_gcr3(domain, pasid); | |
3717 | spin_unlock_irqrestore(&domain->lock, flags); | |
3718 | ||
3719 | return ret; | |
3720 | } | |
3721 | EXPORT_SYMBOL(amd_iommu_domain_clear_gcr3); | |
c99afa25 JR |
3722 | |
3723 | int amd_iommu_complete_ppr(struct pci_dev *pdev, int pasid, | |
3724 | int status, int tag) | |
3725 | { | |
3726 | struct iommu_dev_data *dev_data; | |
3727 | struct amd_iommu *iommu; | |
3728 | struct iommu_cmd cmd; | |
3729 | ||
399be2f5 JR |
3730 | INC_STATS_COUNTER(complete_ppr); |
3731 | ||
c99afa25 JR |
3732 | dev_data = get_dev_data(&pdev->dev); |
3733 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
3734 | ||
3735 | build_complete_ppr(&cmd, dev_data->devid, pasid, status, | |
3736 | tag, dev_data->pri_tlp); | |
3737 | ||
3738 | return iommu_queue_command(iommu, &cmd); | |
3739 | } | |
3740 | EXPORT_SYMBOL(amd_iommu_complete_ppr); | |
f3572db8 JR |
3741 | |
3742 | struct iommu_domain *amd_iommu_get_v2_domain(struct pci_dev *pdev) | |
3743 | { | |
3744 | struct protection_domain *domain; | |
3745 | ||
3746 | domain = get_domain(&pdev->dev); | |
3747 | if (IS_ERR(domain)) | |
3748 | return NULL; | |
3749 | ||
3750 | /* Only return IOMMUv2 domains */ | |
3751 | if (!(domain->flags & PD_IOMMUV2_MASK)) | |
3752 | return NULL; | |
3753 | ||
3754 | return domain->iommu_domain; | |
3755 | } | |
3756 | EXPORT_SYMBOL(amd_iommu_get_v2_domain); | |
6a113ddc JR |
3757 | |
3758 | void amd_iommu_enable_device_erratum(struct pci_dev *pdev, u32 erratum) | |
3759 | { | |
3760 | struct iommu_dev_data *dev_data; | |
3761 | ||
3762 | if (!amd_iommu_v2_supported()) | |
3763 | return; | |
3764 | ||
3765 | dev_data = get_dev_data(&pdev->dev); | |
3766 | dev_data->errata |= (1 << erratum); | |
3767 | } | |
3768 | EXPORT_SYMBOL(amd_iommu_enable_device_erratum); | |
52efdb89 JR |
3769 | |
3770 | int amd_iommu_device_info(struct pci_dev *pdev, | |
3771 | struct amd_iommu_device_info *info) | |
3772 | { | |
3773 | int max_pasids; | |
3774 | int pos; | |
3775 | ||
3776 | if (pdev == NULL || info == NULL) | |
3777 | return -EINVAL; | |
3778 | ||
3779 | if (!amd_iommu_v2_supported()) | |
3780 | return -EINVAL; | |
3781 | ||
3782 | memset(info, 0, sizeof(*info)); | |
3783 | ||
3784 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ATS); | |
3785 | if (pos) | |
3786 | info->flags |= AMD_IOMMU_DEVICE_FLAG_ATS_SUP; | |
3787 | ||
3788 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PRI); | |
3789 | if (pos) | |
3790 | info->flags |= AMD_IOMMU_DEVICE_FLAG_PRI_SUP; | |
3791 | ||
3792 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PASID); | |
3793 | if (pos) { | |
3794 | int features; | |
3795 | ||
3796 | max_pasids = 1 << (9 * (amd_iommu_max_glx_val + 1)); | |
3797 | max_pasids = min(max_pasids, (1 << 20)); | |
3798 | ||
3799 | info->flags |= AMD_IOMMU_DEVICE_FLAG_PASID_SUP; | |
3800 | info->max_pasids = min(pci_max_pasids(pdev), max_pasids); | |
3801 | ||
3802 | features = pci_pasid_features(pdev); | |
3803 | if (features & PCI_PASID_CAP_EXEC) | |
3804 | info->flags |= AMD_IOMMU_DEVICE_FLAG_EXEC_SUP; | |
3805 | if (features & PCI_PASID_CAP_PRIV) | |
3806 | info->flags |= AMD_IOMMU_DEVICE_FLAG_PRIV_SUP; | |
3807 | } | |
3808 | ||
3809 | return 0; | |
3810 | } | |
3811 | EXPORT_SYMBOL(amd_iommu_device_info); | |
2b324506 JR |
3812 | |
3813 | #ifdef CONFIG_IRQ_REMAP | |
3814 | ||
3815 | /***************************************************************************** | |
3816 | * | |
3817 | * Interrupt Remapping Implementation | |
3818 | * | |
3819 | *****************************************************************************/ | |
3820 | ||
3821 | union irte { | |
3822 | u32 val; | |
3823 | struct { | |
3824 | u32 valid : 1, | |
3825 | no_fault : 1, | |
3826 | int_type : 3, | |
3827 | rq_eoi : 1, | |
3828 | dm : 1, | |
3829 | rsvd_1 : 1, | |
3830 | destination : 8, | |
3831 | vector : 8, | |
3832 | rsvd_2 : 8; | |
3833 | } fields; | |
3834 | }; | |
3835 | ||
3836 | #define DTE_IRQ_PHYS_ADDR_MASK (((1ULL << 45)-1) << 6) | |
3837 | #define DTE_IRQ_REMAP_INTCTL (2ULL << 60) | |
3838 | #define DTE_IRQ_TABLE_LEN (8ULL << 1) | |
3839 | #define DTE_IRQ_REMAP_ENABLE 1ULL | |
3840 | ||
3841 | static void set_dte_irq_entry(u16 devid, struct irq_remap_table *table) | |
3842 | { | |
3843 | u64 dte; | |
3844 | ||
3845 | dte = amd_iommu_dev_table[devid].data[2]; | |
3846 | dte &= ~DTE_IRQ_PHYS_ADDR_MASK; | |
3847 | dte |= virt_to_phys(table->table); | |
3848 | dte |= DTE_IRQ_REMAP_INTCTL; | |
3849 | dte |= DTE_IRQ_TABLE_LEN; | |
3850 | dte |= DTE_IRQ_REMAP_ENABLE; | |
3851 | ||
3852 | amd_iommu_dev_table[devid].data[2] = dte; | |
3853 | } | |
3854 | ||
3855 | #define IRTE_ALLOCATED (~1U) | |
3856 | ||
3857 | static struct irq_remap_table *get_irq_table(u16 devid, bool ioapic) | |
3858 | { | |
3859 | struct irq_remap_table *table = NULL; | |
3860 | struct amd_iommu *iommu; | |
3861 | unsigned long flags; | |
3862 | u16 alias; | |
3863 | ||
3864 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
3865 | ||
3866 | iommu = amd_iommu_rlookup_table[devid]; | |
3867 | if (!iommu) | |
3868 | goto out_unlock; | |
3869 | ||
3870 | table = irq_lookup_table[devid]; | |
3871 | if (table) | |
3872 | goto out; | |
3873 | ||
3874 | alias = amd_iommu_alias_table[devid]; | |
3875 | table = irq_lookup_table[alias]; | |
3876 | if (table) { | |
3877 | irq_lookup_table[devid] = table; | |
3878 | set_dte_irq_entry(devid, table); | |
3879 | iommu_flush_dte(iommu, devid); | |
3880 | goto out; | |
3881 | } | |
3882 | ||
3883 | /* Nothing there yet, allocate new irq remapping table */ | |
3884 | table = kzalloc(sizeof(*table), GFP_ATOMIC); | |
3885 | if (!table) | |
3886 | goto out; | |
3887 | ||
197887f0 JR |
3888 | /* Initialize table spin-lock */ |
3889 | spin_lock_init(&table->lock); | |
3890 | ||
2b324506 JR |
3891 | if (ioapic) |
3892 | /* Keep the first 32 indexes free for IOAPIC interrupts */ | |
3893 | table->min_index = 32; | |
3894 | ||
3895 | table->table = kmem_cache_alloc(amd_iommu_irq_cache, GFP_ATOMIC); | |
3896 | if (!table->table) { | |
3897 | kfree(table); | |
821f0f68 | 3898 | table = NULL; |
2b324506 JR |
3899 | goto out; |
3900 | } | |
3901 | ||
3902 | memset(table->table, 0, MAX_IRQS_PER_TABLE * sizeof(u32)); | |
3903 | ||
3904 | if (ioapic) { | |
3905 | int i; | |
3906 | ||
3907 | for (i = 0; i < 32; ++i) | |
3908 | table->table[i] = IRTE_ALLOCATED; | |
3909 | } | |
3910 | ||
3911 | irq_lookup_table[devid] = table; | |
3912 | set_dte_irq_entry(devid, table); | |
3913 | iommu_flush_dte(iommu, devid); | |
3914 | if (devid != alias) { | |
3915 | irq_lookup_table[alias] = table; | |
e028a9e6 | 3916 | set_dte_irq_entry(alias, table); |
2b324506 JR |
3917 | iommu_flush_dte(iommu, alias); |
3918 | } | |
3919 | ||
3920 | out: | |
3921 | iommu_completion_wait(iommu); | |
3922 | ||
3923 | out_unlock: | |
3924 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
3925 | ||
3926 | return table; | |
3927 | } | |
3928 | ||
3929 | static int alloc_irq_index(struct irq_cfg *cfg, u16 devid, int count) | |
3930 | { | |
3931 | struct irq_remap_table *table; | |
3932 | unsigned long flags; | |
3933 | int index, c; | |
3934 | ||
3935 | table = get_irq_table(devid, false); | |
3936 | if (!table) | |
3937 | return -ENODEV; | |
3938 | ||
3939 | spin_lock_irqsave(&table->lock, flags); | |
3940 | ||
3941 | /* Scan table for free entries */ | |
3942 | for (c = 0, index = table->min_index; | |
3943 | index < MAX_IRQS_PER_TABLE; | |
3944 | ++index) { | |
3945 | if (table->table[index] == 0) | |
3946 | c += 1; | |
3947 | else | |
3948 | c = 0; | |
3949 | ||
3950 | if (c == count) { | |
0dfedd61 | 3951 | struct irq_2_irte *irte_info; |
2b324506 JR |
3952 | |
3953 | for (; c != 0; --c) | |
3954 | table->table[index - c + 1] = IRTE_ALLOCATED; | |
3955 | ||
3956 | index -= count - 1; | |
3957 | ||
9b1b0e42 | 3958 | cfg->remapped = 1; |
0dfedd61 JR |
3959 | irte_info = &cfg->irq_2_irte; |
3960 | irte_info->devid = devid; | |
3961 | irte_info->index = index; | |
2b324506 JR |
3962 | |
3963 | goto out; | |
3964 | } | |
3965 | } | |
3966 | ||
3967 | index = -ENOSPC; | |
3968 | ||
3969 | out: | |
3970 | spin_unlock_irqrestore(&table->lock, flags); | |
3971 | ||
3972 | return index; | |
3973 | } | |
3974 | ||
3975 | static int get_irte(u16 devid, int index, union irte *irte) | |
3976 | { | |
3977 | struct irq_remap_table *table; | |
3978 | unsigned long flags; | |
3979 | ||
3980 | table = get_irq_table(devid, false); | |
3981 | if (!table) | |
3982 | return -ENOMEM; | |
3983 | ||
3984 | spin_lock_irqsave(&table->lock, flags); | |
3985 | irte->val = table->table[index]; | |
3986 | spin_unlock_irqrestore(&table->lock, flags); | |
3987 | ||
3988 | return 0; | |
3989 | } | |
3990 | ||
3991 | static int modify_irte(u16 devid, int index, union irte irte) | |
3992 | { | |
3993 | struct irq_remap_table *table; | |
3994 | struct amd_iommu *iommu; | |
3995 | unsigned long flags; | |
3996 | ||
3997 | iommu = amd_iommu_rlookup_table[devid]; | |
3998 | if (iommu == NULL) | |
3999 | return -EINVAL; | |
4000 | ||
4001 | table = get_irq_table(devid, false); | |
4002 | if (!table) | |
4003 | return -ENOMEM; | |
4004 | ||
4005 | spin_lock_irqsave(&table->lock, flags); | |
4006 | table->table[index] = irte.val; | |
4007 | spin_unlock_irqrestore(&table->lock, flags); | |
4008 | ||
4009 | iommu_flush_irt(iommu, devid); | |
4010 | iommu_completion_wait(iommu); | |
4011 | ||
4012 | return 0; | |
4013 | } | |
4014 | ||
4015 | static void free_irte(u16 devid, int index) | |
4016 | { | |
4017 | struct irq_remap_table *table; | |
4018 | struct amd_iommu *iommu; | |
4019 | unsigned long flags; | |
4020 | ||
4021 | iommu = amd_iommu_rlookup_table[devid]; | |
4022 | if (iommu == NULL) | |
4023 | return; | |
4024 | ||
4025 | table = get_irq_table(devid, false); | |
4026 | if (!table) | |
4027 | return; | |
4028 | ||
4029 | spin_lock_irqsave(&table->lock, flags); | |
4030 | table->table[index] = 0; | |
4031 | spin_unlock_irqrestore(&table->lock, flags); | |
4032 | ||
4033 | iommu_flush_irt(iommu, devid); | |
4034 | iommu_completion_wait(iommu); | |
4035 | } | |
4036 | ||
5527de74 JR |
4037 | static int setup_ioapic_entry(int irq, struct IO_APIC_route_entry *entry, |
4038 | unsigned int destination, int vector, | |
4039 | struct io_apic_irq_attr *attr) | |
4040 | { | |
4041 | struct irq_remap_table *table; | |
0dfedd61 | 4042 | struct irq_2_irte *irte_info; |
5527de74 JR |
4043 | struct irq_cfg *cfg; |
4044 | union irte irte; | |
4045 | int ioapic_id; | |
4046 | int index; | |
4047 | int devid; | |
4048 | int ret; | |
4049 | ||
4050 | cfg = irq_get_chip_data(irq); | |
4051 | if (!cfg) | |
4052 | return -EINVAL; | |
4053 | ||
0dfedd61 | 4054 | irte_info = &cfg->irq_2_irte; |
5527de74 JR |
4055 | ioapic_id = mpc_ioapic_id(attr->ioapic); |
4056 | devid = get_ioapic_devid(ioapic_id); | |
4057 | ||
4058 | if (devid < 0) | |
4059 | return devid; | |
4060 | ||
4061 | table = get_irq_table(devid, true); | |
4062 | if (table == NULL) | |
4063 | return -ENOMEM; | |
4064 | ||
4065 | index = attr->ioapic_pin; | |
4066 | ||
4067 | /* Setup IRQ remapping info */ | |
9b1b0e42 | 4068 | cfg->remapped = 1; |
0dfedd61 JR |
4069 | irte_info->devid = devid; |
4070 | irte_info->index = index; | |
5527de74 JR |
4071 | |
4072 | /* Setup IRTE for IOMMU */ | |
4073 | irte.val = 0; | |
4074 | irte.fields.vector = vector; | |
4075 | irte.fields.int_type = apic->irq_delivery_mode; | |
4076 | irte.fields.destination = destination; | |
4077 | irte.fields.dm = apic->irq_dest_mode; | |
4078 | irte.fields.valid = 1; | |
4079 | ||
4080 | ret = modify_irte(devid, index, irte); | |
4081 | if (ret) | |
4082 | return ret; | |
4083 | ||
4084 | /* Setup IOAPIC entry */ | |
4085 | memset(entry, 0, sizeof(*entry)); | |
4086 | ||
4087 | entry->vector = index; | |
4088 | entry->mask = 0; | |
4089 | entry->trigger = attr->trigger; | |
4090 | entry->polarity = attr->polarity; | |
4091 | ||
4092 | /* | |
4093 | * Mask level triggered irqs. | |
5527de74 JR |
4094 | */ |
4095 | if (attr->trigger) | |
4096 | entry->mask = 1; | |
4097 | ||
4098 | return 0; | |
4099 | } | |
4100 | ||
4101 | static int set_affinity(struct irq_data *data, const struct cpumask *mask, | |
4102 | bool force) | |
4103 | { | |
0dfedd61 | 4104 | struct irq_2_irte *irte_info; |
5527de74 JR |
4105 | unsigned int dest, irq; |
4106 | struct irq_cfg *cfg; | |
4107 | union irte irte; | |
4108 | int err; | |
4109 | ||
4110 | if (!config_enabled(CONFIG_SMP)) | |
4111 | return -1; | |
4112 | ||
4113 | cfg = data->chip_data; | |
4114 | irq = data->irq; | |
0dfedd61 | 4115 | irte_info = &cfg->irq_2_irte; |
5527de74 JR |
4116 | |
4117 | if (!cpumask_intersects(mask, cpu_online_mask)) | |
4118 | return -EINVAL; | |
4119 | ||
0dfedd61 | 4120 | if (get_irte(irte_info->devid, irte_info->index, &irte)) |
5527de74 JR |
4121 | return -EBUSY; |
4122 | ||
4123 | if (assign_irq_vector(irq, cfg, mask)) | |
4124 | return -EBUSY; | |
4125 | ||
4126 | err = apic->cpu_mask_to_apicid_and(cfg->domain, mask, &dest); | |
4127 | if (err) { | |
4128 | if (assign_irq_vector(irq, cfg, data->affinity)) | |
4129 | pr_err("AMD-Vi: Failed to recover vector for irq %d\n", irq); | |
4130 | return err; | |
4131 | } | |
4132 | ||
4133 | irte.fields.vector = cfg->vector; | |
4134 | irte.fields.destination = dest; | |
4135 | ||
0dfedd61 | 4136 | modify_irte(irte_info->devid, irte_info->index, irte); |
5527de74 JR |
4137 | |
4138 | if (cfg->move_in_progress) | |
4139 | send_cleanup_vector(cfg); | |
4140 | ||
4141 | cpumask_copy(data->affinity, mask); | |
4142 | ||
4143 | return 0; | |
4144 | } | |
4145 | ||
4146 | static int free_irq(int irq) | |
4147 | { | |
0dfedd61 | 4148 | struct irq_2_irte *irte_info; |
5527de74 JR |
4149 | struct irq_cfg *cfg; |
4150 | ||
4151 | cfg = irq_get_chip_data(irq); | |
4152 | if (!cfg) | |
4153 | return -EINVAL; | |
4154 | ||
0dfedd61 | 4155 | irte_info = &cfg->irq_2_irte; |
5527de74 | 4156 | |
0dfedd61 | 4157 | free_irte(irte_info->devid, irte_info->index); |
5527de74 JR |
4158 | |
4159 | return 0; | |
4160 | } | |
4161 | ||
0b4d48cb JR |
4162 | static void compose_msi_msg(struct pci_dev *pdev, |
4163 | unsigned int irq, unsigned int dest, | |
4164 | struct msi_msg *msg, u8 hpet_id) | |
4165 | { | |
0dfedd61 | 4166 | struct irq_2_irte *irte_info; |
0b4d48cb JR |
4167 | struct irq_cfg *cfg; |
4168 | union irte irte; | |
4169 | ||
4170 | cfg = irq_get_chip_data(irq); | |
4171 | if (!cfg) | |
4172 | return; | |
4173 | ||
0dfedd61 | 4174 | irte_info = &cfg->irq_2_irte; |
0b4d48cb JR |
4175 | |
4176 | irte.val = 0; | |
4177 | irte.fields.vector = cfg->vector; | |
4178 | irte.fields.int_type = apic->irq_delivery_mode; | |
4179 | irte.fields.destination = dest; | |
4180 | irte.fields.dm = apic->irq_dest_mode; | |
4181 | irte.fields.valid = 1; | |
4182 | ||
0dfedd61 | 4183 | modify_irte(irte_info->devid, irte_info->index, irte); |
0b4d48cb JR |
4184 | |
4185 | msg->address_hi = MSI_ADDR_BASE_HI; | |
4186 | msg->address_lo = MSI_ADDR_BASE_LO; | |
0dfedd61 | 4187 | msg->data = irte_info->index; |
0b4d48cb JR |
4188 | } |
4189 | ||
4190 | static int msi_alloc_irq(struct pci_dev *pdev, int irq, int nvec) | |
4191 | { | |
4192 | struct irq_cfg *cfg; | |
4193 | int index; | |
4194 | u16 devid; | |
4195 | ||
4196 | if (!pdev) | |
4197 | return -EINVAL; | |
4198 | ||
4199 | cfg = irq_get_chip_data(irq); | |
4200 | if (!cfg) | |
4201 | return -EINVAL; | |
4202 | ||
4203 | devid = get_device_id(&pdev->dev); | |
4204 | index = alloc_irq_index(cfg, devid, nvec); | |
4205 | ||
4206 | return index < 0 ? MAX_IRQS_PER_TABLE : index; | |
4207 | } | |
4208 | ||
4209 | static int msi_setup_irq(struct pci_dev *pdev, unsigned int irq, | |
4210 | int index, int offset) | |
4211 | { | |
0dfedd61 | 4212 | struct irq_2_irte *irte_info; |
0b4d48cb JR |
4213 | struct irq_cfg *cfg; |
4214 | u16 devid; | |
4215 | ||
4216 | if (!pdev) | |
4217 | return -EINVAL; | |
4218 | ||
4219 | cfg = irq_get_chip_data(irq); | |
4220 | if (!cfg) | |
4221 | return -EINVAL; | |
4222 | ||
4223 | if (index >= MAX_IRQS_PER_TABLE) | |
4224 | return 0; | |
4225 | ||
4226 | devid = get_device_id(&pdev->dev); | |
0dfedd61 | 4227 | irte_info = &cfg->irq_2_irte; |
0b4d48cb | 4228 | |
9b1b0e42 | 4229 | cfg->remapped = 1; |
0dfedd61 JR |
4230 | irte_info->devid = devid; |
4231 | irte_info->index = index + offset; | |
0b4d48cb JR |
4232 | |
4233 | return 0; | |
4234 | } | |
4235 | ||
d976195c JR |
4236 | static int setup_hpet_msi(unsigned int irq, unsigned int id) |
4237 | { | |
0dfedd61 | 4238 | struct irq_2_irte *irte_info; |
d976195c JR |
4239 | struct irq_cfg *cfg; |
4240 | int index, devid; | |
4241 | ||
4242 | cfg = irq_get_chip_data(irq); | |
4243 | if (!cfg) | |
4244 | return -EINVAL; | |
4245 | ||
0dfedd61 | 4246 | irte_info = &cfg->irq_2_irte; |
d976195c JR |
4247 | devid = get_hpet_devid(id); |
4248 | if (devid < 0) | |
4249 | return devid; | |
4250 | ||
4251 | index = alloc_irq_index(cfg, devid, 1); | |
4252 | if (index < 0) | |
4253 | return index; | |
4254 | ||
9b1b0e42 | 4255 | cfg->remapped = 1; |
0dfedd61 JR |
4256 | irte_info->devid = devid; |
4257 | irte_info->index = index; | |
d976195c JR |
4258 | |
4259 | return 0; | |
4260 | } | |
4261 | ||
6b474b82 JR |
4262 | struct irq_remap_ops amd_iommu_irq_ops = { |
4263 | .supported = amd_iommu_supported, | |
4264 | .prepare = amd_iommu_prepare, | |
4265 | .enable = amd_iommu_enable, | |
4266 | .disable = amd_iommu_disable, | |
4267 | .reenable = amd_iommu_reenable, | |
4268 | .enable_faulting = amd_iommu_enable_faulting, | |
4269 | .setup_ioapic_entry = setup_ioapic_entry, | |
4270 | .set_affinity = set_affinity, | |
4271 | .free_irq = free_irq, | |
4272 | .compose_msi_msg = compose_msi_msg, | |
4273 | .msi_alloc_irq = msi_alloc_irq, | |
4274 | .msi_setup_irq = msi_setup_irq, | |
4275 | .setup_hpet_msi = setup_hpet_msi, | |
4276 | }; | |
2b324506 | 4277 | #endif |