]>
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> | |
17f5b569 | 34 | #include <asm/msidef.h> |
b6c02715 | 35 | #include <asm/proto.h> |
46a7fa27 | 36 | #include <asm/iommu.h> |
1d9b16d1 | 37 | #include <asm/gart.h> |
27c2127a | 38 | #include <asm/dma.h> |
403f81d8 JR |
39 | |
40 | #include "amd_iommu_proto.h" | |
41 | #include "amd_iommu_types.h" | |
b6c02715 JR |
42 | |
43 | #define CMD_SET_TYPE(cmd, t) ((cmd)->data[1] |= ((t) << 28)) | |
44 | ||
815b33fd | 45 | #define LOOP_TIMEOUT 100000 |
136f78a1 | 46 | |
aa3de9c0 OBC |
47 | /* |
48 | * This bitmap is used to advertise the page sizes our hardware support | |
49 | * to the IOMMU core, which will then use this information to split | |
50 | * physically contiguous memory regions it is mapping into page sizes | |
51 | * that we support. | |
52 | * | |
53 | * Traditionally the IOMMU core just handed us the mappings directly, | |
54 | * after making sure the size is an order of a 4KiB page and that the | |
55 | * mapping has natural alignment. | |
56 | * | |
57 | * To retain this behavior, we currently advertise that we support | |
58 | * all page sizes that are an order of 4KiB. | |
59 | * | |
60 | * If at some point we'd like to utilize the IOMMU core's new behavior, | |
61 | * we could change this to advertise the real page sizes we support. | |
62 | */ | |
63 | #define AMD_IOMMU_PGSIZES (~0xFFFUL) | |
64 | ||
b6c02715 JR |
65 | static DEFINE_RWLOCK(amd_iommu_devtable_lock); |
66 | ||
bd60b735 JR |
67 | /* A list of preallocated protection domains */ |
68 | static LIST_HEAD(iommu_pd_list); | |
69 | static DEFINE_SPINLOCK(iommu_pd_list_lock); | |
70 | ||
8fa5f802 JR |
71 | /* List of all available dev_data structures */ |
72 | static LIST_HEAD(dev_data_list); | |
73 | static DEFINE_SPINLOCK(dev_data_list_lock); | |
74 | ||
6efed63b JR |
75 | LIST_HEAD(ioapic_map); |
76 | LIST_HEAD(hpet_map); | |
77 | ||
0feae533 JR |
78 | /* |
79 | * Domain for untranslated devices - only allocated | |
80 | * if iommu=pt passed on kernel cmd line. | |
81 | */ | |
82 | static struct protection_domain *pt_domain; | |
83 | ||
26961efe | 84 | static struct iommu_ops amd_iommu_ops; |
26961efe | 85 | |
72e1dcc4 | 86 | static ATOMIC_NOTIFIER_HEAD(ppr_notifier); |
52815b75 | 87 | int amd_iommu_max_glx_val = -1; |
72e1dcc4 | 88 | |
ac1534a5 JR |
89 | static struct dma_map_ops amd_iommu_dma_ops; |
90 | ||
431b2a20 JR |
91 | /* |
92 | * general struct to manage commands send to an IOMMU | |
93 | */ | |
d6449536 | 94 | struct iommu_cmd { |
b6c02715 JR |
95 | u32 data[4]; |
96 | }; | |
97 | ||
05152a04 JR |
98 | struct kmem_cache *amd_iommu_irq_cache; |
99 | ||
04bfdd84 | 100 | static void update_domain(struct protection_domain *domain); |
5abcdba4 | 101 | static int __init alloc_passthrough_domain(void); |
c1eee67b | 102 | |
15898bbc JR |
103 | /**************************************************************************** |
104 | * | |
105 | * Helper functions | |
106 | * | |
107 | ****************************************************************************/ | |
108 | ||
f62dda66 | 109 | static struct iommu_dev_data *alloc_dev_data(u16 devid) |
8fa5f802 JR |
110 | { |
111 | struct iommu_dev_data *dev_data; | |
112 | unsigned long flags; | |
113 | ||
114 | dev_data = kzalloc(sizeof(*dev_data), GFP_KERNEL); | |
115 | if (!dev_data) | |
116 | return NULL; | |
117 | ||
f62dda66 | 118 | dev_data->devid = devid; |
8fa5f802 JR |
119 | atomic_set(&dev_data->bind, 0); |
120 | ||
121 | spin_lock_irqsave(&dev_data_list_lock, flags); | |
122 | list_add_tail(&dev_data->dev_data_list, &dev_data_list); | |
123 | spin_unlock_irqrestore(&dev_data_list_lock, flags); | |
124 | ||
125 | return dev_data; | |
126 | } | |
127 | ||
128 | static void free_dev_data(struct iommu_dev_data *dev_data) | |
129 | { | |
130 | unsigned long flags; | |
131 | ||
132 | spin_lock_irqsave(&dev_data_list_lock, flags); | |
133 | list_del(&dev_data->dev_data_list); | |
134 | spin_unlock_irqrestore(&dev_data_list_lock, flags); | |
135 | ||
136 | kfree(dev_data); | |
137 | } | |
138 | ||
3b03bb74 JR |
139 | static struct iommu_dev_data *search_dev_data(u16 devid) |
140 | { | |
141 | struct iommu_dev_data *dev_data; | |
142 | unsigned long flags; | |
143 | ||
144 | spin_lock_irqsave(&dev_data_list_lock, flags); | |
145 | list_for_each_entry(dev_data, &dev_data_list, dev_data_list) { | |
146 | if (dev_data->devid == devid) | |
147 | goto out_unlock; | |
148 | } | |
149 | ||
150 | dev_data = NULL; | |
151 | ||
152 | out_unlock: | |
153 | spin_unlock_irqrestore(&dev_data_list_lock, flags); | |
154 | ||
155 | return dev_data; | |
156 | } | |
157 | ||
158 | static struct iommu_dev_data *find_dev_data(u16 devid) | |
159 | { | |
160 | struct iommu_dev_data *dev_data; | |
161 | ||
162 | dev_data = search_dev_data(devid); | |
163 | ||
164 | if (dev_data == NULL) | |
165 | dev_data = alloc_dev_data(devid); | |
166 | ||
167 | return dev_data; | |
168 | } | |
169 | ||
15898bbc JR |
170 | static inline u16 get_device_id(struct device *dev) |
171 | { | |
172 | struct pci_dev *pdev = to_pci_dev(dev); | |
173 | ||
174 | return calc_devid(pdev->bus->number, pdev->devfn); | |
175 | } | |
176 | ||
657cbb6b JR |
177 | static struct iommu_dev_data *get_dev_data(struct device *dev) |
178 | { | |
179 | return dev->archdata.iommu; | |
180 | } | |
181 | ||
5abcdba4 JR |
182 | static bool pci_iommuv2_capable(struct pci_dev *pdev) |
183 | { | |
184 | static const int caps[] = { | |
185 | PCI_EXT_CAP_ID_ATS, | |
46277b75 JR |
186 | PCI_EXT_CAP_ID_PRI, |
187 | PCI_EXT_CAP_ID_PASID, | |
5abcdba4 JR |
188 | }; |
189 | int i, pos; | |
190 | ||
191 | for (i = 0; i < 3; ++i) { | |
192 | pos = pci_find_ext_capability(pdev, caps[i]); | |
193 | if (pos == 0) | |
194 | return false; | |
195 | } | |
196 | ||
197 | return true; | |
198 | } | |
199 | ||
6a113ddc JR |
200 | static bool pdev_pri_erratum(struct pci_dev *pdev, u32 erratum) |
201 | { | |
202 | struct iommu_dev_data *dev_data; | |
203 | ||
204 | dev_data = get_dev_data(&pdev->dev); | |
205 | ||
206 | return dev_data->errata & (1 << erratum) ? true : false; | |
207 | } | |
208 | ||
71c70984 JR |
209 | /* |
210 | * In this function the list of preallocated protection domains is traversed to | |
211 | * find the domain for a specific device | |
212 | */ | |
213 | static struct dma_ops_domain *find_protection_domain(u16 devid) | |
214 | { | |
215 | struct dma_ops_domain *entry, *ret = NULL; | |
216 | unsigned long flags; | |
217 | u16 alias = amd_iommu_alias_table[devid]; | |
218 | ||
219 | if (list_empty(&iommu_pd_list)) | |
220 | return NULL; | |
221 | ||
222 | spin_lock_irqsave(&iommu_pd_list_lock, flags); | |
223 | ||
224 | list_for_each_entry(entry, &iommu_pd_list, list) { | |
225 | if (entry->target_dev == devid || | |
226 | entry->target_dev == alias) { | |
227 | ret = entry; | |
228 | break; | |
229 | } | |
230 | } | |
231 | ||
232 | spin_unlock_irqrestore(&iommu_pd_list_lock, flags); | |
233 | ||
234 | return ret; | |
235 | } | |
236 | ||
98fc5a69 JR |
237 | /* |
238 | * This function checks if the driver got a valid device from the caller to | |
239 | * avoid dereferencing invalid pointers. | |
240 | */ | |
241 | static bool check_device(struct device *dev) | |
242 | { | |
243 | u16 devid; | |
244 | ||
245 | if (!dev || !dev->dma_mask) | |
246 | return false; | |
247 | ||
248 | /* No device or no PCI device */ | |
339d3261 | 249 | if (dev->bus != &pci_bus_type) |
98fc5a69 JR |
250 | return false; |
251 | ||
252 | devid = get_device_id(dev); | |
253 | ||
254 | /* Out of our scope? */ | |
255 | if (devid > amd_iommu_last_bdf) | |
256 | return false; | |
257 | ||
258 | if (amd_iommu_rlookup_table[devid] == NULL) | |
259 | return false; | |
260 | ||
261 | return true; | |
262 | } | |
263 | ||
664b6003 AW |
264 | static void swap_pci_ref(struct pci_dev **from, struct pci_dev *to) |
265 | { | |
266 | pci_dev_put(*from); | |
267 | *from = to; | |
268 | } | |
269 | ||
270 | #define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF) | |
271 | ||
657cbb6b JR |
272 | static int iommu_init_device(struct device *dev) |
273 | { | |
9dcd6130 | 274 | struct pci_dev *dma_pdev, *pdev = to_pci_dev(dev); |
657cbb6b | 275 | struct iommu_dev_data *dev_data; |
9dcd6130 | 276 | struct iommu_group *group; |
8fa5f802 | 277 | u16 alias; |
9dcd6130 | 278 | int ret; |
657cbb6b JR |
279 | |
280 | if (dev->archdata.iommu) | |
281 | return 0; | |
282 | ||
3b03bb74 | 283 | dev_data = find_dev_data(get_device_id(dev)); |
657cbb6b JR |
284 | if (!dev_data) |
285 | return -ENOMEM; | |
286 | ||
f62dda66 | 287 | alias = amd_iommu_alias_table[dev_data->devid]; |
2b02b091 | 288 | if (alias != dev_data->devid) { |
71f77580 | 289 | struct iommu_dev_data *alias_data; |
b00d3bcf | 290 | |
71f77580 JR |
291 | alias_data = find_dev_data(alias); |
292 | if (alias_data == NULL) { | |
293 | pr_err("AMD-Vi: Warning: Unhandled device %s\n", | |
294 | dev_name(dev)); | |
2b02b091 JR |
295 | free_dev_data(dev_data); |
296 | return -ENOTSUPP; | |
297 | } | |
71f77580 | 298 | dev_data->alias_data = alias_data; |
9dcd6130 AW |
299 | |
300 | dma_pdev = pci_get_bus_and_slot(alias >> 8, alias & 0xff); | |
301 | } else | |
302 | dma_pdev = pci_dev_get(pdev); | |
303 | ||
31fe9435 | 304 | /* Account for quirked devices */ |
664b6003 AW |
305 | swap_pci_ref(&dma_pdev, pci_get_dma_source(dma_pdev)); |
306 | ||
31fe9435 AW |
307 | /* |
308 | * If it's a multifunction device that does not support our | |
309 | * required ACS flags, add to the same group as function 0. | |
310 | */ | |
664b6003 AW |
311 | if (dma_pdev->multifunction && |
312 | !pci_acs_enabled(dma_pdev, REQ_ACS_FLAGS)) | |
313 | swap_pci_ref(&dma_pdev, | |
314 | pci_get_slot(dma_pdev->bus, | |
315 | PCI_DEVFN(PCI_SLOT(dma_pdev->devfn), | |
316 | 0))); | |
317 | ||
31fe9435 AW |
318 | /* |
319 | * Devices on the root bus go through the iommu. If that's not us, | |
320 | * find the next upstream device and test ACS up to the root bus. | |
321 | * Finding the next device may require skipping virtual buses. | |
322 | */ | |
664b6003 | 323 | while (!pci_is_root_bus(dma_pdev->bus)) { |
31fe9435 AW |
324 | struct pci_bus *bus = dma_pdev->bus; |
325 | ||
326 | while (!bus->self) { | |
327 | if (!pci_is_root_bus(bus)) | |
328 | bus = bus->parent; | |
329 | else | |
330 | goto root_bus; | |
331 | } | |
332 | ||
333 | if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS)) | |
664b6003 AW |
334 | break; |
335 | ||
31fe9435 | 336 | swap_pci_ref(&dma_pdev, pci_dev_get(bus->self)); |
664b6003 AW |
337 | } |
338 | ||
31fe9435 | 339 | root_bus: |
9dcd6130 AW |
340 | group = iommu_group_get(&dma_pdev->dev); |
341 | pci_dev_put(dma_pdev); | |
342 | if (!group) { | |
343 | group = iommu_group_alloc(); | |
344 | if (IS_ERR(group)) | |
345 | return PTR_ERR(group); | |
26018874 | 346 | } |
657cbb6b | 347 | |
9dcd6130 AW |
348 | ret = iommu_group_add_device(group, dev); |
349 | ||
350 | iommu_group_put(group); | |
351 | ||
352 | if (ret) | |
353 | return ret; | |
354 | ||
5abcdba4 JR |
355 | if (pci_iommuv2_capable(pdev)) { |
356 | struct amd_iommu *iommu; | |
357 | ||
358 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
359 | dev_data->iommu_v2 = iommu->is_iommu_v2; | |
360 | } | |
361 | ||
657cbb6b JR |
362 | dev->archdata.iommu = dev_data; |
363 | ||
657cbb6b JR |
364 | return 0; |
365 | } | |
366 | ||
26018874 JR |
367 | static void iommu_ignore_device(struct device *dev) |
368 | { | |
369 | u16 devid, alias; | |
370 | ||
371 | devid = get_device_id(dev); | |
372 | alias = amd_iommu_alias_table[devid]; | |
373 | ||
374 | memset(&amd_iommu_dev_table[devid], 0, sizeof(struct dev_table_entry)); | |
375 | memset(&amd_iommu_dev_table[alias], 0, sizeof(struct dev_table_entry)); | |
376 | ||
377 | amd_iommu_rlookup_table[devid] = NULL; | |
378 | amd_iommu_rlookup_table[alias] = NULL; | |
379 | } | |
380 | ||
657cbb6b JR |
381 | static void iommu_uninit_device(struct device *dev) |
382 | { | |
9dcd6130 AW |
383 | iommu_group_remove_device(dev); |
384 | ||
8fa5f802 JR |
385 | /* |
386 | * Nothing to do here - we keep dev_data around for unplugged devices | |
387 | * and reuse it when the device is re-plugged - not doing so would | |
388 | * introduce a ton of races. | |
389 | */ | |
657cbb6b | 390 | } |
b7cc9554 JR |
391 | |
392 | void __init amd_iommu_uninit_devices(void) | |
393 | { | |
8fa5f802 | 394 | struct iommu_dev_data *dev_data, *n; |
b7cc9554 JR |
395 | struct pci_dev *pdev = NULL; |
396 | ||
397 | for_each_pci_dev(pdev) { | |
398 | ||
399 | if (!check_device(&pdev->dev)) | |
400 | continue; | |
401 | ||
402 | iommu_uninit_device(&pdev->dev); | |
403 | } | |
8fa5f802 JR |
404 | |
405 | /* Free all of our dev_data structures */ | |
406 | list_for_each_entry_safe(dev_data, n, &dev_data_list, dev_data_list) | |
407 | free_dev_data(dev_data); | |
b7cc9554 JR |
408 | } |
409 | ||
410 | int __init amd_iommu_init_devices(void) | |
411 | { | |
412 | struct pci_dev *pdev = NULL; | |
413 | int ret = 0; | |
414 | ||
415 | for_each_pci_dev(pdev) { | |
416 | ||
417 | if (!check_device(&pdev->dev)) | |
418 | continue; | |
419 | ||
420 | ret = iommu_init_device(&pdev->dev); | |
26018874 JR |
421 | if (ret == -ENOTSUPP) |
422 | iommu_ignore_device(&pdev->dev); | |
423 | else if (ret) | |
b7cc9554 JR |
424 | goto out_free; |
425 | } | |
426 | ||
427 | return 0; | |
428 | ||
429 | out_free: | |
430 | ||
431 | amd_iommu_uninit_devices(); | |
432 | ||
433 | return ret; | |
434 | } | |
7f26508b JR |
435 | #ifdef CONFIG_AMD_IOMMU_STATS |
436 | ||
437 | /* | |
438 | * Initialization code for statistics collection | |
439 | */ | |
440 | ||
da49f6df | 441 | DECLARE_STATS_COUNTER(compl_wait); |
0f2a86f2 | 442 | DECLARE_STATS_COUNTER(cnt_map_single); |
146a6917 | 443 | DECLARE_STATS_COUNTER(cnt_unmap_single); |
d03f067a | 444 | DECLARE_STATS_COUNTER(cnt_map_sg); |
55877a6b | 445 | DECLARE_STATS_COUNTER(cnt_unmap_sg); |
c8f0fb36 | 446 | DECLARE_STATS_COUNTER(cnt_alloc_coherent); |
5d31ee7e | 447 | DECLARE_STATS_COUNTER(cnt_free_coherent); |
c1858976 | 448 | DECLARE_STATS_COUNTER(cross_page); |
f57d98ae | 449 | DECLARE_STATS_COUNTER(domain_flush_single); |
18811f55 | 450 | DECLARE_STATS_COUNTER(domain_flush_all); |
5774f7c5 | 451 | DECLARE_STATS_COUNTER(alloced_io_mem); |
8ecaf8f1 | 452 | DECLARE_STATS_COUNTER(total_map_requests); |
399be2f5 JR |
453 | DECLARE_STATS_COUNTER(complete_ppr); |
454 | DECLARE_STATS_COUNTER(invalidate_iotlb); | |
455 | DECLARE_STATS_COUNTER(invalidate_iotlb_all); | |
456 | DECLARE_STATS_COUNTER(pri_requests); | |
457 | ||
7f26508b | 458 | static struct dentry *stats_dir; |
7f26508b JR |
459 | static struct dentry *de_fflush; |
460 | ||
461 | static void amd_iommu_stats_add(struct __iommu_counter *cnt) | |
462 | { | |
463 | if (stats_dir == NULL) | |
464 | return; | |
465 | ||
466 | cnt->dent = debugfs_create_u64(cnt->name, 0444, stats_dir, | |
467 | &cnt->value); | |
468 | } | |
469 | ||
470 | static void amd_iommu_stats_init(void) | |
471 | { | |
472 | stats_dir = debugfs_create_dir("amd-iommu", NULL); | |
473 | if (stats_dir == NULL) | |
474 | return; | |
475 | ||
7f26508b | 476 | de_fflush = debugfs_create_bool("fullflush", 0444, stats_dir, |
3775d481 | 477 | &amd_iommu_unmap_flush); |
da49f6df JR |
478 | |
479 | amd_iommu_stats_add(&compl_wait); | |
0f2a86f2 | 480 | amd_iommu_stats_add(&cnt_map_single); |
146a6917 | 481 | amd_iommu_stats_add(&cnt_unmap_single); |
d03f067a | 482 | amd_iommu_stats_add(&cnt_map_sg); |
55877a6b | 483 | amd_iommu_stats_add(&cnt_unmap_sg); |
c8f0fb36 | 484 | amd_iommu_stats_add(&cnt_alloc_coherent); |
5d31ee7e | 485 | amd_iommu_stats_add(&cnt_free_coherent); |
c1858976 | 486 | amd_iommu_stats_add(&cross_page); |
f57d98ae | 487 | amd_iommu_stats_add(&domain_flush_single); |
18811f55 | 488 | amd_iommu_stats_add(&domain_flush_all); |
5774f7c5 | 489 | amd_iommu_stats_add(&alloced_io_mem); |
8ecaf8f1 | 490 | amd_iommu_stats_add(&total_map_requests); |
399be2f5 JR |
491 | amd_iommu_stats_add(&complete_ppr); |
492 | amd_iommu_stats_add(&invalidate_iotlb); | |
493 | amd_iommu_stats_add(&invalidate_iotlb_all); | |
494 | amd_iommu_stats_add(&pri_requests); | |
7f26508b JR |
495 | } |
496 | ||
497 | #endif | |
498 | ||
a80dc3e0 JR |
499 | /**************************************************************************** |
500 | * | |
501 | * Interrupt handling functions | |
502 | * | |
503 | ****************************************************************************/ | |
504 | ||
e3e59876 JR |
505 | static void dump_dte_entry(u16 devid) |
506 | { | |
507 | int i; | |
508 | ||
ee6c2868 JR |
509 | for (i = 0; i < 4; ++i) |
510 | pr_err("AMD-Vi: DTE[%d]: %016llx\n", i, | |
e3e59876 JR |
511 | amd_iommu_dev_table[devid].data[i]); |
512 | } | |
513 | ||
945b4ac4 JR |
514 | static void dump_command(unsigned long phys_addr) |
515 | { | |
516 | struct iommu_cmd *cmd = phys_to_virt(phys_addr); | |
517 | int i; | |
518 | ||
519 | for (i = 0; i < 4; ++i) | |
520 | pr_err("AMD-Vi: CMD[%d]: %08x\n", i, cmd->data[i]); | |
521 | } | |
522 | ||
a345b23b | 523 | static void iommu_print_event(struct amd_iommu *iommu, void *__evt) |
90008ee4 | 524 | { |
3d06fca8 JR |
525 | int type, devid, domid, flags; |
526 | volatile u32 *event = __evt; | |
527 | int count = 0; | |
528 | u64 address; | |
529 | ||
530 | retry: | |
531 | type = (event[1] >> EVENT_TYPE_SHIFT) & EVENT_TYPE_MASK; | |
532 | devid = (event[0] >> EVENT_DEVID_SHIFT) & EVENT_DEVID_MASK; | |
533 | domid = (event[1] >> EVENT_DOMID_SHIFT) & EVENT_DOMID_MASK; | |
534 | flags = (event[1] >> EVENT_FLAGS_SHIFT) & EVENT_FLAGS_MASK; | |
535 | address = (u64)(((u64)event[3]) << 32) | event[2]; | |
536 | ||
537 | if (type == 0) { | |
538 | /* Did we hit the erratum? */ | |
539 | if (++count == LOOP_TIMEOUT) { | |
540 | pr_err("AMD-Vi: No event written to event log\n"); | |
541 | return; | |
542 | } | |
543 | udelay(1); | |
544 | goto retry; | |
545 | } | |
90008ee4 | 546 | |
4c6f40d4 | 547 | printk(KERN_ERR "AMD-Vi: Event logged ["); |
90008ee4 JR |
548 | |
549 | switch (type) { | |
550 | case EVENT_TYPE_ILL_DEV: | |
551 | printk("ILLEGAL_DEV_TABLE_ENTRY device=%02x:%02x.%x " | |
552 | "address=0x%016llx flags=0x%04x]\n", | |
553 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | |
554 | address, flags); | |
e3e59876 | 555 | dump_dte_entry(devid); |
90008ee4 JR |
556 | break; |
557 | case EVENT_TYPE_IO_FAULT: | |
558 | printk("IO_PAGE_FAULT device=%02x:%02x.%x " | |
559 | "domain=0x%04x address=0x%016llx flags=0x%04x]\n", | |
560 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | |
561 | domid, address, flags); | |
562 | break; | |
563 | case EVENT_TYPE_DEV_TAB_ERR: | |
564 | printk("DEV_TAB_HARDWARE_ERROR device=%02x:%02x.%x " | |
565 | "address=0x%016llx flags=0x%04x]\n", | |
566 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | |
567 | address, flags); | |
568 | break; | |
569 | case EVENT_TYPE_PAGE_TAB_ERR: | |
570 | printk("PAGE_TAB_HARDWARE_ERROR device=%02x:%02x.%x " | |
571 | "domain=0x%04x address=0x%016llx flags=0x%04x]\n", | |
572 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | |
573 | domid, address, flags); | |
574 | break; | |
575 | case EVENT_TYPE_ILL_CMD: | |
576 | printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address); | |
945b4ac4 | 577 | dump_command(address); |
90008ee4 JR |
578 | break; |
579 | case EVENT_TYPE_CMD_HARD_ERR: | |
580 | printk("COMMAND_HARDWARE_ERROR address=0x%016llx " | |
581 | "flags=0x%04x]\n", address, flags); | |
582 | break; | |
583 | case EVENT_TYPE_IOTLB_INV_TO: | |
584 | printk("IOTLB_INV_TIMEOUT device=%02x:%02x.%x " | |
585 | "address=0x%016llx]\n", | |
586 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | |
587 | address); | |
588 | break; | |
589 | case EVENT_TYPE_INV_DEV_REQ: | |
590 | printk("INVALID_DEVICE_REQUEST device=%02x:%02x.%x " | |
591 | "address=0x%016llx flags=0x%04x]\n", | |
592 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | |
593 | address, flags); | |
594 | break; | |
595 | default: | |
596 | printk(KERN_ERR "UNKNOWN type=0x%02x]\n", type); | |
597 | } | |
3d06fca8 JR |
598 | |
599 | memset(__evt, 0, 4 * sizeof(u32)); | |
90008ee4 JR |
600 | } |
601 | ||
602 | static void iommu_poll_events(struct amd_iommu *iommu) | |
603 | { | |
604 | u32 head, tail; | |
605 | unsigned long flags; | |
606 | ||
607 | spin_lock_irqsave(&iommu->lock, flags); | |
608 | ||
609 | head = readl(iommu->mmio_base + MMIO_EVT_HEAD_OFFSET); | |
610 | tail = readl(iommu->mmio_base + MMIO_EVT_TAIL_OFFSET); | |
611 | ||
612 | while (head != tail) { | |
a345b23b | 613 | iommu_print_event(iommu, iommu->evt_buf + head); |
90008ee4 JR |
614 | head = (head + EVENT_ENTRY_SIZE) % iommu->evt_buf_size; |
615 | } | |
616 | ||
617 | writel(head, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET); | |
618 | ||
619 | spin_unlock_irqrestore(&iommu->lock, flags); | |
620 | } | |
621 | ||
eee53537 | 622 | static void iommu_handle_ppr_entry(struct amd_iommu *iommu, u64 *raw) |
72e1dcc4 JR |
623 | { |
624 | struct amd_iommu_fault fault; | |
72e1dcc4 | 625 | |
399be2f5 JR |
626 | INC_STATS_COUNTER(pri_requests); |
627 | ||
72e1dcc4 JR |
628 | if (PPR_REQ_TYPE(raw[0]) != PPR_REQ_FAULT) { |
629 | pr_err_ratelimited("AMD-Vi: Unknown PPR request received\n"); | |
630 | return; | |
631 | } | |
632 | ||
633 | fault.address = raw[1]; | |
634 | fault.pasid = PPR_PASID(raw[0]); | |
635 | fault.device_id = PPR_DEVID(raw[0]); | |
636 | fault.tag = PPR_TAG(raw[0]); | |
637 | fault.flags = PPR_FLAGS(raw[0]); | |
638 | ||
72e1dcc4 JR |
639 | atomic_notifier_call_chain(&ppr_notifier, 0, &fault); |
640 | } | |
641 | ||
642 | static void iommu_poll_ppr_log(struct amd_iommu *iommu) | |
643 | { | |
644 | unsigned long flags; | |
645 | u32 head, tail; | |
646 | ||
647 | if (iommu->ppr_log == NULL) | |
648 | return; | |
649 | ||
eee53537 JR |
650 | /* enable ppr interrupts again */ |
651 | writel(MMIO_STATUS_PPR_INT_MASK, iommu->mmio_base + MMIO_STATUS_OFFSET); | |
652 | ||
72e1dcc4 JR |
653 | spin_lock_irqsave(&iommu->lock, flags); |
654 | ||
655 | head = readl(iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); | |
656 | tail = readl(iommu->mmio_base + MMIO_PPR_TAIL_OFFSET); | |
657 | ||
658 | while (head != tail) { | |
eee53537 JR |
659 | volatile u64 *raw; |
660 | u64 entry[2]; | |
661 | int i; | |
662 | ||
663 | raw = (u64 *)(iommu->ppr_log + head); | |
664 | ||
665 | /* | |
666 | * Hardware bug: Interrupt may arrive before the entry is | |
667 | * written to memory. If this happens we need to wait for the | |
668 | * entry to arrive. | |
669 | */ | |
670 | for (i = 0; i < LOOP_TIMEOUT; ++i) { | |
671 | if (PPR_REQ_TYPE(raw[0]) != 0) | |
672 | break; | |
673 | udelay(1); | |
674 | } | |
72e1dcc4 | 675 | |
eee53537 JR |
676 | /* Avoid memcpy function-call overhead */ |
677 | entry[0] = raw[0]; | |
678 | entry[1] = raw[1]; | |
72e1dcc4 | 679 | |
eee53537 JR |
680 | /* |
681 | * To detect the hardware bug we need to clear the entry | |
682 | * back to zero. | |
683 | */ | |
684 | raw[0] = raw[1] = 0UL; | |
685 | ||
686 | /* Update head pointer of hardware ring-buffer */ | |
72e1dcc4 JR |
687 | head = (head + PPR_ENTRY_SIZE) % PPR_LOG_SIZE; |
688 | writel(head, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); | |
eee53537 JR |
689 | |
690 | /* | |
691 | * Release iommu->lock because ppr-handling might need to | |
692 | * re-aquire it | |
693 | */ | |
694 | spin_unlock_irqrestore(&iommu->lock, flags); | |
695 | ||
696 | /* Handle PPR entry */ | |
697 | iommu_handle_ppr_entry(iommu, entry); | |
698 | ||
699 | spin_lock_irqsave(&iommu->lock, flags); | |
700 | ||
701 | /* Refresh ring-buffer information */ | |
702 | head = readl(iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); | |
72e1dcc4 JR |
703 | tail = readl(iommu->mmio_base + MMIO_PPR_TAIL_OFFSET); |
704 | } | |
705 | ||
72e1dcc4 JR |
706 | spin_unlock_irqrestore(&iommu->lock, flags); |
707 | } | |
708 | ||
72fe00f0 | 709 | irqreturn_t amd_iommu_int_thread(int irq, void *data) |
a80dc3e0 | 710 | { |
90008ee4 JR |
711 | struct amd_iommu *iommu; |
712 | ||
72e1dcc4 | 713 | for_each_iommu(iommu) { |
90008ee4 | 714 | iommu_poll_events(iommu); |
72e1dcc4 JR |
715 | iommu_poll_ppr_log(iommu); |
716 | } | |
90008ee4 JR |
717 | |
718 | return IRQ_HANDLED; | |
a80dc3e0 JR |
719 | } |
720 | ||
72fe00f0 JR |
721 | irqreturn_t amd_iommu_int_handler(int irq, void *data) |
722 | { | |
723 | return IRQ_WAKE_THREAD; | |
724 | } | |
725 | ||
431b2a20 JR |
726 | /**************************************************************************** |
727 | * | |
728 | * IOMMU command queuing functions | |
729 | * | |
730 | ****************************************************************************/ | |
731 | ||
ac0ea6e9 JR |
732 | static int wait_on_sem(volatile u64 *sem) |
733 | { | |
734 | int i = 0; | |
735 | ||
736 | while (*sem == 0 && i < LOOP_TIMEOUT) { | |
737 | udelay(1); | |
738 | i += 1; | |
739 | } | |
740 | ||
741 | if (i == LOOP_TIMEOUT) { | |
742 | pr_alert("AMD-Vi: Completion-Wait loop timed out\n"); | |
743 | return -EIO; | |
744 | } | |
745 | ||
746 | return 0; | |
747 | } | |
748 | ||
749 | static void copy_cmd_to_buffer(struct amd_iommu *iommu, | |
750 | struct iommu_cmd *cmd, | |
751 | u32 tail) | |
a19ae1ec | 752 | { |
a19ae1ec JR |
753 | u8 *target; |
754 | ||
8a7c5ef3 | 755 | target = iommu->cmd_buf + tail; |
ac0ea6e9 JR |
756 | tail = (tail + sizeof(*cmd)) % iommu->cmd_buf_size; |
757 | ||
758 | /* Copy command to buffer */ | |
759 | memcpy(target, cmd, sizeof(*cmd)); | |
760 | ||
761 | /* Tell the IOMMU about it */ | |
a19ae1ec | 762 | writel(tail, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); |
ac0ea6e9 | 763 | } |
a19ae1ec | 764 | |
815b33fd | 765 | static void build_completion_wait(struct iommu_cmd *cmd, u64 address) |
ded46737 | 766 | { |
815b33fd JR |
767 | WARN_ON(address & 0x7ULL); |
768 | ||
ded46737 | 769 | memset(cmd, 0, sizeof(*cmd)); |
815b33fd JR |
770 | cmd->data[0] = lower_32_bits(__pa(address)) | CMD_COMPL_WAIT_STORE_MASK; |
771 | cmd->data[1] = upper_32_bits(__pa(address)); | |
772 | cmd->data[2] = 1; | |
ded46737 JR |
773 | CMD_SET_TYPE(cmd, CMD_COMPL_WAIT); |
774 | } | |
775 | ||
94fe79e2 JR |
776 | static void build_inv_dte(struct iommu_cmd *cmd, u16 devid) |
777 | { | |
778 | memset(cmd, 0, sizeof(*cmd)); | |
779 | cmd->data[0] = devid; | |
780 | CMD_SET_TYPE(cmd, CMD_INV_DEV_ENTRY); | |
781 | } | |
782 | ||
11b6402c JR |
783 | static void build_inv_iommu_pages(struct iommu_cmd *cmd, u64 address, |
784 | size_t size, u16 domid, int pde) | |
785 | { | |
786 | u64 pages; | |
787 | int s; | |
788 | ||
789 | pages = iommu_num_pages(address, size, PAGE_SIZE); | |
790 | s = 0; | |
791 | ||
792 | if (pages > 1) { | |
793 | /* | |
794 | * If we have to flush more than one page, flush all | |
795 | * TLB entries for this domain | |
796 | */ | |
797 | address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS; | |
798 | s = 1; | |
799 | } | |
800 | ||
801 | address &= PAGE_MASK; | |
802 | ||
803 | memset(cmd, 0, sizeof(*cmd)); | |
804 | cmd->data[1] |= domid; | |
805 | cmd->data[2] = lower_32_bits(address); | |
806 | cmd->data[3] = upper_32_bits(address); | |
807 | CMD_SET_TYPE(cmd, CMD_INV_IOMMU_PAGES); | |
808 | if (s) /* size bit - we flush more than one 4kb page */ | |
809 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
810 | if (pde) /* PDE bit - we wan't flush everything not only the PTEs */ | |
811 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK; | |
812 | } | |
813 | ||
cb41ed85 JR |
814 | static void build_inv_iotlb_pages(struct iommu_cmd *cmd, u16 devid, int qdep, |
815 | u64 address, size_t size) | |
816 | { | |
817 | u64 pages; | |
818 | int s; | |
819 | ||
820 | pages = iommu_num_pages(address, size, PAGE_SIZE); | |
821 | s = 0; | |
822 | ||
823 | if (pages > 1) { | |
824 | /* | |
825 | * If we have to flush more than one page, flush all | |
826 | * TLB entries for this domain | |
827 | */ | |
828 | address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS; | |
829 | s = 1; | |
830 | } | |
831 | ||
832 | address &= PAGE_MASK; | |
833 | ||
834 | memset(cmd, 0, sizeof(*cmd)); | |
835 | cmd->data[0] = devid; | |
836 | cmd->data[0] |= (qdep & 0xff) << 24; | |
837 | cmd->data[1] = devid; | |
838 | cmd->data[2] = lower_32_bits(address); | |
839 | cmd->data[3] = upper_32_bits(address); | |
840 | CMD_SET_TYPE(cmd, CMD_INV_IOTLB_PAGES); | |
841 | if (s) | |
842 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
843 | } | |
844 | ||
22e266c7 JR |
845 | static void build_inv_iommu_pasid(struct iommu_cmd *cmd, u16 domid, int pasid, |
846 | u64 address, bool size) | |
847 | { | |
848 | memset(cmd, 0, sizeof(*cmd)); | |
849 | ||
850 | address &= ~(0xfffULL); | |
851 | ||
852 | cmd->data[0] = pasid & PASID_MASK; | |
853 | cmd->data[1] = domid; | |
854 | cmd->data[2] = lower_32_bits(address); | |
855 | cmd->data[3] = upper_32_bits(address); | |
856 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_PDE_MASK; | |
857 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_GN_MASK; | |
858 | if (size) | |
859 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
860 | CMD_SET_TYPE(cmd, CMD_INV_IOMMU_PAGES); | |
861 | } | |
862 | ||
863 | static void build_inv_iotlb_pasid(struct iommu_cmd *cmd, u16 devid, int pasid, | |
864 | int qdep, u64 address, bool size) | |
865 | { | |
866 | memset(cmd, 0, sizeof(*cmd)); | |
867 | ||
868 | address &= ~(0xfffULL); | |
869 | ||
870 | cmd->data[0] = devid; | |
871 | cmd->data[0] |= (pasid & 0xff) << 16; | |
872 | cmd->data[0] |= (qdep & 0xff) << 24; | |
873 | cmd->data[1] = devid; | |
874 | cmd->data[1] |= ((pasid >> 8) & 0xfff) << 16; | |
875 | cmd->data[2] = lower_32_bits(address); | |
876 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_GN_MASK; | |
877 | cmd->data[3] = upper_32_bits(address); | |
878 | if (size) | |
879 | cmd->data[2] |= CMD_INV_IOMMU_PAGES_SIZE_MASK; | |
880 | CMD_SET_TYPE(cmd, CMD_INV_IOTLB_PAGES); | |
881 | } | |
882 | ||
c99afa25 JR |
883 | static void build_complete_ppr(struct iommu_cmd *cmd, u16 devid, int pasid, |
884 | int status, int tag, bool gn) | |
885 | { | |
886 | memset(cmd, 0, sizeof(*cmd)); | |
887 | ||
888 | cmd->data[0] = devid; | |
889 | if (gn) { | |
890 | cmd->data[1] = pasid & PASID_MASK; | |
891 | cmd->data[2] = CMD_INV_IOMMU_PAGES_GN_MASK; | |
892 | } | |
893 | cmd->data[3] = tag & 0x1ff; | |
894 | cmd->data[3] |= (status & PPR_STATUS_MASK) << PPR_STATUS_SHIFT; | |
895 | ||
896 | CMD_SET_TYPE(cmd, CMD_COMPLETE_PPR); | |
897 | } | |
898 | ||
58fc7f14 JR |
899 | static void build_inv_all(struct iommu_cmd *cmd) |
900 | { | |
901 | memset(cmd, 0, sizeof(*cmd)); | |
902 | CMD_SET_TYPE(cmd, CMD_INV_ALL); | |
a19ae1ec JR |
903 | } |
904 | ||
431b2a20 | 905 | /* |
431b2a20 | 906 | * Writes the command to the IOMMUs command buffer and informs the |
ac0ea6e9 | 907 | * hardware about the new command. |
431b2a20 | 908 | */ |
f1ca1512 JR |
909 | static int iommu_queue_command_sync(struct amd_iommu *iommu, |
910 | struct iommu_cmd *cmd, | |
911 | bool sync) | |
a19ae1ec | 912 | { |
ac0ea6e9 | 913 | u32 left, tail, head, next_tail; |
a19ae1ec | 914 | unsigned long flags; |
a19ae1ec | 915 | |
549c90dc | 916 | WARN_ON(iommu->cmd_buf_size & CMD_BUFFER_UNINITIALIZED); |
ac0ea6e9 JR |
917 | |
918 | again: | |
a19ae1ec | 919 | spin_lock_irqsave(&iommu->lock, flags); |
a19ae1ec | 920 | |
ac0ea6e9 JR |
921 | head = readl(iommu->mmio_base + MMIO_CMD_HEAD_OFFSET); |
922 | tail = readl(iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); | |
923 | next_tail = (tail + sizeof(*cmd)) % iommu->cmd_buf_size; | |
924 | left = (head - next_tail) % iommu->cmd_buf_size; | |
a19ae1ec | 925 | |
ac0ea6e9 JR |
926 | if (left <= 2) { |
927 | struct iommu_cmd sync_cmd; | |
928 | volatile u64 sem = 0; | |
929 | int ret; | |
8d201968 | 930 | |
ac0ea6e9 JR |
931 | build_completion_wait(&sync_cmd, (u64)&sem); |
932 | copy_cmd_to_buffer(iommu, &sync_cmd, tail); | |
da49f6df | 933 | |
ac0ea6e9 JR |
934 | spin_unlock_irqrestore(&iommu->lock, flags); |
935 | ||
936 | if ((ret = wait_on_sem(&sem)) != 0) | |
937 | return ret; | |
938 | ||
939 | goto again; | |
8d201968 JR |
940 | } |
941 | ||
ac0ea6e9 JR |
942 | copy_cmd_to_buffer(iommu, cmd, tail); |
943 | ||
944 | /* We need to sync now to make sure all commands are processed */ | |
f1ca1512 | 945 | iommu->need_sync = sync; |
ac0ea6e9 | 946 | |
a19ae1ec | 947 | spin_unlock_irqrestore(&iommu->lock, flags); |
8d201968 | 948 | |
815b33fd | 949 | return 0; |
8d201968 JR |
950 | } |
951 | ||
f1ca1512 JR |
952 | static int iommu_queue_command(struct amd_iommu *iommu, struct iommu_cmd *cmd) |
953 | { | |
954 | return iommu_queue_command_sync(iommu, cmd, true); | |
955 | } | |
956 | ||
8d201968 JR |
957 | /* |
958 | * This function queues a completion wait command into the command | |
959 | * buffer of an IOMMU | |
960 | */ | |
a19ae1ec | 961 | static int iommu_completion_wait(struct amd_iommu *iommu) |
8d201968 JR |
962 | { |
963 | struct iommu_cmd cmd; | |
815b33fd | 964 | volatile u64 sem = 0; |
ac0ea6e9 | 965 | int ret; |
8d201968 | 966 | |
09ee17eb | 967 | if (!iommu->need_sync) |
815b33fd | 968 | return 0; |
09ee17eb | 969 | |
815b33fd | 970 | build_completion_wait(&cmd, (u64)&sem); |
a19ae1ec | 971 | |
f1ca1512 | 972 | ret = iommu_queue_command_sync(iommu, &cmd, false); |
a19ae1ec | 973 | if (ret) |
815b33fd | 974 | return ret; |
8d201968 | 975 | |
ac0ea6e9 | 976 | return wait_on_sem(&sem); |
8d201968 JR |
977 | } |
978 | ||
d8c13085 | 979 | static int iommu_flush_dte(struct amd_iommu *iommu, u16 devid) |
a19ae1ec | 980 | { |
d8c13085 | 981 | struct iommu_cmd cmd; |
a19ae1ec | 982 | |
d8c13085 | 983 | build_inv_dte(&cmd, devid); |
7e4f88da | 984 | |
d8c13085 JR |
985 | return iommu_queue_command(iommu, &cmd); |
986 | } | |
09ee17eb | 987 | |
7d0c5cc5 JR |
988 | static void iommu_flush_dte_all(struct amd_iommu *iommu) |
989 | { | |
990 | u32 devid; | |
09ee17eb | 991 | |
7d0c5cc5 JR |
992 | for (devid = 0; devid <= 0xffff; ++devid) |
993 | iommu_flush_dte(iommu, devid); | |
a19ae1ec | 994 | |
7d0c5cc5 JR |
995 | iommu_completion_wait(iommu); |
996 | } | |
84df8175 | 997 | |
7d0c5cc5 JR |
998 | /* |
999 | * This function uses heavy locking and may disable irqs for some time. But | |
1000 | * this is no issue because it is only called during resume. | |
1001 | */ | |
1002 | static void iommu_flush_tlb_all(struct amd_iommu *iommu) | |
1003 | { | |
1004 | u32 dom_id; | |
a19ae1ec | 1005 | |
7d0c5cc5 JR |
1006 | for (dom_id = 0; dom_id <= 0xffff; ++dom_id) { |
1007 | struct iommu_cmd cmd; | |
1008 | build_inv_iommu_pages(&cmd, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, | |
1009 | dom_id, 1); | |
1010 | iommu_queue_command(iommu, &cmd); | |
1011 | } | |
8eed9833 | 1012 | |
7d0c5cc5 | 1013 | iommu_completion_wait(iommu); |
a19ae1ec JR |
1014 | } |
1015 | ||
58fc7f14 | 1016 | static void iommu_flush_all(struct amd_iommu *iommu) |
0518a3a4 | 1017 | { |
58fc7f14 | 1018 | struct iommu_cmd cmd; |
0518a3a4 | 1019 | |
58fc7f14 | 1020 | build_inv_all(&cmd); |
0518a3a4 | 1021 | |
58fc7f14 JR |
1022 | iommu_queue_command(iommu, &cmd); |
1023 | iommu_completion_wait(iommu); | |
1024 | } | |
1025 | ||
7d0c5cc5 JR |
1026 | void iommu_flush_all_caches(struct amd_iommu *iommu) |
1027 | { | |
58fc7f14 JR |
1028 | if (iommu_feature(iommu, FEATURE_IA)) { |
1029 | iommu_flush_all(iommu); | |
1030 | } else { | |
1031 | iommu_flush_dte_all(iommu); | |
1032 | iommu_flush_tlb_all(iommu); | |
0518a3a4 JR |
1033 | } |
1034 | } | |
1035 | ||
431b2a20 | 1036 | /* |
cb41ed85 | 1037 | * Command send function for flushing on-device TLB |
431b2a20 | 1038 | */ |
6c542047 JR |
1039 | static int device_flush_iotlb(struct iommu_dev_data *dev_data, |
1040 | u64 address, size_t size) | |
3fa43655 JR |
1041 | { |
1042 | struct amd_iommu *iommu; | |
b00d3bcf | 1043 | struct iommu_cmd cmd; |
cb41ed85 | 1044 | int qdep; |
3fa43655 | 1045 | |
ea61cddb JR |
1046 | qdep = dev_data->ats.qdep; |
1047 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
3fa43655 | 1048 | |
ea61cddb | 1049 | build_inv_iotlb_pages(&cmd, dev_data->devid, qdep, address, size); |
b00d3bcf JR |
1050 | |
1051 | return iommu_queue_command(iommu, &cmd); | |
3fa43655 JR |
1052 | } |
1053 | ||
431b2a20 | 1054 | /* |
431b2a20 | 1055 | * Command send function for invalidating a device table entry |
431b2a20 | 1056 | */ |
6c542047 | 1057 | static int device_flush_dte(struct iommu_dev_data *dev_data) |
a19ae1ec | 1058 | { |
3fa43655 | 1059 | struct amd_iommu *iommu; |
ee2fa743 | 1060 | int ret; |
a19ae1ec | 1061 | |
6c542047 | 1062 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
a19ae1ec | 1063 | |
f62dda66 | 1064 | ret = iommu_flush_dte(iommu, dev_data->devid); |
cb41ed85 JR |
1065 | if (ret) |
1066 | return ret; | |
1067 | ||
ea61cddb | 1068 | if (dev_data->ats.enabled) |
6c542047 | 1069 | ret = device_flush_iotlb(dev_data, 0, ~0UL); |
ee2fa743 | 1070 | |
ee2fa743 | 1071 | return ret; |
a19ae1ec JR |
1072 | } |
1073 | ||
431b2a20 JR |
1074 | /* |
1075 | * TLB invalidation function which is called from the mapping functions. | |
1076 | * It invalidates a single PTE if the range to flush is within a single | |
1077 | * page. Otherwise it flushes the whole TLB of the IOMMU. | |
1078 | */ | |
17b124bf JR |
1079 | static void __domain_flush_pages(struct protection_domain *domain, |
1080 | u64 address, size_t size, int pde) | |
a19ae1ec | 1081 | { |
cb41ed85 | 1082 | struct iommu_dev_data *dev_data; |
11b6402c JR |
1083 | struct iommu_cmd cmd; |
1084 | int ret = 0, i; | |
a19ae1ec | 1085 | |
11b6402c | 1086 | build_inv_iommu_pages(&cmd, address, size, domain->id, pde); |
999ba417 | 1087 | |
6de8ad9b JR |
1088 | for (i = 0; i < amd_iommus_present; ++i) { |
1089 | if (!domain->dev_iommu[i]) | |
1090 | continue; | |
1091 | ||
1092 | /* | |
1093 | * Devices of this domain are behind this IOMMU | |
1094 | * We need a TLB flush | |
1095 | */ | |
11b6402c | 1096 | ret |= iommu_queue_command(amd_iommus[i], &cmd); |
6de8ad9b JR |
1097 | } |
1098 | ||
cb41ed85 | 1099 | list_for_each_entry(dev_data, &domain->dev_list, list) { |
cb41ed85 | 1100 | |
ea61cddb | 1101 | if (!dev_data->ats.enabled) |
cb41ed85 JR |
1102 | continue; |
1103 | ||
6c542047 | 1104 | ret |= device_flush_iotlb(dev_data, address, size); |
cb41ed85 JR |
1105 | } |
1106 | ||
11b6402c | 1107 | WARN_ON(ret); |
6de8ad9b JR |
1108 | } |
1109 | ||
17b124bf JR |
1110 | static void domain_flush_pages(struct protection_domain *domain, |
1111 | u64 address, size_t size) | |
6de8ad9b | 1112 | { |
17b124bf | 1113 | __domain_flush_pages(domain, address, size, 0); |
a19ae1ec | 1114 | } |
b6c02715 | 1115 | |
1c655773 | 1116 | /* Flush the whole IO/TLB for a given protection domain */ |
17b124bf | 1117 | static void domain_flush_tlb(struct protection_domain *domain) |
1c655773 | 1118 | { |
17b124bf | 1119 | __domain_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 0); |
1c655773 JR |
1120 | } |
1121 | ||
42a49f96 | 1122 | /* Flush the whole IO/TLB for a given protection domain - including PDE */ |
17b124bf | 1123 | static void domain_flush_tlb_pde(struct protection_domain *domain) |
42a49f96 | 1124 | { |
17b124bf | 1125 | __domain_flush_pages(domain, 0, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, 1); |
42a49f96 CW |
1126 | } |
1127 | ||
17b124bf | 1128 | static void domain_flush_complete(struct protection_domain *domain) |
b00d3bcf | 1129 | { |
17b124bf | 1130 | int i; |
18811f55 | 1131 | |
17b124bf JR |
1132 | for (i = 0; i < amd_iommus_present; ++i) { |
1133 | if (!domain->dev_iommu[i]) | |
1134 | continue; | |
bfd1be18 | 1135 | |
17b124bf JR |
1136 | /* |
1137 | * Devices of this domain are behind this IOMMU | |
1138 | * We need to wait for completion of all commands. | |
1139 | */ | |
1140 | iommu_completion_wait(amd_iommus[i]); | |
bfd1be18 | 1141 | } |
e394d72a JR |
1142 | } |
1143 | ||
b00d3bcf | 1144 | |
09b42804 | 1145 | /* |
b00d3bcf | 1146 | * This function flushes the DTEs for all devices in domain |
09b42804 | 1147 | */ |
17b124bf | 1148 | static void domain_flush_devices(struct protection_domain *domain) |
e394d72a | 1149 | { |
b00d3bcf | 1150 | struct iommu_dev_data *dev_data; |
b26e81b8 | 1151 | |
b00d3bcf | 1152 | list_for_each_entry(dev_data, &domain->dev_list, list) |
6c542047 | 1153 | device_flush_dte(dev_data); |
a345b23b JR |
1154 | } |
1155 | ||
431b2a20 JR |
1156 | /**************************************************************************** |
1157 | * | |
1158 | * The functions below are used the create the page table mappings for | |
1159 | * unity mapped regions. | |
1160 | * | |
1161 | ****************************************************************************/ | |
1162 | ||
308973d3 JR |
1163 | /* |
1164 | * This function is used to add another level to an IO page table. Adding | |
1165 | * another level increases the size of the address space by 9 bits to a size up | |
1166 | * to 64 bits. | |
1167 | */ | |
1168 | static bool increase_address_space(struct protection_domain *domain, | |
1169 | gfp_t gfp) | |
1170 | { | |
1171 | u64 *pte; | |
1172 | ||
1173 | if (domain->mode == PAGE_MODE_6_LEVEL) | |
1174 | /* address space already 64 bit large */ | |
1175 | return false; | |
1176 | ||
1177 | pte = (void *)get_zeroed_page(gfp); | |
1178 | if (!pte) | |
1179 | return false; | |
1180 | ||
1181 | *pte = PM_LEVEL_PDE(domain->mode, | |
1182 | virt_to_phys(domain->pt_root)); | |
1183 | domain->pt_root = pte; | |
1184 | domain->mode += 1; | |
1185 | domain->updated = true; | |
1186 | ||
1187 | return true; | |
1188 | } | |
1189 | ||
1190 | static u64 *alloc_pte(struct protection_domain *domain, | |
1191 | unsigned long address, | |
cbb9d729 | 1192 | unsigned long page_size, |
308973d3 JR |
1193 | u64 **pte_page, |
1194 | gfp_t gfp) | |
1195 | { | |
cbb9d729 | 1196 | int level, end_lvl; |
308973d3 | 1197 | u64 *pte, *page; |
cbb9d729 JR |
1198 | |
1199 | BUG_ON(!is_power_of_2(page_size)); | |
308973d3 JR |
1200 | |
1201 | while (address > PM_LEVEL_SIZE(domain->mode)) | |
1202 | increase_address_space(domain, gfp); | |
1203 | ||
cbb9d729 JR |
1204 | level = domain->mode - 1; |
1205 | pte = &domain->pt_root[PM_LEVEL_INDEX(level, address)]; | |
1206 | address = PAGE_SIZE_ALIGN(address, page_size); | |
1207 | end_lvl = PAGE_SIZE_LEVEL(page_size); | |
308973d3 JR |
1208 | |
1209 | while (level > end_lvl) { | |
1210 | if (!IOMMU_PTE_PRESENT(*pte)) { | |
1211 | page = (u64 *)get_zeroed_page(gfp); | |
1212 | if (!page) | |
1213 | return NULL; | |
1214 | *pte = PM_LEVEL_PDE(level, virt_to_phys(page)); | |
1215 | } | |
1216 | ||
cbb9d729 JR |
1217 | /* No level skipping support yet */ |
1218 | if (PM_PTE_LEVEL(*pte) != level) | |
1219 | return NULL; | |
1220 | ||
308973d3 JR |
1221 | level -= 1; |
1222 | ||
1223 | pte = IOMMU_PTE_PAGE(*pte); | |
1224 | ||
1225 | if (pte_page && level == end_lvl) | |
1226 | *pte_page = pte; | |
1227 | ||
1228 | pte = &pte[PM_LEVEL_INDEX(level, address)]; | |
1229 | } | |
1230 | ||
1231 | return pte; | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * This function checks if there is a PTE for a given dma address. If | |
1236 | * there is one, it returns the pointer to it. | |
1237 | */ | |
24cd7723 | 1238 | static u64 *fetch_pte(struct protection_domain *domain, unsigned long address) |
308973d3 JR |
1239 | { |
1240 | int level; | |
1241 | u64 *pte; | |
1242 | ||
24cd7723 JR |
1243 | if (address > PM_LEVEL_SIZE(domain->mode)) |
1244 | return NULL; | |
1245 | ||
1246 | level = domain->mode - 1; | |
1247 | pte = &domain->pt_root[PM_LEVEL_INDEX(level, address)]; | |
308973d3 | 1248 | |
24cd7723 JR |
1249 | while (level > 0) { |
1250 | ||
1251 | /* Not Present */ | |
308973d3 JR |
1252 | if (!IOMMU_PTE_PRESENT(*pte)) |
1253 | return NULL; | |
1254 | ||
24cd7723 JR |
1255 | /* Large PTE */ |
1256 | if (PM_PTE_LEVEL(*pte) == 0x07) { | |
1257 | unsigned long pte_mask, __pte; | |
1258 | ||
1259 | /* | |
1260 | * If we have a series of large PTEs, make | |
1261 | * sure to return a pointer to the first one. | |
1262 | */ | |
1263 | pte_mask = PTE_PAGE_SIZE(*pte); | |
1264 | pte_mask = ~((PAGE_SIZE_PTE_COUNT(pte_mask) << 3) - 1); | |
1265 | __pte = ((unsigned long)pte) & pte_mask; | |
1266 | ||
1267 | return (u64 *)__pte; | |
1268 | } | |
1269 | ||
1270 | /* No level skipping support yet */ | |
1271 | if (PM_PTE_LEVEL(*pte) != level) | |
1272 | return NULL; | |
1273 | ||
308973d3 JR |
1274 | level -= 1; |
1275 | ||
24cd7723 | 1276 | /* Walk to the next level */ |
308973d3 JR |
1277 | pte = IOMMU_PTE_PAGE(*pte); |
1278 | pte = &pte[PM_LEVEL_INDEX(level, address)]; | |
308973d3 JR |
1279 | } |
1280 | ||
1281 | return pte; | |
1282 | } | |
1283 | ||
431b2a20 JR |
1284 | /* |
1285 | * Generic mapping functions. It maps a physical address into a DMA | |
1286 | * address space. It allocates the page table pages if necessary. | |
1287 | * In the future it can be extended to a generic mapping function | |
1288 | * supporting all features of AMD IOMMU page tables like level skipping | |
1289 | * and full 64 bit address spaces. | |
1290 | */ | |
38e817fe JR |
1291 | static int iommu_map_page(struct protection_domain *dom, |
1292 | unsigned long bus_addr, | |
1293 | unsigned long phys_addr, | |
abdc5eb3 | 1294 | int prot, |
cbb9d729 | 1295 | unsigned long page_size) |
bd0e5211 | 1296 | { |
8bda3092 | 1297 | u64 __pte, *pte; |
cbb9d729 | 1298 | int i, count; |
abdc5eb3 | 1299 | |
bad1cac2 | 1300 | if (!(prot & IOMMU_PROT_MASK)) |
bd0e5211 JR |
1301 | return -EINVAL; |
1302 | ||
cbb9d729 JR |
1303 | bus_addr = PAGE_ALIGN(bus_addr); |
1304 | phys_addr = PAGE_ALIGN(phys_addr); | |
1305 | count = PAGE_SIZE_PTE_COUNT(page_size); | |
1306 | pte = alloc_pte(dom, bus_addr, page_size, NULL, GFP_KERNEL); | |
1307 | ||
1308 | for (i = 0; i < count; ++i) | |
1309 | if (IOMMU_PTE_PRESENT(pte[i])) | |
1310 | return -EBUSY; | |
bd0e5211 | 1311 | |
cbb9d729 JR |
1312 | if (page_size > PAGE_SIZE) { |
1313 | __pte = PAGE_SIZE_PTE(phys_addr, page_size); | |
1314 | __pte |= PM_LEVEL_ENC(7) | IOMMU_PTE_P | IOMMU_PTE_FC; | |
1315 | } else | |
1316 | __pte = phys_addr | IOMMU_PTE_P | IOMMU_PTE_FC; | |
bd0e5211 | 1317 | |
bd0e5211 JR |
1318 | if (prot & IOMMU_PROT_IR) |
1319 | __pte |= IOMMU_PTE_IR; | |
1320 | if (prot & IOMMU_PROT_IW) | |
1321 | __pte |= IOMMU_PTE_IW; | |
1322 | ||
cbb9d729 JR |
1323 | for (i = 0; i < count; ++i) |
1324 | pte[i] = __pte; | |
bd0e5211 | 1325 | |
04bfdd84 JR |
1326 | update_domain(dom); |
1327 | ||
bd0e5211 JR |
1328 | return 0; |
1329 | } | |
1330 | ||
24cd7723 JR |
1331 | static unsigned long iommu_unmap_page(struct protection_domain *dom, |
1332 | unsigned long bus_addr, | |
1333 | unsigned long page_size) | |
eb74ff6c | 1334 | { |
24cd7723 JR |
1335 | unsigned long long unmap_size, unmapped; |
1336 | u64 *pte; | |
1337 | ||
1338 | BUG_ON(!is_power_of_2(page_size)); | |
1339 | ||
1340 | unmapped = 0; | |
eb74ff6c | 1341 | |
24cd7723 JR |
1342 | while (unmapped < page_size) { |
1343 | ||
1344 | pte = fetch_pte(dom, bus_addr); | |
1345 | ||
1346 | if (!pte) { | |
1347 | /* | |
1348 | * No PTE for this address | |
1349 | * move forward in 4kb steps | |
1350 | */ | |
1351 | unmap_size = PAGE_SIZE; | |
1352 | } else if (PM_PTE_LEVEL(*pte) == 0) { | |
1353 | /* 4kb PTE found for this address */ | |
1354 | unmap_size = PAGE_SIZE; | |
1355 | *pte = 0ULL; | |
1356 | } else { | |
1357 | int count, i; | |
1358 | ||
1359 | /* Large PTE found which maps this address */ | |
1360 | unmap_size = PTE_PAGE_SIZE(*pte); | |
1361 | count = PAGE_SIZE_PTE_COUNT(unmap_size); | |
1362 | for (i = 0; i < count; i++) | |
1363 | pte[i] = 0ULL; | |
1364 | } | |
1365 | ||
1366 | bus_addr = (bus_addr & ~(unmap_size - 1)) + unmap_size; | |
1367 | unmapped += unmap_size; | |
1368 | } | |
1369 | ||
1370 | BUG_ON(!is_power_of_2(unmapped)); | |
eb74ff6c | 1371 | |
24cd7723 | 1372 | return unmapped; |
eb74ff6c | 1373 | } |
eb74ff6c | 1374 | |
431b2a20 JR |
1375 | /* |
1376 | * This function checks if a specific unity mapping entry is needed for | |
1377 | * this specific IOMMU. | |
1378 | */ | |
bd0e5211 JR |
1379 | static int iommu_for_unity_map(struct amd_iommu *iommu, |
1380 | struct unity_map_entry *entry) | |
1381 | { | |
1382 | u16 bdf, i; | |
1383 | ||
1384 | for (i = entry->devid_start; i <= entry->devid_end; ++i) { | |
1385 | bdf = amd_iommu_alias_table[i]; | |
1386 | if (amd_iommu_rlookup_table[bdf] == iommu) | |
1387 | return 1; | |
1388 | } | |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | ||
431b2a20 JR |
1393 | /* |
1394 | * This function actually applies the mapping to the page table of the | |
1395 | * dma_ops domain. | |
1396 | */ | |
bd0e5211 JR |
1397 | static int dma_ops_unity_map(struct dma_ops_domain *dma_dom, |
1398 | struct unity_map_entry *e) | |
1399 | { | |
1400 | u64 addr; | |
1401 | int ret; | |
1402 | ||
1403 | for (addr = e->address_start; addr < e->address_end; | |
1404 | addr += PAGE_SIZE) { | |
abdc5eb3 | 1405 | ret = iommu_map_page(&dma_dom->domain, addr, addr, e->prot, |
cbb9d729 | 1406 | PAGE_SIZE); |
bd0e5211 JR |
1407 | if (ret) |
1408 | return ret; | |
1409 | /* | |
1410 | * if unity mapping is in aperture range mark the page | |
1411 | * as allocated in the aperture | |
1412 | */ | |
1413 | if (addr < dma_dom->aperture_size) | |
c3239567 | 1414 | __set_bit(addr >> PAGE_SHIFT, |
384de729 | 1415 | dma_dom->aperture[0]->bitmap); |
bd0e5211 JR |
1416 | } |
1417 | ||
1418 | return 0; | |
1419 | } | |
1420 | ||
171e7b37 JR |
1421 | /* |
1422 | * Init the unity mappings for a specific IOMMU in the system | |
1423 | * | |
1424 | * Basically iterates over all unity mapping entries and applies them to | |
1425 | * the default domain DMA of that IOMMU if necessary. | |
1426 | */ | |
1427 | static int iommu_init_unity_mappings(struct amd_iommu *iommu) | |
1428 | { | |
1429 | struct unity_map_entry *entry; | |
1430 | int ret; | |
1431 | ||
1432 | list_for_each_entry(entry, &amd_iommu_unity_map, list) { | |
1433 | if (!iommu_for_unity_map(iommu, entry)) | |
1434 | continue; | |
1435 | ret = dma_ops_unity_map(iommu->default_dom, entry); | |
1436 | if (ret) | |
1437 | return ret; | |
1438 | } | |
1439 | ||
1440 | return 0; | |
1441 | } | |
1442 | ||
431b2a20 JR |
1443 | /* |
1444 | * Inits the unity mappings required for a specific device | |
1445 | */ | |
bd0e5211 JR |
1446 | static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom, |
1447 | u16 devid) | |
1448 | { | |
1449 | struct unity_map_entry *e; | |
1450 | int ret; | |
1451 | ||
1452 | list_for_each_entry(e, &amd_iommu_unity_map, list) { | |
1453 | if (!(devid >= e->devid_start && devid <= e->devid_end)) | |
1454 | continue; | |
1455 | ret = dma_ops_unity_map(dma_dom, e); | |
1456 | if (ret) | |
1457 | return ret; | |
1458 | } | |
1459 | ||
1460 | return 0; | |
1461 | } | |
1462 | ||
431b2a20 JR |
1463 | /**************************************************************************** |
1464 | * | |
1465 | * The next functions belong to the address allocator for the dma_ops | |
1466 | * interface functions. They work like the allocators in the other IOMMU | |
1467 | * drivers. Its basically a bitmap which marks the allocated pages in | |
1468 | * the aperture. Maybe it could be enhanced in the future to a more | |
1469 | * efficient allocator. | |
1470 | * | |
1471 | ****************************************************************************/ | |
d3086444 | 1472 | |
431b2a20 | 1473 | /* |
384de729 | 1474 | * The address allocator core functions. |
431b2a20 JR |
1475 | * |
1476 | * called with domain->lock held | |
1477 | */ | |
384de729 | 1478 | |
171e7b37 JR |
1479 | /* |
1480 | * Used to reserve address ranges in the aperture (e.g. for exclusion | |
1481 | * ranges. | |
1482 | */ | |
1483 | static void dma_ops_reserve_addresses(struct dma_ops_domain *dom, | |
1484 | unsigned long start_page, | |
1485 | unsigned int pages) | |
1486 | { | |
1487 | unsigned int i, last_page = dom->aperture_size >> PAGE_SHIFT; | |
1488 | ||
1489 | if (start_page + pages > last_page) | |
1490 | pages = last_page - start_page; | |
1491 | ||
1492 | for (i = start_page; i < start_page + pages; ++i) { | |
1493 | int index = i / APERTURE_RANGE_PAGES; | |
1494 | int page = i % APERTURE_RANGE_PAGES; | |
1495 | __set_bit(page, dom->aperture[index]->bitmap); | |
1496 | } | |
1497 | } | |
1498 | ||
9cabe89b JR |
1499 | /* |
1500 | * This function is used to add a new aperture range to an existing | |
1501 | * aperture in case of dma_ops domain allocation or address allocation | |
1502 | * failure. | |
1503 | */ | |
576175c2 | 1504 | static int alloc_new_range(struct dma_ops_domain *dma_dom, |
9cabe89b JR |
1505 | bool populate, gfp_t gfp) |
1506 | { | |
1507 | int index = dma_dom->aperture_size >> APERTURE_RANGE_SHIFT; | |
576175c2 | 1508 | struct amd_iommu *iommu; |
17f5b569 | 1509 | unsigned long i, old_size; |
9cabe89b | 1510 | |
f5e9705c JR |
1511 | #ifdef CONFIG_IOMMU_STRESS |
1512 | populate = false; | |
1513 | #endif | |
1514 | ||
9cabe89b JR |
1515 | if (index >= APERTURE_MAX_RANGES) |
1516 | return -ENOMEM; | |
1517 | ||
1518 | dma_dom->aperture[index] = kzalloc(sizeof(struct aperture_range), gfp); | |
1519 | if (!dma_dom->aperture[index]) | |
1520 | return -ENOMEM; | |
1521 | ||
1522 | dma_dom->aperture[index]->bitmap = (void *)get_zeroed_page(gfp); | |
1523 | if (!dma_dom->aperture[index]->bitmap) | |
1524 | goto out_free; | |
1525 | ||
1526 | dma_dom->aperture[index]->offset = dma_dom->aperture_size; | |
1527 | ||
1528 | if (populate) { | |
1529 | unsigned long address = dma_dom->aperture_size; | |
1530 | int i, num_ptes = APERTURE_RANGE_PAGES / 512; | |
1531 | u64 *pte, *pte_page; | |
1532 | ||
1533 | for (i = 0; i < num_ptes; ++i) { | |
cbb9d729 | 1534 | pte = alloc_pte(&dma_dom->domain, address, PAGE_SIZE, |
9cabe89b JR |
1535 | &pte_page, gfp); |
1536 | if (!pte) | |
1537 | goto out_free; | |
1538 | ||
1539 | dma_dom->aperture[index]->pte_pages[i] = pte_page; | |
1540 | ||
1541 | address += APERTURE_RANGE_SIZE / 64; | |
1542 | } | |
1543 | } | |
1544 | ||
17f5b569 | 1545 | old_size = dma_dom->aperture_size; |
9cabe89b JR |
1546 | dma_dom->aperture_size += APERTURE_RANGE_SIZE; |
1547 | ||
17f5b569 JR |
1548 | /* Reserve address range used for MSI messages */ |
1549 | if (old_size < MSI_ADDR_BASE_LO && | |
1550 | dma_dom->aperture_size > MSI_ADDR_BASE_LO) { | |
1551 | unsigned long spage; | |
1552 | int pages; | |
1553 | ||
1554 | pages = iommu_num_pages(MSI_ADDR_BASE_LO, 0x10000, PAGE_SIZE); | |
1555 | spage = MSI_ADDR_BASE_LO >> PAGE_SHIFT; | |
1556 | ||
1557 | dma_ops_reserve_addresses(dma_dom, spage, pages); | |
1558 | } | |
1559 | ||
b595076a | 1560 | /* Initialize the exclusion range if necessary */ |
576175c2 JR |
1561 | for_each_iommu(iommu) { |
1562 | if (iommu->exclusion_start && | |
1563 | iommu->exclusion_start >= dma_dom->aperture[index]->offset | |
1564 | && iommu->exclusion_start < dma_dom->aperture_size) { | |
1565 | unsigned long startpage; | |
1566 | int pages = iommu_num_pages(iommu->exclusion_start, | |
1567 | iommu->exclusion_length, | |
1568 | PAGE_SIZE); | |
1569 | startpage = iommu->exclusion_start >> PAGE_SHIFT; | |
1570 | dma_ops_reserve_addresses(dma_dom, startpage, pages); | |
1571 | } | |
00cd122a JR |
1572 | } |
1573 | ||
1574 | /* | |
1575 | * Check for areas already mapped as present in the new aperture | |
1576 | * range and mark those pages as reserved in the allocator. Such | |
1577 | * mappings may already exist as a result of requested unity | |
1578 | * mappings for devices. | |
1579 | */ | |
1580 | for (i = dma_dom->aperture[index]->offset; | |
1581 | i < dma_dom->aperture_size; | |
1582 | i += PAGE_SIZE) { | |
24cd7723 | 1583 | u64 *pte = fetch_pte(&dma_dom->domain, i); |
00cd122a JR |
1584 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) |
1585 | continue; | |
1586 | ||
fcd0861d | 1587 | dma_ops_reserve_addresses(dma_dom, i >> PAGE_SHIFT, 1); |
00cd122a JR |
1588 | } |
1589 | ||
04bfdd84 JR |
1590 | update_domain(&dma_dom->domain); |
1591 | ||
9cabe89b JR |
1592 | return 0; |
1593 | ||
1594 | out_free: | |
04bfdd84 JR |
1595 | update_domain(&dma_dom->domain); |
1596 | ||
9cabe89b JR |
1597 | free_page((unsigned long)dma_dom->aperture[index]->bitmap); |
1598 | ||
1599 | kfree(dma_dom->aperture[index]); | |
1600 | dma_dom->aperture[index] = NULL; | |
1601 | ||
1602 | return -ENOMEM; | |
1603 | } | |
1604 | ||
384de729 JR |
1605 | static unsigned long dma_ops_area_alloc(struct device *dev, |
1606 | struct dma_ops_domain *dom, | |
1607 | unsigned int pages, | |
1608 | unsigned long align_mask, | |
1609 | u64 dma_mask, | |
1610 | unsigned long start) | |
1611 | { | |
803b8cb4 | 1612 | unsigned long next_bit = dom->next_address % APERTURE_RANGE_SIZE; |
384de729 JR |
1613 | int max_index = dom->aperture_size >> APERTURE_RANGE_SHIFT; |
1614 | int i = start >> APERTURE_RANGE_SHIFT; | |
1615 | unsigned long boundary_size; | |
1616 | unsigned long address = -1; | |
1617 | unsigned long limit; | |
1618 | ||
803b8cb4 JR |
1619 | next_bit >>= PAGE_SHIFT; |
1620 | ||
384de729 JR |
1621 | boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1, |
1622 | PAGE_SIZE) >> PAGE_SHIFT; | |
1623 | ||
1624 | for (;i < max_index; ++i) { | |
1625 | unsigned long offset = dom->aperture[i]->offset >> PAGE_SHIFT; | |
1626 | ||
1627 | if (dom->aperture[i]->offset >= dma_mask) | |
1628 | break; | |
1629 | ||
1630 | limit = iommu_device_max_index(APERTURE_RANGE_PAGES, offset, | |
1631 | dma_mask >> PAGE_SHIFT); | |
1632 | ||
1633 | address = iommu_area_alloc(dom->aperture[i]->bitmap, | |
1634 | limit, next_bit, pages, 0, | |
1635 | boundary_size, align_mask); | |
1636 | if (address != -1) { | |
1637 | address = dom->aperture[i]->offset + | |
1638 | (address << PAGE_SHIFT); | |
803b8cb4 | 1639 | dom->next_address = address + (pages << PAGE_SHIFT); |
384de729 JR |
1640 | break; |
1641 | } | |
1642 | ||
1643 | next_bit = 0; | |
1644 | } | |
1645 | ||
1646 | return address; | |
1647 | } | |
1648 | ||
d3086444 JR |
1649 | static unsigned long dma_ops_alloc_addresses(struct device *dev, |
1650 | struct dma_ops_domain *dom, | |
6d4f343f | 1651 | unsigned int pages, |
832a90c3 JR |
1652 | unsigned long align_mask, |
1653 | u64 dma_mask) | |
d3086444 | 1654 | { |
d3086444 | 1655 | unsigned long address; |
d3086444 | 1656 | |
fe16f088 JR |
1657 | #ifdef CONFIG_IOMMU_STRESS |
1658 | dom->next_address = 0; | |
1659 | dom->need_flush = true; | |
1660 | #endif | |
d3086444 | 1661 | |
384de729 | 1662 | address = dma_ops_area_alloc(dev, dom, pages, align_mask, |
803b8cb4 | 1663 | dma_mask, dom->next_address); |
d3086444 | 1664 | |
1c655773 | 1665 | if (address == -1) { |
803b8cb4 | 1666 | dom->next_address = 0; |
384de729 JR |
1667 | address = dma_ops_area_alloc(dev, dom, pages, align_mask, |
1668 | dma_mask, 0); | |
1c655773 JR |
1669 | dom->need_flush = true; |
1670 | } | |
d3086444 | 1671 | |
384de729 | 1672 | if (unlikely(address == -1)) |
8fd524b3 | 1673 | address = DMA_ERROR_CODE; |
d3086444 JR |
1674 | |
1675 | WARN_ON((address + (PAGE_SIZE*pages)) > dom->aperture_size); | |
1676 | ||
1677 | return address; | |
1678 | } | |
1679 | ||
431b2a20 JR |
1680 | /* |
1681 | * The address free function. | |
1682 | * | |
1683 | * called with domain->lock held | |
1684 | */ | |
d3086444 JR |
1685 | static void dma_ops_free_addresses(struct dma_ops_domain *dom, |
1686 | unsigned long address, | |
1687 | unsigned int pages) | |
1688 | { | |
384de729 JR |
1689 | unsigned i = address >> APERTURE_RANGE_SHIFT; |
1690 | struct aperture_range *range = dom->aperture[i]; | |
80be308d | 1691 | |
384de729 JR |
1692 | BUG_ON(i >= APERTURE_MAX_RANGES || range == NULL); |
1693 | ||
47bccd6b JR |
1694 | #ifdef CONFIG_IOMMU_STRESS |
1695 | if (i < 4) | |
1696 | return; | |
1697 | #endif | |
80be308d | 1698 | |
803b8cb4 | 1699 | if (address >= dom->next_address) |
80be308d | 1700 | dom->need_flush = true; |
384de729 JR |
1701 | |
1702 | address = (address % APERTURE_RANGE_SIZE) >> PAGE_SHIFT; | |
803b8cb4 | 1703 | |
a66022c4 | 1704 | bitmap_clear(range->bitmap, address, pages); |
384de729 | 1705 | |
d3086444 JR |
1706 | } |
1707 | ||
431b2a20 JR |
1708 | /**************************************************************************** |
1709 | * | |
1710 | * The next functions belong to the domain allocation. A domain is | |
1711 | * allocated for every IOMMU as the default domain. If device isolation | |
1712 | * is enabled, every device get its own domain. The most important thing | |
1713 | * about domains is the page table mapping the DMA address space they | |
1714 | * contain. | |
1715 | * | |
1716 | ****************************************************************************/ | |
1717 | ||
aeb26f55 JR |
1718 | /* |
1719 | * This function adds a protection domain to the global protection domain list | |
1720 | */ | |
1721 | static void add_domain_to_list(struct protection_domain *domain) | |
1722 | { | |
1723 | unsigned long flags; | |
1724 | ||
1725 | spin_lock_irqsave(&amd_iommu_pd_lock, flags); | |
1726 | list_add(&domain->list, &amd_iommu_pd_list); | |
1727 | spin_unlock_irqrestore(&amd_iommu_pd_lock, flags); | |
1728 | } | |
1729 | ||
1730 | /* | |
1731 | * This function removes a protection domain to the global | |
1732 | * protection domain list | |
1733 | */ | |
1734 | static void del_domain_from_list(struct protection_domain *domain) | |
1735 | { | |
1736 | unsigned long flags; | |
1737 | ||
1738 | spin_lock_irqsave(&amd_iommu_pd_lock, flags); | |
1739 | list_del(&domain->list); | |
1740 | spin_unlock_irqrestore(&amd_iommu_pd_lock, flags); | |
1741 | } | |
1742 | ||
ec487d1a JR |
1743 | static u16 domain_id_alloc(void) |
1744 | { | |
1745 | unsigned long flags; | |
1746 | int id; | |
1747 | ||
1748 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
1749 | id = find_first_zero_bit(amd_iommu_pd_alloc_bitmap, MAX_DOMAIN_ID); | |
1750 | BUG_ON(id == 0); | |
1751 | if (id > 0 && id < MAX_DOMAIN_ID) | |
1752 | __set_bit(id, amd_iommu_pd_alloc_bitmap); | |
1753 | else | |
1754 | id = 0; | |
1755 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
1756 | ||
1757 | return id; | |
1758 | } | |
1759 | ||
a2acfb75 JR |
1760 | static void domain_id_free(int id) |
1761 | { | |
1762 | unsigned long flags; | |
1763 | ||
1764 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
1765 | if (id > 0 && id < MAX_DOMAIN_ID) | |
1766 | __clear_bit(id, amd_iommu_pd_alloc_bitmap); | |
1767 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
1768 | } | |
a2acfb75 | 1769 | |
86db2e5d | 1770 | static void free_pagetable(struct protection_domain *domain) |
ec487d1a JR |
1771 | { |
1772 | int i, j; | |
1773 | u64 *p1, *p2, *p3; | |
1774 | ||
86db2e5d | 1775 | p1 = domain->pt_root; |
ec487d1a JR |
1776 | |
1777 | if (!p1) | |
1778 | return; | |
1779 | ||
1780 | for (i = 0; i < 512; ++i) { | |
1781 | if (!IOMMU_PTE_PRESENT(p1[i])) | |
1782 | continue; | |
1783 | ||
1784 | p2 = IOMMU_PTE_PAGE(p1[i]); | |
3cc3d84b | 1785 | for (j = 0; j < 512; ++j) { |
ec487d1a JR |
1786 | if (!IOMMU_PTE_PRESENT(p2[j])) |
1787 | continue; | |
1788 | p3 = IOMMU_PTE_PAGE(p2[j]); | |
1789 | free_page((unsigned long)p3); | |
1790 | } | |
1791 | ||
1792 | free_page((unsigned long)p2); | |
1793 | } | |
1794 | ||
1795 | free_page((unsigned long)p1); | |
86db2e5d JR |
1796 | |
1797 | domain->pt_root = NULL; | |
ec487d1a JR |
1798 | } |
1799 | ||
b16137b1 JR |
1800 | static void free_gcr3_tbl_level1(u64 *tbl) |
1801 | { | |
1802 | u64 *ptr; | |
1803 | int i; | |
1804 | ||
1805 | for (i = 0; i < 512; ++i) { | |
1806 | if (!(tbl[i] & GCR3_VALID)) | |
1807 | continue; | |
1808 | ||
1809 | ptr = __va(tbl[i] & PAGE_MASK); | |
1810 | ||
1811 | free_page((unsigned long)ptr); | |
1812 | } | |
1813 | } | |
1814 | ||
1815 | static void free_gcr3_tbl_level2(u64 *tbl) | |
1816 | { | |
1817 | u64 *ptr; | |
1818 | int i; | |
1819 | ||
1820 | for (i = 0; i < 512; ++i) { | |
1821 | if (!(tbl[i] & GCR3_VALID)) | |
1822 | continue; | |
1823 | ||
1824 | ptr = __va(tbl[i] & PAGE_MASK); | |
1825 | ||
1826 | free_gcr3_tbl_level1(ptr); | |
1827 | } | |
1828 | } | |
1829 | ||
52815b75 JR |
1830 | static void free_gcr3_table(struct protection_domain *domain) |
1831 | { | |
b16137b1 JR |
1832 | if (domain->glx == 2) |
1833 | free_gcr3_tbl_level2(domain->gcr3_tbl); | |
1834 | else if (domain->glx == 1) | |
1835 | free_gcr3_tbl_level1(domain->gcr3_tbl); | |
1836 | else if (domain->glx != 0) | |
1837 | BUG(); | |
1838 | ||
52815b75 JR |
1839 | free_page((unsigned long)domain->gcr3_tbl); |
1840 | } | |
1841 | ||
431b2a20 JR |
1842 | /* |
1843 | * Free a domain, only used if something went wrong in the | |
1844 | * allocation path and we need to free an already allocated page table | |
1845 | */ | |
ec487d1a JR |
1846 | static void dma_ops_domain_free(struct dma_ops_domain *dom) |
1847 | { | |
384de729 JR |
1848 | int i; |
1849 | ||
ec487d1a JR |
1850 | if (!dom) |
1851 | return; | |
1852 | ||
aeb26f55 JR |
1853 | del_domain_from_list(&dom->domain); |
1854 | ||
86db2e5d | 1855 | free_pagetable(&dom->domain); |
ec487d1a | 1856 | |
384de729 JR |
1857 | for (i = 0; i < APERTURE_MAX_RANGES; ++i) { |
1858 | if (!dom->aperture[i]) | |
1859 | continue; | |
1860 | free_page((unsigned long)dom->aperture[i]->bitmap); | |
1861 | kfree(dom->aperture[i]); | |
1862 | } | |
ec487d1a JR |
1863 | |
1864 | kfree(dom); | |
1865 | } | |
1866 | ||
431b2a20 JR |
1867 | /* |
1868 | * Allocates a new protection domain usable for the dma_ops functions. | |
b595076a | 1869 | * It also initializes the page table and the address allocator data |
431b2a20 JR |
1870 | * structures required for the dma_ops interface |
1871 | */ | |
87a64d52 | 1872 | static struct dma_ops_domain *dma_ops_domain_alloc(void) |
ec487d1a JR |
1873 | { |
1874 | struct dma_ops_domain *dma_dom; | |
ec487d1a JR |
1875 | |
1876 | dma_dom = kzalloc(sizeof(struct dma_ops_domain), GFP_KERNEL); | |
1877 | if (!dma_dom) | |
1878 | return NULL; | |
1879 | ||
1880 | spin_lock_init(&dma_dom->domain.lock); | |
1881 | ||
1882 | dma_dom->domain.id = domain_id_alloc(); | |
1883 | if (dma_dom->domain.id == 0) | |
1884 | goto free_dma_dom; | |
7c392cbe | 1885 | INIT_LIST_HEAD(&dma_dom->domain.dev_list); |
8f7a017c | 1886 | dma_dom->domain.mode = PAGE_MODE_2_LEVEL; |
ec487d1a | 1887 | dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL); |
9fdb19d6 | 1888 | dma_dom->domain.flags = PD_DMA_OPS_MASK; |
ec487d1a JR |
1889 | dma_dom->domain.priv = dma_dom; |
1890 | if (!dma_dom->domain.pt_root) | |
1891 | goto free_dma_dom; | |
ec487d1a | 1892 | |
1c655773 | 1893 | dma_dom->need_flush = false; |
bd60b735 | 1894 | dma_dom->target_dev = 0xffff; |
1c655773 | 1895 | |
aeb26f55 JR |
1896 | add_domain_to_list(&dma_dom->domain); |
1897 | ||
576175c2 | 1898 | if (alloc_new_range(dma_dom, true, GFP_KERNEL)) |
ec487d1a | 1899 | goto free_dma_dom; |
ec487d1a | 1900 | |
431b2a20 | 1901 | /* |
ec487d1a JR |
1902 | * mark the first page as allocated so we never return 0 as |
1903 | * a valid dma-address. So we can use 0 as error value | |
431b2a20 | 1904 | */ |
384de729 | 1905 | dma_dom->aperture[0]->bitmap[0] = 1; |
803b8cb4 | 1906 | dma_dom->next_address = 0; |
ec487d1a | 1907 | |
ec487d1a JR |
1908 | |
1909 | return dma_dom; | |
1910 | ||
1911 | free_dma_dom: | |
1912 | dma_ops_domain_free(dma_dom); | |
1913 | ||
1914 | return NULL; | |
1915 | } | |
1916 | ||
5b28df6f JR |
1917 | /* |
1918 | * little helper function to check whether a given protection domain is a | |
1919 | * dma_ops domain | |
1920 | */ | |
1921 | static bool dma_ops_domain(struct protection_domain *domain) | |
1922 | { | |
1923 | return domain->flags & PD_DMA_OPS_MASK; | |
1924 | } | |
1925 | ||
fd7b5535 | 1926 | static void set_dte_entry(u16 devid, struct protection_domain *domain, bool ats) |
b20ac0d4 | 1927 | { |
132bd68f | 1928 | u64 pte_root = 0; |
ee6c2868 | 1929 | u64 flags = 0; |
863c74eb | 1930 | |
132bd68f JR |
1931 | if (domain->mode != PAGE_MODE_NONE) |
1932 | pte_root = virt_to_phys(domain->pt_root); | |
1933 | ||
38ddf41b JR |
1934 | pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK) |
1935 | << DEV_ENTRY_MODE_SHIFT; | |
1936 | pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV; | |
b20ac0d4 | 1937 | |
ee6c2868 JR |
1938 | flags = amd_iommu_dev_table[devid].data[1]; |
1939 | ||
fd7b5535 JR |
1940 | if (ats) |
1941 | flags |= DTE_FLAG_IOTLB; | |
1942 | ||
52815b75 JR |
1943 | if (domain->flags & PD_IOMMUV2_MASK) { |
1944 | u64 gcr3 = __pa(domain->gcr3_tbl); | |
1945 | u64 glx = domain->glx; | |
1946 | u64 tmp; | |
1947 | ||
1948 | pte_root |= DTE_FLAG_GV; | |
1949 | pte_root |= (glx & DTE_GLX_MASK) << DTE_GLX_SHIFT; | |
1950 | ||
1951 | /* First mask out possible old values for GCR3 table */ | |
1952 | tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B; | |
1953 | flags &= ~tmp; | |
1954 | ||
1955 | tmp = DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C; | |
1956 | flags &= ~tmp; | |
1957 | ||
1958 | /* Encode GCR3 table into DTE */ | |
1959 | tmp = DTE_GCR3_VAL_A(gcr3) << DTE_GCR3_SHIFT_A; | |
1960 | pte_root |= tmp; | |
1961 | ||
1962 | tmp = DTE_GCR3_VAL_B(gcr3) << DTE_GCR3_SHIFT_B; | |
1963 | flags |= tmp; | |
1964 | ||
1965 | tmp = DTE_GCR3_VAL_C(gcr3) << DTE_GCR3_SHIFT_C; | |
1966 | flags |= tmp; | |
1967 | } | |
1968 | ||
ee6c2868 JR |
1969 | flags &= ~(0xffffUL); |
1970 | flags |= domain->id; | |
1971 | ||
1972 | amd_iommu_dev_table[devid].data[1] = flags; | |
1973 | amd_iommu_dev_table[devid].data[0] = pte_root; | |
15898bbc JR |
1974 | } |
1975 | ||
1976 | static void clear_dte_entry(u16 devid) | |
1977 | { | |
15898bbc JR |
1978 | /* remove entry from the device table seen by the hardware */ |
1979 | amd_iommu_dev_table[devid].data[0] = IOMMU_PTE_P | IOMMU_PTE_TV; | |
1980 | amd_iommu_dev_table[devid].data[1] = 0; | |
15898bbc JR |
1981 | |
1982 | amd_iommu_apply_erratum_63(devid); | |
7f760ddd JR |
1983 | } |
1984 | ||
ec9e79ef JR |
1985 | static void do_attach(struct iommu_dev_data *dev_data, |
1986 | struct protection_domain *domain) | |
7f760ddd | 1987 | { |
7f760ddd | 1988 | struct amd_iommu *iommu; |
ec9e79ef | 1989 | bool ats; |
fd7b5535 | 1990 | |
ec9e79ef JR |
1991 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
1992 | ats = dev_data->ats.enabled; | |
7f760ddd JR |
1993 | |
1994 | /* Update data structures */ | |
1995 | dev_data->domain = domain; | |
1996 | list_add(&dev_data->list, &domain->dev_list); | |
f62dda66 | 1997 | set_dte_entry(dev_data->devid, domain, ats); |
7f760ddd JR |
1998 | |
1999 | /* Do reference counting */ | |
2000 | domain->dev_iommu[iommu->index] += 1; | |
2001 | domain->dev_cnt += 1; | |
2002 | ||
2003 | /* Flush the DTE entry */ | |
6c542047 | 2004 | device_flush_dte(dev_data); |
7f760ddd JR |
2005 | } |
2006 | ||
ec9e79ef | 2007 | static void do_detach(struct iommu_dev_data *dev_data) |
7f760ddd | 2008 | { |
7f760ddd | 2009 | struct amd_iommu *iommu; |
7f760ddd | 2010 | |
ec9e79ef | 2011 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
15898bbc JR |
2012 | |
2013 | /* decrease reference counters */ | |
7f760ddd JR |
2014 | dev_data->domain->dev_iommu[iommu->index] -= 1; |
2015 | dev_data->domain->dev_cnt -= 1; | |
2016 | ||
2017 | /* Update data structures */ | |
2018 | dev_data->domain = NULL; | |
2019 | list_del(&dev_data->list); | |
f62dda66 | 2020 | clear_dte_entry(dev_data->devid); |
15898bbc | 2021 | |
7f760ddd | 2022 | /* Flush the DTE entry */ |
6c542047 | 2023 | device_flush_dte(dev_data); |
2b681faf JR |
2024 | } |
2025 | ||
2026 | /* | |
2027 | * If a device is not yet associated with a domain, this function does | |
2028 | * assigns it visible for the hardware | |
2029 | */ | |
ec9e79ef | 2030 | static int __attach_device(struct iommu_dev_data *dev_data, |
15898bbc | 2031 | struct protection_domain *domain) |
2b681faf | 2032 | { |
84fe6c19 | 2033 | int ret; |
657cbb6b | 2034 | |
2b681faf JR |
2035 | /* lock domain */ |
2036 | spin_lock(&domain->lock); | |
2037 | ||
71f77580 JR |
2038 | if (dev_data->alias_data != NULL) { |
2039 | struct iommu_dev_data *alias_data = dev_data->alias_data; | |
15898bbc | 2040 | |
2b02b091 JR |
2041 | /* Some sanity checks */ |
2042 | ret = -EBUSY; | |
2043 | if (alias_data->domain != NULL && | |
2044 | alias_data->domain != domain) | |
2045 | goto out_unlock; | |
eba6ac60 | 2046 | |
2b02b091 JR |
2047 | if (dev_data->domain != NULL && |
2048 | dev_data->domain != domain) | |
2049 | goto out_unlock; | |
15898bbc | 2050 | |
2b02b091 | 2051 | /* Do real assignment */ |
7f760ddd | 2052 | if (alias_data->domain == NULL) |
ec9e79ef | 2053 | do_attach(alias_data, domain); |
24100055 JR |
2054 | |
2055 | atomic_inc(&alias_data->bind); | |
657cbb6b | 2056 | } |
15898bbc | 2057 | |
7f760ddd | 2058 | if (dev_data->domain == NULL) |
ec9e79ef | 2059 | do_attach(dev_data, domain); |
eba6ac60 | 2060 | |
24100055 JR |
2061 | atomic_inc(&dev_data->bind); |
2062 | ||
84fe6c19 JL |
2063 | ret = 0; |
2064 | ||
2065 | out_unlock: | |
2066 | ||
eba6ac60 JR |
2067 | /* ready */ |
2068 | spin_unlock(&domain->lock); | |
15898bbc | 2069 | |
84fe6c19 | 2070 | return ret; |
0feae533 | 2071 | } |
b20ac0d4 | 2072 | |
52815b75 JR |
2073 | |
2074 | static void pdev_iommuv2_disable(struct pci_dev *pdev) | |
2075 | { | |
2076 | pci_disable_ats(pdev); | |
2077 | pci_disable_pri(pdev); | |
2078 | pci_disable_pasid(pdev); | |
2079 | } | |
2080 | ||
6a113ddc JR |
2081 | /* FIXME: Change generic reset-function to do the same */ |
2082 | static int pri_reset_while_enabled(struct pci_dev *pdev) | |
2083 | { | |
2084 | u16 control; | |
2085 | int pos; | |
2086 | ||
46277b75 | 2087 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PRI); |
6a113ddc JR |
2088 | if (!pos) |
2089 | return -EINVAL; | |
2090 | ||
46277b75 JR |
2091 | pci_read_config_word(pdev, pos + PCI_PRI_CTRL, &control); |
2092 | control |= PCI_PRI_CTRL_RESET; | |
2093 | pci_write_config_word(pdev, pos + PCI_PRI_CTRL, control); | |
6a113ddc JR |
2094 | |
2095 | return 0; | |
2096 | } | |
2097 | ||
52815b75 JR |
2098 | static int pdev_iommuv2_enable(struct pci_dev *pdev) |
2099 | { | |
6a113ddc JR |
2100 | bool reset_enable; |
2101 | int reqs, ret; | |
2102 | ||
2103 | /* FIXME: Hardcode number of outstanding requests for now */ | |
2104 | reqs = 32; | |
2105 | if (pdev_pri_erratum(pdev, AMD_PRI_DEV_ERRATUM_LIMIT_REQ_ONE)) | |
2106 | reqs = 1; | |
2107 | reset_enable = pdev_pri_erratum(pdev, AMD_PRI_DEV_ERRATUM_ENABLE_RESET); | |
52815b75 JR |
2108 | |
2109 | /* Only allow access to user-accessible pages */ | |
2110 | ret = pci_enable_pasid(pdev, 0); | |
2111 | if (ret) | |
2112 | goto out_err; | |
2113 | ||
2114 | /* First reset the PRI state of the device */ | |
2115 | ret = pci_reset_pri(pdev); | |
2116 | if (ret) | |
2117 | goto out_err; | |
2118 | ||
6a113ddc JR |
2119 | /* Enable PRI */ |
2120 | ret = pci_enable_pri(pdev, reqs); | |
52815b75 JR |
2121 | if (ret) |
2122 | goto out_err; | |
2123 | ||
6a113ddc JR |
2124 | if (reset_enable) { |
2125 | ret = pri_reset_while_enabled(pdev); | |
2126 | if (ret) | |
2127 | goto out_err; | |
2128 | } | |
2129 | ||
52815b75 JR |
2130 | ret = pci_enable_ats(pdev, PAGE_SHIFT); |
2131 | if (ret) | |
2132 | goto out_err; | |
2133 | ||
2134 | return 0; | |
2135 | ||
2136 | out_err: | |
2137 | pci_disable_pri(pdev); | |
2138 | pci_disable_pasid(pdev); | |
2139 | ||
2140 | return ret; | |
2141 | } | |
2142 | ||
c99afa25 | 2143 | /* FIXME: Move this to PCI code */ |
a3b93121 | 2144 | #define PCI_PRI_TLP_OFF (1 << 15) |
c99afa25 | 2145 | |
98f1ad25 | 2146 | static bool pci_pri_tlp_required(struct pci_dev *pdev) |
c99afa25 | 2147 | { |
a3b93121 | 2148 | u16 status; |
c99afa25 JR |
2149 | int pos; |
2150 | ||
46277b75 | 2151 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PRI); |
c99afa25 JR |
2152 | if (!pos) |
2153 | return false; | |
2154 | ||
a3b93121 | 2155 | pci_read_config_word(pdev, pos + PCI_PRI_STATUS, &status); |
c99afa25 | 2156 | |
a3b93121 | 2157 | return (status & PCI_PRI_TLP_OFF) ? true : false; |
c99afa25 JR |
2158 | } |
2159 | ||
407d733e JR |
2160 | /* |
2161 | * If a device is not yet associated with a domain, this function does | |
2162 | * assigns it visible for the hardware | |
2163 | */ | |
15898bbc JR |
2164 | static int attach_device(struct device *dev, |
2165 | struct protection_domain *domain) | |
0feae533 | 2166 | { |
fd7b5535 | 2167 | struct pci_dev *pdev = to_pci_dev(dev); |
ea61cddb | 2168 | struct iommu_dev_data *dev_data; |
eba6ac60 | 2169 | unsigned long flags; |
15898bbc | 2170 | int ret; |
eba6ac60 | 2171 | |
ea61cddb JR |
2172 | dev_data = get_dev_data(dev); |
2173 | ||
52815b75 JR |
2174 | if (domain->flags & PD_IOMMUV2_MASK) { |
2175 | if (!dev_data->iommu_v2 || !dev_data->passthrough) | |
2176 | return -EINVAL; | |
2177 | ||
2178 | if (pdev_iommuv2_enable(pdev) != 0) | |
2179 | return -EINVAL; | |
2180 | ||
2181 | dev_data->ats.enabled = true; | |
2182 | dev_data->ats.qdep = pci_ats_queue_depth(pdev); | |
c99afa25 | 2183 | dev_data->pri_tlp = pci_pri_tlp_required(pdev); |
52815b75 JR |
2184 | } else if (amd_iommu_iotlb_sup && |
2185 | pci_enable_ats(pdev, PAGE_SHIFT) == 0) { | |
ea61cddb JR |
2186 | dev_data->ats.enabled = true; |
2187 | dev_data->ats.qdep = pci_ats_queue_depth(pdev); | |
2188 | } | |
fd7b5535 | 2189 | |
eba6ac60 | 2190 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); |
ec9e79ef | 2191 | ret = __attach_device(dev_data, domain); |
b20ac0d4 JR |
2192 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); |
2193 | ||
0feae533 JR |
2194 | /* |
2195 | * We might boot into a crash-kernel here. The crashed kernel | |
2196 | * left the caches in the IOMMU dirty. So we have to flush | |
2197 | * here to evict all dirty stuff. | |
2198 | */ | |
17b124bf | 2199 | domain_flush_tlb_pde(domain); |
15898bbc JR |
2200 | |
2201 | return ret; | |
b20ac0d4 JR |
2202 | } |
2203 | ||
355bf553 JR |
2204 | /* |
2205 | * Removes a device from a protection domain (unlocked) | |
2206 | */ | |
ec9e79ef | 2207 | static void __detach_device(struct iommu_dev_data *dev_data) |
355bf553 | 2208 | { |
2ca76279 | 2209 | struct protection_domain *domain; |
7c392cbe | 2210 | unsigned long flags; |
c4596114 | 2211 | |
7f760ddd | 2212 | BUG_ON(!dev_data->domain); |
355bf553 | 2213 | |
2ca76279 JR |
2214 | domain = dev_data->domain; |
2215 | ||
2216 | spin_lock_irqsave(&domain->lock, flags); | |
24100055 | 2217 | |
71f77580 JR |
2218 | if (dev_data->alias_data != NULL) { |
2219 | struct iommu_dev_data *alias_data = dev_data->alias_data; | |
2220 | ||
7f760ddd | 2221 | if (atomic_dec_and_test(&alias_data->bind)) |
ec9e79ef | 2222 | do_detach(alias_data); |
24100055 JR |
2223 | } |
2224 | ||
7f760ddd | 2225 | if (atomic_dec_and_test(&dev_data->bind)) |
ec9e79ef | 2226 | do_detach(dev_data); |
7f760ddd | 2227 | |
2ca76279 | 2228 | spin_unlock_irqrestore(&domain->lock, flags); |
21129f78 JR |
2229 | |
2230 | /* | |
2231 | * If we run in passthrough mode the device must be assigned to the | |
d3ad9373 JR |
2232 | * passthrough domain if it is detached from any other domain. |
2233 | * Make sure we can deassign from the pt_domain itself. | |
21129f78 | 2234 | */ |
5abcdba4 | 2235 | if (dev_data->passthrough && |
d3ad9373 | 2236 | (dev_data->domain == NULL && domain != pt_domain)) |
ec9e79ef | 2237 | __attach_device(dev_data, pt_domain); |
355bf553 JR |
2238 | } |
2239 | ||
2240 | /* | |
2241 | * Removes a device from a protection domain (with devtable_lock held) | |
2242 | */ | |
15898bbc | 2243 | static void detach_device(struct device *dev) |
355bf553 | 2244 | { |
52815b75 | 2245 | struct protection_domain *domain; |
ea61cddb | 2246 | struct iommu_dev_data *dev_data; |
355bf553 JR |
2247 | unsigned long flags; |
2248 | ||
ec9e79ef | 2249 | dev_data = get_dev_data(dev); |
52815b75 | 2250 | domain = dev_data->domain; |
ec9e79ef | 2251 | |
355bf553 JR |
2252 | /* lock device table */ |
2253 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
ec9e79ef | 2254 | __detach_device(dev_data); |
355bf553 | 2255 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); |
fd7b5535 | 2256 | |
52815b75 JR |
2257 | if (domain->flags & PD_IOMMUV2_MASK) |
2258 | pdev_iommuv2_disable(to_pci_dev(dev)); | |
2259 | else if (dev_data->ats.enabled) | |
ea61cddb | 2260 | pci_disable_ats(to_pci_dev(dev)); |
52815b75 JR |
2261 | |
2262 | dev_data->ats.enabled = false; | |
355bf553 | 2263 | } |
e275a2a0 | 2264 | |
15898bbc JR |
2265 | /* |
2266 | * Find out the protection domain structure for a given PCI device. This | |
2267 | * will give us the pointer to the page table root for example. | |
2268 | */ | |
2269 | static struct protection_domain *domain_for_device(struct device *dev) | |
2270 | { | |
71f77580 | 2271 | struct iommu_dev_data *dev_data; |
2b02b091 | 2272 | struct protection_domain *dom = NULL; |
15898bbc | 2273 | unsigned long flags; |
15898bbc | 2274 | |
657cbb6b | 2275 | dev_data = get_dev_data(dev); |
15898bbc | 2276 | |
2b02b091 JR |
2277 | if (dev_data->domain) |
2278 | return dev_data->domain; | |
15898bbc | 2279 | |
71f77580 JR |
2280 | if (dev_data->alias_data != NULL) { |
2281 | struct iommu_dev_data *alias_data = dev_data->alias_data; | |
2b02b091 JR |
2282 | |
2283 | read_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
2284 | if (alias_data->domain != NULL) { | |
2285 | __attach_device(dev_data, alias_data->domain); | |
2286 | dom = alias_data->domain; | |
2287 | } | |
2288 | read_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
2289 | } | |
15898bbc JR |
2290 | |
2291 | return dom; | |
2292 | } | |
2293 | ||
e275a2a0 JR |
2294 | static int device_change_notifier(struct notifier_block *nb, |
2295 | unsigned long action, void *data) | |
2296 | { | |
e275a2a0 | 2297 | struct dma_ops_domain *dma_domain; |
5abcdba4 JR |
2298 | struct protection_domain *domain; |
2299 | struct iommu_dev_data *dev_data; | |
2300 | struct device *dev = data; | |
e275a2a0 | 2301 | struct amd_iommu *iommu; |
1ac4cbbc | 2302 | unsigned long flags; |
5abcdba4 | 2303 | u16 devid; |
e275a2a0 | 2304 | |
98fc5a69 JR |
2305 | if (!check_device(dev)) |
2306 | return 0; | |
e275a2a0 | 2307 | |
5abcdba4 JR |
2308 | devid = get_device_id(dev); |
2309 | iommu = amd_iommu_rlookup_table[devid]; | |
2310 | dev_data = get_dev_data(dev); | |
e275a2a0 JR |
2311 | |
2312 | switch (action) { | |
c1eee67b | 2313 | case BUS_NOTIFY_UNBOUND_DRIVER: |
657cbb6b JR |
2314 | |
2315 | domain = domain_for_device(dev); | |
2316 | ||
e275a2a0 JR |
2317 | if (!domain) |
2318 | goto out; | |
5abcdba4 | 2319 | if (dev_data->passthrough) |
a1ca331c | 2320 | break; |
15898bbc | 2321 | detach_device(dev); |
1ac4cbbc JR |
2322 | break; |
2323 | case BUS_NOTIFY_ADD_DEVICE: | |
657cbb6b JR |
2324 | |
2325 | iommu_init_device(dev); | |
2326 | ||
2c9195e9 JR |
2327 | /* |
2328 | * dev_data is still NULL and | |
2329 | * got initialized in iommu_init_device | |
2330 | */ | |
2331 | dev_data = get_dev_data(dev); | |
2332 | ||
2333 | if (iommu_pass_through || dev_data->iommu_v2) { | |
2334 | dev_data->passthrough = true; | |
2335 | attach_device(dev, pt_domain); | |
2336 | break; | |
2337 | } | |
2338 | ||
657cbb6b JR |
2339 | domain = domain_for_device(dev); |
2340 | ||
1ac4cbbc JR |
2341 | /* allocate a protection domain if a device is added */ |
2342 | dma_domain = find_protection_domain(devid); | |
2343 | if (dma_domain) | |
2344 | goto out; | |
87a64d52 | 2345 | dma_domain = dma_ops_domain_alloc(); |
1ac4cbbc JR |
2346 | if (!dma_domain) |
2347 | goto out; | |
2348 | dma_domain->target_dev = devid; | |
2349 | ||
2350 | spin_lock_irqsave(&iommu_pd_list_lock, flags); | |
2351 | list_add_tail(&dma_domain->list, &iommu_pd_list); | |
2352 | spin_unlock_irqrestore(&iommu_pd_list_lock, flags); | |
2353 | ||
ac1534a5 JR |
2354 | dev_data = get_dev_data(dev); |
2355 | ||
2c9195e9 | 2356 | dev->archdata.dma_ops = &amd_iommu_dma_ops; |
ac1534a5 | 2357 | |
e275a2a0 | 2358 | break; |
657cbb6b JR |
2359 | case BUS_NOTIFY_DEL_DEVICE: |
2360 | ||
2361 | iommu_uninit_device(dev); | |
2362 | ||
e275a2a0 JR |
2363 | default: |
2364 | goto out; | |
2365 | } | |
2366 | ||
e275a2a0 JR |
2367 | iommu_completion_wait(iommu); |
2368 | ||
2369 | out: | |
2370 | return 0; | |
2371 | } | |
2372 | ||
b25ae679 | 2373 | static struct notifier_block device_nb = { |
e275a2a0 JR |
2374 | .notifier_call = device_change_notifier, |
2375 | }; | |
355bf553 | 2376 | |
8638c491 JR |
2377 | void amd_iommu_init_notifier(void) |
2378 | { | |
2379 | bus_register_notifier(&pci_bus_type, &device_nb); | |
2380 | } | |
2381 | ||
431b2a20 JR |
2382 | /***************************************************************************** |
2383 | * | |
2384 | * The next functions belong to the dma_ops mapping/unmapping code. | |
2385 | * | |
2386 | *****************************************************************************/ | |
2387 | ||
2388 | /* | |
2389 | * In the dma_ops path we only have the struct device. This function | |
2390 | * finds the corresponding IOMMU, the protection domain and the | |
2391 | * requestor id for a given device. | |
2392 | * If the device is not yet associated with a domain this is also done | |
2393 | * in this function. | |
2394 | */ | |
94f6d190 | 2395 | static struct protection_domain *get_domain(struct device *dev) |
b20ac0d4 | 2396 | { |
94f6d190 | 2397 | struct protection_domain *domain; |
b20ac0d4 | 2398 | struct dma_ops_domain *dma_dom; |
94f6d190 | 2399 | u16 devid = get_device_id(dev); |
b20ac0d4 | 2400 | |
f99c0f1c | 2401 | if (!check_device(dev)) |
94f6d190 | 2402 | return ERR_PTR(-EINVAL); |
b20ac0d4 | 2403 | |
94f6d190 JR |
2404 | domain = domain_for_device(dev); |
2405 | if (domain != NULL && !dma_ops_domain(domain)) | |
2406 | return ERR_PTR(-EBUSY); | |
f99c0f1c | 2407 | |
94f6d190 JR |
2408 | if (domain != NULL) |
2409 | return domain; | |
b20ac0d4 | 2410 | |
15898bbc | 2411 | /* Device not bount yet - bind it */ |
94f6d190 | 2412 | dma_dom = find_protection_domain(devid); |
15898bbc | 2413 | if (!dma_dom) |
94f6d190 JR |
2414 | dma_dom = amd_iommu_rlookup_table[devid]->default_dom; |
2415 | attach_device(dev, &dma_dom->domain); | |
15898bbc | 2416 | DUMP_printk("Using protection domain %d for device %s\n", |
94f6d190 | 2417 | dma_dom->domain.id, dev_name(dev)); |
f91ba190 | 2418 | |
94f6d190 | 2419 | return &dma_dom->domain; |
b20ac0d4 JR |
2420 | } |
2421 | ||
04bfdd84 JR |
2422 | static void update_device_table(struct protection_domain *domain) |
2423 | { | |
492667da | 2424 | struct iommu_dev_data *dev_data; |
04bfdd84 | 2425 | |
ea61cddb JR |
2426 | list_for_each_entry(dev_data, &domain->dev_list, list) |
2427 | set_dte_entry(dev_data->devid, domain, dev_data->ats.enabled); | |
04bfdd84 JR |
2428 | } |
2429 | ||
2430 | static void update_domain(struct protection_domain *domain) | |
2431 | { | |
2432 | if (!domain->updated) | |
2433 | return; | |
2434 | ||
2435 | update_device_table(domain); | |
17b124bf JR |
2436 | |
2437 | domain_flush_devices(domain); | |
2438 | domain_flush_tlb_pde(domain); | |
04bfdd84 JR |
2439 | |
2440 | domain->updated = false; | |
2441 | } | |
2442 | ||
8bda3092 JR |
2443 | /* |
2444 | * This function fetches the PTE for a given address in the aperture | |
2445 | */ | |
2446 | static u64* dma_ops_get_pte(struct dma_ops_domain *dom, | |
2447 | unsigned long address) | |
2448 | { | |
384de729 | 2449 | struct aperture_range *aperture; |
8bda3092 JR |
2450 | u64 *pte, *pte_page; |
2451 | ||
384de729 JR |
2452 | aperture = dom->aperture[APERTURE_RANGE_INDEX(address)]; |
2453 | if (!aperture) | |
2454 | return NULL; | |
2455 | ||
2456 | pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)]; | |
8bda3092 | 2457 | if (!pte) { |
cbb9d729 | 2458 | pte = alloc_pte(&dom->domain, address, PAGE_SIZE, &pte_page, |
abdc5eb3 | 2459 | GFP_ATOMIC); |
384de729 JR |
2460 | aperture->pte_pages[APERTURE_PAGE_INDEX(address)] = pte_page; |
2461 | } else | |
8c8c143c | 2462 | pte += PM_LEVEL_INDEX(0, address); |
8bda3092 | 2463 | |
04bfdd84 | 2464 | update_domain(&dom->domain); |
8bda3092 JR |
2465 | |
2466 | return pte; | |
2467 | } | |
2468 | ||
431b2a20 JR |
2469 | /* |
2470 | * This is the generic map function. It maps one 4kb page at paddr to | |
2471 | * the given address in the DMA address space for the domain. | |
2472 | */ | |
680525e0 | 2473 | static dma_addr_t dma_ops_domain_map(struct dma_ops_domain *dom, |
cb76c322 JR |
2474 | unsigned long address, |
2475 | phys_addr_t paddr, | |
2476 | int direction) | |
2477 | { | |
2478 | u64 *pte, __pte; | |
2479 | ||
2480 | WARN_ON(address > dom->aperture_size); | |
2481 | ||
2482 | paddr &= PAGE_MASK; | |
2483 | ||
8bda3092 | 2484 | pte = dma_ops_get_pte(dom, address); |
53812c11 | 2485 | if (!pte) |
8fd524b3 | 2486 | return DMA_ERROR_CODE; |
cb76c322 JR |
2487 | |
2488 | __pte = paddr | IOMMU_PTE_P | IOMMU_PTE_FC; | |
2489 | ||
2490 | if (direction == DMA_TO_DEVICE) | |
2491 | __pte |= IOMMU_PTE_IR; | |
2492 | else if (direction == DMA_FROM_DEVICE) | |
2493 | __pte |= IOMMU_PTE_IW; | |
2494 | else if (direction == DMA_BIDIRECTIONAL) | |
2495 | __pte |= IOMMU_PTE_IR | IOMMU_PTE_IW; | |
2496 | ||
2497 | WARN_ON(*pte); | |
2498 | ||
2499 | *pte = __pte; | |
2500 | ||
2501 | return (dma_addr_t)address; | |
2502 | } | |
2503 | ||
431b2a20 JR |
2504 | /* |
2505 | * The generic unmapping function for on page in the DMA address space. | |
2506 | */ | |
680525e0 | 2507 | static void dma_ops_domain_unmap(struct dma_ops_domain *dom, |
cb76c322 JR |
2508 | unsigned long address) |
2509 | { | |
384de729 | 2510 | struct aperture_range *aperture; |
cb76c322 JR |
2511 | u64 *pte; |
2512 | ||
2513 | if (address >= dom->aperture_size) | |
2514 | return; | |
2515 | ||
384de729 JR |
2516 | aperture = dom->aperture[APERTURE_RANGE_INDEX(address)]; |
2517 | if (!aperture) | |
2518 | return; | |
2519 | ||
2520 | pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)]; | |
2521 | if (!pte) | |
2522 | return; | |
cb76c322 | 2523 | |
8c8c143c | 2524 | pte += PM_LEVEL_INDEX(0, address); |
cb76c322 JR |
2525 | |
2526 | WARN_ON(!*pte); | |
2527 | ||
2528 | *pte = 0ULL; | |
2529 | } | |
2530 | ||
431b2a20 JR |
2531 | /* |
2532 | * This function contains common code for mapping of a physically | |
24f81160 JR |
2533 | * contiguous memory region into DMA address space. It is used by all |
2534 | * mapping functions provided with this IOMMU driver. | |
431b2a20 JR |
2535 | * Must be called with the domain lock held. |
2536 | */ | |
cb76c322 | 2537 | static dma_addr_t __map_single(struct device *dev, |
cb76c322 JR |
2538 | struct dma_ops_domain *dma_dom, |
2539 | phys_addr_t paddr, | |
2540 | size_t size, | |
6d4f343f | 2541 | int dir, |
832a90c3 JR |
2542 | bool align, |
2543 | u64 dma_mask) | |
cb76c322 JR |
2544 | { |
2545 | dma_addr_t offset = paddr & ~PAGE_MASK; | |
53812c11 | 2546 | dma_addr_t address, start, ret; |
cb76c322 | 2547 | unsigned int pages; |
6d4f343f | 2548 | unsigned long align_mask = 0; |
cb76c322 JR |
2549 | int i; |
2550 | ||
e3c449f5 | 2551 | pages = iommu_num_pages(paddr, size, PAGE_SIZE); |
cb76c322 JR |
2552 | paddr &= PAGE_MASK; |
2553 | ||
8ecaf8f1 JR |
2554 | INC_STATS_COUNTER(total_map_requests); |
2555 | ||
c1858976 JR |
2556 | if (pages > 1) |
2557 | INC_STATS_COUNTER(cross_page); | |
2558 | ||
6d4f343f JR |
2559 | if (align) |
2560 | align_mask = (1UL << get_order(size)) - 1; | |
2561 | ||
11b83888 | 2562 | retry: |
832a90c3 JR |
2563 | address = dma_ops_alloc_addresses(dev, dma_dom, pages, align_mask, |
2564 | dma_mask); | |
8fd524b3 | 2565 | if (unlikely(address == DMA_ERROR_CODE)) { |
11b83888 JR |
2566 | /* |
2567 | * setting next_address here will let the address | |
2568 | * allocator only scan the new allocated range in the | |
2569 | * first run. This is a small optimization. | |
2570 | */ | |
2571 | dma_dom->next_address = dma_dom->aperture_size; | |
2572 | ||
576175c2 | 2573 | if (alloc_new_range(dma_dom, false, GFP_ATOMIC)) |
11b83888 JR |
2574 | goto out; |
2575 | ||
2576 | /* | |
af901ca1 | 2577 | * aperture was successfully enlarged by 128 MB, try |
11b83888 JR |
2578 | * allocation again |
2579 | */ | |
2580 | goto retry; | |
2581 | } | |
cb76c322 JR |
2582 | |
2583 | start = address; | |
2584 | for (i = 0; i < pages; ++i) { | |
680525e0 | 2585 | ret = dma_ops_domain_map(dma_dom, start, paddr, dir); |
8fd524b3 | 2586 | if (ret == DMA_ERROR_CODE) |
53812c11 JR |
2587 | goto out_unmap; |
2588 | ||
cb76c322 JR |
2589 | paddr += PAGE_SIZE; |
2590 | start += PAGE_SIZE; | |
2591 | } | |
2592 | address += offset; | |
2593 | ||
5774f7c5 JR |
2594 | ADD_STATS_COUNTER(alloced_io_mem, size); |
2595 | ||
afa9fdc2 | 2596 | if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) { |
17b124bf | 2597 | domain_flush_tlb(&dma_dom->domain); |
1c655773 | 2598 | dma_dom->need_flush = false; |
318afd41 | 2599 | } else if (unlikely(amd_iommu_np_cache)) |
17b124bf | 2600 | domain_flush_pages(&dma_dom->domain, address, size); |
270cab24 | 2601 | |
cb76c322 JR |
2602 | out: |
2603 | return address; | |
53812c11 JR |
2604 | |
2605 | out_unmap: | |
2606 | ||
2607 | for (--i; i >= 0; --i) { | |
2608 | start -= PAGE_SIZE; | |
680525e0 | 2609 | dma_ops_domain_unmap(dma_dom, start); |
53812c11 JR |
2610 | } |
2611 | ||
2612 | dma_ops_free_addresses(dma_dom, address, pages); | |
2613 | ||
8fd524b3 | 2614 | return DMA_ERROR_CODE; |
cb76c322 JR |
2615 | } |
2616 | ||
431b2a20 JR |
2617 | /* |
2618 | * Does the reverse of the __map_single function. Must be called with | |
2619 | * the domain lock held too | |
2620 | */ | |
cd8c82e8 | 2621 | static void __unmap_single(struct dma_ops_domain *dma_dom, |
cb76c322 JR |
2622 | dma_addr_t dma_addr, |
2623 | size_t size, | |
2624 | int dir) | |
2625 | { | |
04e0463e | 2626 | dma_addr_t flush_addr; |
cb76c322 JR |
2627 | dma_addr_t i, start; |
2628 | unsigned int pages; | |
2629 | ||
8fd524b3 | 2630 | if ((dma_addr == DMA_ERROR_CODE) || |
b8d9905d | 2631 | (dma_addr + size > dma_dom->aperture_size)) |
cb76c322 JR |
2632 | return; |
2633 | ||
04e0463e | 2634 | flush_addr = dma_addr; |
e3c449f5 | 2635 | pages = iommu_num_pages(dma_addr, size, PAGE_SIZE); |
cb76c322 JR |
2636 | dma_addr &= PAGE_MASK; |
2637 | start = dma_addr; | |
2638 | ||
2639 | for (i = 0; i < pages; ++i) { | |
680525e0 | 2640 | dma_ops_domain_unmap(dma_dom, start); |
cb76c322 JR |
2641 | start += PAGE_SIZE; |
2642 | } | |
2643 | ||
5774f7c5 JR |
2644 | SUB_STATS_COUNTER(alloced_io_mem, size); |
2645 | ||
cb76c322 | 2646 | dma_ops_free_addresses(dma_dom, dma_addr, pages); |
270cab24 | 2647 | |
80be308d | 2648 | if (amd_iommu_unmap_flush || dma_dom->need_flush) { |
17b124bf | 2649 | domain_flush_pages(&dma_dom->domain, flush_addr, size); |
80be308d JR |
2650 | dma_dom->need_flush = false; |
2651 | } | |
cb76c322 JR |
2652 | } |
2653 | ||
431b2a20 JR |
2654 | /* |
2655 | * The exported map_single function for dma_ops. | |
2656 | */ | |
51491367 FT |
2657 | static dma_addr_t map_page(struct device *dev, struct page *page, |
2658 | unsigned long offset, size_t size, | |
2659 | enum dma_data_direction dir, | |
2660 | struct dma_attrs *attrs) | |
4da70b9e JR |
2661 | { |
2662 | unsigned long flags; | |
4da70b9e | 2663 | struct protection_domain *domain; |
4da70b9e | 2664 | dma_addr_t addr; |
832a90c3 | 2665 | u64 dma_mask; |
51491367 | 2666 | phys_addr_t paddr = page_to_phys(page) + offset; |
4da70b9e | 2667 | |
0f2a86f2 JR |
2668 | INC_STATS_COUNTER(cnt_map_single); |
2669 | ||
94f6d190 JR |
2670 | domain = get_domain(dev); |
2671 | if (PTR_ERR(domain) == -EINVAL) | |
4da70b9e | 2672 | return (dma_addr_t)paddr; |
94f6d190 JR |
2673 | else if (IS_ERR(domain)) |
2674 | return DMA_ERROR_CODE; | |
4da70b9e | 2675 | |
f99c0f1c JR |
2676 | dma_mask = *dev->dma_mask; |
2677 | ||
4da70b9e | 2678 | spin_lock_irqsave(&domain->lock, flags); |
94f6d190 | 2679 | |
cd8c82e8 | 2680 | addr = __map_single(dev, domain->priv, paddr, size, dir, false, |
832a90c3 | 2681 | dma_mask); |
8fd524b3 | 2682 | if (addr == DMA_ERROR_CODE) |
4da70b9e JR |
2683 | goto out; |
2684 | ||
17b124bf | 2685 | domain_flush_complete(domain); |
4da70b9e JR |
2686 | |
2687 | out: | |
2688 | spin_unlock_irqrestore(&domain->lock, flags); | |
2689 | ||
2690 | return addr; | |
2691 | } | |
2692 | ||
431b2a20 JR |
2693 | /* |
2694 | * The exported unmap_single function for dma_ops. | |
2695 | */ | |
51491367 FT |
2696 | static void unmap_page(struct device *dev, dma_addr_t dma_addr, size_t size, |
2697 | enum dma_data_direction dir, struct dma_attrs *attrs) | |
4da70b9e JR |
2698 | { |
2699 | unsigned long flags; | |
4da70b9e | 2700 | struct protection_domain *domain; |
4da70b9e | 2701 | |
146a6917 JR |
2702 | INC_STATS_COUNTER(cnt_unmap_single); |
2703 | ||
94f6d190 JR |
2704 | domain = get_domain(dev); |
2705 | if (IS_ERR(domain)) | |
5b28df6f JR |
2706 | return; |
2707 | ||
4da70b9e JR |
2708 | spin_lock_irqsave(&domain->lock, flags); |
2709 | ||
cd8c82e8 | 2710 | __unmap_single(domain->priv, dma_addr, size, dir); |
4da70b9e | 2711 | |
17b124bf | 2712 | domain_flush_complete(domain); |
4da70b9e JR |
2713 | |
2714 | spin_unlock_irqrestore(&domain->lock, flags); | |
2715 | } | |
2716 | ||
431b2a20 JR |
2717 | /* |
2718 | * This is a special map_sg function which is used if we should map a | |
2719 | * device which is not handled by an AMD IOMMU in the system. | |
2720 | */ | |
65b050ad JR |
2721 | static int map_sg_no_iommu(struct device *dev, struct scatterlist *sglist, |
2722 | int nelems, int dir) | |
2723 | { | |
2724 | struct scatterlist *s; | |
2725 | int i; | |
2726 | ||
2727 | for_each_sg(sglist, s, nelems, i) { | |
2728 | s->dma_address = (dma_addr_t)sg_phys(s); | |
2729 | s->dma_length = s->length; | |
2730 | } | |
2731 | ||
2732 | return nelems; | |
2733 | } | |
2734 | ||
431b2a20 JR |
2735 | /* |
2736 | * The exported map_sg function for dma_ops (handles scatter-gather | |
2737 | * lists). | |
2738 | */ | |
65b050ad | 2739 | static int map_sg(struct device *dev, struct scatterlist *sglist, |
160c1d8e FT |
2740 | int nelems, enum dma_data_direction dir, |
2741 | struct dma_attrs *attrs) | |
65b050ad JR |
2742 | { |
2743 | unsigned long flags; | |
65b050ad | 2744 | struct protection_domain *domain; |
65b050ad JR |
2745 | int i; |
2746 | struct scatterlist *s; | |
2747 | phys_addr_t paddr; | |
2748 | int mapped_elems = 0; | |
832a90c3 | 2749 | u64 dma_mask; |
65b050ad | 2750 | |
d03f067a JR |
2751 | INC_STATS_COUNTER(cnt_map_sg); |
2752 | ||
94f6d190 JR |
2753 | domain = get_domain(dev); |
2754 | if (PTR_ERR(domain) == -EINVAL) | |
f99c0f1c | 2755 | return map_sg_no_iommu(dev, sglist, nelems, dir); |
94f6d190 JR |
2756 | else if (IS_ERR(domain)) |
2757 | return 0; | |
dbcc112e | 2758 | |
832a90c3 | 2759 | dma_mask = *dev->dma_mask; |
65b050ad | 2760 | |
65b050ad JR |
2761 | spin_lock_irqsave(&domain->lock, flags); |
2762 | ||
2763 | for_each_sg(sglist, s, nelems, i) { | |
2764 | paddr = sg_phys(s); | |
2765 | ||
cd8c82e8 | 2766 | s->dma_address = __map_single(dev, domain->priv, |
832a90c3 JR |
2767 | paddr, s->length, dir, false, |
2768 | dma_mask); | |
65b050ad JR |
2769 | |
2770 | if (s->dma_address) { | |
2771 | s->dma_length = s->length; | |
2772 | mapped_elems++; | |
2773 | } else | |
2774 | goto unmap; | |
65b050ad JR |
2775 | } |
2776 | ||
17b124bf | 2777 | domain_flush_complete(domain); |
65b050ad JR |
2778 | |
2779 | out: | |
2780 | spin_unlock_irqrestore(&domain->lock, flags); | |
2781 | ||
2782 | return mapped_elems; | |
2783 | unmap: | |
2784 | for_each_sg(sglist, s, mapped_elems, i) { | |
2785 | if (s->dma_address) | |
cd8c82e8 | 2786 | __unmap_single(domain->priv, s->dma_address, |
65b050ad JR |
2787 | s->dma_length, dir); |
2788 | s->dma_address = s->dma_length = 0; | |
2789 | } | |
2790 | ||
2791 | mapped_elems = 0; | |
2792 | ||
2793 | goto out; | |
2794 | } | |
2795 | ||
431b2a20 JR |
2796 | /* |
2797 | * The exported map_sg function for dma_ops (handles scatter-gather | |
2798 | * lists). | |
2799 | */ | |
65b050ad | 2800 | static void unmap_sg(struct device *dev, struct scatterlist *sglist, |
160c1d8e FT |
2801 | int nelems, enum dma_data_direction dir, |
2802 | struct dma_attrs *attrs) | |
65b050ad JR |
2803 | { |
2804 | unsigned long flags; | |
65b050ad JR |
2805 | struct protection_domain *domain; |
2806 | struct scatterlist *s; | |
65b050ad JR |
2807 | int i; |
2808 | ||
55877a6b JR |
2809 | INC_STATS_COUNTER(cnt_unmap_sg); |
2810 | ||
94f6d190 JR |
2811 | domain = get_domain(dev); |
2812 | if (IS_ERR(domain)) | |
5b28df6f JR |
2813 | return; |
2814 | ||
65b050ad JR |
2815 | spin_lock_irqsave(&domain->lock, flags); |
2816 | ||
2817 | for_each_sg(sglist, s, nelems, i) { | |
cd8c82e8 | 2818 | __unmap_single(domain->priv, s->dma_address, |
65b050ad | 2819 | s->dma_length, dir); |
65b050ad JR |
2820 | s->dma_address = s->dma_length = 0; |
2821 | } | |
2822 | ||
17b124bf | 2823 | domain_flush_complete(domain); |
65b050ad JR |
2824 | |
2825 | spin_unlock_irqrestore(&domain->lock, flags); | |
2826 | } | |
2827 | ||
431b2a20 JR |
2828 | /* |
2829 | * The exported alloc_coherent function for dma_ops. | |
2830 | */ | |
5d8b53cf | 2831 | static void *alloc_coherent(struct device *dev, size_t size, |
baa676fc AP |
2832 | dma_addr_t *dma_addr, gfp_t flag, |
2833 | struct dma_attrs *attrs) | |
5d8b53cf JR |
2834 | { |
2835 | unsigned long flags; | |
2836 | void *virt_addr; | |
5d8b53cf | 2837 | struct protection_domain *domain; |
5d8b53cf | 2838 | phys_addr_t paddr; |
832a90c3 | 2839 | u64 dma_mask = dev->coherent_dma_mask; |
5d8b53cf | 2840 | |
c8f0fb36 JR |
2841 | INC_STATS_COUNTER(cnt_alloc_coherent); |
2842 | ||
94f6d190 JR |
2843 | domain = get_domain(dev); |
2844 | if (PTR_ERR(domain) == -EINVAL) { | |
f99c0f1c JR |
2845 | virt_addr = (void *)__get_free_pages(flag, get_order(size)); |
2846 | *dma_addr = __pa(virt_addr); | |
2847 | return virt_addr; | |
94f6d190 JR |
2848 | } else if (IS_ERR(domain)) |
2849 | return NULL; | |
5d8b53cf | 2850 | |
f99c0f1c JR |
2851 | dma_mask = dev->coherent_dma_mask; |
2852 | flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32); | |
2853 | flag |= __GFP_ZERO; | |
5d8b53cf JR |
2854 | |
2855 | virt_addr = (void *)__get_free_pages(flag, get_order(size)); | |
2856 | if (!virt_addr) | |
b25ae679 | 2857 | return NULL; |
5d8b53cf | 2858 | |
5d8b53cf JR |
2859 | paddr = virt_to_phys(virt_addr); |
2860 | ||
832a90c3 JR |
2861 | if (!dma_mask) |
2862 | dma_mask = *dev->dma_mask; | |
2863 | ||
5d8b53cf JR |
2864 | spin_lock_irqsave(&domain->lock, flags); |
2865 | ||
cd8c82e8 | 2866 | *dma_addr = __map_single(dev, domain->priv, paddr, |
832a90c3 | 2867 | size, DMA_BIDIRECTIONAL, true, dma_mask); |
5d8b53cf | 2868 | |
8fd524b3 | 2869 | if (*dma_addr == DMA_ERROR_CODE) { |
367d04c4 | 2870 | spin_unlock_irqrestore(&domain->lock, flags); |
5b28df6f | 2871 | goto out_free; |
367d04c4 | 2872 | } |
5d8b53cf | 2873 | |
17b124bf | 2874 | domain_flush_complete(domain); |
5d8b53cf | 2875 | |
5d8b53cf JR |
2876 | spin_unlock_irqrestore(&domain->lock, flags); |
2877 | ||
2878 | return virt_addr; | |
5b28df6f JR |
2879 | |
2880 | out_free: | |
2881 | ||
2882 | free_pages((unsigned long)virt_addr, get_order(size)); | |
2883 | ||
2884 | return NULL; | |
5d8b53cf JR |
2885 | } |
2886 | ||
431b2a20 JR |
2887 | /* |
2888 | * The exported free_coherent function for dma_ops. | |
431b2a20 | 2889 | */ |
5d8b53cf | 2890 | static void free_coherent(struct device *dev, size_t size, |
baa676fc AP |
2891 | void *virt_addr, dma_addr_t dma_addr, |
2892 | struct dma_attrs *attrs) | |
5d8b53cf JR |
2893 | { |
2894 | unsigned long flags; | |
5d8b53cf | 2895 | struct protection_domain *domain; |
5d8b53cf | 2896 | |
5d31ee7e JR |
2897 | INC_STATS_COUNTER(cnt_free_coherent); |
2898 | ||
94f6d190 JR |
2899 | domain = get_domain(dev); |
2900 | if (IS_ERR(domain)) | |
5b28df6f JR |
2901 | goto free_mem; |
2902 | ||
5d8b53cf JR |
2903 | spin_lock_irqsave(&domain->lock, flags); |
2904 | ||
cd8c82e8 | 2905 | __unmap_single(domain->priv, dma_addr, size, DMA_BIDIRECTIONAL); |
5d8b53cf | 2906 | |
17b124bf | 2907 | domain_flush_complete(domain); |
5d8b53cf JR |
2908 | |
2909 | spin_unlock_irqrestore(&domain->lock, flags); | |
2910 | ||
2911 | free_mem: | |
2912 | free_pages((unsigned long)virt_addr, get_order(size)); | |
2913 | } | |
2914 | ||
b39ba6ad JR |
2915 | /* |
2916 | * This function is called by the DMA layer to find out if we can handle a | |
2917 | * particular device. It is part of the dma_ops. | |
2918 | */ | |
2919 | static int amd_iommu_dma_supported(struct device *dev, u64 mask) | |
2920 | { | |
420aef8a | 2921 | return check_device(dev); |
b39ba6ad JR |
2922 | } |
2923 | ||
c432f3df | 2924 | /* |
431b2a20 JR |
2925 | * The function for pre-allocating protection domains. |
2926 | * | |
c432f3df JR |
2927 | * If the driver core informs the DMA layer if a driver grabs a device |
2928 | * we don't need to preallocate the protection domains anymore. | |
2929 | * For now we have to. | |
2930 | */ | |
943bc7e1 | 2931 | static void __init prealloc_protection_domains(void) |
c432f3df | 2932 | { |
5abcdba4 | 2933 | struct iommu_dev_data *dev_data; |
c432f3df | 2934 | struct dma_ops_domain *dma_dom; |
5abcdba4 | 2935 | struct pci_dev *dev = NULL; |
98fc5a69 | 2936 | u16 devid; |
c432f3df | 2937 | |
d18c69d3 | 2938 | for_each_pci_dev(dev) { |
98fc5a69 JR |
2939 | |
2940 | /* Do we handle this device? */ | |
2941 | if (!check_device(&dev->dev)) | |
c432f3df | 2942 | continue; |
98fc5a69 | 2943 | |
5abcdba4 JR |
2944 | dev_data = get_dev_data(&dev->dev); |
2945 | if (!amd_iommu_force_isolation && dev_data->iommu_v2) { | |
2946 | /* Make sure passthrough domain is allocated */ | |
2947 | alloc_passthrough_domain(); | |
2948 | dev_data->passthrough = true; | |
2949 | attach_device(&dev->dev, pt_domain); | |
2950 | pr_info("AMD-Vi: Using passthough domain for device %s\n", | |
2951 | dev_name(&dev->dev)); | |
2952 | } | |
2953 | ||
98fc5a69 | 2954 | /* Is there already any domain for it? */ |
15898bbc | 2955 | if (domain_for_device(&dev->dev)) |
c432f3df | 2956 | continue; |
98fc5a69 JR |
2957 | |
2958 | devid = get_device_id(&dev->dev); | |
2959 | ||
87a64d52 | 2960 | dma_dom = dma_ops_domain_alloc(); |
c432f3df JR |
2961 | if (!dma_dom) |
2962 | continue; | |
2963 | init_unity_mappings_for_device(dma_dom, devid); | |
bd60b735 JR |
2964 | dma_dom->target_dev = devid; |
2965 | ||
15898bbc | 2966 | attach_device(&dev->dev, &dma_dom->domain); |
be831297 | 2967 | |
bd60b735 | 2968 | list_add_tail(&dma_dom->list, &iommu_pd_list); |
c432f3df JR |
2969 | } |
2970 | } | |
2971 | ||
160c1d8e | 2972 | static struct dma_map_ops amd_iommu_dma_ops = { |
baa676fc AP |
2973 | .alloc = alloc_coherent, |
2974 | .free = free_coherent, | |
51491367 FT |
2975 | .map_page = map_page, |
2976 | .unmap_page = unmap_page, | |
6631ee9d JR |
2977 | .map_sg = map_sg, |
2978 | .unmap_sg = unmap_sg, | |
b39ba6ad | 2979 | .dma_supported = amd_iommu_dma_supported, |
6631ee9d JR |
2980 | }; |
2981 | ||
27c2127a JR |
2982 | static unsigned device_dma_ops_init(void) |
2983 | { | |
5abcdba4 | 2984 | struct iommu_dev_data *dev_data; |
27c2127a JR |
2985 | struct pci_dev *pdev = NULL; |
2986 | unsigned unhandled = 0; | |
2987 | ||
2988 | for_each_pci_dev(pdev) { | |
2989 | if (!check_device(&pdev->dev)) { | |
af1be049 JR |
2990 | |
2991 | iommu_ignore_device(&pdev->dev); | |
2992 | ||
27c2127a JR |
2993 | unhandled += 1; |
2994 | continue; | |
2995 | } | |
2996 | ||
5abcdba4 JR |
2997 | dev_data = get_dev_data(&pdev->dev); |
2998 | ||
2999 | if (!dev_data->passthrough) | |
3000 | pdev->dev.archdata.dma_ops = &amd_iommu_dma_ops; | |
3001 | else | |
3002 | pdev->dev.archdata.dma_ops = &nommu_dma_ops; | |
27c2127a JR |
3003 | } |
3004 | ||
3005 | return unhandled; | |
3006 | } | |
3007 | ||
431b2a20 JR |
3008 | /* |
3009 | * The function which clues the AMD IOMMU driver into dma_ops. | |
3010 | */ | |
f5325094 JR |
3011 | |
3012 | void __init amd_iommu_init_api(void) | |
3013 | { | |
2cc21c42 | 3014 | bus_set_iommu(&pci_bus_type, &amd_iommu_ops); |
f5325094 JR |
3015 | } |
3016 | ||
6631ee9d JR |
3017 | int __init amd_iommu_init_dma_ops(void) |
3018 | { | |
3019 | struct amd_iommu *iommu; | |
27c2127a | 3020 | int ret, unhandled; |
6631ee9d | 3021 | |
431b2a20 JR |
3022 | /* |
3023 | * first allocate a default protection domain for every IOMMU we | |
3024 | * found in the system. Devices not assigned to any other | |
3025 | * protection domain will be assigned to the default one. | |
3026 | */ | |
3bd22172 | 3027 | for_each_iommu(iommu) { |
87a64d52 | 3028 | iommu->default_dom = dma_ops_domain_alloc(); |
6631ee9d JR |
3029 | if (iommu->default_dom == NULL) |
3030 | return -ENOMEM; | |
e2dc14a2 | 3031 | iommu->default_dom->domain.flags |= PD_DEFAULT_MASK; |
6631ee9d JR |
3032 | ret = iommu_init_unity_mappings(iommu); |
3033 | if (ret) | |
3034 | goto free_domains; | |
3035 | } | |
3036 | ||
431b2a20 | 3037 | /* |
8793abeb | 3038 | * Pre-allocate the protection domains for each device. |
431b2a20 | 3039 | */ |
8793abeb | 3040 | prealloc_protection_domains(); |
6631ee9d JR |
3041 | |
3042 | iommu_detected = 1; | |
75f1cdf1 | 3043 | swiotlb = 0; |
6631ee9d | 3044 | |
431b2a20 | 3045 | /* Make the driver finally visible to the drivers */ |
27c2127a JR |
3046 | unhandled = device_dma_ops_init(); |
3047 | if (unhandled && max_pfn > MAX_DMA32_PFN) { | |
3048 | /* There are unhandled devices - initialize swiotlb for them */ | |
3049 | swiotlb = 1; | |
3050 | } | |
6631ee9d | 3051 | |
7f26508b JR |
3052 | amd_iommu_stats_init(); |
3053 | ||
62410eeb JR |
3054 | if (amd_iommu_unmap_flush) |
3055 | pr_info("AMD-Vi: IO/TLB flush on unmap enabled\n"); | |
3056 | else | |
3057 | pr_info("AMD-Vi: Lazy IO/TLB flushing enabled\n"); | |
3058 | ||
6631ee9d JR |
3059 | return 0; |
3060 | ||
3061 | free_domains: | |
3062 | ||
3bd22172 | 3063 | for_each_iommu(iommu) { |
6631ee9d JR |
3064 | if (iommu->default_dom) |
3065 | dma_ops_domain_free(iommu->default_dom); | |
3066 | } | |
3067 | ||
3068 | return ret; | |
3069 | } | |
6d98cd80 JR |
3070 | |
3071 | /***************************************************************************** | |
3072 | * | |
3073 | * The following functions belong to the exported interface of AMD IOMMU | |
3074 | * | |
3075 | * This interface allows access to lower level functions of the IOMMU | |
3076 | * like protection domain handling and assignement of devices to domains | |
3077 | * which is not possible with the dma_ops interface. | |
3078 | * | |
3079 | *****************************************************************************/ | |
3080 | ||
6d98cd80 JR |
3081 | static void cleanup_domain(struct protection_domain *domain) |
3082 | { | |
492667da | 3083 | struct iommu_dev_data *dev_data, *next; |
6d98cd80 | 3084 | unsigned long flags; |
6d98cd80 JR |
3085 | |
3086 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | |
3087 | ||
492667da | 3088 | list_for_each_entry_safe(dev_data, next, &domain->dev_list, list) { |
ec9e79ef | 3089 | __detach_device(dev_data); |
492667da JR |
3090 | atomic_set(&dev_data->bind, 0); |
3091 | } | |
6d98cd80 JR |
3092 | |
3093 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | |
3094 | } | |
3095 | ||
2650815f JR |
3096 | static void protection_domain_free(struct protection_domain *domain) |
3097 | { | |
3098 | if (!domain) | |
3099 | return; | |
3100 | ||
aeb26f55 JR |
3101 | del_domain_from_list(domain); |
3102 | ||
2650815f JR |
3103 | if (domain->id) |
3104 | domain_id_free(domain->id); | |
3105 | ||
3106 | kfree(domain); | |
3107 | } | |
3108 | ||
3109 | static struct protection_domain *protection_domain_alloc(void) | |
c156e347 JR |
3110 | { |
3111 | struct protection_domain *domain; | |
3112 | ||
3113 | domain = kzalloc(sizeof(*domain), GFP_KERNEL); | |
3114 | if (!domain) | |
2650815f | 3115 | return NULL; |
c156e347 JR |
3116 | |
3117 | spin_lock_init(&domain->lock); | |
5d214fe6 | 3118 | mutex_init(&domain->api_lock); |
c156e347 JR |
3119 | domain->id = domain_id_alloc(); |
3120 | if (!domain->id) | |
2650815f | 3121 | goto out_err; |
7c392cbe | 3122 | INIT_LIST_HEAD(&domain->dev_list); |
2650815f | 3123 | |
aeb26f55 JR |
3124 | add_domain_to_list(domain); |
3125 | ||
2650815f JR |
3126 | return domain; |
3127 | ||
3128 | out_err: | |
3129 | kfree(domain); | |
3130 | ||
3131 | return NULL; | |
3132 | } | |
3133 | ||
5abcdba4 JR |
3134 | static int __init alloc_passthrough_domain(void) |
3135 | { | |
3136 | if (pt_domain != NULL) | |
3137 | return 0; | |
3138 | ||
3139 | /* allocate passthrough domain */ | |
3140 | pt_domain = protection_domain_alloc(); | |
3141 | if (!pt_domain) | |
3142 | return -ENOMEM; | |
3143 | ||
3144 | pt_domain->mode = PAGE_MODE_NONE; | |
3145 | ||
3146 | return 0; | |
3147 | } | |
2650815f JR |
3148 | static int amd_iommu_domain_init(struct iommu_domain *dom) |
3149 | { | |
3150 | struct protection_domain *domain; | |
3151 | ||
3152 | domain = protection_domain_alloc(); | |
3153 | if (!domain) | |
c156e347 | 3154 | goto out_free; |
2650815f JR |
3155 | |
3156 | domain->mode = PAGE_MODE_3_LEVEL; | |
c156e347 JR |
3157 | domain->pt_root = (void *)get_zeroed_page(GFP_KERNEL); |
3158 | if (!domain->pt_root) | |
3159 | goto out_free; | |
3160 | ||
f3572db8 JR |
3161 | domain->iommu_domain = dom; |
3162 | ||
c156e347 JR |
3163 | dom->priv = domain; |
3164 | ||
0ff64f80 JR |
3165 | dom->geometry.aperture_start = 0; |
3166 | dom->geometry.aperture_end = ~0ULL; | |
3167 | dom->geometry.force_aperture = true; | |
3168 | ||
c156e347 JR |
3169 | return 0; |
3170 | ||
3171 | out_free: | |
2650815f | 3172 | protection_domain_free(domain); |
c156e347 JR |
3173 | |
3174 | return -ENOMEM; | |
3175 | } | |
3176 | ||
98383fc3 JR |
3177 | static void amd_iommu_domain_destroy(struct iommu_domain *dom) |
3178 | { | |
3179 | struct protection_domain *domain = dom->priv; | |
3180 | ||
3181 | if (!domain) | |
3182 | return; | |
3183 | ||
3184 | if (domain->dev_cnt > 0) | |
3185 | cleanup_domain(domain); | |
3186 | ||
3187 | BUG_ON(domain->dev_cnt != 0); | |
3188 | ||
132bd68f JR |
3189 | if (domain->mode != PAGE_MODE_NONE) |
3190 | free_pagetable(domain); | |
98383fc3 | 3191 | |
52815b75 JR |
3192 | if (domain->flags & PD_IOMMUV2_MASK) |
3193 | free_gcr3_table(domain); | |
3194 | ||
8b408fe4 | 3195 | protection_domain_free(domain); |
98383fc3 JR |
3196 | |
3197 | dom->priv = NULL; | |
3198 | } | |
3199 | ||
684f2888 JR |
3200 | static void amd_iommu_detach_device(struct iommu_domain *dom, |
3201 | struct device *dev) | |
3202 | { | |
657cbb6b | 3203 | struct iommu_dev_data *dev_data = dev->archdata.iommu; |
684f2888 | 3204 | struct amd_iommu *iommu; |
684f2888 JR |
3205 | u16 devid; |
3206 | ||
98fc5a69 | 3207 | if (!check_device(dev)) |
684f2888 JR |
3208 | return; |
3209 | ||
98fc5a69 | 3210 | devid = get_device_id(dev); |
684f2888 | 3211 | |
657cbb6b | 3212 | if (dev_data->domain != NULL) |
15898bbc | 3213 | detach_device(dev); |
684f2888 JR |
3214 | |
3215 | iommu = amd_iommu_rlookup_table[devid]; | |
3216 | if (!iommu) | |
3217 | return; | |
3218 | ||
684f2888 JR |
3219 | iommu_completion_wait(iommu); |
3220 | } | |
3221 | ||
01106066 JR |
3222 | static int amd_iommu_attach_device(struct iommu_domain *dom, |
3223 | struct device *dev) | |
3224 | { | |
3225 | struct protection_domain *domain = dom->priv; | |
657cbb6b | 3226 | struct iommu_dev_data *dev_data; |
01106066 | 3227 | struct amd_iommu *iommu; |
15898bbc | 3228 | int ret; |
01106066 | 3229 | |
98fc5a69 | 3230 | if (!check_device(dev)) |
01106066 JR |
3231 | return -EINVAL; |
3232 | ||
657cbb6b JR |
3233 | dev_data = dev->archdata.iommu; |
3234 | ||
f62dda66 | 3235 | iommu = amd_iommu_rlookup_table[dev_data->devid]; |
01106066 JR |
3236 | if (!iommu) |
3237 | return -EINVAL; | |
3238 | ||
657cbb6b | 3239 | if (dev_data->domain) |
15898bbc | 3240 | detach_device(dev); |
01106066 | 3241 | |
15898bbc | 3242 | ret = attach_device(dev, domain); |
01106066 JR |
3243 | |
3244 | iommu_completion_wait(iommu); | |
3245 | ||
15898bbc | 3246 | return ret; |
01106066 JR |
3247 | } |
3248 | ||
468e2366 | 3249 | static int amd_iommu_map(struct iommu_domain *dom, unsigned long iova, |
5009065d | 3250 | phys_addr_t paddr, size_t page_size, int iommu_prot) |
c6229ca6 JR |
3251 | { |
3252 | struct protection_domain *domain = dom->priv; | |
c6229ca6 JR |
3253 | int prot = 0; |
3254 | int ret; | |
3255 | ||
132bd68f JR |
3256 | if (domain->mode == PAGE_MODE_NONE) |
3257 | return -EINVAL; | |
3258 | ||
c6229ca6 JR |
3259 | if (iommu_prot & IOMMU_READ) |
3260 | prot |= IOMMU_PROT_IR; | |
3261 | if (iommu_prot & IOMMU_WRITE) | |
3262 | prot |= IOMMU_PROT_IW; | |
3263 | ||
5d214fe6 | 3264 | mutex_lock(&domain->api_lock); |
795e74f7 | 3265 | ret = iommu_map_page(domain, iova, paddr, prot, page_size); |
5d214fe6 JR |
3266 | mutex_unlock(&domain->api_lock); |
3267 | ||
795e74f7 | 3268 | return ret; |
c6229ca6 JR |
3269 | } |
3270 | ||
5009065d OBC |
3271 | static size_t amd_iommu_unmap(struct iommu_domain *dom, unsigned long iova, |
3272 | size_t page_size) | |
eb74ff6c | 3273 | { |
eb74ff6c | 3274 | struct protection_domain *domain = dom->priv; |
5009065d | 3275 | size_t unmap_size; |
eb74ff6c | 3276 | |
132bd68f JR |
3277 | if (domain->mode == PAGE_MODE_NONE) |
3278 | return -EINVAL; | |
3279 | ||
5d214fe6 | 3280 | mutex_lock(&domain->api_lock); |
468e2366 | 3281 | unmap_size = iommu_unmap_page(domain, iova, page_size); |
795e74f7 | 3282 | mutex_unlock(&domain->api_lock); |
eb74ff6c | 3283 | |
17b124bf | 3284 | domain_flush_tlb_pde(domain); |
5d214fe6 | 3285 | |
5009065d | 3286 | return unmap_size; |
eb74ff6c JR |
3287 | } |
3288 | ||
645c4c8d JR |
3289 | static phys_addr_t amd_iommu_iova_to_phys(struct iommu_domain *dom, |
3290 | unsigned long iova) | |
3291 | { | |
3292 | struct protection_domain *domain = dom->priv; | |
f03152bb | 3293 | unsigned long offset_mask; |
645c4c8d | 3294 | phys_addr_t paddr; |
f03152bb | 3295 | u64 *pte, __pte; |
645c4c8d | 3296 | |
132bd68f JR |
3297 | if (domain->mode == PAGE_MODE_NONE) |
3298 | return iova; | |
3299 | ||
24cd7723 | 3300 | pte = fetch_pte(domain, iova); |
645c4c8d | 3301 | |
a6d41a40 | 3302 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) |
645c4c8d JR |
3303 | return 0; |
3304 | ||
f03152bb JR |
3305 | if (PM_PTE_LEVEL(*pte) == 0) |
3306 | offset_mask = PAGE_SIZE - 1; | |
3307 | else | |
3308 | offset_mask = PTE_PAGE_SIZE(*pte) - 1; | |
3309 | ||
3310 | __pte = *pte & PM_ADDR_MASK; | |
3311 | paddr = (__pte & ~offset_mask) | (iova & offset_mask); | |
645c4c8d JR |
3312 | |
3313 | return paddr; | |
3314 | } | |
3315 | ||
dbb9fd86 SY |
3316 | static int amd_iommu_domain_has_cap(struct iommu_domain *domain, |
3317 | unsigned long cap) | |
3318 | { | |
80a506b8 JR |
3319 | switch (cap) { |
3320 | case IOMMU_CAP_CACHE_COHERENCY: | |
3321 | return 1; | |
3322 | } | |
3323 | ||
dbb9fd86 SY |
3324 | return 0; |
3325 | } | |
3326 | ||
26961efe JR |
3327 | static struct iommu_ops amd_iommu_ops = { |
3328 | .domain_init = amd_iommu_domain_init, | |
3329 | .domain_destroy = amd_iommu_domain_destroy, | |
3330 | .attach_dev = amd_iommu_attach_device, | |
3331 | .detach_dev = amd_iommu_detach_device, | |
468e2366 JR |
3332 | .map = amd_iommu_map, |
3333 | .unmap = amd_iommu_unmap, | |
26961efe | 3334 | .iova_to_phys = amd_iommu_iova_to_phys, |
dbb9fd86 | 3335 | .domain_has_cap = amd_iommu_domain_has_cap, |
aa3de9c0 | 3336 | .pgsize_bitmap = AMD_IOMMU_PGSIZES, |
26961efe JR |
3337 | }; |
3338 | ||
0feae533 JR |
3339 | /***************************************************************************** |
3340 | * | |
3341 | * The next functions do a basic initialization of IOMMU for pass through | |
3342 | * mode | |
3343 | * | |
3344 | * In passthrough mode the IOMMU is initialized and enabled but not used for | |
3345 | * DMA-API translation. | |
3346 | * | |
3347 | *****************************************************************************/ | |
3348 | ||
3349 | int __init amd_iommu_init_passthrough(void) | |
3350 | { | |
5abcdba4 | 3351 | struct iommu_dev_data *dev_data; |
0feae533 | 3352 | struct pci_dev *dev = NULL; |
5abcdba4 | 3353 | struct amd_iommu *iommu; |
15898bbc | 3354 | u16 devid; |
5abcdba4 | 3355 | int ret; |
0feae533 | 3356 | |
5abcdba4 JR |
3357 | ret = alloc_passthrough_domain(); |
3358 | if (ret) | |
3359 | return ret; | |
0feae533 | 3360 | |
6c54aabd | 3361 | for_each_pci_dev(dev) { |
98fc5a69 | 3362 | if (!check_device(&dev->dev)) |
0feae533 JR |
3363 | continue; |
3364 | ||
5abcdba4 JR |
3365 | dev_data = get_dev_data(&dev->dev); |
3366 | dev_data->passthrough = true; | |
3367 | ||
98fc5a69 JR |
3368 | devid = get_device_id(&dev->dev); |
3369 | ||
15898bbc | 3370 | iommu = amd_iommu_rlookup_table[devid]; |
0feae533 JR |
3371 | if (!iommu) |
3372 | continue; | |
3373 | ||
15898bbc | 3374 | attach_device(&dev->dev, pt_domain); |
0feae533 JR |
3375 | } |
3376 | ||
2655d7a2 JR |
3377 | amd_iommu_stats_init(); |
3378 | ||
0feae533 JR |
3379 | pr_info("AMD-Vi: Initialized for Passthrough Mode\n"); |
3380 | ||
3381 | return 0; | |
3382 | } | |
72e1dcc4 JR |
3383 | |
3384 | /* IOMMUv2 specific functions */ | |
3385 | int amd_iommu_register_ppr_notifier(struct notifier_block *nb) | |
3386 | { | |
3387 | return atomic_notifier_chain_register(&ppr_notifier, nb); | |
3388 | } | |
3389 | EXPORT_SYMBOL(amd_iommu_register_ppr_notifier); | |
3390 | ||
3391 | int amd_iommu_unregister_ppr_notifier(struct notifier_block *nb) | |
3392 | { | |
3393 | return atomic_notifier_chain_unregister(&ppr_notifier, nb); | |
3394 | } | |
3395 | EXPORT_SYMBOL(amd_iommu_unregister_ppr_notifier); | |
132bd68f JR |
3396 | |
3397 | void amd_iommu_domain_direct_map(struct iommu_domain *dom) | |
3398 | { | |
3399 | struct protection_domain *domain = dom->priv; | |
3400 | unsigned long flags; | |
3401 | ||
3402 | spin_lock_irqsave(&domain->lock, flags); | |
3403 | ||
3404 | /* Update data structure */ | |
3405 | domain->mode = PAGE_MODE_NONE; | |
3406 | domain->updated = true; | |
3407 | ||
3408 | /* Make changes visible to IOMMUs */ | |
3409 | update_domain(domain); | |
3410 | ||
3411 | /* Page-table is not visible to IOMMU anymore, so free it */ | |
3412 | free_pagetable(domain); | |
3413 | ||
3414 | spin_unlock_irqrestore(&domain->lock, flags); | |
3415 | } | |
3416 | EXPORT_SYMBOL(amd_iommu_domain_direct_map); | |
52815b75 JR |
3417 | |
3418 | int amd_iommu_domain_enable_v2(struct iommu_domain *dom, int pasids) | |
3419 | { | |
3420 | struct protection_domain *domain = dom->priv; | |
3421 | unsigned long flags; | |
3422 | int levels, ret; | |
3423 | ||
3424 | if (pasids <= 0 || pasids > (PASID_MASK + 1)) | |
3425 | return -EINVAL; | |
3426 | ||
3427 | /* Number of GCR3 table levels required */ | |
3428 | for (levels = 0; (pasids - 1) & ~0x1ff; pasids >>= 9) | |
3429 | levels += 1; | |
3430 | ||
3431 | if (levels > amd_iommu_max_glx_val) | |
3432 | return -EINVAL; | |
3433 | ||
3434 | spin_lock_irqsave(&domain->lock, flags); | |
3435 | ||
3436 | /* | |
3437 | * Save us all sanity checks whether devices already in the | |
3438 | * domain support IOMMUv2. Just force that the domain has no | |
3439 | * devices attached when it is switched into IOMMUv2 mode. | |
3440 | */ | |
3441 | ret = -EBUSY; | |
3442 | if (domain->dev_cnt > 0 || domain->flags & PD_IOMMUV2_MASK) | |
3443 | goto out; | |
3444 | ||
3445 | ret = -ENOMEM; | |
3446 | domain->gcr3_tbl = (void *)get_zeroed_page(GFP_ATOMIC); | |
3447 | if (domain->gcr3_tbl == NULL) | |
3448 | goto out; | |
3449 | ||
3450 | domain->glx = levels; | |
3451 | domain->flags |= PD_IOMMUV2_MASK; | |
3452 | domain->updated = true; | |
3453 | ||
3454 | update_domain(domain); | |
3455 | ||
3456 | ret = 0; | |
3457 | ||
3458 | out: | |
3459 | spin_unlock_irqrestore(&domain->lock, flags); | |
3460 | ||
3461 | return ret; | |
3462 | } | |
3463 | EXPORT_SYMBOL(amd_iommu_domain_enable_v2); | |
22e266c7 JR |
3464 | |
3465 | static int __flush_pasid(struct protection_domain *domain, int pasid, | |
3466 | u64 address, bool size) | |
3467 | { | |
3468 | struct iommu_dev_data *dev_data; | |
3469 | struct iommu_cmd cmd; | |
3470 | int i, ret; | |
3471 | ||
3472 | if (!(domain->flags & PD_IOMMUV2_MASK)) | |
3473 | return -EINVAL; | |
3474 | ||
3475 | build_inv_iommu_pasid(&cmd, domain->id, pasid, address, size); | |
3476 | ||
3477 | /* | |
3478 | * IOMMU TLB needs to be flushed before Device TLB to | |
3479 | * prevent device TLB refill from IOMMU TLB | |
3480 | */ | |
3481 | for (i = 0; i < amd_iommus_present; ++i) { | |
3482 | if (domain->dev_iommu[i] == 0) | |
3483 | continue; | |
3484 | ||
3485 | ret = iommu_queue_command(amd_iommus[i], &cmd); | |
3486 | if (ret != 0) | |
3487 | goto out; | |
3488 | } | |
3489 | ||
3490 | /* Wait until IOMMU TLB flushes are complete */ | |
3491 | domain_flush_complete(domain); | |
3492 | ||
3493 | /* Now flush device TLBs */ | |
3494 | list_for_each_entry(dev_data, &domain->dev_list, list) { | |
3495 | struct amd_iommu *iommu; | |
3496 | int qdep; | |
3497 | ||
3498 | BUG_ON(!dev_data->ats.enabled); | |
3499 | ||
3500 | qdep = dev_data->ats.qdep; | |
3501 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
3502 | ||
3503 | build_inv_iotlb_pasid(&cmd, dev_data->devid, pasid, | |
3504 | qdep, address, size); | |
3505 | ||
3506 | ret = iommu_queue_command(iommu, &cmd); | |
3507 | if (ret != 0) | |
3508 | goto out; | |
3509 | } | |
3510 | ||
3511 | /* Wait until all device TLBs are flushed */ | |
3512 | domain_flush_complete(domain); | |
3513 | ||
3514 | ret = 0; | |
3515 | ||
3516 | out: | |
3517 | ||
3518 | return ret; | |
3519 | } | |
3520 | ||
3521 | static int __amd_iommu_flush_page(struct protection_domain *domain, int pasid, | |
3522 | u64 address) | |
3523 | { | |
399be2f5 JR |
3524 | INC_STATS_COUNTER(invalidate_iotlb); |
3525 | ||
22e266c7 JR |
3526 | return __flush_pasid(domain, pasid, address, false); |
3527 | } | |
3528 | ||
3529 | int amd_iommu_flush_page(struct iommu_domain *dom, int pasid, | |
3530 | u64 address) | |
3531 | { | |
3532 | struct protection_domain *domain = dom->priv; | |
3533 | unsigned long flags; | |
3534 | int ret; | |
3535 | ||
3536 | spin_lock_irqsave(&domain->lock, flags); | |
3537 | ret = __amd_iommu_flush_page(domain, pasid, address); | |
3538 | spin_unlock_irqrestore(&domain->lock, flags); | |
3539 | ||
3540 | return ret; | |
3541 | } | |
3542 | EXPORT_SYMBOL(amd_iommu_flush_page); | |
3543 | ||
3544 | static int __amd_iommu_flush_tlb(struct protection_domain *domain, int pasid) | |
3545 | { | |
399be2f5 JR |
3546 | INC_STATS_COUNTER(invalidate_iotlb_all); |
3547 | ||
22e266c7 JR |
3548 | return __flush_pasid(domain, pasid, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, |
3549 | true); | |
3550 | } | |
3551 | ||
3552 | int amd_iommu_flush_tlb(struct iommu_domain *dom, int pasid) | |
3553 | { | |
3554 | struct protection_domain *domain = dom->priv; | |
3555 | unsigned long flags; | |
3556 | int ret; | |
3557 | ||
3558 | spin_lock_irqsave(&domain->lock, flags); | |
3559 | ret = __amd_iommu_flush_tlb(domain, pasid); | |
3560 | spin_unlock_irqrestore(&domain->lock, flags); | |
3561 | ||
3562 | return ret; | |
3563 | } | |
3564 | EXPORT_SYMBOL(amd_iommu_flush_tlb); | |
3565 | ||
b16137b1 JR |
3566 | static u64 *__get_gcr3_pte(u64 *root, int level, int pasid, bool alloc) |
3567 | { | |
3568 | int index; | |
3569 | u64 *pte; | |
3570 | ||
3571 | while (true) { | |
3572 | ||
3573 | index = (pasid >> (9 * level)) & 0x1ff; | |
3574 | pte = &root[index]; | |
3575 | ||
3576 | if (level == 0) | |
3577 | break; | |
3578 | ||
3579 | if (!(*pte & GCR3_VALID)) { | |
3580 | if (!alloc) | |
3581 | return NULL; | |
3582 | ||
3583 | root = (void *)get_zeroed_page(GFP_ATOMIC); | |
3584 | if (root == NULL) | |
3585 | return NULL; | |
3586 | ||
3587 | *pte = __pa(root) | GCR3_VALID; | |
3588 | } | |
3589 | ||
3590 | root = __va(*pte & PAGE_MASK); | |
3591 | ||
3592 | level -= 1; | |
3593 | } | |
3594 | ||
3595 | return pte; | |
3596 | } | |
3597 | ||
3598 | static int __set_gcr3(struct protection_domain *domain, int pasid, | |
3599 | unsigned long cr3) | |
3600 | { | |
3601 | u64 *pte; | |
3602 | ||
3603 | if (domain->mode != PAGE_MODE_NONE) | |
3604 | return -EINVAL; | |
3605 | ||
3606 | pte = __get_gcr3_pte(domain->gcr3_tbl, domain->glx, pasid, true); | |
3607 | if (pte == NULL) | |
3608 | return -ENOMEM; | |
3609 | ||
3610 | *pte = (cr3 & PAGE_MASK) | GCR3_VALID; | |
3611 | ||
3612 | return __amd_iommu_flush_tlb(domain, pasid); | |
3613 | } | |
3614 | ||
3615 | static int __clear_gcr3(struct protection_domain *domain, int pasid) | |
3616 | { | |
3617 | u64 *pte; | |
3618 | ||
3619 | if (domain->mode != PAGE_MODE_NONE) | |
3620 | return -EINVAL; | |
3621 | ||
3622 | pte = __get_gcr3_pte(domain->gcr3_tbl, domain->glx, pasid, false); | |
3623 | if (pte == NULL) | |
3624 | return 0; | |
3625 | ||
3626 | *pte = 0; | |
3627 | ||
3628 | return __amd_iommu_flush_tlb(domain, pasid); | |
3629 | } | |
3630 | ||
3631 | int amd_iommu_domain_set_gcr3(struct iommu_domain *dom, int pasid, | |
3632 | unsigned long cr3) | |
3633 | { | |
3634 | struct protection_domain *domain = dom->priv; | |
3635 | unsigned long flags; | |
3636 | int ret; | |
3637 | ||
3638 | spin_lock_irqsave(&domain->lock, flags); | |
3639 | ret = __set_gcr3(domain, pasid, cr3); | |
3640 | spin_unlock_irqrestore(&domain->lock, flags); | |
3641 | ||
3642 | return ret; | |
3643 | } | |
3644 | EXPORT_SYMBOL(amd_iommu_domain_set_gcr3); | |
3645 | ||
3646 | int amd_iommu_domain_clear_gcr3(struct iommu_domain *dom, int pasid) | |
3647 | { | |
3648 | struct protection_domain *domain = dom->priv; | |
3649 | unsigned long flags; | |
3650 | int ret; | |
3651 | ||
3652 | spin_lock_irqsave(&domain->lock, flags); | |
3653 | ret = __clear_gcr3(domain, pasid); | |
3654 | spin_unlock_irqrestore(&domain->lock, flags); | |
3655 | ||
3656 | return ret; | |
3657 | } | |
3658 | EXPORT_SYMBOL(amd_iommu_domain_clear_gcr3); | |
c99afa25 JR |
3659 | |
3660 | int amd_iommu_complete_ppr(struct pci_dev *pdev, int pasid, | |
3661 | int status, int tag) | |
3662 | { | |
3663 | struct iommu_dev_data *dev_data; | |
3664 | struct amd_iommu *iommu; | |
3665 | struct iommu_cmd cmd; | |
3666 | ||
399be2f5 JR |
3667 | INC_STATS_COUNTER(complete_ppr); |
3668 | ||
c99afa25 JR |
3669 | dev_data = get_dev_data(&pdev->dev); |
3670 | iommu = amd_iommu_rlookup_table[dev_data->devid]; | |
3671 | ||
3672 | build_complete_ppr(&cmd, dev_data->devid, pasid, status, | |
3673 | tag, dev_data->pri_tlp); | |
3674 | ||
3675 | return iommu_queue_command(iommu, &cmd); | |
3676 | } | |
3677 | EXPORT_SYMBOL(amd_iommu_complete_ppr); | |
f3572db8 JR |
3678 | |
3679 | struct iommu_domain *amd_iommu_get_v2_domain(struct pci_dev *pdev) | |
3680 | { | |
3681 | struct protection_domain *domain; | |
3682 | ||
3683 | domain = get_domain(&pdev->dev); | |
3684 | if (IS_ERR(domain)) | |
3685 | return NULL; | |
3686 | ||
3687 | /* Only return IOMMUv2 domains */ | |
3688 | if (!(domain->flags & PD_IOMMUV2_MASK)) | |
3689 | return NULL; | |
3690 | ||
3691 | return domain->iommu_domain; | |
3692 | } | |
3693 | EXPORT_SYMBOL(amd_iommu_get_v2_domain); | |
6a113ddc JR |
3694 | |
3695 | void amd_iommu_enable_device_erratum(struct pci_dev *pdev, u32 erratum) | |
3696 | { | |
3697 | struct iommu_dev_data *dev_data; | |
3698 | ||
3699 | if (!amd_iommu_v2_supported()) | |
3700 | return; | |
3701 | ||
3702 | dev_data = get_dev_data(&pdev->dev); | |
3703 | dev_data->errata |= (1 << erratum); | |
3704 | } | |
3705 | EXPORT_SYMBOL(amd_iommu_enable_device_erratum); | |
52efdb89 JR |
3706 | |
3707 | int amd_iommu_device_info(struct pci_dev *pdev, | |
3708 | struct amd_iommu_device_info *info) | |
3709 | { | |
3710 | int max_pasids; | |
3711 | int pos; | |
3712 | ||
3713 | if (pdev == NULL || info == NULL) | |
3714 | return -EINVAL; | |
3715 | ||
3716 | if (!amd_iommu_v2_supported()) | |
3717 | return -EINVAL; | |
3718 | ||
3719 | memset(info, 0, sizeof(*info)); | |
3720 | ||
3721 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ATS); | |
3722 | if (pos) | |
3723 | info->flags |= AMD_IOMMU_DEVICE_FLAG_ATS_SUP; | |
3724 | ||
3725 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PRI); | |
3726 | if (pos) | |
3727 | info->flags |= AMD_IOMMU_DEVICE_FLAG_PRI_SUP; | |
3728 | ||
3729 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_PASID); | |
3730 | if (pos) { | |
3731 | int features; | |
3732 | ||
3733 | max_pasids = 1 << (9 * (amd_iommu_max_glx_val + 1)); | |
3734 | max_pasids = min(max_pasids, (1 << 20)); | |
3735 | ||
3736 | info->flags |= AMD_IOMMU_DEVICE_FLAG_PASID_SUP; | |
3737 | info->max_pasids = min(pci_max_pasids(pdev), max_pasids); | |
3738 | ||
3739 | features = pci_pasid_features(pdev); | |
3740 | if (features & PCI_PASID_CAP_EXEC) | |
3741 | info->flags |= AMD_IOMMU_DEVICE_FLAG_EXEC_SUP; | |
3742 | if (features & PCI_PASID_CAP_PRIV) | |
3743 | info->flags |= AMD_IOMMU_DEVICE_FLAG_PRIV_SUP; | |
3744 | } | |
3745 | ||
3746 | return 0; | |
3747 | } | |
3748 | EXPORT_SYMBOL(amd_iommu_device_info); |