Commit | Line | Data |
---|---|---|
ba4e7d97 TH |
1 | /************************************************************************** |
2 | * | |
3 | * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA | |
4 | * All Rights Reserved. | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a | |
7 | * copy of this software and associated documentation files (the | |
8 | * "Software"), to deal in the Software without restriction, including | |
9 | * without limitation the rights to use, copy, modify, merge, publish, | |
10 | * distribute, sub license, and/or sell copies of the Software, and to | |
11 | * permit persons to whom the Software is furnished to do so, subject to | |
12 | * the following conditions: | |
13 | * | |
14 | * The above copyright notice and this permission notice (including the | |
15 | * next paragraph) shall be included in all copies or substantial portions | |
16 | * of the Software. | |
17 | * | |
18 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
19 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
20 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL | |
21 | * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, | |
22 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR | |
23 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE | |
24 | * USE OR OTHER DEALINGS IN THE SOFTWARE. | |
25 | * | |
26 | **************************************************************************/ | |
27 | /* | |
28 | * Authors: Thomas Hellstrom <thellstrom-at-vmware-dot-com> | |
29 | */ | |
30 | ||
31 | #include <ttm/ttm_module.h> | |
32 | #include <ttm/ttm_bo_driver.h> | |
33 | #include <ttm/ttm_placement.h> | |
34 | #include <linux/mm.h> | |
ba4e7d97 TH |
35 | #include <linux/rbtree.h> |
36 | #include <linux/module.h> | |
37 | #include <linux/uaccess.h> | |
38 | ||
39 | #define TTM_BO_VM_NUM_PREFAULT 16 | |
40 | ||
41 | static struct ttm_buffer_object *ttm_bo_vm_lookup_rb(struct ttm_bo_device *bdev, | |
42 | unsigned long page_start, | |
43 | unsigned long num_pages) | |
44 | { | |
45 | struct rb_node *cur = bdev->addr_space_rb.rb_node; | |
46 | unsigned long cur_offset; | |
47 | struct ttm_buffer_object *bo; | |
48 | struct ttm_buffer_object *best_bo = NULL; | |
49 | ||
50 | while (likely(cur != NULL)) { | |
51 | bo = rb_entry(cur, struct ttm_buffer_object, vm_rb); | |
52 | cur_offset = bo->vm_node->start; | |
53 | if (page_start >= cur_offset) { | |
54 | cur = cur->rb_right; | |
55 | best_bo = bo; | |
56 | if (page_start == cur_offset) | |
57 | break; | |
58 | } else | |
59 | cur = cur->rb_left; | |
60 | } | |
61 | ||
62 | if (unlikely(best_bo == NULL)) | |
63 | return NULL; | |
64 | ||
65 | if (unlikely((best_bo->vm_node->start + best_bo->num_pages) < | |
66 | (page_start + num_pages))) | |
67 | return NULL; | |
68 | ||
69 | return best_bo; | |
70 | } | |
71 | ||
72 | static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
73 | { | |
74 | struct ttm_buffer_object *bo = (struct ttm_buffer_object *) | |
75 | vma->vm_private_data; | |
76 | struct ttm_bo_device *bdev = bo->bdev; | |
77 | unsigned long bus_base; | |
78 | unsigned long bus_offset; | |
79 | unsigned long bus_size; | |
80 | unsigned long page_offset; | |
81 | unsigned long page_last; | |
82 | unsigned long pfn; | |
83 | struct ttm_tt *ttm = NULL; | |
84 | struct page *page; | |
85 | int ret; | |
86 | int i; | |
87 | bool is_iomem; | |
88 | unsigned long address = (unsigned long)vmf->virtual_address; | |
89 | int retval = VM_FAULT_NOPAGE; | |
90 | ||
91 | /* | |
92 | * Work around locking order reversal in fault / nopfn | |
93 | * between mmap_sem and bo_reserve: Perform a trylock operation | |
94 | * for reserve, and if it fails, retry the fault after scheduling. | |
95 | */ | |
96 | ||
97 | ret = ttm_bo_reserve(bo, true, true, false, 0); | |
98 | if (unlikely(ret != 0)) { | |
99 | if (ret == -EBUSY) | |
100 | set_need_resched(); | |
101 | return VM_FAULT_NOPAGE; | |
102 | } | |
103 | ||
104 | /* | |
105 | * Wait for buffer data in transit, due to a pipelined | |
106 | * move. | |
107 | */ | |
108 | ||
109 | spin_lock(&bo->lock); | |
110 | if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) { | |
111 | ret = ttm_bo_wait(bo, false, true, false); | |
112 | spin_unlock(&bo->lock); | |
113 | if (unlikely(ret != 0)) { | |
114 | retval = (ret != -ERESTART) ? | |
115 | VM_FAULT_SIGBUS : VM_FAULT_NOPAGE; | |
116 | goto out_unlock; | |
117 | } | |
118 | } else | |
119 | spin_unlock(&bo->lock); | |
120 | ||
121 | ||
122 | ret = ttm_bo_pci_offset(bdev, &bo->mem, &bus_base, &bus_offset, | |
123 | &bus_size); | |
124 | if (unlikely(ret != 0)) { | |
125 | retval = VM_FAULT_SIGBUS; | |
126 | goto out_unlock; | |
127 | } | |
128 | ||
129 | is_iomem = (bus_size != 0); | |
130 | ||
131 | page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) + | |
132 | bo->vm_node->start - vma->vm_pgoff; | |
133 | page_last = ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) + | |
134 | bo->vm_node->start - vma->vm_pgoff; | |
135 | ||
136 | if (unlikely(page_offset >= bo->num_pages)) { | |
137 | retval = VM_FAULT_SIGBUS; | |
138 | goto out_unlock; | |
139 | } | |
140 | ||
141 | /* | |
142 | * Strictly, we're not allowed to modify vma->vm_page_prot here, | |
143 | * since the mmap_sem is only held in read mode. However, we | |
144 | * modify only the caching bits of vma->vm_page_prot and | |
145 | * consider those bits protected by | |
146 | * the bo->mutex, as we should be the only writers. | |
147 | * There shouldn't really be any readers of these bits except | |
148 | * within vm_insert_mixed()? fork? | |
149 | * | |
150 | * TODO: Add a list of vmas to the bo, and change the | |
151 | * vma->vm_page_prot when the object changes caching policy, with | |
152 | * the correct locks held. | |
153 | */ | |
154 | ||
155 | if (is_iomem) { | |
156 | vma->vm_page_prot = ttm_io_prot(bo->mem.placement, | |
157 | vma->vm_page_prot); | |
158 | } else { | |
159 | ttm = bo->ttm; | |
160 | vma->vm_page_prot = (bo->mem.placement & TTM_PL_FLAG_CACHED) ? | |
161 | vm_get_page_prot(vma->vm_flags) : | |
162 | ttm_io_prot(bo->mem.placement, vma->vm_page_prot); | |
163 | } | |
164 | ||
165 | /* | |
166 | * Speculatively prefault a number of pages. Only error on | |
167 | * first page. | |
168 | */ | |
169 | ||
170 | for (i = 0; i < TTM_BO_VM_NUM_PREFAULT; ++i) { | |
171 | ||
172 | if (is_iomem) | |
173 | pfn = ((bus_base + bus_offset) >> PAGE_SHIFT) + | |
174 | page_offset; | |
175 | else { | |
176 | page = ttm_tt_get_page(ttm, page_offset); | |
177 | if (unlikely(!page && i == 0)) { | |
178 | retval = VM_FAULT_OOM; | |
179 | goto out_unlock; | |
180 | } else if (unlikely(!page)) { | |
181 | break; | |
182 | } | |
183 | pfn = page_to_pfn(page); | |
184 | } | |
185 | ||
186 | ret = vm_insert_mixed(vma, address, pfn); | |
187 | /* | |
188 | * Somebody beat us to this PTE or prefaulting to | |
189 | * an already populated PTE, or prefaulting error. | |
190 | */ | |
191 | ||
192 | if (unlikely((ret == -EBUSY) || (ret != 0 && i > 0))) | |
193 | break; | |
194 | else if (unlikely(ret != 0)) { | |
195 | retval = | |
196 | (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; | |
197 | goto out_unlock; | |
198 | ||
199 | } | |
200 | ||
201 | address += PAGE_SIZE; | |
202 | if (unlikely(++page_offset >= page_last)) | |
203 | break; | |
204 | } | |
205 | ||
206 | out_unlock: | |
207 | ttm_bo_unreserve(bo); | |
208 | return retval; | |
209 | } | |
210 | ||
211 | static void ttm_bo_vm_open(struct vm_area_struct *vma) | |
212 | { | |
213 | struct ttm_buffer_object *bo = | |
214 | (struct ttm_buffer_object *)vma->vm_private_data; | |
215 | ||
216 | (void)ttm_bo_reference(bo); | |
217 | } | |
218 | ||
219 | static void ttm_bo_vm_close(struct vm_area_struct *vma) | |
220 | { | |
221 | struct ttm_buffer_object *bo = | |
222 | (struct ttm_buffer_object *)vma->vm_private_data; | |
223 | ||
224 | ttm_bo_unref(&bo); | |
225 | vma->vm_private_data = NULL; | |
226 | } | |
227 | ||
228 | static struct vm_operations_struct ttm_bo_vm_ops = { | |
229 | .fault = ttm_bo_vm_fault, | |
230 | .open = ttm_bo_vm_open, | |
231 | .close = ttm_bo_vm_close | |
232 | }; | |
233 | ||
234 | int ttm_bo_mmap(struct file *filp, struct vm_area_struct *vma, | |
235 | struct ttm_bo_device *bdev) | |
236 | { | |
237 | struct ttm_bo_driver *driver; | |
238 | struct ttm_buffer_object *bo; | |
239 | int ret; | |
240 | ||
241 | read_lock(&bdev->vm_lock); | |
242 | bo = ttm_bo_vm_lookup_rb(bdev, vma->vm_pgoff, | |
243 | (vma->vm_end - vma->vm_start) >> PAGE_SHIFT); | |
244 | if (likely(bo != NULL)) | |
245 | ttm_bo_reference(bo); | |
246 | read_unlock(&bdev->vm_lock); | |
247 | ||
248 | if (unlikely(bo == NULL)) { | |
249 | printk(KERN_ERR TTM_PFX | |
250 | "Could not find buffer object to map.\n"); | |
251 | return -EINVAL; | |
252 | } | |
253 | ||
254 | driver = bo->bdev->driver; | |
255 | if (unlikely(!driver->verify_access)) { | |
256 | ret = -EPERM; | |
257 | goto out_unref; | |
258 | } | |
259 | ret = driver->verify_access(bo, filp); | |
260 | if (unlikely(ret != 0)) | |
261 | goto out_unref; | |
262 | ||
263 | vma->vm_ops = &ttm_bo_vm_ops; | |
264 | ||
265 | /* | |
266 | * Note: We're transferring the bo reference to | |
267 | * vma->vm_private_data here. | |
268 | */ | |
269 | ||
270 | vma->vm_private_data = bo; | |
271 | vma->vm_flags |= VM_RESERVED | VM_IO | VM_MIXEDMAP | VM_DONTEXPAND; | |
272 | return 0; | |
273 | out_unref: | |
274 | ttm_bo_unref(&bo); | |
275 | return ret; | |
276 | } | |
277 | EXPORT_SYMBOL(ttm_bo_mmap); | |
278 | ||
279 | int ttm_fbdev_mmap(struct vm_area_struct *vma, struct ttm_buffer_object *bo) | |
280 | { | |
281 | if (vma->vm_pgoff != 0) | |
282 | return -EACCES; | |
283 | ||
284 | vma->vm_ops = &ttm_bo_vm_ops; | |
285 | vma->vm_private_data = ttm_bo_reference(bo); | |
286 | vma->vm_flags |= VM_RESERVED | VM_IO | VM_MIXEDMAP | VM_DONTEXPAND; | |
287 | return 0; | |
288 | } | |
289 | EXPORT_SYMBOL(ttm_fbdev_mmap); | |
290 | ||
291 | ||
292 | ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp, | |
293 | const char __user *wbuf, char __user *rbuf, size_t count, | |
294 | loff_t *f_pos, bool write) | |
295 | { | |
296 | struct ttm_buffer_object *bo; | |
297 | struct ttm_bo_driver *driver; | |
298 | struct ttm_bo_kmap_obj map; | |
299 | unsigned long dev_offset = (*f_pos >> PAGE_SHIFT); | |
300 | unsigned long kmap_offset; | |
301 | unsigned long kmap_end; | |
302 | unsigned long kmap_num; | |
303 | size_t io_size; | |
304 | unsigned int page_offset; | |
305 | char *virtual; | |
306 | int ret; | |
307 | bool no_wait = false; | |
308 | bool dummy; | |
309 | ||
310 | read_lock(&bdev->vm_lock); | |
311 | bo = ttm_bo_vm_lookup_rb(bdev, dev_offset, 1); | |
312 | if (likely(bo != NULL)) | |
313 | ttm_bo_reference(bo); | |
314 | read_unlock(&bdev->vm_lock); | |
315 | ||
316 | if (unlikely(bo == NULL)) | |
317 | return -EFAULT; | |
318 | ||
319 | driver = bo->bdev->driver; | |
320 | if (unlikely(driver->verify_access)) { | |
321 | ret = -EPERM; | |
322 | goto out_unref; | |
323 | } | |
324 | ||
325 | ret = driver->verify_access(bo, filp); | |
326 | if (unlikely(ret != 0)) | |
327 | goto out_unref; | |
328 | ||
329 | kmap_offset = dev_offset - bo->vm_node->start; | |
330 | if (unlikely(kmap_offset) >= bo->num_pages) { | |
331 | ret = -EFBIG; | |
332 | goto out_unref; | |
333 | } | |
334 | ||
335 | page_offset = *f_pos & ~PAGE_MASK; | |
336 | io_size = bo->num_pages - kmap_offset; | |
337 | io_size = (io_size << PAGE_SHIFT) - page_offset; | |
338 | if (count < io_size) | |
339 | io_size = count; | |
340 | ||
341 | kmap_end = (*f_pos + count - 1) >> PAGE_SHIFT; | |
342 | kmap_num = kmap_end - kmap_offset + 1; | |
343 | ||
344 | ret = ttm_bo_reserve(bo, true, no_wait, false, 0); | |
345 | ||
346 | switch (ret) { | |
347 | case 0: | |
348 | break; | |
349 | case -ERESTART: | |
350 | ret = -EINTR; | |
351 | goto out_unref; | |
352 | case -EBUSY: | |
353 | ret = -EAGAIN; | |
354 | goto out_unref; | |
355 | default: | |
356 | goto out_unref; | |
357 | } | |
358 | ||
359 | ret = ttm_bo_kmap(bo, kmap_offset, kmap_num, &map); | |
360 | if (unlikely(ret != 0)) { | |
361 | ttm_bo_unreserve(bo); | |
362 | goto out_unref; | |
363 | } | |
364 | ||
365 | virtual = ttm_kmap_obj_virtual(&map, &dummy); | |
366 | virtual += page_offset; | |
367 | ||
368 | if (write) | |
369 | ret = copy_from_user(virtual, wbuf, io_size); | |
370 | else | |
371 | ret = copy_to_user(rbuf, virtual, io_size); | |
372 | ||
373 | ttm_bo_kunmap(&map); | |
374 | ttm_bo_unreserve(bo); | |
375 | ttm_bo_unref(&bo); | |
376 | ||
377 | if (unlikely(ret != 0)) | |
378 | return -EFBIG; | |
379 | ||
380 | *f_pos += io_size; | |
381 | ||
382 | return io_size; | |
383 | out_unref: | |
384 | ttm_bo_unref(&bo); | |
385 | return ret; | |
386 | } | |
387 | ||
388 | ssize_t ttm_bo_fbdev_io(struct ttm_buffer_object *bo, const char __user *wbuf, | |
389 | char __user *rbuf, size_t count, loff_t *f_pos, | |
390 | bool write) | |
391 | { | |
392 | struct ttm_bo_kmap_obj map; | |
393 | unsigned long kmap_offset; | |
394 | unsigned long kmap_end; | |
395 | unsigned long kmap_num; | |
396 | size_t io_size; | |
397 | unsigned int page_offset; | |
398 | char *virtual; | |
399 | int ret; | |
400 | bool no_wait = false; | |
401 | bool dummy; | |
402 | ||
403 | kmap_offset = (*f_pos >> PAGE_SHIFT); | |
404 | if (unlikely(kmap_offset) >= bo->num_pages) | |
405 | return -EFBIG; | |
406 | ||
407 | page_offset = *f_pos & ~PAGE_MASK; | |
408 | io_size = bo->num_pages - kmap_offset; | |
409 | io_size = (io_size << PAGE_SHIFT) - page_offset; | |
410 | if (count < io_size) | |
411 | io_size = count; | |
412 | ||
413 | kmap_end = (*f_pos + count - 1) >> PAGE_SHIFT; | |
414 | kmap_num = kmap_end - kmap_offset + 1; | |
415 | ||
416 | ret = ttm_bo_reserve(bo, true, no_wait, false, 0); | |
417 | ||
418 | switch (ret) { | |
419 | case 0: | |
420 | break; | |
421 | case -ERESTART: | |
422 | return -EINTR; | |
423 | case -EBUSY: | |
424 | return -EAGAIN; | |
425 | default: | |
426 | return ret; | |
427 | } | |
428 | ||
429 | ret = ttm_bo_kmap(bo, kmap_offset, kmap_num, &map); | |
430 | if (unlikely(ret != 0)) { | |
431 | ttm_bo_unreserve(bo); | |
432 | return ret; | |
433 | } | |
434 | ||
435 | virtual = ttm_kmap_obj_virtual(&map, &dummy); | |
436 | virtual += page_offset; | |
437 | ||
438 | if (write) | |
439 | ret = copy_from_user(virtual, wbuf, io_size); | |
440 | else | |
441 | ret = copy_to_user(rbuf, virtual, io_size); | |
442 | ||
443 | ttm_bo_kunmap(&map); | |
444 | ttm_bo_unreserve(bo); | |
445 | ttm_bo_unref(&bo); | |
446 | ||
447 | if (unlikely(ret != 0)) | |
448 | return ret; | |
449 | ||
450 | *f_pos += io_size; | |
451 | ||
452 | return io_size; | |
453 | } |