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> |
6e84f315 | 23 | #include <linux/sched/mm.h> |
3f07c014 | 24 | #include <linux/sched/signal.h> |
6e84f315 | 25 | |
5ffd229c AK |
26 | #include <asm/iommu.h> |
27 | #include <asm/tce.h> | |
2157e7b8 | 28 | #include <asm/mmu_context.h> |
5ffd229c AK |
29 | |
30 | #define DRIVER_VERSION "0.1" | |
31 | #define DRIVER_AUTHOR "aik@ozlabs.ru" | |
32 | #define DRIVER_DESC "VFIO IOMMU SPAPR TCE" | |
33 | ||
34 | static void tce_iommu_detach_group(void *iommu_data, | |
35 | struct iommu_group *iommu_group); | |
36 | ||
bc82d122 | 37 | static long try_increment_locked_vm(struct mm_struct *mm, long npages) |
2d270df8 AK |
38 | { |
39 | long ret = 0, locked, lock_limit; | |
40 | ||
bc82d122 AK |
41 | if (WARN_ON_ONCE(!mm)) |
42 | return -EPERM; | |
2d270df8 AK |
43 | |
44 | if (!npages) | |
45 | return 0; | |
46 | ||
bc82d122 AK |
47 | down_write(&mm->mmap_sem); |
48 | locked = mm->locked_vm + npages; | |
2d270df8 AK |
49 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; |
50 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
51 | ret = -ENOMEM; | |
52 | else | |
bc82d122 | 53 | mm->locked_vm += npages; |
2d270df8 AK |
54 | |
55 | pr_debug("[%d] RLIMIT_MEMLOCK +%ld %ld/%ld%s\n", current->pid, | |
56 | npages << PAGE_SHIFT, | |
bc82d122 | 57 | mm->locked_vm << PAGE_SHIFT, |
2d270df8 AK |
58 | rlimit(RLIMIT_MEMLOCK), |
59 | ret ? " - exceeded" : ""); | |
60 | ||
bc82d122 | 61 | up_write(&mm->mmap_sem); |
2d270df8 AK |
62 | |
63 | return ret; | |
64 | } | |
65 | ||
bc82d122 | 66 | static void decrement_locked_vm(struct mm_struct *mm, long npages) |
2d270df8 | 67 | { |
bc82d122 AK |
68 | if (!mm || !npages) |
69 | return; | |
2d270df8 | 70 | |
bc82d122 AK |
71 | down_write(&mm->mmap_sem); |
72 | if (WARN_ON_ONCE(npages > mm->locked_vm)) | |
73 | npages = mm->locked_vm; | |
74 | mm->locked_vm -= npages; | |
2d270df8 AK |
75 | pr_debug("[%d] RLIMIT_MEMLOCK -%ld %ld/%ld\n", current->pid, |
76 | npages << PAGE_SHIFT, | |
bc82d122 | 77 | mm->locked_vm << PAGE_SHIFT, |
2d270df8 | 78 | rlimit(RLIMIT_MEMLOCK)); |
bc82d122 | 79 | up_write(&mm->mmap_sem); |
2d270df8 AK |
80 | } |
81 | ||
5ffd229c AK |
82 | /* |
83 | * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation | |
84 | * | |
85 | * This code handles mapping and unmapping of user data buffers | |
86 | * into DMA'ble space using the IOMMU | |
87 | */ | |
88 | ||
2157e7b8 AK |
89 | struct tce_iommu_group { |
90 | struct list_head next; | |
91 | struct iommu_group *grp; | |
92 | }; | |
93 | ||
4b6fad70 AK |
94 | /* |
95 | * A container needs to remember which preregistered region it has | |
96 | * referenced to do proper cleanup at the userspace process exit. | |
97 | */ | |
98 | struct tce_iommu_prereg { | |
99 | struct list_head next; | |
100 | struct mm_iommu_table_group_mem_t *mem; | |
101 | }; | |
102 | ||
5ffd229c AK |
103 | /* |
104 | * The container descriptor supports only a single group per container. | |
105 | * Required by the API as the container is not supplied with the IOMMU group | |
106 | * at the moment of initialization. | |
107 | */ | |
108 | struct tce_container { | |
109 | struct mutex lock; | |
5ffd229c | 110 | bool enabled; |
2157e7b8 | 111 | bool v2; |
d9c72894 | 112 | bool def_window_pending; |
2d270df8 | 113 | unsigned long locked_pages; |
bc82d122 | 114 | struct mm_struct *mm; |
2157e7b8 AK |
115 | struct iommu_table *tables[IOMMU_TABLE_GROUP_MAX_TABLES]; |
116 | struct list_head group_list; | |
4b6fad70 | 117 | struct list_head prereg_list; |
5ffd229c AK |
118 | }; |
119 | ||
bc82d122 AK |
120 | static long tce_iommu_mm_set(struct tce_container *container) |
121 | { | |
122 | if (container->mm) { | |
123 | if (container->mm == current->mm) | |
124 | return 0; | |
125 | return -EPERM; | |
126 | } | |
127 | BUG_ON(!current->mm); | |
128 | container->mm = current->mm; | |
129 | atomic_inc(&container->mm->mm_count); | |
130 | ||
131 | return 0; | |
132 | } | |
133 | ||
4b6fad70 AK |
134 | static long tce_iommu_prereg_free(struct tce_container *container, |
135 | struct tce_iommu_prereg *tcemem) | |
136 | { | |
137 | long ret; | |
138 | ||
139 | ret = mm_iommu_put(container->mm, tcemem->mem); | |
140 | if (ret) | |
141 | return ret; | |
142 | ||
143 | list_del(&tcemem->next); | |
144 | kfree(tcemem); | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
2157e7b8 AK |
149 | static long tce_iommu_unregister_pages(struct tce_container *container, |
150 | __u64 vaddr, __u64 size) | |
151 | { | |
152 | struct mm_iommu_table_group_mem_t *mem; | |
4b6fad70 AK |
153 | struct tce_iommu_prereg *tcemem; |
154 | bool found = false; | |
2157e7b8 AK |
155 | |
156 | if ((vaddr & ~PAGE_MASK) || (size & ~PAGE_MASK)) | |
157 | return -EINVAL; | |
158 | ||
bc82d122 | 159 | mem = mm_iommu_find(container->mm, vaddr, size >> PAGE_SHIFT); |
2157e7b8 AK |
160 | if (!mem) |
161 | return -ENOENT; | |
162 | ||
4b6fad70 AK |
163 | list_for_each_entry(tcemem, &container->prereg_list, next) { |
164 | if (tcemem->mem == mem) { | |
165 | found = true; | |
166 | break; | |
167 | } | |
168 | } | |
169 | ||
170 | if (!found) | |
171 | return -ENOENT; | |
172 | ||
173 | return tce_iommu_prereg_free(container, tcemem); | |
2157e7b8 AK |
174 | } |
175 | ||
176 | static long tce_iommu_register_pages(struct tce_container *container, | |
177 | __u64 vaddr, __u64 size) | |
178 | { | |
179 | long ret = 0; | |
180 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
4b6fad70 | 181 | struct tce_iommu_prereg *tcemem; |
2157e7b8 AK |
182 | unsigned long entries = size >> PAGE_SHIFT; |
183 | ||
184 | if ((vaddr & ~PAGE_MASK) || (size & ~PAGE_MASK) || | |
185 | ((vaddr + size) < vaddr)) | |
186 | return -EINVAL; | |
187 | ||
4b6fad70 AK |
188 | mem = mm_iommu_find(container->mm, vaddr, entries); |
189 | if (mem) { | |
190 | list_for_each_entry(tcemem, &container->prereg_list, next) { | |
191 | if (tcemem->mem == mem) | |
192 | return -EBUSY; | |
193 | } | |
194 | } | |
195 | ||
bc82d122 | 196 | ret = mm_iommu_get(container->mm, vaddr, entries, &mem); |
2157e7b8 AK |
197 | if (ret) |
198 | return ret; | |
199 | ||
4b6fad70 | 200 | tcemem = kzalloc(sizeof(*tcemem), GFP_KERNEL); |
3393af24 AK |
201 | if (!tcemem) { |
202 | mm_iommu_put(container->mm, mem); | |
203 | return -ENOMEM; | |
204 | } | |
205 | ||
4b6fad70 AK |
206 | tcemem->mem = mem; |
207 | list_add(&tcemem->next, &container->prereg_list); | |
208 | ||
2157e7b8 AK |
209 | container->enabled = true; |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
bc82d122 AK |
214 | static long tce_iommu_userspace_view_alloc(struct iommu_table *tbl, |
215 | struct mm_struct *mm) | |
2157e7b8 AK |
216 | { |
217 | unsigned long cb = _ALIGN_UP(sizeof(tbl->it_userspace[0]) * | |
218 | tbl->it_size, PAGE_SIZE); | |
219 | unsigned long *uas; | |
220 | long ret; | |
221 | ||
222 | BUG_ON(tbl->it_userspace); | |
223 | ||
bc82d122 | 224 | ret = try_increment_locked_vm(mm, cb >> PAGE_SHIFT); |
2157e7b8 AK |
225 | if (ret) |
226 | return ret; | |
227 | ||
228 | uas = vzalloc(cb); | |
229 | if (!uas) { | |
bc82d122 | 230 | decrement_locked_vm(mm, cb >> PAGE_SHIFT); |
2157e7b8 AK |
231 | return -ENOMEM; |
232 | } | |
233 | tbl->it_userspace = uas; | |
234 | ||
235 | return 0; | |
236 | } | |
237 | ||
bc82d122 AK |
238 | static void tce_iommu_userspace_view_free(struct iommu_table *tbl, |
239 | struct mm_struct *mm) | |
2157e7b8 AK |
240 | { |
241 | unsigned long cb = _ALIGN_UP(sizeof(tbl->it_userspace[0]) * | |
242 | tbl->it_size, PAGE_SIZE); | |
243 | ||
244 | if (!tbl->it_userspace) | |
245 | return; | |
246 | ||
247 | vfree(tbl->it_userspace); | |
248 | tbl->it_userspace = NULL; | |
bc82d122 | 249 | decrement_locked_vm(mm, cb >> PAGE_SHIFT); |
2157e7b8 AK |
250 | } |
251 | ||
e432bc7e AK |
252 | static bool tce_page_is_contained(struct page *page, unsigned page_shift) |
253 | { | |
254 | /* | |
255 | * Check that the TCE table granularity is not bigger than the size of | |
256 | * a page we just found. Otherwise the hardware can get access to | |
257 | * a bigger memory chunk that it should. | |
258 | */ | |
259 | return (PAGE_SHIFT + compound_order(compound_head(page))) >= page_shift; | |
260 | } | |
261 | ||
2157e7b8 AK |
262 | static inline bool tce_groups_attached(struct tce_container *container) |
263 | { | |
264 | return !list_empty(&container->group_list); | |
265 | } | |
266 | ||
0eaf4def AK |
267 | static long tce_iommu_find_table(struct tce_container *container, |
268 | phys_addr_t ioba, struct iommu_table **ptbl) | |
269 | { | |
270 | long i; | |
0eaf4def AK |
271 | |
272 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
2157e7b8 | 273 | struct iommu_table *tbl = container->tables[i]; |
0eaf4def AK |
274 | |
275 | if (tbl) { | |
276 | unsigned long entry = ioba >> tbl->it_page_shift; | |
277 | unsigned long start = tbl->it_offset; | |
278 | unsigned long end = start + tbl->it_size; | |
279 | ||
280 | if ((start <= entry) && (entry < end)) { | |
281 | *ptbl = tbl; | |
282 | return i; | |
283 | } | |
284 | } | |
285 | } | |
286 | ||
287 | return -1; | |
288 | } | |
289 | ||
e633bc86 AK |
290 | static int tce_iommu_find_free_table(struct tce_container *container) |
291 | { | |
292 | int i; | |
293 | ||
294 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
295 | if (!container->tables[i]) | |
296 | return i; | |
297 | } | |
298 | ||
299 | return -ENOSPC; | |
300 | } | |
301 | ||
5ffd229c AK |
302 | static int tce_iommu_enable(struct tce_container *container) |
303 | { | |
304 | int ret = 0; | |
2d270df8 | 305 | unsigned long locked; |
0eaf4def | 306 | struct iommu_table_group *table_group; |
2157e7b8 | 307 | struct tce_iommu_group *tcegrp; |
5ffd229c | 308 | |
5ffd229c AK |
309 | if (container->enabled) |
310 | return -EBUSY; | |
311 | ||
312 | /* | |
313 | * When userspace pages are mapped into the IOMMU, they are effectively | |
314 | * locked memory, so, theoretically, we need to update the accounting | |
315 | * of locked pages on each map and unmap. For powerpc, the map unmap | |
316 | * paths can be very hot, though, and the accounting would kill | |
317 | * performance, especially since it would be difficult to impossible | |
318 | * to handle the accounting in real mode only. | |
319 | * | |
320 | * To address that, rather than precisely accounting every page, we | |
321 | * instead account for a worst case on locked memory when the iommu is | |
322 | * enabled and disabled. The worst case upper bound on locked memory | |
323 | * is the size of the whole iommu window, which is usually relatively | |
324 | * small (compared to total memory sizes) on POWER hardware. | |
325 | * | |
326 | * Also we don't have a nice way to fail on H_PUT_TCE due to ulimits, | |
327 | * that would effectively kill the guest at random points, much better | |
328 | * enforcing the limit based on the max that the guest can map. | |
2d270df8 AK |
329 | * |
330 | * Unfortunately at the moment it counts whole tables, no matter how | |
331 | * much memory the guest has. I.e. for 4GB guest and 4 IOMMU groups | |
332 | * each with 2GB DMA window, 8GB will be counted here. The reason for | |
333 | * this is that we cannot tell here the amount of RAM used by the guest | |
334 | * as this information is only available from KVM and VFIO is | |
335 | * KVM agnostic. | |
4793d65d AK |
336 | * |
337 | * So we do not allow enabling a container without a group attached | |
338 | * as there is no way to know how much we should increment | |
339 | * the locked_vm counter. | |
5ffd229c | 340 | */ |
2157e7b8 AK |
341 | if (!tce_groups_attached(container)) |
342 | return -ENODEV; | |
343 | ||
344 | tcegrp = list_first_entry(&container->group_list, | |
345 | struct tce_iommu_group, next); | |
346 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
0eaf4def AK |
347 | if (!table_group) |
348 | return -ENODEV; | |
349 | ||
4793d65d AK |
350 | if (!table_group->tce32_size) |
351 | return -EPERM; | |
352 | ||
bc82d122 AK |
353 | ret = tce_iommu_mm_set(container); |
354 | if (ret) | |
355 | return ret; | |
356 | ||
4793d65d | 357 | locked = table_group->tce32_size >> PAGE_SHIFT; |
bc82d122 | 358 | ret = try_increment_locked_vm(container->mm, locked); |
2d270df8 AK |
359 | if (ret) |
360 | return ret; | |
5ffd229c | 361 | |
2d270df8 AK |
362 | container->locked_pages = locked; |
363 | ||
364 | container->enabled = true; | |
5ffd229c AK |
365 | |
366 | return ret; | |
367 | } | |
368 | ||
369 | static void tce_iommu_disable(struct tce_container *container) | |
370 | { | |
371 | if (!container->enabled) | |
372 | return; | |
373 | ||
374 | container->enabled = false; | |
375 | ||
bc82d122 AK |
376 | BUG_ON(!container->mm); |
377 | decrement_locked_vm(container->mm, container->locked_pages); | |
5ffd229c AK |
378 | } |
379 | ||
380 | static void *tce_iommu_open(unsigned long arg) | |
381 | { | |
382 | struct tce_container *container; | |
383 | ||
2157e7b8 | 384 | if ((arg != VFIO_SPAPR_TCE_IOMMU) && (arg != VFIO_SPAPR_TCE_v2_IOMMU)) { |
5ffd229c AK |
385 | pr_err("tce_vfio: Wrong IOMMU type\n"); |
386 | return ERR_PTR(-EINVAL); | |
387 | } | |
388 | ||
389 | container = kzalloc(sizeof(*container), GFP_KERNEL); | |
390 | if (!container) | |
391 | return ERR_PTR(-ENOMEM); | |
392 | ||
393 | mutex_init(&container->lock); | |
2157e7b8 | 394 | INIT_LIST_HEAD_RCU(&container->group_list); |
4b6fad70 | 395 | INIT_LIST_HEAD_RCU(&container->prereg_list); |
2157e7b8 AK |
396 | |
397 | container->v2 = arg == VFIO_SPAPR_TCE_v2_IOMMU; | |
5ffd229c AK |
398 | |
399 | return container; | |
400 | } | |
401 | ||
2157e7b8 AK |
402 | static int tce_iommu_clear(struct tce_container *container, |
403 | struct iommu_table *tbl, | |
404 | unsigned long entry, unsigned long pages); | |
bc82d122 AK |
405 | static void tce_iommu_free_table(struct tce_container *container, |
406 | struct iommu_table *tbl); | |
2157e7b8 | 407 | |
5ffd229c AK |
408 | static void tce_iommu_release(void *iommu_data) |
409 | { | |
410 | struct tce_container *container = iommu_data; | |
2157e7b8 AK |
411 | struct tce_iommu_group *tcegrp; |
412 | long i; | |
5ffd229c | 413 | |
2157e7b8 AK |
414 | while (tce_groups_attached(container)) { |
415 | tcegrp = list_first_entry(&container->group_list, | |
416 | struct tce_iommu_group, next); | |
2157e7b8 AK |
417 | tce_iommu_detach_group(iommu_data, tcegrp->grp); |
418 | } | |
5ffd229c | 419 | |
2157e7b8 AK |
420 | /* |
421 | * If VFIO created a table, it was not disposed | |
422 | * by tce_iommu_detach_group() so do it now. | |
423 | */ | |
424 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
425 | struct iommu_table *tbl = container->tables[i]; | |
426 | ||
427 | if (!tbl) | |
428 | continue; | |
429 | ||
430 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
bc82d122 | 431 | tce_iommu_free_table(container, tbl); |
2157e7b8 | 432 | } |
5ffd229c | 433 | |
4b6fad70 AK |
434 | while (!list_empty(&container->prereg_list)) { |
435 | struct tce_iommu_prereg *tcemem; | |
436 | ||
437 | tcemem = list_first_entry(&container->prereg_list, | |
438 | struct tce_iommu_prereg, next); | |
439 | WARN_ON_ONCE(tce_iommu_prereg_free(container, tcemem)); | |
440 | } | |
441 | ||
649354b7 | 442 | tce_iommu_disable(container); |
bc82d122 AK |
443 | if (container->mm) |
444 | mmdrop(container->mm); | |
5ffd229c AK |
445 | mutex_destroy(&container->lock); |
446 | ||
447 | kfree(container); | |
448 | } | |
449 | ||
649354b7 | 450 | static void tce_iommu_unuse_page(struct tce_container *container, |
05c6cfb9 | 451 | unsigned long hpa) |
649354b7 AK |
452 | { |
453 | struct page *page; | |
454 | ||
05c6cfb9 | 455 | page = pfn_to_page(hpa >> PAGE_SHIFT); |
649354b7 AK |
456 | put_page(page); |
457 | } | |
458 | ||
bc82d122 AK |
459 | static int tce_iommu_prereg_ua_to_hpa(struct tce_container *container, |
460 | unsigned long tce, unsigned long size, | |
2157e7b8 AK |
461 | unsigned long *phpa, struct mm_iommu_table_group_mem_t **pmem) |
462 | { | |
463 | long ret = 0; | |
464 | struct mm_iommu_table_group_mem_t *mem; | |
465 | ||
bc82d122 | 466 | mem = mm_iommu_lookup(container->mm, tce, size); |
2157e7b8 AK |
467 | if (!mem) |
468 | return -EINVAL; | |
469 | ||
470 | ret = mm_iommu_ua_to_hpa(mem, tce, phpa); | |
471 | if (ret) | |
472 | return -EINVAL; | |
473 | ||
474 | *pmem = mem; | |
475 | ||
476 | return 0; | |
477 | } | |
478 | ||
bc82d122 AK |
479 | static void tce_iommu_unuse_page_v2(struct tce_container *container, |
480 | struct iommu_table *tbl, unsigned long entry) | |
2157e7b8 AK |
481 | { |
482 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
483 | int ret; | |
484 | unsigned long hpa = 0; | |
485 | unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry); | |
486 | ||
bc82d122 | 487 | if (!pua) |
2157e7b8 AK |
488 | return; |
489 | ||
bc82d122 | 490 | ret = tce_iommu_prereg_ua_to_hpa(container, *pua, IOMMU_PAGE_SIZE(tbl), |
2157e7b8 AK |
491 | &hpa, &mem); |
492 | if (ret) | |
493 | pr_debug("%s: tce %lx at #%lx was not cached, ret=%d\n", | |
494 | __func__, *pua, entry, ret); | |
495 | if (mem) | |
496 | mm_iommu_mapped_dec(mem); | |
497 | ||
498 | *pua = 0; | |
499 | } | |
500 | ||
9b14a1ff AK |
501 | static int tce_iommu_clear(struct tce_container *container, |
502 | struct iommu_table *tbl, | |
503 | unsigned long entry, unsigned long pages) | |
504 | { | |
05c6cfb9 AK |
505 | unsigned long oldhpa; |
506 | long ret; | |
507 | enum dma_data_direction direction; | |
9b14a1ff AK |
508 | |
509 | for ( ; pages; --pages, ++entry) { | |
5c2fefd8 AK |
510 | cond_resched(); |
511 | ||
05c6cfb9 AK |
512 | direction = DMA_NONE; |
513 | oldhpa = 0; | |
514 | ret = iommu_tce_xchg(tbl, entry, &oldhpa, &direction); | |
515 | if (ret) | |
516 | continue; | |
517 | ||
518 | if (direction == DMA_NONE) | |
9b14a1ff AK |
519 | continue; |
520 | ||
2157e7b8 | 521 | if (container->v2) { |
bc82d122 | 522 | tce_iommu_unuse_page_v2(container, tbl, entry); |
2157e7b8 AK |
523 | continue; |
524 | } | |
525 | ||
05c6cfb9 | 526 | tce_iommu_unuse_page(container, oldhpa); |
9b14a1ff AK |
527 | } |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
649354b7 AK |
532 | static int tce_iommu_use_page(unsigned long tce, unsigned long *hpa) |
533 | { | |
534 | struct page *page = NULL; | |
535 | enum dma_data_direction direction = iommu_tce_direction(tce); | |
536 | ||
537 | if (get_user_pages_fast(tce & PAGE_MASK, 1, | |
538 | direction != DMA_TO_DEVICE, &page) != 1) | |
539 | return -EFAULT; | |
540 | ||
541 | *hpa = __pa((unsigned long) page_address(page)); | |
542 | ||
543 | return 0; | |
544 | } | |
545 | ||
9b14a1ff AK |
546 | static long tce_iommu_build(struct tce_container *container, |
547 | struct iommu_table *tbl, | |
05c6cfb9 AK |
548 | unsigned long entry, unsigned long tce, unsigned long pages, |
549 | enum dma_data_direction direction) | |
9b14a1ff AK |
550 | { |
551 | long i, ret = 0; | |
649354b7 AK |
552 | struct page *page; |
553 | unsigned long hpa; | |
05c6cfb9 | 554 | enum dma_data_direction dirtmp; |
9b14a1ff AK |
555 | |
556 | for (i = 0; i < pages; ++i) { | |
557 | unsigned long offset = tce & IOMMU_PAGE_MASK(tbl) & ~PAGE_MASK; | |
558 | ||
649354b7 AK |
559 | ret = tce_iommu_use_page(tce, &hpa); |
560 | if (ret) | |
9b14a1ff | 561 | break; |
e432bc7e | 562 | |
649354b7 | 563 | page = pfn_to_page(hpa >> PAGE_SHIFT); |
e432bc7e AK |
564 | if (!tce_page_is_contained(page, tbl->it_page_shift)) { |
565 | ret = -EPERM; | |
566 | break; | |
567 | } | |
568 | ||
649354b7 | 569 | hpa |= offset; |
05c6cfb9 AK |
570 | dirtmp = direction; |
571 | ret = iommu_tce_xchg(tbl, entry + i, &hpa, &dirtmp); | |
9b14a1ff | 572 | if (ret) { |
649354b7 | 573 | tce_iommu_unuse_page(container, hpa); |
9b14a1ff AK |
574 | pr_err("iommu_tce: %s failed ioba=%lx, tce=%lx, ret=%ld\n", |
575 | __func__, entry << tbl->it_page_shift, | |
576 | tce, ret); | |
577 | break; | |
578 | } | |
05c6cfb9 AK |
579 | |
580 | if (dirtmp != DMA_NONE) | |
581 | tce_iommu_unuse_page(container, hpa); | |
582 | ||
00663d4e | 583 | tce += IOMMU_PAGE_SIZE(tbl); |
9b14a1ff AK |
584 | } |
585 | ||
586 | if (ret) | |
587 | tce_iommu_clear(container, tbl, entry, i); | |
588 | ||
589 | return ret; | |
590 | } | |
591 | ||
2157e7b8 AK |
592 | static long tce_iommu_build_v2(struct tce_container *container, |
593 | struct iommu_table *tbl, | |
594 | unsigned long entry, unsigned long tce, unsigned long pages, | |
595 | enum dma_data_direction direction) | |
596 | { | |
597 | long i, ret = 0; | |
598 | struct page *page; | |
599 | unsigned long hpa; | |
600 | enum dma_data_direction dirtmp; | |
601 | ||
39701e56 | 602 | if (!tbl->it_userspace) { |
bc82d122 | 603 | ret = tce_iommu_userspace_view_alloc(tbl, container->mm); |
39701e56 AK |
604 | if (ret) |
605 | return ret; | |
606 | } | |
607 | ||
2157e7b8 AK |
608 | for (i = 0; i < pages; ++i) { |
609 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
610 | unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, | |
611 | entry + i); | |
612 | ||
bc82d122 AK |
613 | ret = tce_iommu_prereg_ua_to_hpa(container, |
614 | tce, IOMMU_PAGE_SIZE(tbl), &hpa, &mem); | |
2157e7b8 AK |
615 | if (ret) |
616 | break; | |
617 | ||
618 | page = pfn_to_page(hpa >> PAGE_SHIFT); | |
619 | if (!tce_page_is_contained(page, tbl->it_page_shift)) { | |
620 | ret = -EPERM; | |
621 | break; | |
622 | } | |
623 | ||
624 | /* Preserve offset within IOMMU page */ | |
625 | hpa |= tce & IOMMU_PAGE_MASK(tbl) & ~PAGE_MASK; | |
626 | dirtmp = direction; | |
627 | ||
628 | /* The registered region is being unregistered */ | |
629 | if (mm_iommu_mapped_inc(mem)) | |
630 | break; | |
631 | ||
632 | ret = iommu_tce_xchg(tbl, entry + i, &hpa, &dirtmp); | |
633 | if (ret) { | |
634 | /* dirtmp cannot be DMA_NONE here */ | |
bc82d122 | 635 | tce_iommu_unuse_page_v2(container, tbl, entry + i); |
2157e7b8 AK |
636 | pr_err("iommu_tce: %s failed ioba=%lx, tce=%lx, ret=%ld\n", |
637 | __func__, entry << tbl->it_page_shift, | |
638 | tce, ret); | |
639 | break; | |
640 | } | |
641 | ||
642 | if (dirtmp != DMA_NONE) | |
bc82d122 | 643 | tce_iommu_unuse_page_v2(container, tbl, entry + i); |
2157e7b8 AK |
644 | |
645 | *pua = tce; | |
646 | ||
647 | tce += IOMMU_PAGE_SIZE(tbl); | |
648 | } | |
649 | ||
650 | if (ret) | |
651 | tce_iommu_clear(container, tbl, entry, i); | |
652 | ||
653 | return ret; | |
654 | } | |
655 | ||
46d3e1e1 AK |
656 | static long tce_iommu_create_table(struct tce_container *container, |
657 | struct iommu_table_group *table_group, | |
658 | int num, | |
659 | __u32 page_shift, | |
660 | __u64 window_size, | |
661 | __u32 levels, | |
662 | struct iommu_table **ptbl) | |
663 | { | |
664 | long ret, table_size; | |
665 | ||
666 | table_size = table_group->ops->get_table_size(page_shift, window_size, | |
667 | levels); | |
668 | if (!table_size) | |
669 | return -EINVAL; | |
670 | ||
bc82d122 | 671 | ret = try_increment_locked_vm(container->mm, table_size >> PAGE_SHIFT); |
46d3e1e1 AK |
672 | if (ret) |
673 | return ret; | |
674 | ||
675 | ret = table_group->ops->create_table(table_group, num, | |
676 | page_shift, window_size, levels, ptbl); | |
677 | ||
678 | WARN_ON(!ret && !(*ptbl)->it_ops->free); | |
679 | WARN_ON(!ret && ((*ptbl)->it_allocated_size != table_size)); | |
680 | ||
46d3e1e1 AK |
681 | return ret; |
682 | } | |
683 | ||
bc82d122 AK |
684 | static void tce_iommu_free_table(struct tce_container *container, |
685 | struct iommu_table *tbl) | |
46d3e1e1 AK |
686 | { |
687 | unsigned long pages = tbl->it_allocated_size >> PAGE_SHIFT; | |
688 | ||
bc82d122 | 689 | tce_iommu_userspace_view_free(tbl, container->mm); |
e5afdf9d | 690 | iommu_tce_table_put(tbl); |
bc82d122 | 691 | decrement_locked_vm(container->mm, pages); |
46d3e1e1 AK |
692 | } |
693 | ||
e633bc86 AK |
694 | static long tce_iommu_create_window(struct tce_container *container, |
695 | __u32 page_shift, __u64 window_size, __u32 levels, | |
696 | __u64 *start_addr) | |
697 | { | |
698 | struct tce_iommu_group *tcegrp; | |
699 | struct iommu_table_group *table_group; | |
700 | struct iommu_table *tbl = NULL; | |
701 | long ret, num; | |
702 | ||
703 | num = tce_iommu_find_free_table(container); | |
704 | if (num < 0) | |
705 | return num; | |
706 | ||
707 | /* Get the first group for ops::create_table */ | |
708 | tcegrp = list_first_entry(&container->group_list, | |
709 | struct tce_iommu_group, next); | |
710 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
711 | if (!table_group) | |
712 | return -EFAULT; | |
713 | ||
714 | if (!(table_group->pgsizes & (1ULL << page_shift))) | |
715 | return -EINVAL; | |
716 | ||
717 | if (!table_group->ops->set_window || !table_group->ops->unset_window || | |
718 | !table_group->ops->get_table_size || | |
719 | !table_group->ops->create_table) | |
720 | return -EPERM; | |
721 | ||
722 | /* Create TCE table */ | |
723 | ret = tce_iommu_create_table(container, table_group, num, | |
724 | page_shift, window_size, levels, &tbl); | |
725 | if (ret) | |
726 | return ret; | |
727 | ||
728 | BUG_ON(!tbl->it_ops->free); | |
729 | ||
730 | /* | |
731 | * Program the table to every group. | |
732 | * Groups have been tested for compatibility at the attach time. | |
733 | */ | |
734 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
735 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
736 | ||
737 | ret = table_group->ops->set_window(table_group, num, tbl); | |
738 | if (ret) | |
739 | goto unset_exit; | |
740 | } | |
741 | ||
742 | container->tables[num] = tbl; | |
743 | ||
744 | /* Return start address assigned by platform in create_table() */ | |
745 | *start_addr = tbl->it_offset << tbl->it_page_shift; | |
746 | ||
747 | return 0; | |
748 | ||
749 | unset_exit: | |
750 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
751 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
752 | table_group->ops->unset_window(table_group, num); | |
753 | } | |
bc82d122 | 754 | tce_iommu_free_table(container, tbl); |
e633bc86 AK |
755 | |
756 | return ret; | |
757 | } | |
758 | ||
759 | static long tce_iommu_remove_window(struct tce_container *container, | |
760 | __u64 start_addr) | |
761 | { | |
762 | struct iommu_table_group *table_group = NULL; | |
763 | struct iommu_table *tbl; | |
764 | struct tce_iommu_group *tcegrp; | |
765 | int num; | |
766 | ||
767 | num = tce_iommu_find_table(container, start_addr, &tbl); | |
768 | if (num < 0) | |
769 | return -EINVAL; | |
770 | ||
771 | BUG_ON(!tbl->it_size); | |
772 | ||
773 | /* Detach groups from IOMMUs */ | |
774 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
775 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
776 | ||
777 | /* | |
778 | * SPAPR TCE IOMMU exposes the default DMA window to | |
779 | * the guest via dma32_window_start/size of | |
780 | * VFIO_IOMMU_SPAPR_TCE_GET_INFO. Some platforms allow | |
781 | * the userspace to remove this window, some do not so | |
782 | * here we check for the platform capability. | |
783 | */ | |
784 | if (!table_group->ops || !table_group->ops->unset_window) | |
785 | return -EPERM; | |
786 | ||
787 | table_group->ops->unset_window(table_group, num); | |
788 | } | |
789 | ||
790 | /* Free table */ | |
791 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
bc82d122 | 792 | tce_iommu_free_table(container, tbl); |
e633bc86 AK |
793 | container->tables[num] = NULL; |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
6f01cc69 AK |
798 | static long tce_iommu_create_default_window(struct tce_container *container) |
799 | { | |
800 | long ret; | |
801 | __u64 start_addr = 0; | |
802 | struct tce_iommu_group *tcegrp; | |
803 | struct iommu_table_group *table_group; | |
804 | ||
d9c72894 AK |
805 | if (!container->def_window_pending) |
806 | return 0; | |
807 | ||
6f01cc69 AK |
808 | if (!tce_groups_attached(container)) |
809 | return -ENODEV; | |
810 | ||
811 | tcegrp = list_first_entry(&container->group_list, | |
812 | struct tce_iommu_group, next); | |
813 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
814 | if (!table_group) | |
815 | return -ENODEV; | |
816 | ||
817 | ret = tce_iommu_create_window(container, IOMMU_PAGE_SHIFT_4K, | |
818 | table_group->tce32_size, 1, &start_addr); | |
819 | WARN_ON_ONCE(!ret && start_addr); | |
820 | ||
d9c72894 AK |
821 | if (!ret) |
822 | container->def_window_pending = false; | |
823 | ||
6f01cc69 AK |
824 | return ret; |
825 | } | |
826 | ||
5ffd229c AK |
827 | static long tce_iommu_ioctl(void *iommu_data, |
828 | unsigned int cmd, unsigned long arg) | |
829 | { | |
830 | struct tce_container *container = iommu_data; | |
e633bc86 | 831 | unsigned long minsz, ddwsz; |
5ffd229c AK |
832 | long ret; |
833 | ||
834 | switch (cmd) { | |
835 | case VFIO_CHECK_EXTENSION: | |
1b69be5e GS |
836 | switch (arg) { |
837 | case VFIO_SPAPR_TCE_IOMMU: | |
2157e7b8 | 838 | case VFIO_SPAPR_TCE_v2_IOMMU: |
1b69be5e GS |
839 | ret = 1; |
840 | break; | |
841 | default: | |
842 | ret = vfio_spapr_iommu_eeh_ioctl(NULL, cmd, arg); | |
843 | break; | |
844 | } | |
845 | ||
846 | return (ret < 0) ? 0 : ret; | |
bc82d122 AK |
847 | } |
848 | ||
849 | /* | |
850 | * Sanity check to prevent one userspace from manipulating | |
851 | * another userspace mm. | |
852 | */ | |
853 | BUG_ON(!container); | |
854 | if (container->mm && container->mm != current->mm) | |
855 | return -EPERM; | |
5ffd229c | 856 | |
bc82d122 | 857 | switch (cmd) { |
5ffd229c AK |
858 | case VFIO_IOMMU_SPAPR_TCE_GET_INFO: { |
859 | struct vfio_iommu_spapr_tce_info info; | |
2157e7b8 | 860 | struct tce_iommu_group *tcegrp; |
0eaf4def AK |
861 | struct iommu_table_group *table_group; |
862 | ||
2157e7b8 | 863 | if (!tce_groups_attached(container)) |
0eaf4def AK |
864 | return -ENXIO; |
865 | ||
2157e7b8 AK |
866 | tcegrp = list_first_entry(&container->group_list, |
867 | struct tce_iommu_group, next); | |
868 | table_group = iommu_group_get_iommudata(tcegrp->grp); | |
5ffd229c | 869 | |
4793d65d | 870 | if (!table_group) |
5ffd229c AK |
871 | return -ENXIO; |
872 | ||
873 | minsz = offsetofend(struct vfio_iommu_spapr_tce_info, | |
874 | dma32_window_size); | |
875 | ||
876 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
877 | return -EFAULT; | |
878 | ||
879 | if (info.argsz < minsz) | |
880 | return -EINVAL; | |
881 | ||
4793d65d AK |
882 | info.dma32_window_start = table_group->tce32_start; |
883 | info.dma32_window_size = table_group->tce32_size; | |
5ffd229c | 884 | info.flags = 0; |
e633bc86 AK |
885 | memset(&info.ddw, 0, sizeof(info.ddw)); |
886 | ||
887 | if (table_group->max_dynamic_windows_supported && | |
888 | container->v2) { | |
889 | info.flags |= VFIO_IOMMU_SPAPR_INFO_DDW; | |
890 | info.ddw.pgsizes = table_group->pgsizes; | |
891 | info.ddw.max_dynamic_windows_supported = | |
892 | table_group->max_dynamic_windows_supported; | |
893 | info.ddw.levels = table_group->max_levels; | |
894 | } | |
895 | ||
896 | ddwsz = offsetofend(struct vfio_iommu_spapr_tce_info, ddw); | |
897 | ||
898 | if (info.argsz >= ddwsz) | |
899 | minsz = ddwsz; | |
5ffd229c AK |
900 | |
901 | if (copy_to_user((void __user *)arg, &info, minsz)) | |
902 | return -EFAULT; | |
903 | ||
904 | return 0; | |
905 | } | |
906 | case VFIO_IOMMU_MAP_DMA: { | |
907 | struct vfio_iommu_type1_dma_map param; | |
0eaf4def | 908 | struct iommu_table *tbl = NULL; |
0eaf4def | 909 | long num; |
05c6cfb9 | 910 | enum dma_data_direction direction; |
5ffd229c | 911 | |
3c56e822 AK |
912 | if (!container->enabled) |
913 | return -EPERM; | |
914 | ||
5ffd229c AK |
915 | minsz = offsetofend(struct vfio_iommu_type1_dma_map, size); |
916 | ||
917 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
918 | return -EFAULT; | |
919 | ||
920 | if (param.argsz < minsz) | |
921 | return -EINVAL; | |
922 | ||
923 | if (param.flags & ~(VFIO_DMA_MAP_FLAG_READ | | |
924 | VFIO_DMA_MAP_FLAG_WRITE)) | |
925 | return -EINVAL; | |
926 | ||
d9c72894 AK |
927 | ret = tce_iommu_create_default_window(container); |
928 | if (ret) | |
929 | return ret; | |
930 | ||
0eaf4def AK |
931 | num = tce_iommu_find_table(container, param.iova, &tbl); |
932 | if (num < 0) | |
933 | return -ENXIO; | |
934 | ||
00663d4e AK |
935 | if ((param.size & ~IOMMU_PAGE_MASK(tbl)) || |
936 | (param.vaddr & ~IOMMU_PAGE_MASK(tbl))) | |
5ffd229c AK |
937 | return -EINVAL; |
938 | ||
939 | /* iova is checked by the IOMMU API */ | |
05c6cfb9 AK |
940 | if (param.flags & VFIO_DMA_MAP_FLAG_READ) { |
941 | if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) | |
942 | direction = DMA_BIDIRECTIONAL; | |
943 | else | |
944 | direction = DMA_TO_DEVICE; | |
945 | } else { | |
946 | if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) | |
947 | direction = DMA_FROM_DEVICE; | |
948 | else | |
949 | return -EINVAL; | |
950 | } | |
5ffd229c | 951 | |
05c6cfb9 | 952 | ret = iommu_tce_put_param_check(tbl, param.iova, param.vaddr); |
5ffd229c AK |
953 | if (ret) |
954 | return ret; | |
955 | ||
2157e7b8 AK |
956 | if (container->v2) |
957 | ret = tce_iommu_build_v2(container, tbl, | |
958 | param.iova >> tbl->it_page_shift, | |
959 | param.vaddr, | |
960 | param.size >> tbl->it_page_shift, | |
961 | direction); | |
962 | else | |
963 | ret = tce_iommu_build(container, tbl, | |
964 | param.iova >> tbl->it_page_shift, | |
965 | param.vaddr, | |
966 | param.size >> tbl->it_page_shift, | |
967 | direction); | |
5ffd229c AK |
968 | |
969 | iommu_flush_tce(tbl); | |
970 | ||
971 | return ret; | |
972 | } | |
973 | case VFIO_IOMMU_UNMAP_DMA: { | |
974 | struct vfio_iommu_type1_dma_unmap param; | |
0eaf4def AK |
975 | struct iommu_table *tbl = NULL; |
976 | long num; | |
5ffd229c | 977 | |
3c56e822 AK |
978 | if (!container->enabled) |
979 | return -EPERM; | |
980 | ||
5ffd229c AK |
981 | minsz = offsetofend(struct vfio_iommu_type1_dma_unmap, |
982 | size); | |
983 | ||
984 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
985 | return -EFAULT; | |
986 | ||
987 | if (param.argsz < minsz) | |
988 | return -EINVAL; | |
989 | ||
990 | /* No flag is supported now */ | |
991 | if (param.flags) | |
992 | return -EINVAL; | |
993 | ||
d9c72894 AK |
994 | ret = tce_iommu_create_default_window(container); |
995 | if (ret) | |
996 | return ret; | |
997 | ||
0eaf4def AK |
998 | num = tce_iommu_find_table(container, param.iova, &tbl); |
999 | if (num < 0) | |
1000 | return -ENXIO; | |
1001 | ||
00663d4e | 1002 | if (param.size & ~IOMMU_PAGE_MASK(tbl)) |
5ffd229c AK |
1003 | return -EINVAL; |
1004 | ||
1005 | ret = iommu_tce_clear_param_check(tbl, param.iova, 0, | |
00663d4e | 1006 | param.size >> tbl->it_page_shift); |
5ffd229c AK |
1007 | if (ret) |
1008 | return ret; | |
1009 | ||
9b14a1ff | 1010 | ret = tce_iommu_clear(container, tbl, |
00663d4e AK |
1011 | param.iova >> tbl->it_page_shift, |
1012 | param.size >> tbl->it_page_shift); | |
5ffd229c AK |
1013 | iommu_flush_tce(tbl); |
1014 | ||
1015 | return ret; | |
1016 | } | |
2157e7b8 AK |
1017 | case VFIO_IOMMU_SPAPR_REGISTER_MEMORY: { |
1018 | struct vfio_iommu_spapr_register_memory param; | |
1019 | ||
1020 | if (!container->v2) | |
1021 | break; | |
1022 | ||
1023 | minsz = offsetofend(struct vfio_iommu_spapr_register_memory, | |
1024 | size); | |
1025 | ||
bc82d122 AK |
1026 | ret = tce_iommu_mm_set(container); |
1027 | if (ret) | |
1028 | return ret; | |
1029 | ||
2157e7b8 AK |
1030 | if (copy_from_user(¶m, (void __user *)arg, minsz)) |
1031 | return -EFAULT; | |
1032 | ||
1033 | if (param.argsz < minsz) | |
1034 | return -EINVAL; | |
1035 | ||
1036 | /* No flag is supported now */ | |
1037 | if (param.flags) | |
1038 | return -EINVAL; | |
1039 | ||
1040 | mutex_lock(&container->lock); | |
1041 | ret = tce_iommu_register_pages(container, param.vaddr, | |
1042 | param.size); | |
1043 | mutex_unlock(&container->lock); | |
1044 | ||
1045 | return ret; | |
1046 | } | |
1047 | case VFIO_IOMMU_SPAPR_UNREGISTER_MEMORY: { | |
1048 | struct vfio_iommu_spapr_register_memory param; | |
1049 | ||
1050 | if (!container->v2) | |
1051 | break; | |
1052 | ||
bc82d122 AK |
1053 | if (!container->mm) |
1054 | return -EPERM; | |
1055 | ||
2157e7b8 AK |
1056 | minsz = offsetofend(struct vfio_iommu_spapr_register_memory, |
1057 | size); | |
1058 | ||
1059 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
1060 | return -EFAULT; | |
1061 | ||
1062 | if (param.argsz < minsz) | |
1063 | return -EINVAL; | |
1064 | ||
1065 | /* No flag is supported now */ | |
1066 | if (param.flags) | |
1067 | return -EINVAL; | |
1068 | ||
1069 | mutex_lock(&container->lock); | |
1070 | ret = tce_iommu_unregister_pages(container, param.vaddr, | |
1071 | param.size); | |
1072 | mutex_unlock(&container->lock); | |
1073 | ||
1074 | return ret; | |
1075 | } | |
5ffd229c | 1076 | case VFIO_IOMMU_ENABLE: |
2157e7b8 AK |
1077 | if (container->v2) |
1078 | break; | |
1079 | ||
5ffd229c AK |
1080 | mutex_lock(&container->lock); |
1081 | ret = tce_iommu_enable(container); | |
1082 | mutex_unlock(&container->lock); | |
1083 | return ret; | |
1084 | ||
1085 | ||
1086 | case VFIO_IOMMU_DISABLE: | |
2157e7b8 AK |
1087 | if (container->v2) |
1088 | break; | |
1089 | ||
5ffd229c AK |
1090 | mutex_lock(&container->lock); |
1091 | tce_iommu_disable(container); | |
1092 | mutex_unlock(&container->lock); | |
1093 | return 0; | |
1b69be5e | 1094 | |
2157e7b8 AK |
1095 | case VFIO_EEH_PE_OP: { |
1096 | struct tce_iommu_group *tcegrp; | |
1097 | ||
1098 | ret = 0; | |
1099 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
1100 | ret = vfio_spapr_iommu_eeh_ioctl(tcegrp->grp, | |
1101 | cmd, arg); | |
1102 | if (ret) | |
1103 | return ret; | |
1104 | } | |
1105 | return ret; | |
1106 | } | |
1107 | ||
e633bc86 AK |
1108 | case VFIO_IOMMU_SPAPR_TCE_CREATE: { |
1109 | struct vfio_iommu_spapr_tce_create create; | |
1110 | ||
1111 | if (!container->v2) | |
1112 | break; | |
1113 | ||
bc82d122 AK |
1114 | ret = tce_iommu_mm_set(container); |
1115 | if (ret) | |
1116 | return ret; | |
1117 | ||
e633bc86 AK |
1118 | if (!tce_groups_attached(container)) |
1119 | return -ENXIO; | |
1120 | ||
1121 | minsz = offsetofend(struct vfio_iommu_spapr_tce_create, | |
1122 | start_addr); | |
1123 | ||
1124 | if (copy_from_user(&create, (void __user *)arg, minsz)) | |
1125 | return -EFAULT; | |
1126 | ||
1127 | if (create.argsz < minsz) | |
1128 | return -EINVAL; | |
1129 | ||
1130 | if (create.flags) | |
1131 | return -EINVAL; | |
1132 | ||
1133 | mutex_lock(&container->lock); | |
1134 | ||
d9c72894 | 1135 | ret = tce_iommu_create_default_window(container); |
2da64d20 AK |
1136 | if (!ret) |
1137 | ret = tce_iommu_create_window(container, | |
1138 | create.page_shift, | |
1139 | create.window_size, create.levels, | |
1140 | &create.start_addr); | |
e633bc86 AK |
1141 | |
1142 | mutex_unlock(&container->lock); | |
1143 | ||
1144 | if (!ret && copy_to_user((void __user *)arg, &create, minsz)) | |
1145 | ret = -EFAULT; | |
1146 | ||
1147 | return ret; | |
1148 | } | |
1149 | case VFIO_IOMMU_SPAPR_TCE_REMOVE: { | |
1150 | struct vfio_iommu_spapr_tce_remove remove; | |
1151 | ||
1152 | if (!container->v2) | |
1153 | break; | |
1154 | ||
bc82d122 AK |
1155 | ret = tce_iommu_mm_set(container); |
1156 | if (ret) | |
1157 | return ret; | |
1158 | ||
e633bc86 AK |
1159 | if (!tce_groups_attached(container)) |
1160 | return -ENXIO; | |
1161 | ||
1162 | minsz = offsetofend(struct vfio_iommu_spapr_tce_remove, | |
1163 | start_addr); | |
1164 | ||
1165 | if (copy_from_user(&remove, (void __user *)arg, minsz)) | |
1166 | return -EFAULT; | |
1167 | ||
1168 | if (remove.argsz < minsz) | |
1169 | return -EINVAL; | |
1170 | ||
1171 | if (remove.flags) | |
1172 | return -EINVAL; | |
1173 | ||
d9c72894 AK |
1174 | if (container->def_window_pending && !remove.start_addr) { |
1175 | container->def_window_pending = false; | |
1176 | return 0; | |
1177 | } | |
1178 | ||
e633bc86 AK |
1179 | mutex_lock(&container->lock); |
1180 | ||
1181 | ret = tce_iommu_remove_window(container, remove.start_addr); | |
1182 | ||
1183 | mutex_unlock(&container->lock); | |
1184 | ||
1185 | return ret; | |
1186 | } | |
5ffd229c AK |
1187 | } |
1188 | ||
1189 | return -ENOTTY; | |
1190 | } | |
1191 | ||
f87a8864 AK |
1192 | static void tce_iommu_release_ownership(struct tce_container *container, |
1193 | struct iommu_table_group *table_group) | |
1194 | { | |
1195 | int i; | |
1196 | ||
1197 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
2157e7b8 | 1198 | struct iommu_table *tbl = container->tables[i]; |
f87a8864 AK |
1199 | |
1200 | if (!tbl) | |
1201 | continue; | |
1202 | ||
1203 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
bc82d122 | 1204 | tce_iommu_userspace_view_free(tbl, container->mm); |
f87a8864 AK |
1205 | if (tbl->it_map) |
1206 | iommu_release_ownership(tbl); | |
2157e7b8 AK |
1207 | |
1208 | container->tables[i] = NULL; | |
f87a8864 AK |
1209 | } |
1210 | } | |
1211 | ||
1212 | static int tce_iommu_take_ownership(struct tce_container *container, | |
1213 | struct iommu_table_group *table_group) | |
1214 | { | |
1215 | int i, j, rc = 0; | |
1216 | ||
1217 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
1218 | struct iommu_table *tbl = table_group->tables[i]; | |
1219 | ||
1220 | if (!tbl || !tbl->it_map) | |
1221 | continue; | |
1222 | ||
39701e56 | 1223 | rc = iommu_take_ownership(tbl); |
f87a8864 AK |
1224 | if (rc) { |
1225 | for (j = 0; j < i; ++j) | |
1226 | iommu_release_ownership( | |
1227 | table_group->tables[j]); | |
1228 | ||
1229 | return rc; | |
1230 | } | |
1231 | } | |
1232 | ||
2157e7b8 AK |
1233 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) |
1234 | container->tables[i] = table_group->tables[i]; | |
1235 | ||
f87a8864 AK |
1236 | return 0; |
1237 | } | |
1238 | ||
1239 | static void tce_iommu_release_ownership_ddw(struct tce_container *container, | |
1240 | struct iommu_table_group *table_group) | |
1241 | { | |
46d3e1e1 AK |
1242 | long i; |
1243 | ||
1244 | if (!table_group->ops->unset_window) { | |
1245 | WARN_ON_ONCE(1); | |
1246 | return; | |
1247 | } | |
1248 | ||
2157e7b8 | 1249 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) |
46d3e1e1 | 1250 | table_group->ops->unset_window(table_group, i); |
46d3e1e1 | 1251 | |
f87a8864 AK |
1252 | table_group->ops->release_ownership(table_group); |
1253 | } | |
1254 | ||
1255 | static long tce_iommu_take_ownership_ddw(struct tce_container *container, | |
1256 | struct iommu_table_group *table_group) | |
1257 | { | |
930a42de AK |
1258 | long i, ret = 0; |
1259 | ||
46d3e1e1 AK |
1260 | if (!table_group->ops->create_table || !table_group->ops->set_window || |
1261 | !table_group->ops->release_ownership) { | |
1262 | WARN_ON_ONCE(1); | |
1263 | return -EFAULT; | |
1264 | } | |
1265 | ||
f87a8864 AK |
1266 | table_group->ops->take_ownership(table_group); |
1267 | ||
930a42de AK |
1268 | /* Set all windows to the new group */ |
1269 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) { | |
1270 | struct iommu_table *tbl = container->tables[i]; | |
1271 | ||
1272 | if (!tbl) | |
1273 | continue; | |
1274 | ||
1275 | ret = table_group->ops->set_window(table_group, i, tbl); | |
1276 | if (ret) | |
1277 | goto release_exit; | |
1278 | } | |
1279 | ||
2157e7b8 | 1280 | return 0; |
930a42de AK |
1281 | |
1282 | release_exit: | |
1283 | for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) | |
1284 | table_group->ops->unset_window(table_group, i); | |
1285 | ||
1286 | table_group->ops->release_ownership(table_group); | |
1287 | ||
1288 | return ret; | |
f87a8864 AK |
1289 | } |
1290 | ||
5ffd229c AK |
1291 | static int tce_iommu_attach_group(void *iommu_data, |
1292 | struct iommu_group *iommu_group) | |
1293 | { | |
1294 | int ret; | |
1295 | struct tce_container *container = iommu_data; | |
0eaf4def | 1296 | struct iommu_table_group *table_group; |
2157e7b8 | 1297 | struct tce_iommu_group *tcegrp = NULL; |
5ffd229c | 1298 | |
5ffd229c AK |
1299 | mutex_lock(&container->lock); |
1300 | ||
1301 | /* pr_debug("tce_vfio: Attaching group #%u to iommu %p\n", | |
1302 | iommu_group_id(iommu_group), iommu_group); */ | |
2157e7b8 | 1303 | table_group = iommu_group_get_iommudata(iommu_group); |
bd00fdf1 GK |
1304 | if (!table_group) { |
1305 | ret = -ENODEV; | |
1306 | goto unlock_exit; | |
1307 | } | |
2157e7b8 AK |
1308 | |
1309 | if (tce_groups_attached(container) && (!table_group->ops || | |
1310 | !table_group->ops->take_ownership || | |
1311 | !table_group->ops->release_ownership)) { | |
5ffd229c | 1312 | ret = -EBUSY; |
22af4859 AK |
1313 | goto unlock_exit; |
1314 | } | |
1315 | ||
2157e7b8 AK |
1316 | /* Check if new group has the same iommu_ops (i.e. compatible) */ |
1317 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
1318 | struct iommu_table_group *table_group_tmp; | |
1319 | ||
1320 | if (tcegrp->grp == iommu_group) { | |
1321 | pr_warn("tce_vfio: Group %d is already attached\n", | |
1322 | iommu_group_id(iommu_group)); | |
1323 | ret = -EBUSY; | |
1324 | goto unlock_exit; | |
1325 | } | |
1326 | table_group_tmp = iommu_group_get_iommudata(tcegrp->grp); | |
54de285b AK |
1327 | if (table_group_tmp->ops->create_table != |
1328 | table_group->ops->create_table) { | |
2157e7b8 AK |
1329 | pr_warn("tce_vfio: Group %d is incompatible with group %d\n", |
1330 | iommu_group_id(iommu_group), | |
1331 | iommu_group_id(tcegrp->grp)); | |
1332 | ret = -EPERM; | |
1333 | goto unlock_exit; | |
1334 | } | |
5ffd229c AK |
1335 | } |
1336 | ||
2157e7b8 AK |
1337 | tcegrp = kzalloc(sizeof(*tcegrp), GFP_KERNEL); |
1338 | if (!tcegrp) { | |
1339 | ret = -ENOMEM; | |
0eaf4def AK |
1340 | goto unlock_exit; |
1341 | } | |
1342 | ||
f87a8864 | 1343 | if (!table_group->ops || !table_group->ops->take_ownership || |
6f01cc69 | 1344 | !table_group->ops->release_ownership) { |
1282ba7f AK |
1345 | if (container->v2) { |
1346 | ret = -EPERM; | |
1347 | goto unlock_exit; | |
1348 | } | |
f87a8864 | 1349 | ret = tce_iommu_take_ownership(container, table_group); |
6f01cc69 | 1350 | } else { |
1282ba7f AK |
1351 | if (!container->v2) { |
1352 | ret = -EPERM; | |
1353 | goto unlock_exit; | |
1354 | } | |
f87a8864 | 1355 | ret = tce_iommu_take_ownership_ddw(container, table_group); |
6f01cc69 | 1356 | if (!tce_groups_attached(container) && !container->tables[0]) |
d9c72894 | 1357 | container->def_window_pending = true; |
6f01cc69 | 1358 | } |
f87a8864 | 1359 | |
2157e7b8 AK |
1360 | if (!ret) { |
1361 | tcegrp->grp = iommu_group; | |
1362 | list_add(&tcegrp->next, &container->group_list); | |
1363 | } | |
22af4859 AK |
1364 | |
1365 | unlock_exit: | |
2157e7b8 AK |
1366 | if (ret && tcegrp) |
1367 | kfree(tcegrp); | |
1368 | ||
5ffd229c AK |
1369 | mutex_unlock(&container->lock); |
1370 | ||
1371 | return ret; | |
1372 | } | |
1373 | ||
1374 | static void tce_iommu_detach_group(void *iommu_data, | |
1375 | struct iommu_group *iommu_group) | |
1376 | { | |
1377 | struct tce_container *container = iommu_data; | |
0eaf4def | 1378 | struct iommu_table_group *table_group; |
2157e7b8 AK |
1379 | bool found = false; |
1380 | struct tce_iommu_group *tcegrp; | |
5ffd229c | 1381 | |
5ffd229c | 1382 | mutex_lock(&container->lock); |
2157e7b8 AK |
1383 | |
1384 | list_for_each_entry(tcegrp, &container->group_list, next) { | |
1385 | if (tcegrp->grp == iommu_group) { | |
1386 | found = true; | |
1387 | break; | |
1388 | } | |
22af4859 | 1389 | } |
5ffd229c | 1390 | |
2157e7b8 AK |
1391 | if (!found) { |
1392 | pr_warn("tce_vfio: detaching unattached group #%u\n", | |
1393 | iommu_group_id(iommu_group)); | |
1394 | goto unlock_exit; | |
5ffd229c | 1395 | } |
22af4859 | 1396 | |
2157e7b8 AK |
1397 | list_del(&tcegrp->next); |
1398 | kfree(tcegrp); | |
0eaf4def AK |
1399 | |
1400 | table_group = iommu_group_get_iommudata(iommu_group); | |
1401 | BUG_ON(!table_group); | |
1402 | ||
f87a8864 AK |
1403 | if (!table_group->ops || !table_group->ops->release_ownership) |
1404 | tce_iommu_release_ownership(container, table_group); | |
1405 | else | |
1406 | tce_iommu_release_ownership_ddw(container, table_group); | |
22af4859 AK |
1407 | |
1408 | unlock_exit: | |
5ffd229c AK |
1409 | mutex_unlock(&container->lock); |
1410 | } | |
1411 | ||
1412 | const struct vfio_iommu_driver_ops tce_iommu_driver_ops = { | |
1413 | .name = "iommu-vfio-powerpc", | |
1414 | .owner = THIS_MODULE, | |
1415 | .open = tce_iommu_open, | |
1416 | .release = tce_iommu_release, | |
1417 | .ioctl = tce_iommu_ioctl, | |
1418 | .attach_group = tce_iommu_attach_group, | |
1419 | .detach_group = tce_iommu_detach_group, | |
1420 | }; | |
1421 | ||
1422 | static int __init tce_iommu_init(void) | |
1423 | { | |
1424 | return vfio_register_iommu_driver(&tce_iommu_driver_ops); | |
1425 | } | |
1426 | ||
1427 | static void __exit tce_iommu_cleanup(void) | |
1428 | { | |
1429 | vfio_unregister_iommu_driver(&tce_iommu_driver_ops); | |
1430 | } | |
1431 | ||
1432 | module_init(tce_iommu_init); | |
1433 | module_exit(tce_iommu_cleanup); | |
1434 | ||
1435 | MODULE_VERSION(DRIVER_VERSION); | |
1436 | MODULE_LICENSE("GPL v2"); | |
1437 | MODULE_AUTHOR(DRIVER_AUTHOR); | |
1438 | MODULE_DESCRIPTION(DRIVER_DESC); | |
1439 |