]>
Commit | Line | Data |
---|---|---|
5ffd229c AK |
1 | /* |
2 | * VFIO: IOMMU DMA mapping support for TCE on POWER | |
3 | * | |
4 | * Copyright (C) 2013 IBM Corp. All rights reserved. | |
5 | * Author: Alexey Kardashevskiy <aik@ozlabs.ru> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | * | |
11 | * Derived from original vfio_iommu_type1.c: | |
12 | * Copyright (C) 2012 Red Hat, Inc. All rights reserved. | |
13 | * Author: Alex Williamson <alex.williamson@redhat.com> | |
14 | */ | |
15 | ||
16 | #include <linux/module.h> | |
17 | #include <linux/pci.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/uaccess.h> | |
20 | #include <linux/err.h> | |
21 | #include <linux/vfio.h> | |
2157e7b8 | 22 | #include <linux/vmalloc.h> |
5ffd229c AK |
23 | #include <asm/iommu.h> |
24 | #include <asm/tce.h> | |
2157e7b8 | 25 | #include <asm/mmu_context.h> |
5ffd229c AK |
26 | |
27 | #define DRIVER_VERSION "0.1" | |
28 | #define DRIVER_AUTHOR "aik@ozlabs.ru" | |
29 | #define DRIVER_DESC "VFIO IOMMU SPAPR TCE" | |
30 | ||
31 | static void tce_iommu_detach_group(void *iommu_data, | |
32 | struct iommu_group *iommu_group); | |
33 | ||
bc82d122 | 34 | static long try_increment_locked_vm(struct mm_struct *mm, long npages) |
2d270df8 AK |
35 | { |
36 | long ret = 0, locked, lock_limit; | |
37 | ||
bc82d122 AK |
38 | if (WARN_ON_ONCE(!mm)) |
39 | return -EPERM; | |
2d270df8 AK |
40 | |
41 | if (!npages) | |
42 | return 0; | |
43 | ||
bc82d122 AK |
44 | down_write(&mm->mmap_sem); |
45 | locked = mm->locked_vm + npages; | |
2d270df8 AK |
46 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; |
47 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
48 | ret = -ENOMEM; | |
49 | else | |
bc82d122 | 50 | mm->locked_vm += npages; |
2d270df8 AK |
51 | |
52 | pr_debug("[%d] RLIMIT_MEMLOCK +%ld %ld/%ld%s\n", current->pid, | |
53 | npages << PAGE_SHIFT, | |
bc82d122 | 54 | mm->locked_vm << PAGE_SHIFT, |
2d270df8 AK |
55 | rlimit(RLIMIT_MEMLOCK), |
56 | ret ? " - exceeded" : ""); | |
57 | ||
bc82d122 | 58 | up_write(&mm->mmap_sem); |
2d270df8 AK |
59 | |
60 | return ret; | |
61 | } | |
62 | ||
bc82d122 | 63 | static void decrement_locked_vm(struct mm_struct *mm, long npages) |
2d270df8 | 64 | { |
bc82d122 AK |
65 | if (!mm || !npages) |
66 | return; | |
2d270df8 | 67 | |
bc82d122 AK |
68 | down_write(&mm->mmap_sem); |
69 | if (WARN_ON_ONCE(npages > mm->locked_vm)) | |
70 | npages = mm->locked_vm; | |
71 | mm->locked_vm -= npages; | |
2d270df8 AK |
72 | pr_debug("[%d] RLIMIT_MEMLOCK -%ld %ld/%ld\n", current->pid, |
73 | npages << PAGE_SHIFT, | |
bc82d122 | 74 | mm->locked_vm << PAGE_SHIFT, |
2d270df8 | 75 | rlimit(RLIMIT_MEMLOCK)); |
bc82d122 | 76 | up_write(&mm->mmap_sem); |
2d270df8 AK |
77 | } |
78 | ||
5ffd229c AK |
79 | /* |
80 | * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation | |
81 | * | |
82 | * This code handles mapping and unmapping of user data buffers | |
83 | * into DMA'ble space using the IOMMU | |
84 | */ | |
85 | ||
2157e7b8 AK |
86 | struct tce_iommu_group { |
87 | struct list_head next; | |
88 | struct iommu_group *grp; | |
89 | }; | |
90 | ||
4b6fad70 AK |
91 | /* |
92 | * A container needs to remember which preregistered region it has | |
93 | * referenced to do proper cleanup at the userspace process exit. | |
94 | */ | |
95 | struct tce_iommu_prereg { | |
96 | struct list_head next; | |
97 | struct mm_iommu_table_group_mem_t *mem; | |
98 | }; | |
99 | ||
5ffd229c AK |
100 | /* |
101 | * The container descriptor supports only a single group per container. | |
102 | * Required by the API as the container is not supplied with the IOMMU group | |
103 | * at the moment of initialization. | |
104 | */ | |
105 | struct tce_container { | |
106 | struct mutex lock; | |
5ffd229c | 107 | bool enabled; |
2157e7b8 | 108 | bool v2; |
d9c72894 | 109 | bool def_window_pending; |
2d270df8 | 110 | unsigned long locked_pages; |
bc82d122 | 111 | struct mm_struct *mm; |
2157e7b8 AK |
112 | struct iommu_table *tables[IOMMU_TABLE_GROUP_MAX_TABLES]; |
113 | struct list_head group_list; | |
4b6fad70 | 114 | struct list_head prereg_list; |
5ffd229c AK |
115 | }; |
116 | ||
bc82d122 AK |
117 | static long tce_iommu_mm_set(struct tce_container *container) |
118 | { | |
119 | if (container->mm) { | |
120 | if (container->mm == current->mm) | |
121 | return 0; | |
122 | return -EPERM; | |
123 | } | |
124 | BUG_ON(!current->mm); | |
125 | container->mm = current->mm; | |
126 | atomic_inc(&container->mm->mm_count); | |
127 | ||
128 | return 0; | |
129 | } | |
130 | ||
4b6fad70 AK |
131 | static long tce_iommu_prereg_free(struct tce_container *container, |
132 | struct tce_iommu_prereg *tcemem) | |
133 | { | |
134 | long ret; | |
135 | ||
136 | ret = mm_iommu_put(container->mm, tcemem->mem); | |
137 | if (ret) | |
138 | return ret; | |
139 | ||
140 | list_del(&tcemem->next); | |
141 | kfree(tcemem); | |
142 | ||
143 | return 0; | |
144 | } | |
145 | ||
2157e7b8 AK |
146 | static long tce_iommu_unregister_pages(struct tce_container *container, |
147 | __u64 vaddr, __u64 size) | |
148 | { | |
149 | struct mm_iommu_table_group_mem_t *mem; | |
4b6fad70 AK |
150 | struct tce_iommu_prereg *tcemem; |
151 | bool found = false; | |
2157e7b8 AK |
152 | |
153 | if ((vaddr & ~PAGE_MASK) || (size & ~PAGE_MASK)) | |
154 | return -EINVAL; | |
155 | ||
bc82d122 | 156 | mem = mm_iommu_find(container->mm, vaddr, size >> PAGE_SHIFT); |
2157e7b8 AK |
157 | if (!mem) |
158 | return -ENOENT; | |
159 | ||
4b6fad70 AK |
160 | list_for_each_entry(tcemem, &container->prereg_list, next) { |
161 | if (tcemem->mem == mem) { | |
162 | found = true; | |
163 | break; | |
164 | } | |
165 | } | |
166 | ||
167 | if (!found) | |
168 | return -ENOENT; | |
169 | ||
170 | return tce_iommu_prereg_free(container, tcemem); | |
2157e7b8 AK |
171 | } |
172 | ||
173 | static long tce_iommu_register_pages(struct tce_container *container, | |
174 | __u64 vaddr, __u64 size) | |
175 | { | |
176 | long ret = 0; | |
177 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
4b6fad70 | 178 | struct tce_iommu_prereg *tcemem; |
2157e7b8 AK |
179 | unsigned long entries = size >> PAGE_SHIFT; |
180 | ||
181 | if ((vaddr & ~PAGE_MASK) || (size & ~PAGE_MASK) || | |
182 | ((vaddr + size) < vaddr)) | |
183 | return -EINVAL; | |
184 | ||
4b6fad70 AK |
185 | mem = mm_iommu_find(container->mm, vaddr, entries); |
186 | if (mem) { | |
187 | list_for_each_entry(tcemem, &container->prereg_list, next) { | |
188 | if (tcemem->mem == mem) | |
189 | return -EBUSY; | |
190 | } | |
191 | } | |
192 | ||
bc82d122 | 193 | ret = mm_iommu_get(container->mm, vaddr, entries, &mem); |
2157e7b8 AK |
194 | if (ret) |
195 | return ret; | |
196 | ||
4b6fad70 AK |
197 | tcemem = kzalloc(sizeof(*tcemem), GFP_KERNEL); |
198 | tcemem->mem = mem; | |
199 | list_add(&tcemem->next, &container->prereg_list); | |
200 | ||
2157e7b8 AK |
201 | container->enabled = true; |
202 | ||
203 | return 0; | |
204 | } | |
205 | ||
bc82d122 AK |
206 | static long tce_iommu_userspace_view_alloc(struct iommu_table *tbl, |
207 | struct mm_struct *mm) | |
2157e7b8 AK |
208 | { |
209 | unsigned long cb = _ALIGN_UP(sizeof(tbl->it_userspace[0]) * | |
210 | tbl->it_size, PAGE_SIZE); | |
211 | unsigned long *uas; | |
212 | long ret; | |
213 | ||
214 | BUG_ON(tbl->it_userspace); | |
215 | ||
bc82d122 | 216 | ret = try_increment_locked_vm(mm, cb >> PAGE_SHIFT); |
2157e7b8 AK |
217 | if (ret) |
218 | return ret; | |
219 | ||
220 | uas = vzalloc(cb); | |
221 | if (!uas) { | |
bc82d122 | 222 | decrement_locked_vm(mm, cb >> PAGE_SHIFT); |
2157e7b8 AK |
223 | return -ENOMEM; |
224 | } | |
225 | tbl->it_userspace = uas; | |
226 | ||
227 | return 0; | |
228 | } | |
229 | ||
bc82d122 AK |
230 | static void tce_iommu_userspace_view_free(struct iommu_table *tbl, |
231 | struct mm_struct *mm) | |
2157e7b8 AK |
232 | { |
233 | unsigned long cb = _ALIGN_UP(sizeof(tbl->it_userspace[0]) * | |
234 | tbl->it_size, PAGE_SIZE); | |
235 | ||
236 | if (!tbl->it_userspace) | |
237 | return; | |
238 | ||
239 | vfree(tbl->it_userspace); | |
240 | tbl->it_userspace = NULL; | |
bc82d122 | 241 | decrement_locked_vm(mm, cb >> PAGE_SHIFT); |
2157e7b8 AK |
242 | } |
243 | ||
e432bc7e AK |
244 | static bool tce_page_is_contained(struct page *page, unsigned page_shift) |
245 | { | |
246 | /* | |
247 | * Check that the TCE table granularity is not bigger than the size of | |
248 | * a page we just found. Otherwise the hardware can get access to | |
249 | * a bigger memory chunk that it should. | |
250 | */ | |
251 | return (PAGE_SHIFT + compound_order(compound_head(page))) >= page_shift; | |
252 | } | |
253 | ||
2157e7b8 AK |
254 | static inline bool tce_groups_attached(struct tce_container *container) |
255 | { | |
256 | return !list_empty(&container->group_list); | |
257 | } | |
258 | ||
0eaf4def AK |
259 | static long tce_iommu_find_table(struct tce_container *container, |
260 | phys_addr_t ioba, struct iommu_table **ptbl) | |
261 | { | |
262 | long i; | |
0eaf4def AK |
263 | |
264 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
2157e7b8 | 265 | struct iommu_table *tbl = container->tables[i]; |
0eaf4def AK |
266 | |
267 | if (tbl) { | |
268 | unsigned long entry = ioba >> tbl->it_page_shift; | |
269 | unsigned long start = tbl->it_offset; | |
270 | unsigned long end = start + tbl->it_size; | |
271 | ||
272 | if ((start <= entry) && (entry < end)) { | |
273 | *ptbl = tbl; | |
274 | return i; | |
275 | } | |
276 | } | |
277 | } | |
278 | ||
279 | return -1; | |
280 | } | |
281 | ||
e633bc86 AK |
282 | static int tce_iommu_find_free_table(struct tce_container *container) |
283 | { | |
284 | int i; | |
285 | ||
286 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
287 | if (!container->tables[i]) | |
288 | return i; | |
289 | } | |
290 | ||
291 | return -ENOSPC; | |
292 | } | |
293 | ||
5ffd229c AK |
294 | static int tce_iommu_enable(struct tce_container *container) |
295 | { | |
296 | int ret = 0; | |
2d270df8 | 297 | unsigned long locked; |
0eaf4def | 298 | struct iommu_table_group *table_group; |
2157e7b8 | 299 | struct tce_iommu_group *tcegrp; |
5ffd229c | 300 | |
5ffd229c AK |
301 | if (container->enabled) |
302 | return -EBUSY; | |
303 | ||
304 | /* | |
305 | * When userspace pages are mapped into the IOMMU, they are effectively | |
306 | * locked memory, so, theoretically, we need to update the accounting | |
307 | * of locked pages on each map and unmap. For powerpc, the map unmap | |
308 | * paths can be very hot, though, and the accounting would kill | |
309 | * performance, especially since it would be difficult to impossible | |
310 | * to handle the accounting in real mode only. | |
311 | * | |
312 | * To address that, rather than precisely accounting every page, we | |
313 | * instead account for a worst case on locked memory when the iommu is | |
314 | * enabled and disabled. The worst case upper bound on locked memory | |
315 | * is the size of the whole iommu window, which is usually relatively | |
316 | * small (compared to total memory sizes) on POWER hardware. | |
317 | * | |
318 | * Also we don't have a nice way to fail on H_PUT_TCE due to ulimits, | |
319 | * that would effectively kill the guest at random points, much better | |
320 | * enforcing the limit based on the max that the guest can map. | |
2d270df8 AK |
321 | * |
322 | * Unfortunately at the moment it counts whole tables, no matter how | |
323 | * much memory the guest has. I.e. for 4GB guest and 4 IOMMU groups | |
324 | * each with 2GB DMA window, 8GB will be counted here. The reason for | |
325 | * this is that we cannot tell here the amount of RAM used by the guest | |
326 | * as this information is only available from KVM and VFIO is | |
327 | * KVM agnostic. | |
4793d65d AK |
328 | * |
329 | * So we do not allow enabling a container without a group attached | |
330 | * as there is no way to know how much we should increment | |
331 | * the locked_vm counter. | |
5ffd229c | 332 | */ |
2157e7b8 AK |
333 | if (!tce_groups_attached(container)) |
334 | return -ENODEV; | |
335 | ||
336 | tcegrp = list_first_entry(&container->group_list, | |
337 | struct tce_iommu_group, next); | |
338 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
0eaf4def AK |
339 | if (!table_group) |
340 | return -ENODEV; | |
341 | ||
4793d65d AK |
342 | if (!table_group->tce32_size) |
343 | return -EPERM; | |
344 | ||
bc82d122 AK |
345 | ret = tce_iommu_mm_set(container); |
346 | if (ret) | |
347 | return ret; | |
348 | ||
4793d65d | 349 | locked = table_group->tce32_size >> PAGE_SHIFT; |
bc82d122 | 350 | ret = try_increment_locked_vm(container->mm, locked); |
2d270df8 AK |
351 | if (ret) |
352 | return ret; | |
5ffd229c | 353 | |
2d270df8 AK |
354 | container->locked_pages = locked; |
355 | ||
356 | container->enabled = true; | |
5ffd229c AK |
357 | |
358 | return ret; | |
359 | } | |
360 | ||
361 | static void tce_iommu_disable(struct tce_container *container) | |
362 | { | |
363 | if (!container->enabled) | |
364 | return; | |
365 | ||
366 | container->enabled = false; | |
367 | ||
bc82d122 AK |
368 | BUG_ON(!container->mm); |
369 | decrement_locked_vm(container->mm, container->locked_pages); | |
5ffd229c AK |
370 | } |
371 | ||
372 | static void *tce_iommu_open(unsigned long arg) | |
373 | { | |
374 | struct tce_container *container; | |
375 | ||
2157e7b8 | 376 | if ((arg != VFIO_SPAPR_TCE_IOMMU) && (arg != VFIO_SPAPR_TCE_v2_IOMMU)) { |
5ffd229c AK |
377 | pr_err("tce_vfio: Wrong IOMMU type\n"); |
378 | return ERR_PTR(-EINVAL); | |
379 | } | |
380 | ||
381 | container = kzalloc(sizeof(*container), GFP_KERNEL); | |
382 | if (!container) | |
383 | return ERR_PTR(-ENOMEM); | |
384 | ||
385 | mutex_init(&container->lock); | |
2157e7b8 | 386 | INIT_LIST_HEAD_RCU(&container->group_list); |
4b6fad70 | 387 | INIT_LIST_HEAD_RCU(&container->prereg_list); |
2157e7b8 AK |
388 | |
389 | container->v2 = arg == VFIO_SPAPR_TCE_v2_IOMMU; | |
5ffd229c AK |
390 | |
391 | return container; | |
392 | } | |
393 | ||
2157e7b8 AK |
394 | static int tce_iommu_clear(struct tce_container *container, |
395 | struct iommu_table *tbl, | |
396 | unsigned long entry, unsigned long pages); | |
bc82d122 AK |
397 | static void tce_iommu_free_table(struct tce_container *container, |
398 | struct iommu_table *tbl); | |
2157e7b8 | 399 | |
5ffd229c AK |
400 | static void tce_iommu_release(void *iommu_data) |
401 | { | |
402 | struct tce_container *container = iommu_data; | |
2157e7b8 AK |
403 | struct tce_iommu_group *tcegrp; |
404 | long i; | |
5ffd229c | 405 | |
2157e7b8 AK |
406 | while (tce_groups_attached(container)) { |
407 | tcegrp = list_first_entry(&container->group_list, | |
408 | struct tce_iommu_group, next); | |
2157e7b8 AK |
409 | tce_iommu_detach_group(iommu_data, tcegrp->grp); |
410 | } | |
5ffd229c | 411 | |
2157e7b8 AK |
412 | /* |
413 | * If VFIO created a table, it was not disposed | |
414 | * by tce_iommu_detach_group() so do it now. | |
415 | */ | |
416 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
417 | struct iommu_table *tbl = container->tables[i]; | |
418 | ||
419 | if (!tbl) | |
420 | continue; | |
421 | ||
422 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
bc82d122 | 423 | tce_iommu_free_table(container, tbl); |
2157e7b8 | 424 | } |
5ffd229c | 425 | |
4b6fad70 AK |
426 | while (!list_empty(&container->prereg_list)) { |
427 | struct tce_iommu_prereg *tcemem; | |
428 | ||
429 | tcemem = list_first_entry(&container->prereg_list, | |
430 | struct tce_iommu_prereg, next); | |
431 | WARN_ON_ONCE(tce_iommu_prereg_free(container, tcemem)); | |
432 | } | |
433 | ||
649354b7 | 434 | tce_iommu_disable(container); |
bc82d122 AK |
435 | if (container->mm) |
436 | mmdrop(container->mm); | |
5ffd229c AK |
437 | mutex_destroy(&container->lock); |
438 | ||
439 | kfree(container); | |
440 | } | |
441 | ||
649354b7 | 442 | static void tce_iommu_unuse_page(struct tce_container *container, |
05c6cfb9 | 443 | unsigned long hpa) |
649354b7 AK |
444 | { |
445 | struct page *page; | |
446 | ||
05c6cfb9 | 447 | page = pfn_to_page(hpa >> PAGE_SHIFT); |
649354b7 AK |
448 | put_page(page); |
449 | } | |
450 | ||
bc82d122 AK |
451 | static int tce_iommu_prereg_ua_to_hpa(struct tce_container *container, |
452 | unsigned long tce, unsigned long size, | |
2157e7b8 AK |
453 | unsigned long *phpa, struct mm_iommu_table_group_mem_t **pmem) |
454 | { | |
455 | long ret = 0; | |
456 | struct mm_iommu_table_group_mem_t *mem; | |
457 | ||
bc82d122 | 458 | mem = mm_iommu_lookup(container->mm, tce, size); |
2157e7b8 AK |
459 | if (!mem) |
460 | return -EINVAL; | |
461 | ||
462 | ret = mm_iommu_ua_to_hpa(mem, tce, phpa); | |
463 | if (ret) | |
464 | return -EINVAL; | |
465 | ||
466 | *pmem = mem; | |
467 | ||
468 | return 0; | |
469 | } | |
470 | ||
bc82d122 AK |
471 | static void tce_iommu_unuse_page_v2(struct tce_container *container, |
472 | struct iommu_table *tbl, unsigned long entry) | |
2157e7b8 AK |
473 | { |
474 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
475 | int ret; | |
476 | unsigned long hpa = 0; | |
477 | unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry); | |
478 | ||
bc82d122 | 479 | if (!pua) |
2157e7b8 AK |
480 | return; |
481 | ||
bc82d122 | 482 | ret = tce_iommu_prereg_ua_to_hpa(container, *pua, IOMMU_PAGE_SIZE(tbl), |
2157e7b8 AK |
483 | &hpa, &mem); |
484 | if (ret) | |
485 | pr_debug("%s: tce %lx at #%lx was not cached, ret=%d\n", | |
486 | __func__, *pua, entry, ret); | |
487 | if (mem) | |
488 | mm_iommu_mapped_dec(mem); | |
489 | ||
490 | *pua = 0; | |
491 | } | |
492 | ||
9b14a1ff AK |
493 | static int tce_iommu_clear(struct tce_container *container, |
494 | struct iommu_table *tbl, | |
495 | unsigned long entry, unsigned long pages) | |
496 | { | |
05c6cfb9 AK |
497 | unsigned long oldhpa; |
498 | long ret; | |
499 | enum dma_data_direction direction; | |
9b14a1ff AK |
500 | |
501 | for ( ; pages; --pages, ++entry) { | |
05c6cfb9 AK |
502 | direction = DMA_NONE; |
503 | oldhpa = 0; | |
504 | ret = iommu_tce_xchg(tbl, entry, &oldhpa, &direction); | |
505 | if (ret) | |
506 | continue; | |
507 | ||
508 | if (direction == DMA_NONE) | |
9b14a1ff AK |
509 | continue; |
510 | ||
2157e7b8 | 511 | if (container->v2) { |
bc82d122 | 512 | tce_iommu_unuse_page_v2(container, tbl, entry); |
2157e7b8 AK |
513 | continue; |
514 | } | |
515 | ||
05c6cfb9 | 516 | tce_iommu_unuse_page(container, oldhpa); |
9b14a1ff AK |
517 | } |
518 | ||
519 | return 0; | |
520 | } | |
521 | ||
649354b7 AK |
522 | static int tce_iommu_use_page(unsigned long tce, unsigned long *hpa) |
523 | { | |
524 | struct page *page = NULL; | |
525 | enum dma_data_direction direction = iommu_tce_direction(tce); | |
526 | ||
527 | if (get_user_pages_fast(tce & PAGE_MASK, 1, | |
528 | direction != DMA_TO_DEVICE, &page) != 1) | |
529 | return -EFAULT; | |
530 | ||
531 | *hpa = __pa((unsigned long) page_address(page)); | |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
9b14a1ff AK |
536 | static long tce_iommu_build(struct tce_container *container, |
537 | struct iommu_table *tbl, | |
05c6cfb9 AK |
538 | unsigned long entry, unsigned long tce, unsigned long pages, |
539 | enum dma_data_direction direction) | |
9b14a1ff AK |
540 | { |
541 | long i, ret = 0; | |
649354b7 AK |
542 | struct page *page; |
543 | unsigned long hpa; | |
05c6cfb9 | 544 | enum dma_data_direction dirtmp; |
9b14a1ff AK |
545 | |
546 | for (i = 0; i < pages; ++i) { | |
547 | unsigned long offset = tce & IOMMU_PAGE_MASK(tbl) & ~PAGE_MASK; | |
548 | ||
649354b7 AK |
549 | ret = tce_iommu_use_page(tce, &hpa); |
550 | if (ret) | |
9b14a1ff | 551 | break; |
e432bc7e | 552 | |
649354b7 | 553 | page = pfn_to_page(hpa >> PAGE_SHIFT); |
e432bc7e AK |
554 | if (!tce_page_is_contained(page, tbl->it_page_shift)) { |
555 | ret = -EPERM; | |
556 | break; | |
557 | } | |
558 | ||
649354b7 | 559 | hpa |= offset; |
05c6cfb9 AK |
560 | dirtmp = direction; |
561 | ret = iommu_tce_xchg(tbl, entry + i, &hpa, &dirtmp); | |
9b14a1ff | 562 | if (ret) { |
649354b7 | 563 | tce_iommu_unuse_page(container, hpa); |
9b14a1ff AK |
564 | pr_err("iommu_tce: %s failed ioba=%lx, tce=%lx, ret=%ld\n", |
565 | __func__, entry << tbl->it_page_shift, | |
566 | tce, ret); | |
567 | break; | |
568 | } | |
05c6cfb9 AK |
569 | |
570 | if (dirtmp != DMA_NONE) | |
571 | tce_iommu_unuse_page(container, hpa); | |
572 | ||
00663d4e | 573 | tce += IOMMU_PAGE_SIZE(tbl); |
9b14a1ff AK |
574 | } |
575 | ||
576 | if (ret) | |
577 | tce_iommu_clear(container, tbl, entry, i); | |
578 | ||
579 | return ret; | |
580 | } | |
581 | ||
2157e7b8 AK |
582 | static long tce_iommu_build_v2(struct tce_container *container, |
583 | struct iommu_table *tbl, | |
584 | unsigned long entry, unsigned long tce, unsigned long pages, | |
585 | enum dma_data_direction direction) | |
586 | { | |
587 | long i, ret = 0; | |
588 | struct page *page; | |
589 | unsigned long hpa; | |
590 | enum dma_data_direction dirtmp; | |
591 | ||
39701e56 | 592 | if (!tbl->it_userspace) { |
bc82d122 | 593 | ret = tce_iommu_userspace_view_alloc(tbl, container->mm); |
39701e56 AK |
594 | if (ret) |
595 | return ret; | |
596 | } | |
597 | ||
2157e7b8 AK |
598 | for (i = 0; i < pages; ++i) { |
599 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
600 | unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, | |
601 | entry + i); | |
602 | ||
bc82d122 AK |
603 | ret = tce_iommu_prereg_ua_to_hpa(container, |
604 | tce, IOMMU_PAGE_SIZE(tbl), &hpa, &mem); | |
2157e7b8 AK |
605 | if (ret) |
606 | break; | |
607 | ||
608 | page = pfn_to_page(hpa >> PAGE_SHIFT); | |
609 | if (!tce_page_is_contained(page, tbl->it_page_shift)) { | |
610 | ret = -EPERM; | |
611 | break; | |
612 | } | |
613 | ||
614 | /* Preserve offset within IOMMU page */ | |
615 | hpa |= tce & IOMMU_PAGE_MASK(tbl) & ~PAGE_MASK; | |
616 | dirtmp = direction; | |
617 | ||
618 | /* The registered region is being unregistered */ | |
619 | if (mm_iommu_mapped_inc(mem)) | |
620 | break; | |
621 | ||
622 | ret = iommu_tce_xchg(tbl, entry + i, &hpa, &dirtmp); | |
623 | if (ret) { | |
624 | /* dirtmp cannot be DMA_NONE here */ | |
bc82d122 | 625 | tce_iommu_unuse_page_v2(container, tbl, entry + i); |
2157e7b8 AK |
626 | pr_err("iommu_tce: %s failed ioba=%lx, tce=%lx, ret=%ld\n", |
627 | __func__, entry << tbl->it_page_shift, | |
628 | tce, ret); | |
629 | break; | |
630 | } | |
631 | ||
632 | if (dirtmp != DMA_NONE) | |
bc82d122 | 633 | tce_iommu_unuse_page_v2(container, tbl, entry + i); |
2157e7b8 AK |
634 | |
635 | *pua = tce; | |
636 | ||
637 | tce += IOMMU_PAGE_SIZE(tbl); | |
638 | } | |
639 | ||
640 | if (ret) | |
641 | tce_iommu_clear(container, tbl, entry, i); | |
642 | ||
643 | return ret; | |
644 | } | |
645 | ||
46d3e1e1 AK |
646 | static long tce_iommu_create_table(struct tce_container *container, |
647 | struct iommu_table_group *table_group, | |
648 | int num, | |
649 | __u32 page_shift, | |
650 | __u64 window_size, | |
651 | __u32 levels, | |
652 | struct iommu_table **ptbl) | |
653 | { | |
654 | long ret, table_size; | |
655 | ||
656 | table_size = table_group->ops->get_table_size(page_shift, window_size, | |
657 | levels); | |
658 | if (!table_size) | |
659 | return -EINVAL; | |
660 | ||
bc82d122 | 661 | ret = try_increment_locked_vm(container->mm, table_size >> PAGE_SHIFT); |
46d3e1e1 AK |
662 | if (ret) |
663 | return ret; | |
664 | ||
665 | ret = table_group->ops->create_table(table_group, num, | |
666 | page_shift, window_size, levels, ptbl); | |
667 | ||
668 | WARN_ON(!ret && !(*ptbl)->it_ops->free); | |
669 | WARN_ON(!ret && ((*ptbl)->it_allocated_size != table_size)); | |
670 | ||
46d3e1e1 AK |
671 | return ret; |
672 | } | |
673 | ||
bc82d122 AK |
674 | static void tce_iommu_free_table(struct tce_container *container, |
675 | struct iommu_table *tbl) | |
46d3e1e1 AK |
676 | { |
677 | unsigned long pages = tbl->it_allocated_size >> PAGE_SHIFT; | |
678 | ||
bc82d122 | 679 | tce_iommu_userspace_view_free(tbl, container->mm); |
46d3e1e1 | 680 | tbl->it_ops->free(tbl); |
bc82d122 | 681 | decrement_locked_vm(container->mm, pages); |
46d3e1e1 AK |
682 | } |
683 | ||
e633bc86 AK |
684 | static long tce_iommu_create_window(struct tce_container *container, |
685 | __u32 page_shift, __u64 window_size, __u32 levels, | |
686 | __u64 *start_addr) | |
687 | { | |
688 | struct tce_iommu_group *tcegrp; | |
689 | struct iommu_table_group *table_group; | |
690 | struct iommu_table *tbl = NULL; | |
691 | long ret, num; | |
692 | ||
693 | num = tce_iommu_find_free_table(container); | |
694 | if (num < 0) | |
695 | return num; | |
696 | ||
697 | /* Get the first group for ops::create_table */ | |
698 | tcegrp = list_first_entry(&container->group_list, | |
699 | struct tce_iommu_group, next); | |
700 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
701 | if (!table_group) | |
702 | return -EFAULT; | |
703 | ||
704 | if (!(table_group->pgsizes & (1ULL << page_shift))) | |
705 | return -EINVAL; | |
706 | ||
707 | if (!table_group->ops->set_window || !table_group->ops->unset_window || | |
708 | !table_group->ops->get_table_size || | |
709 | !table_group->ops->create_table) | |
710 | return -EPERM; | |
711 | ||
712 | /* Create TCE table */ | |
713 | ret = tce_iommu_create_table(container, table_group, num, | |
714 | page_shift, window_size, levels, &tbl); | |
715 | if (ret) | |
716 | return ret; | |
717 | ||
718 | BUG_ON(!tbl->it_ops->free); | |
719 | ||
720 | /* | |
721 | * Program the table to every group. | |
722 | * Groups have been tested for compatibility at the attach time. | |
723 | */ | |
724 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
725 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
726 | ||
727 | ret = table_group->ops->set_window(table_group, num, tbl); | |
728 | if (ret) | |
729 | goto unset_exit; | |
730 | } | |
731 | ||
732 | container->tables[num] = tbl; | |
733 | ||
734 | /* Return start address assigned by platform in create_table() */ | |
735 | *start_addr = tbl->it_offset << tbl->it_page_shift; | |
736 | ||
737 | return 0; | |
738 | ||
739 | unset_exit: | |
740 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
741 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
742 | table_group->ops->unset_window(table_group, num); | |
743 | } | |
bc82d122 | 744 | tce_iommu_free_table(container, tbl); |
e633bc86 AK |
745 | |
746 | return ret; | |
747 | } | |
748 | ||
749 | static long tce_iommu_remove_window(struct tce_container *container, | |
750 | __u64 start_addr) | |
751 | { | |
752 | struct iommu_table_group *table_group = NULL; | |
753 | struct iommu_table *tbl; | |
754 | struct tce_iommu_group *tcegrp; | |
755 | int num; | |
756 | ||
757 | num = tce_iommu_find_table(container, start_addr, &tbl); | |
758 | if (num < 0) | |
759 | return -EINVAL; | |
760 | ||
761 | BUG_ON(!tbl->it_size); | |
762 | ||
763 | /* Detach groups from IOMMUs */ | |
764 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
765 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
766 | ||
767 | /* | |
768 | * SPAPR TCE IOMMU exposes the default DMA window to | |
769 | * the guest via dma32_window_start/size of | |
770 | * VFIO_IOMMU_SPAPR_TCE_GET_INFO. Some platforms allow | |
771 | * the userspace to remove this window, some do not so | |
772 | * here we check for the platform capability. | |
773 | */ | |
774 | if (!table_group->ops || !table_group->ops->unset_window) | |
775 | return -EPERM; | |
776 | ||
777 | table_group->ops->unset_window(table_group, num); | |
778 | } | |
779 | ||
780 | /* Free table */ | |
781 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
bc82d122 | 782 | tce_iommu_free_table(container, tbl); |
e633bc86 AK |
783 | container->tables[num] = NULL; |
784 | ||
785 | return 0; | |
786 | } | |
787 | ||
6f01cc69 AK |
788 | static long tce_iommu_create_default_window(struct tce_container *container) |
789 | { | |
790 | long ret; | |
791 | __u64 start_addr = 0; | |
792 | struct tce_iommu_group *tcegrp; | |
793 | struct iommu_table_group *table_group; | |
794 | ||
d9c72894 AK |
795 | if (!container->def_window_pending) |
796 | return 0; | |
797 | ||
6f01cc69 AK |
798 | if (!tce_groups_attached(container)) |
799 | return -ENODEV; | |
800 | ||
801 | tcegrp = list_first_entry(&container->group_list, | |
802 | struct tce_iommu_group, next); | |
803 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
804 | if (!table_group) | |
805 | return -ENODEV; | |
806 | ||
807 | ret = tce_iommu_create_window(container, IOMMU_PAGE_SHIFT_4K, | |
808 | table_group->tce32_size, 1, &start_addr); | |
809 | WARN_ON_ONCE(!ret && start_addr); | |
810 | ||
d9c72894 AK |
811 | if (!ret) |
812 | container->def_window_pending = false; | |
813 | ||
6f01cc69 AK |
814 | return ret; |
815 | } | |
816 | ||
5ffd229c AK |
817 | static long tce_iommu_ioctl(void *iommu_data, |
818 | unsigned int cmd, unsigned long arg) | |
819 | { | |
820 | struct tce_container *container = iommu_data; | |
e633bc86 | 821 | unsigned long minsz, ddwsz; |
5ffd229c AK |
822 | long ret; |
823 | ||
824 | switch (cmd) { | |
825 | case VFIO_CHECK_EXTENSION: | |
1b69be5e GS |
826 | switch (arg) { |
827 | case VFIO_SPAPR_TCE_IOMMU: | |
2157e7b8 | 828 | case VFIO_SPAPR_TCE_v2_IOMMU: |
1b69be5e GS |
829 | ret = 1; |
830 | break; | |
831 | default: | |
832 | ret = vfio_spapr_iommu_eeh_ioctl(NULL, cmd, arg); | |
833 | break; | |
834 | } | |
835 | ||
836 | return (ret < 0) ? 0 : ret; | |
bc82d122 AK |
837 | } |
838 | ||
839 | /* | |
840 | * Sanity check to prevent one userspace from manipulating | |
841 | * another userspace mm. | |
842 | */ | |
843 | BUG_ON(!container); | |
844 | if (container->mm && container->mm != current->mm) | |
845 | return -EPERM; | |
5ffd229c | 846 | |
bc82d122 | 847 | switch (cmd) { |
5ffd229c AK |
848 | case VFIO_IOMMU_SPAPR_TCE_GET_INFO: { |
849 | struct vfio_iommu_spapr_tce_info info; | |
2157e7b8 | 850 | struct tce_iommu_group *tcegrp; |
0eaf4def AK |
851 | struct iommu_table_group *table_group; |
852 | ||
2157e7b8 | 853 | if (!tce_groups_attached(container)) |
0eaf4def AK |
854 | return -ENXIO; |
855 | ||
2157e7b8 AK |
856 | tcegrp = list_first_entry(&container->group_list, |
857 | struct tce_iommu_group, next); | |
858 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
5ffd229c | 859 | |
4793d65d | 860 | if (!table_group) |
5ffd229c AK |
861 | return -ENXIO; |
862 | ||
863 | minsz = offsetofend(struct vfio_iommu_spapr_tce_info, | |
864 | dma32_window_size); | |
865 | ||
866 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
867 | return -EFAULT; | |
868 | ||
869 | if (info.argsz < minsz) | |
870 | return -EINVAL; | |
871 | ||
4793d65d AK |
872 | info.dma32_window_start = table_group->tce32_start; |
873 | info.dma32_window_size = table_group->tce32_size; | |
5ffd229c | 874 | info.flags = 0; |
e633bc86 AK |
875 | memset(&info.ddw, 0, sizeof(info.ddw)); |
876 | ||
877 | if (table_group->max_dynamic_windows_supported && | |
878 | container->v2) { | |
879 | info.flags |= VFIO_IOMMU_SPAPR_INFO_DDW; | |
880 | info.ddw.pgsizes = table_group->pgsizes; | |
881 | info.ddw.max_dynamic_windows_supported = | |
882 | table_group->max_dynamic_windows_supported; | |
883 | info.ddw.levels = table_group->max_levels; | |
884 | } | |
885 | ||
886 | ddwsz = offsetofend(struct vfio_iommu_spapr_tce_info, ddw); | |
887 | ||
888 | if (info.argsz >= ddwsz) | |
889 | minsz = ddwsz; | |
5ffd229c AK |
890 | |
891 | if (copy_to_user((void __user *)arg, &info, minsz)) | |
892 | return -EFAULT; | |
893 | ||
894 | return 0; | |
895 | } | |
896 | case VFIO_IOMMU_MAP_DMA: { | |
897 | struct vfio_iommu_type1_dma_map param; | |
0eaf4def | 898 | struct iommu_table *tbl = NULL; |
0eaf4def | 899 | long num; |
05c6cfb9 | 900 | enum dma_data_direction direction; |
5ffd229c | 901 | |
3c56e822 AK |
902 | if (!container->enabled) |
903 | return -EPERM; | |
904 | ||
5ffd229c AK |
905 | minsz = offsetofend(struct vfio_iommu_type1_dma_map, size); |
906 | ||
907 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
908 | return -EFAULT; | |
909 | ||
910 | if (param.argsz < minsz) | |
911 | return -EINVAL; | |
912 | ||
913 | if (param.flags & ~(VFIO_DMA_MAP_FLAG_READ | | |
914 | VFIO_DMA_MAP_FLAG_WRITE)) | |
915 | return -EINVAL; | |
916 | ||
d9c72894 AK |
917 | ret = tce_iommu_create_default_window(container); |
918 | if (ret) | |
919 | return ret; | |
920 | ||
0eaf4def AK |
921 | num = tce_iommu_find_table(container, param.iova, &tbl); |
922 | if (num < 0) | |
923 | return -ENXIO; | |
924 | ||
00663d4e AK |
925 | if ((param.size & ~IOMMU_PAGE_MASK(tbl)) || |
926 | (param.vaddr & ~IOMMU_PAGE_MASK(tbl))) | |
5ffd229c AK |
927 | return -EINVAL; |
928 | ||
929 | /* iova is checked by the IOMMU API */ | |
05c6cfb9 AK |
930 | if (param.flags & VFIO_DMA_MAP_FLAG_READ) { |
931 | if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) | |
932 | direction = DMA_BIDIRECTIONAL; | |
933 | else | |
934 | direction = DMA_TO_DEVICE; | |
935 | } else { | |
936 | if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) | |
937 | direction = DMA_FROM_DEVICE; | |
938 | else | |
939 | return -EINVAL; | |
940 | } | |
5ffd229c | 941 | |
05c6cfb9 | 942 | ret = iommu_tce_put_param_check(tbl, param.iova, param.vaddr); |
5ffd229c AK |
943 | if (ret) |
944 | return ret; | |
945 | ||
2157e7b8 AK |
946 | if (container->v2) |
947 | ret = tce_iommu_build_v2(container, tbl, | |
948 | param.iova >> tbl->it_page_shift, | |
949 | param.vaddr, | |
950 | param.size >> tbl->it_page_shift, | |
951 | direction); | |
952 | else | |
953 | ret = tce_iommu_build(container, tbl, | |
954 | param.iova >> tbl->it_page_shift, | |
955 | param.vaddr, | |
956 | param.size >> tbl->it_page_shift, | |
957 | direction); | |
5ffd229c AK |
958 | |
959 | iommu_flush_tce(tbl); | |
960 | ||
961 | return ret; | |
962 | } | |
963 | case VFIO_IOMMU_UNMAP_DMA: { | |
964 | struct vfio_iommu_type1_dma_unmap param; | |
0eaf4def AK |
965 | struct iommu_table *tbl = NULL; |
966 | long num; | |
5ffd229c | 967 | |
3c56e822 AK |
968 | if (!container->enabled) |
969 | return -EPERM; | |
970 | ||
5ffd229c AK |
971 | minsz = offsetofend(struct vfio_iommu_type1_dma_unmap, |
972 | size); | |
973 | ||
974 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
975 | return -EFAULT; | |
976 | ||
977 | if (param.argsz < minsz) | |
978 | return -EINVAL; | |
979 | ||
980 | /* No flag is supported now */ | |
981 | if (param.flags) | |
982 | return -EINVAL; | |
983 | ||
d9c72894 AK |
984 | ret = tce_iommu_create_default_window(container); |
985 | if (ret) | |
986 | return ret; | |
987 | ||
0eaf4def AK |
988 | num = tce_iommu_find_table(container, param.iova, &tbl); |
989 | if (num < 0) | |
990 | return -ENXIO; | |
991 | ||
00663d4e | 992 | if (param.size & ~IOMMU_PAGE_MASK(tbl)) |
5ffd229c AK |
993 | return -EINVAL; |
994 | ||
995 | ret = iommu_tce_clear_param_check(tbl, param.iova, 0, | |
00663d4e | 996 | param.size >> tbl->it_page_shift); |
5ffd229c AK |
997 | if (ret) |
998 | return ret; | |
999 | ||
9b14a1ff | 1000 | ret = tce_iommu_clear(container, tbl, |
00663d4e AK |
1001 | param.iova >> tbl->it_page_shift, |
1002 | param.size >> tbl->it_page_shift); | |
5ffd229c AK |
1003 | iommu_flush_tce(tbl); |
1004 | ||
1005 | return ret; | |
1006 | } | |
2157e7b8 AK |
1007 | case VFIO_IOMMU_SPAPR_REGISTER_MEMORY: { |
1008 | struct vfio_iommu_spapr_register_memory param; | |
1009 | ||
1010 | if (!container->v2) | |
1011 | break; | |
1012 | ||
1013 | minsz = offsetofend(struct vfio_iommu_spapr_register_memory, | |
1014 | size); | |
1015 | ||
bc82d122 AK |
1016 | ret = tce_iommu_mm_set(container); |
1017 | if (ret) | |
1018 | return ret; | |
1019 | ||
2157e7b8 AK |
1020 | if (copy_from_user(¶m, (void __user *)arg, minsz)) |
1021 | return -EFAULT; | |
1022 | ||
1023 | if (param.argsz < minsz) | |
1024 | return -EINVAL; | |
1025 | ||
1026 | /* No flag is supported now */ | |
1027 | if (param.flags) | |
1028 | return -EINVAL; | |
1029 | ||
1030 | mutex_lock(&container->lock); | |
1031 | ret = tce_iommu_register_pages(container, param.vaddr, | |
1032 | param.size); | |
1033 | mutex_unlock(&container->lock); | |
1034 | ||
1035 | return ret; | |
1036 | } | |
1037 | case VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY: { | |
1038 | struct vfio_iommu_spapr_register_memory param; | |
1039 | ||
1040 | if (!container->v2) | |
1041 | break; | |
1042 | ||
bc82d122 AK |
1043 | if (!container->mm) |
1044 | return -EPERM; | |
1045 | ||
2157e7b8 AK |
1046 | minsz = offsetofend(struct vfio_iommu_spapr_register_memory, |
1047 | size); | |
1048 | ||
1049 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
1050 | return -EFAULT; | |
1051 | ||
1052 | if (param.argsz < minsz) | |
1053 | return -EINVAL; | |
1054 | ||
1055 | /* No flag is supported now */ | |
1056 | if (param.flags) | |
1057 | return -EINVAL; | |
1058 | ||
1059 | mutex_lock(&container->lock); | |
1060 | ret = tce_iommu_unregister_pages(container, param.vaddr, | |
1061 | param.size); | |
1062 | mutex_unlock(&container->lock); | |
1063 | ||
1064 | return ret; | |
1065 | } | |
5ffd229c | 1066 | case VFIO_IOMMU_ENABLE: |
2157e7b8 AK |
1067 | if (container->v2) |
1068 | break; | |
1069 | ||
5ffd229c AK |
1070 | mutex_lock(&container->lock); |
1071 | ret = tce_iommu_enable(container); | |
1072 | mutex_unlock(&container->lock); | |
1073 | return ret; | |
1074 | ||
1075 | ||
1076 | case VFIO_IOMMU_DISABLE: | |
2157e7b8 AK |
1077 | if (container->v2) |
1078 | break; | |
1079 | ||
5ffd229c AK |
1080 | mutex_lock(&container->lock); |
1081 | tce_iommu_disable(container); | |
1082 | mutex_unlock(&container->lock); | |
1083 | return 0; | |
1b69be5e | 1084 | |
2157e7b8 AK |
1085 | case VFIO_EEH_PE_OP: { |
1086 | struct tce_iommu_group *tcegrp; | |
1087 | ||
1088 | ret = 0; | |
1089 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
1090 | ret = vfio_spapr_iommu_eeh_ioctl(tcegrp->grp, | |
1091 | cmd, arg); | |
1092 | if (ret) | |
1093 | return ret; | |
1094 | } | |
1095 | return ret; | |
1096 | } | |
1097 | ||
e633bc86 AK |
1098 | case VFIO_IOMMU_SPAPR_TCE_CREATE: { |
1099 | struct vfio_iommu_spapr_tce_create create; | |
1100 | ||
1101 | if (!container->v2) | |
1102 | break; | |
1103 | ||
bc82d122 AK |
1104 | ret = tce_iommu_mm_set(container); |
1105 | if (ret) | |
1106 | return ret; | |
1107 | ||
e633bc86 AK |
1108 | if (!tce_groups_attached(container)) |
1109 | return -ENXIO; | |
1110 | ||
1111 | minsz = offsetofend(struct vfio_iommu_spapr_tce_create, | |
1112 | start_addr); | |
1113 | ||
1114 | if (copy_from_user(&create, (void __user *)arg, minsz)) | |
1115 | return -EFAULT; | |
1116 | ||
1117 | if (create.argsz < minsz) | |
1118 | return -EINVAL; | |
1119 | ||
1120 | if (create.flags) | |
1121 | return -EINVAL; | |
1122 | ||
1123 | mutex_lock(&container->lock); | |
1124 | ||
d9c72894 | 1125 | ret = tce_iommu_create_default_window(container); |
2da64d20 AK |
1126 | if (!ret) |
1127 | ret = tce_iommu_create_window(container, | |
1128 | create.page_shift, | |
1129 | create.window_size, create.levels, | |
1130 | &create.start_addr); | |
e633bc86 AK |
1131 | |
1132 | mutex_unlock(&container->lock); | |
1133 | ||
1134 | if (!ret && copy_to_user((void __user *)arg, &create, minsz)) | |
1135 | ret = -EFAULT; | |
1136 | ||
1137 | return ret; | |
1138 | } | |
1139 | case VFIO_IOMMU_SPAPR_TCE_REMOVE: { | |
1140 | struct vfio_iommu_spapr_tce_remove remove; | |
1141 | ||
1142 | if (!container->v2) | |
1143 | break; | |
1144 | ||
bc82d122 AK |
1145 | ret = tce_iommu_mm_set(container); |
1146 | if (ret) | |
1147 | return ret; | |
1148 | ||
e633bc86 AK |
1149 | if (!tce_groups_attached(container)) |
1150 | return -ENXIO; | |
1151 | ||
1152 | minsz = offsetofend(struct vfio_iommu_spapr_tce_remove, | |
1153 | start_addr); | |
1154 | ||
1155 | if (copy_from_user(&remove, (void __user *)arg, minsz)) | |
1156 | return -EFAULT; | |
1157 | ||
1158 | if (remove.argsz < minsz) | |
1159 | return -EINVAL; | |
1160 | ||
1161 | if (remove.flags) | |
1162 | return -EINVAL; | |
1163 | ||
d9c72894 AK |
1164 | if (container->def_window_pending && !remove.start_addr) { |
1165 | container->def_window_pending = false; | |
1166 | return 0; | |
1167 | } | |
1168 | ||
e633bc86 AK |
1169 | mutex_lock(&container->lock); |
1170 | ||
1171 | ret = tce_iommu_remove_window(container, remove.start_addr); | |
1172 | ||
1173 | mutex_unlock(&container->lock); | |
1174 | ||
1175 | return ret; | |
1176 | } | |
5ffd229c AK |
1177 | } |
1178 | ||
1179 | return -ENOTTY; | |
1180 | } | |
1181 | ||
f87a8864 AK |
1182 | static void tce_iommu_release_ownership(struct tce_container *container, |
1183 | struct iommu_table_group *table_group) | |
1184 | { | |
1185 | int i; | |
1186 | ||
1187 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
2157e7b8 | 1188 | struct iommu_table *tbl = container->tables[i]; |
f87a8864 AK |
1189 | |
1190 | if (!tbl) | |
1191 | continue; | |
1192 | ||
1193 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
bc82d122 | 1194 | tce_iommu_userspace_view_free(tbl, container->mm); |
f87a8864 AK |
1195 | if (tbl->it_map) |
1196 | iommu_release_ownership(tbl); | |
2157e7b8 AK |
1197 | |
1198 | container->tables[i] = NULL; | |
f87a8864 AK |
1199 | } |
1200 | } | |
1201 | ||
1202 | static int tce_iommu_take_ownership(struct tce_container *container, | |
1203 | struct iommu_table_group *table_group) | |
1204 | { | |
1205 | int i, j, rc = 0; | |
1206 | ||
1207 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
1208 | struct iommu_table *tbl = table_group->tables[i]; | |
1209 | ||
1210 | if (!tbl || !tbl->it_map) | |
1211 | continue; | |
1212 | ||
39701e56 | 1213 | rc = iommu_take_ownership(tbl); |
f87a8864 AK |
1214 | if (rc) { |
1215 | for (j = 0; j < i; ++j) | |
1216 | iommu_release_ownership( | |
1217 | table_group->tables[j]); | |
1218 | ||
1219 | return rc; | |
1220 | } | |
1221 | } | |
1222 | ||
2157e7b8 AK |
1223 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) |
1224 | container->tables[i] = table_group->tables[i]; | |
1225 | ||
f87a8864 AK |
1226 | return 0; |
1227 | } | |
1228 | ||
1229 | static void tce_iommu_release_ownership_ddw(struct tce_container *container, | |
1230 | struct iommu_table_group *table_group) | |
1231 | { | |
46d3e1e1 AK |
1232 | long i; |
1233 | ||
1234 | if (!table_group->ops->unset_window) { | |
1235 | WARN_ON_ONCE(1); | |
1236 | return; | |
1237 | } | |
1238 | ||
2157e7b8 | 1239 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) |
46d3e1e1 | 1240 | table_group->ops->unset_window(table_group, i); |
46d3e1e1 | 1241 | |
f87a8864 AK |
1242 | table_group->ops->release_ownership(table_group); |
1243 | } | |
1244 | ||
1245 | static long tce_iommu_take_ownership_ddw(struct tce_container *container, | |
1246 | struct iommu_table_group *table_group) | |
1247 | { | |
930a42de AK |
1248 | long i, ret = 0; |
1249 | ||
46d3e1e1 AK |
1250 | if (!table_group->ops->create_table || !table_group->ops->set_window || |
1251 | !table_group->ops->release_ownership) { | |
1252 | WARN_ON_ONCE(1); | |
1253 | return -EFAULT; | |
1254 | } | |
1255 | ||
f87a8864 AK |
1256 | table_group->ops->take_ownership(table_group); |
1257 | ||
930a42de AK |
1258 | /* Set all windows to the new group */ |
1259 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
1260 | struct iommu_table *tbl = container->tables[i]; | |
1261 | ||
1262 | if (!tbl) | |
1263 | continue; | |
1264 | ||
1265 | ret = table_group->ops->set_window(table_group, i, tbl); | |
1266 | if (ret) | |
1267 | goto release_exit; | |
1268 | } | |
1269 | ||
2157e7b8 | 1270 | return 0; |
930a42de AK |
1271 | |
1272 | release_exit: | |
1273 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) | |
1274 | table_group->ops->unset_window(table_group, i); | |
1275 | ||
1276 | table_group->ops->release_ownership(table_group); | |
1277 | ||
1278 | return ret; | |
f87a8864 AK |
1279 | } |
1280 | ||
5ffd229c AK |
1281 | static int tce_iommu_attach_group(void *iommu_data, |
1282 | struct iommu_group *iommu_group) | |
1283 | { | |
1284 | int ret; | |
1285 | struct tce_container *container = iommu_data; | |
0eaf4def | 1286 | struct iommu_table_group *table_group; |
2157e7b8 | 1287 | struct tce_iommu_group *tcegrp = NULL; |
5ffd229c | 1288 | |
5ffd229c AK |
1289 | mutex_lock(&container->lock); |
1290 | ||
1291 | /* pr_debug("tce_vfio: Attaching group #%u to iommu %p\n", | |
1292 | iommu_group_id(iommu_group), iommu_group); */ | |
2157e7b8 | 1293 | table_group = iommu_group_get_iommudata(iommu_group); |
bd00fdf1 GK |
1294 | if (!table_group) { |
1295 | ret = -ENODEV; | |
1296 | goto unlock_exit; | |
1297 | } | |
2157e7b8 AK |
1298 | |
1299 | if (tce_groups_attached(container) && (!table_group->ops || | |
1300 | !table_group->ops->take_ownership || | |
1301 | !table_group->ops->release_ownership)) { | |
5ffd229c | 1302 | ret = -EBUSY; |
22af4859 AK |
1303 | goto unlock_exit; |
1304 | } | |
1305 | ||
2157e7b8 AK |
1306 | /* Check if new group has the same iommu_ops (i.e. compatible) */ |
1307 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
1308 | struct iommu_table_group *table_group_tmp; | |
1309 | ||
1310 | if (tcegrp->grp == iommu_group) { | |
1311 | pr_warn("tce_vfio: Group %d is already attached\n", | |
1312 | iommu_group_id(iommu_group)); | |
1313 | ret = -EBUSY; | |
1314 | goto unlock_exit; | |
1315 | } | |
1316 | table_group_tmp = iommu_group_get_iommudata(tcegrp->grp); | |
54de285b AK |
1317 | if (table_group_tmp->ops->create_table != |
1318 | table_group->ops->create_table) { | |
2157e7b8 AK |
1319 | pr_warn("tce_vfio: Group %d is incompatible with group %d\n", |
1320 | iommu_group_id(iommu_group), | |
1321 | iommu_group_id(tcegrp->grp)); | |
1322 | ret = -EPERM; | |
1323 | goto unlock_exit; | |
1324 | } | |
5ffd229c AK |
1325 | } |
1326 | ||
2157e7b8 AK |
1327 | tcegrp = kzalloc(sizeof(*tcegrp), GFP_KERNEL); |
1328 | if (!tcegrp) { | |
1329 | ret = -ENOMEM; | |
0eaf4def AK |
1330 | goto unlock_exit; |
1331 | } | |
1332 | ||
f87a8864 | 1333 | if (!table_group->ops || !table_group->ops->take_ownership || |
6f01cc69 | 1334 | !table_group->ops->release_ownership) { |
f87a8864 | 1335 | ret = tce_iommu_take_ownership(container, table_group); |
6f01cc69 | 1336 | } else { |
f87a8864 | 1337 | ret = tce_iommu_take_ownership_ddw(container, table_group); |
6f01cc69 | 1338 | if (!tce_groups_attached(container) && !container->tables[0]) |
d9c72894 | 1339 | container->def_window_pending = true; |
6f01cc69 | 1340 | } |
f87a8864 | 1341 | |
2157e7b8 AK |
1342 | if (!ret) { |
1343 | tcegrp->grp = iommu_group; | |
1344 | list_add(&tcegrp->next, &container->group_list); | |
1345 | } | |
22af4859 AK |
1346 | |
1347 | unlock_exit: | |
2157e7b8 AK |
1348 | if (ret && tcegrp) |
1349 | kfree(tcegrp); | |
1350 | ||
5ffd229c AK |
1351 | mutex_unlock(&container->lock); |
1352 | ||
1353 | return ret; | |
1354 | } | |
1355 | ||
1356 | static void tce_iommu_detach_group(void *iommu_data, | |
1357 | struct iommu_group *iommu_group) | |
1358 | { | |
1359 | struct tce_container *container = iommu_data; | |
0eaf4def | 1360 | struct iommu_table_group *table_group; |
2157e7b8 AK |
1361 | bool found = false; |
1362 | struct tce_iommu_group *tcegrp; | |
5ffd229c | 1363 | |
5ffd229c | 1364 | mutex_lock(&container->lock); |
2157e7b8 AK |
1365 | |
1366 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
1367 | if (tcegrp->grp == iommu_group) { | |
1368 | found = true; | |
1369 | break; | |
1370 | } | |
22af4859 | 1371 | } |
5ffd229c | 1372 | |
2157e7b8 AK |
1373 | if (!found) { |
1374 | pr_warn("tce_vfio: detaching unattached group #%u\n", | |
1375 | iommu_group_id(iommu_group)); | |
1376 | goto unlock_exit; | |
5ffd229c | 1377 | } |
22af4859 | 1378 | |
2157e7b8 AK |
1379 | list_del(&tcegrp->next); |
1380 | kfree(tcegrp); | |
0eaf4def AK |
1381 | |
1382 | table_group = iommu_group_get_iommudata(iommu_group); | |
1383 | BUG_ON(!table_group); | |
1384 | ||
f87a8864 AK |
1385 | if (!table_group->ops || !table_group->ops->release_ownership) |
1386 | tce_iommu_release_ownership(container, table_group); | |
1387 | else | |
1388 | tce_iommu_release_ownership_ddw(container, table_group); | |
22af4859 AK |
1389 | |
1390 | unlock_exit: | |
5ffd229c AK |
1391 | mutex_unlock(&container->lock); |
1392 | } | |
1393 | ||
1394 | const struct vfio_iommu_driver_ops tce_iommu_driver_ops = { | |
1395 | .name = "iommu-vfio-powerpc", | |
1396 | .owner = THIS_MODULE, | |
1397 | .open = tce_iommu_open, | |
1398 | .release = tce_iommu_release, | |
1399 | .ioctl = tce_iommu_ioctl, | |
1400 | .attach_group = tce_iommu_attach_group, | |
1401 | .detach_group = tce_iommu_detach_group, | |
1402 | }; | |
1403 | ||
1404 | static int __init tce_iommu_init(void) | |
1405 | { | |
1406 | return vfio_register_iommu_driver(&tce_iommu_driver_ops); | |
1407 | } | |
1408 | ||
1409 | static void __exit tce_iommu_cleanup(void) | |
1410 | { | |
1411 | vfio_unregister_iommu_driver(&tce_iommu_driver_ops); | |
1412 | } | |
1413 | ||
1414 | module_init(tce_iommu_init); | |
1415 | module_exit(tce_iommu_cleanup); | |
1416 | ||
1417 | MODULE_VERSION(DRIVER_VERSION); | |
1418 | MODULE_LICENSE("GPL v2"); | |
1419 | MODULE_AUTHOR(DRIVER_AUTHOR); | |
1420 | MODULE_DESCRIPTION(DRIVER_DESC); | |
1421 |