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> | |
22 | #include <asm/iommu.h> | |
23 | #include <asm/tce.h> | |
24 | ||
25 | #define DRIVER_VERSION "0.1" | |
26 | #define DRIVER_AUTHOR "aik@ozlabs.ru" | |
27 | #define DRIVER_DESC "VFIO IOMMU SPAPR TCE" | |
28 | ||
29 | static void tce_iommu_detach_group(void *iommu_data, | |
30 | struct iommu_group *iommu_group); | |
31 | ||
2d270df8 AK |
32 | static long try_increment_locked_vm(long npages) |
33 | { | |
34 | long ret = 0, locked, lock_limit; | |
35 | ||
36 | if (!current || !current->mm) | |
37 | return -ESRCH; /* process exited */ | |
38 | ||
39 | if (!npages) | |
40 | return 0; | |
41 | ||
42 | down_write(¤t->mm->mmap_sem); | |
43 | locked = current->mm->locked_vm + npages; | |
44 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
45 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
46 | ret = -ENOMEM; | |
47 | else | |
48 | current->mm->locked_vm += npages; | |
49 | ||
50 | pr_debug("[%d] RLIMIT_MEMLOCK +%ld %ld/%ld%s\n", current->pid, | |
51 | npages << PAGE_SHIFT, | |
52 | current->mm->locked_vm << PAGE_SHIFT, | |
53 | rlimit(RLIMIT_MEMLOCK), | |
54 | ret ? " - exceeded" : ""); | |
55 | ||
56 | up_write(¤t->mm->mmap_sem); | |
57 | ||
58 | return ret; | |
59 | } | |
60 | ||
61 | static void decrement_locked_vm(long npages) | |
62 | { | |
63 | if (!current || !current->mm || !npages) | |
64 | return; /* process exited */ | |
65 | ||
66 | down_write(¤t->mm->mmap_sem); | |
67 | if (WARN_ON_ONCE(npages > current->mm->locked_vm)) | |
68 | npages = current->mm->locked_vm; | |
69 | current->mm->locked_vm -= npages; | |
70 | pr_debug("[%d] RLIMIT_MEMLOCK -%ld %ld/%ld\n", current->pid, | |
71 | npages << PAGE_SHIFT, | |
72 | current->mm->locked_vm << PAGE_SHIFT, | |
73 | rlimit(RLIMIT_MEMLOCK)); | |
74 | up_write(¤t->mm->mmap_sem); | |
75 | } | |
76 | ||
5ffd229c AK |
77 | /* |
78 | * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation | |
79 | * | |
80 | * This code handles mapping and unmapping of user data buffers | |
81 | * into DMA'ble space using the IOMMU | |
82 | */ | |
83 | ||
84 | /* | |
85 | * The container descriptor supports only a single group per container. | |
86 | * Required by the API as the container is not supplied with the IOMMU group | |
87 | * at the moment of initialization. | |
88 | */ | |
89 | struct tce_container { | |
90 | struct mutex lock; | |
91 | struct iommu_table *tbl; | |
92 | bool enabled; | |
2d270df8 | 93 | unsigned long locked_pages; |
5ffd229c AK |
94 | }; |
95 | ||
e432bc7e AK |
96 | static bool tce_page_is_contained(struct page *page, unsigned page_shift) |
97 | { | |
98 | /* | |
99 | * Check that the TCE table granularity is not bigger than the size of | |
100 | * a page we just found. Otherwise the hardware can get access to | |
101 | * a bigger memory chunk that it should. | |
102 | */ | |
103 | return (PAGE_SHIFT + compound_order(compound_head(page))) >= page_shift; | |
104 | } | |
105 | ||
5ffd229c AK |
106 | static int tce_iommu_enable(struct tce_container *container) |
107 | { | |
108 | int ret = 0; | |
2d270df8 | 109 | unsigned long locked; |
5ffd229c AK |
110 | struct iommu_table *tbl = container->tbl; |
111 | ||
112 | if (!container->tbl) | |
113 | return -ENXIO; | |
114 | ||
115 | if (!current->mm) | |
116 | return -ESRCH; /* process exited */ | |
117 | ||
118 | if (container->enabled) | |
119 | return -EBUSY; | |
120 | ||
121 | /* | |
122 | * When userspace pages are mapped into the IOMMU, they are effectively | |
123 | * locked memory, so, theoretically, we need to update the accounting | |
124 | * of locked pages on each map and unmap. For powerpc, the map unmap | |
125 | * paths can be very hot, though, and the accounting would kill | |
126 | * performance, especially since it would be difficult to impossible | |
127 | * to handle the accounting in real mode only. | |
128 | * | |
129 | * To address that, rather than precisely accounting every page, we | |
130 | * instead account for a worst case on locked memory when the iommu is | |
131 | * enabled and disabled. The worst case upper bound on locked memory | |
132 | * is the size of the whole iommu window, which is usually relatively | |
133 | * small (compared to total memory sizes) on POWER hardware. | |
134 | * | |
135 | * Also we don't have a nice way to fail on H_PUT_TCE due to ulimits, | |
136 | * that would effectively kill the guest at random points, much better | |
137 | * enforcing the limit based on the max that the guest can map. | |
2d270df8 AK |
138 | * |
139 | * Unfortunately at the moment it counts whole tables, no matter how | |
140 | * much memory the guest has. I.e. for 4GB guest and 4 IOMMU groups | |
141 | * each with 2GB DMA window, 8GB will be counted here. The reason for | |
142 | * this is that we cannot tell here the amount of RAM used by the guest | |
143 | * as this information is only available from KVM and VFIO is | |
144 | * KVM agnostic. | |
5ffd229c | 145 | */ |
2d270df8 AK |
146 | locked = (tbl->it_size << tbl->it_page_shift) >> PAGE_SHIFT; |
147 | ret = try_increment_locked_vm(locked); | |
148 | if (ret) | |
149 | return ret; | |
5ffd229c | 150 | |
2d270df8 AK |
151 | container->locked_pages = locked; |
152 | ||
153 | container->enabled = true; | |
5ffd229c AK |
154 | |
155 | return ret; | |
156 | } | |
157 | ||
158 | static void tce_iommu_disable(struct tce_container *container) | |
159 | { | |
160 | if (!container->enabled) | |
161 | return; | |
162 | ||
163 | container->enabled = false; | |
164 | ||
2d270df8 | 165 | if (!current->mm) |
5ffd229c AK |
166 | return; |
167 | ||
2d270df8 | 168 | decrement_locked_vm(container->locked_pages); |
5ffd229c AK |
169 | } |
170 | ||
171 | static void *tce_iommu_open(unsigned long arg) | |
172 | { | |
173 | struct tce_container *container; | |
174 | ||
175 | if (arg != VFIO_SPAPR_TCE_IOMMU) { | |
176 | pr_err("tce_vfio: Wrong IOMMU type\n"); | |
177 | return ERR_PTR(-EINVAL); | |
178 | } | |
179 | ||
180 | container = kzalloc(sizeof(*container), GFP_KERNEL); | |
181 | if (!container) | |
182 | return ERR_PTR(-ENOMEM); | |
183 | ||
184 | mutex_init(&container->lock); | |
185 | ||
186 | return container; | |
187 | } | |
188 | ||
189 | static void tce_iommu_release(void *iommu_data) | |
190 | { | |
191 | struct tce_container *container = iommu_data; | |
192 | ||
b348aa65 | 193 | WARN_ON(container->tbl && !container->tbl->it_table_group->group); |
5ffd229c | 194 | |
b348aa65 AK |
195 | if (container->tbl && container->tbl->it_table_group->group) |
196 | tce_iommu_detach_group(iommu_data, | |
197 | container->tbl->it_table_group->group); | |
5ffd229c | 198 | |
649354b7 | 199 | tce_iommu_disable(container); |
5ffd229c AK |
200 | mutex_destroy(&container->lock); |
201 | ||
202 | kfree(container); | |
203 | } | |
204 | ||
649354b7 AK |
205 | static void tce_iommu_unuse_page(struct tce_container *container, |
206 | unsigned long oldtce) | |
207 | { | |
208 | struct page *page; | |
209 | ||
210 | if (!(oldtce & (TCE_PCI_READ | TCE_PCI_WRITE))) | |
211 | return; | |
212 | ||
213 | page = pfn_to_page(oldtce >> PAGE_SHIFT); | |
214 | ||
215 | if (oldtce & TCE_PCI_WRITE) | |
216 | SetPageDirty(page); | |
217 | ||
218 | put_page(page); | |
219 | } | |
220 | ||
9b14a1ff AK |
221 | static int tce_iommu_clear(struct tce_container *container, |
222 | struct iommu_table *tbl, | |
223 | unsigned long entry, unsigned long pages) | |
224 | { | |
225 | unsigned long oldtce; | |
9b14a1ff AK |
226 | |
227 | for ( ; pages; --pages, ++entry) { | |
228 | oldtce = iommu_clear_tce(tbl, entry); | |
229 | if (!oldtce) | |
230 | continue; | |
231 | ||
649354b7 | 232 | tce_iommu_unuse_page(container, oldtce); |
9b14a1ff AK |
233 | } |
234 | ||
235 | return 0; | |
236 | } | |
237 | ||
649354b7 AK |
238 | static int tce_iommu_use_page(unsigned long tce, unsigned long *hpa) |
239 | { | |
240 | struct page *page = NULL; | |
241 | enum dma_data_direction direction = iommu_tce_direction(tce); | |
242 | ||
243 | if (get_user_pages_fast(tce & PAGE_MASK, 1, | |
244 | direction != DMA_TO_DEVICE, &page) != 1) | |
245 | return -EFAULT; | |
246 | ||
247 | *hpa = __pa((unsigned long) page_address(page)); | |
248 | ||
249 | return 0; | |
250 | } | |
251 | ||
9b14a1ff AK |
252 | static long tce_iommu_build(struct tce_container *container, |
253 | struct iommu_table *tbl, | |
254 | unsigned long entry, unsigned long tce, unsigned long pages) | |
255 | { | |
256 | long i, ret = 0; | |
649354b7 AK |
257 | struct page *page; |
258 | unsigned long hpa; | |
9b14a1ff AK |
259 | enum dma_data_direction direction = iommu_tce_direction(tce); |
260 | ||
261 | for (i = 0; i < pages; ++i) { | |
262 | unsigned long offset = tce & IOMMU_PAGE_MASK(tbl) & ~PAGE_MASK; | |
263 | ||
649354b7 AK |
264 | ret = tce_iommu_use_page(tce, &hpa); |
265 | if (ret) | |
9b14a1ff | 266 | break; |
e432bc7e | 267 | |
649354b7 | 268 | page = pfn_to_page(hpa >> PAGE_SHIFT); |
e432bc7e AK |
269 | if (!tce_page_is_contained(page, tbl->it_page_shift)) { |
270 | ret = -EPERM; | |
271 | break; | |
272 | } | |
273 | ||
649354b7 AK |
274 | hpa |= offset; |
275 | ret = iommu_tce_build(tbl, entry + i, (unsigned long) __va(hpa), | |
276 | direction); | |
9b14a1ff | 277 | if (ret) { |
649354b7 | 278 | tce_iommu_unuse_page(container, hpa); |
9b14a1ff AK |
279 | pr_err("iommu_tce: %s failed ioba=%lx, tce=%lx, ret=%ld\n", |
280 | __func__, entry << tbl->it_page_shift, | |
281 | tce, ret); | |
282 | break; | |
283 | } | |
00663d4e | 284 | tce += IOMMU_PAGE_SIZE(tbl); |
9b14a1ff AK |
285 | } |
286 | ||
287 | if (ret) | |
288 | tce_iommu_clear(container, tbl, entry, i); | |
289 | ||
290 | return ret; | |
291 | } | |
292 | ||
5ffd229c AK |
293 | static long tce_iommu_ioctl(void *iommu_data, |
294 | unsigned int cmd, unsigned long arg) | |
295 | { | |
296 | struct tce_container *container = iommu_data; | |
297 | unsigned long minsz; | |
298 | long ret; | |
299 | ||
300 | switch (cmd) { | |
301 | case VFIO_CHECK_EXTENSION: | |
1b69be5e GS |
302 | switch (arg) { |
303 | case VFIO_SPAPR_TCE_IOMMU: | |
304 | ret = 1; | |
305 | break; | |
306 | default: | |
307 | ret = vfio_spapr_iommu_eeh_ioctl(NULL, cmd, arg); | |
308 | break; | |
309 | } | |
310 | ||
311 | return (ret < 0) ? 0 : ret; | |
5ffd229c AK |
312 | |
313 | case VFIO_IOMMU_SPAPR_TCE_GET_INFO: { | |
314 | struct vfio_iommu_spapr_tce_info info; | |
315 | struct iommu_table *tbl = container->tbl; | |
316 | ||
317 | if (WARN_ON(!tbl)) | |
318 | return -ENXIO; | |
319 | ||
320 | minsz = offsetofend(struct vfio_iommu_spapr_tce_info, | |
321 | dma32_window_size); | |
322 | ||
323 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
324 | return -EFAULT; | |
325 | ||
326 | if (info.argsz < minsz) | |
327 | return -EINVAL; | |
328 | ||
00663d4e AK |
329 | info.dma32_window_start = tbl->it_offset << tbl->it_page_shift; |
330 | info.dma32_window_size = tbl->it_size << tbl->it_page_shift; | |
5ffd229c AK |
331 | info.flags = 0; |
332 | ||
333 | if (copy_to_user((void __user *)arg, &info, minsz)) | |
334 | return -EFAULT; | |
335 | ||
336 | return 0; | |
337 | } | |
338 | case VFIO_IOMMU_MAP_DMA: { | |
339 | struct vfio_iommu_type1_dma_map param; | |
340 | struct iommu_table *tbl = container->tbl; | |
9b14a1ff | 341 | unsigned long tce; |
5ffd229c | 342 | |
3c56e822 AK |
343 | if (!container->enabled) |
344 | return -EPERM; | |
345 | ||
5ffd229c AK |
346 | if (!tbl) |
347 | return -ENXIO; | |
348 | ||
b348aa65 | 349 | BUG_ON(!tbl->it_table_group->group); |
5ffd229c AK |
350 | |
351 | minsz = offsetofend(struct vfio_iommu_type1_dma_map, size); | |
352 | ||
353 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
354 | return -EFAULT; | |
355 | ||
356 | if (param.argsz < minsz) | |
357 | return -EINVAL; | |
358 | ||
359 | if (param.flags & ~(VFIO_DMA_MAP_FLAG_READ | | |
360 | VFIO_DMA_MAP_FLAG_WRITE)) | |
361 | return -EINVAL; | |
362 | ||
00663d4e AK |
363 | if ((param.size & ~IOMMU_PAGE_MASK(tbl)) || |
364 | (param.vaddr & ~IOMMU_PAGE_MASK(tbl))) | |
5ffd229c AK |
365 | return -EINVAL; |
366 | ||
367 | /* iova is checked by the IOMMU API */ | |
368 | tce = param.vaddr; | |
369 | if (param.flags & VFIO_DMA_MAP_FLAG_READ) | |
370 | tce |= TCE_PCI_READ; | |
371 | if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) | |
372 | tce |= TCE_PCI_WRITE; | |
373 | ||
374 | ret = iommu_tce_put_param_check(tbl, param.iova, tce); | |
375 | if (ret) | |
376 | return ret; | |
377 | ||
9b14a1ff | 378 | ret = tce_iommu_build(container, tbl, |
00663d4e AK |
379 | param.iova >> tbl->it_page_shift, |
380 | tce, param.size >> tbl->it_page_shift); | |
5ffd229c AK |
381 | |
382 | iommu_flush_tce(tbl); | |
383 | ||
384 | return ret; | |
385 | } | |
386 | case VFIO_IOMMU_UNMAP_DMA: { | |
387 | struct vfio_iommu_type1_dma_unmap param; | |
388 | struct iommu_table *tbl = container->tbl; | |
389 | ||
3c56e822 AK |
390 | if (!container->enabled) |
391 | return -EPERM; | |
392 | ||
5ffd229c AK |
393 | if (WARN_ON(!tbl)) |
394 | return -ENXIO; | |
395 | ||
396 | minsz = offsetofend(struct vfio_iommu_type1_dma_unmap, | |
397 | size); | |
398 | ||
399 | if (copy_from_user(¶m, (void __user *)arg, minsz)) | |
400 | return -EFAULT; | |
401 | ||
402 | if (param.argsz < minsz) | |
403 | return -EINVAL; | |
404 | ||
405 | /* No flag is supported now */ | |
406 | if (param.flags) | |
407 | return -EINVAL; | |
408 | ||
00663d4e | 409 | if (param.size & ~IOMMU_PAGE_MASK(tbl)) |
5ffd229c AK |
410 | return -EINVAL; |
411 | ||
412 | ret = iommu_tce_clear_param_check(tbl, param.iova, 0, | |
00663d4e | 413 | param.size >> tbl->it_page_shift); |
5ffd229c AK |
414 | if (ret) |
415 | return ret; | |
416 | ||
9b14a1ff | 417 | ret = tce_iommu_clear(container, tbl, |
00663d4e AK |
418 | param.iova >> tbl->it_page_shift, |
419 | param.size >> tbl->it_page_shift); | |
5ffd229c AK |
420 | iommu_flush_tce(tbl); |
421 | ||
422 | return ret; | |
423 | } | |
424 | case VFIO_IOMMU_ENABLE: | |
425 | mutex_lock(&container->lock); | |
426 | ret = tce_iommu_enable(container); | |
427 | mutex_unlock(&container->lock); | |
428 | return ret; | |
429 | ||
430 | ||
431 | case VFIO_IOMMU_DISABLE: | |
432 | mutex_lock(&container->lock); | |
433 | tce_iommu_disable(container); | |
434 | mutex_unlock(&container->lock); | |
435 | return 0; | |
1b69be5e | 436 | case VFIO_EEH_PE_OP: |
b348aa65 | 437 | if (!container->tbl || !container->tbl->it_table_group->group) |
1b69be5e GS |
438 | return -ENODEV; |
439 | ||
b348aa65 AK |
440 | return vfio_spapr_iommu_eeh_ioctl( |
441 | container->tbl->it_table_group->group, | |
442 | cmd, arg); | |
5ffd229c AK |
443 | } |
444 | ||
445 | return -ENOTTY; | |
446 | } | |
447 | ||
448 | static int tce_iommu_attach_group(void *iommu_data, | |
449 | struct iommu_group *iommu_group) | |
450 | { | |
451 | int ret; | |
452 | struct tce_container *container = iommu_data; | |
453 | struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group); | |
454 | ||
455 | BUG_ON(!tbl); | |
456 | mutex_lock(&container->lock); | |
457 | ||
458 | /* pr_debug("tce_vfio: Attaching group #%u to iommu %p\n", | |
459 | iommu_group_id(iommu_group), iommu_group); */ | |
460 | if (container->tbl) { | |
461 | pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n", | |
b348aa65 AK |
462 | iommu_group_id(container->tbl-> |
463 | it_table_group->group), | |
5ffd229c AK |
464 | iommu_group_id(iommu_group)); |
465 | ret = -EBUSY; | |
22af4859 AK |
466 | goto unlock_exit; |
467 | } | |
468 | ||
469 | if (container->enabled) { | |
5ffd229c AK |
470 | pr_err("tce_vfio: attaching group #%u to enabled container\n", |
471 | iommu_group_id(iommu_group)); | |
472 | ret = -EBUSY; | |
22af4859 | 473 | goto unlock_exit; |
5ffd229c AK |
474 | } |
475 | ||
22af4859 AK |
476 | ret = iommu_take_ownership(tbl); |
477 | if (!ret) | |
478 | container->tbl = tbl; | |
479 | ||
480 | unlock_exit: | |
5ffd229c AK |
481 | mutex_unlock(&container->lock); |
482 | ||
483 | return ret; | |
484 | } | |
485 | ||
486 | static void tce_iommu_detach_group(void *iommu_data, | |
487 | struct iommu_group *iommu_group) | |
488 | { | |
489 | struct tce_container *container = iommu_data; | |
490 | struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group); | |
491 | ||
492 | BUG_ON(!tbl); | |
493 | mutex_lock(&container->lock); | |
494 | if (tbl != container->tbl) { | |
495 | pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n", | |
496 | iommu_group_id(iommu_group), | |
b348aa65 | 497 | iommu_group_id(tbl->it_table_group->group)); |
22af4859 AK |
498 | goto unlock_exit; |
499 | } | |
5ffd229c | 500 | |
22af4859 AK |
501 | if (container->enabled) { |
502 | pr_warn("tce_vfio: detaching group #%u from enabled container, forcing disable\n", | |
b348aa65 | 503 | iommu_group_id(tbl->it_table_group->group)); |
22af4859 | 504 | tce_iommu_disable(container); |
5ffd229c | 505 | } |
22af4859 AK |
506 | |
507 | /* pr_debug("tce_vfio: detaching group #%u from iommu %p\n", | |
508 | iommu_group_id(iommu_group), iommu_group); */ | |
509 | container->tbl = NULL; | |
510 | tce_iommu_clear(container, tbl, tbl->it_offset, tbl->it_size); | |
511 | iommu_release_ownership(tbl); | |
512 | ||
513 | unlock_exit: | |
5ffd229c AK |
514 | mutex_unlock(&container->lock); |
515 | } | |
516 | ||
517 | const struct vfio_iommu_driver_ops tce_iommu_driver_ops = { | |
518 | .name = "iommu-vfio-powerpc", | |
519 | .owner = THIS_MODULE, | |
520 | .open = tce_iommu_open, | |
521 | .release = tce_iommu_release, | |
522 | .ioctl = tce_iommu_ioctl, | |
523 | .attach_group = tce_iommu_attach_group, | |
524 | .detach_group = tce_iommu_detach_group, | |
525 | }; | |
526 | ||
527 | static int __init tce_iommu_init(void) | |
528 | { | |
529 | return vfio_register_iommu_driver(&tce_iommu_driver_ops); | |
530 | } | |
531 | ||
532 | static void __exit tce_iommu_cleanup(void) | |
533 | { | |
534 | vfio_unregister_iommu_driver(&tce_iommu_driver_ops); | |
535 | } | |
536 | ||
537 | module_init(tce_iommu_init); | |
538 | module_exit(tce_iommu_cleanup); | |
539 | ||
540 | MODULE_VERSION(DRIVER_VERSION); | |
541 | MODULE_LICENSE("GPL v2"); | |
542 | MODULE_AUTHOR(DRIVER_AUTHOR); | |
543 | MODULE_DESCRIPTION(DRIVER_DESC); | |
544 |