Commit | Line | Data |
---|---|---|
a11c3198 BS |
1 | /* |
2 | * Copyright 2010 Red Hat Inc. | |
3 | * | |
4 | * Permission is hereby granted, free of charge, to any person obtaining a | |
5 | * copy of this software and associated documentation files (the "Software"), | |
6 | * to deal in the Software without restriction, including without limitation | |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
8 | * and/or sell copies of the Software, and to permit persons to whom the | |
9 | * Software is furnished to do so, subject to the following conditions: | |
10 | * | |
11 | * The above copyright notice and this permission notice shall be included in | |
12 | * all copies or substantial portions of the Software. | |
13 | * | |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
20 | * OTHER DEALINGS IN THE SOFTWARE. | |
21 | * | |
22 | * Authors: Ben Skeggs | |
23 | */ | |
24 | ||
3863c9bc | 25 | #include <core/gpuobj.h> |
02a841d4 | 26 | #include <core/mm.h> |
3863c9bc BS |
27 | |
28 | #include <subdev/fb.h> | |
02a841d4 | 29 | #include <subdev/vm.h> |
a11c3198 BS |
30 | |
31 | void | |
d5f42394 | 32 | nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node) |
a11c3198 BS |
33 | { |
34 | struct nouveau_vm *vm = vma->vm; | |
3863c9bc | 35 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 | 36 | struct nouveau_mm_node *r; |
3863c9bc | 37 | int big = vma->node->type != vmm->spg_shift; |
a11c3198 BS |
38 | u32 offset = vma->node->offset + (delta >> 12); |
39 | u32 bits = vma->node->type - 12; | |
3863c9bc BS |
40 | u32 pde = (offset >> vmm->pgt_bits) - vm->fpde; |
41 | u32 pte = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits; | |
42 | u32 max = 1 << (vmm->pgt_bits - bits); | |
a11c3198 BS |
43 | u32 end, len; |
44 | ||
8f7286f8 | 45 | delta = 0; |
d5f42394 | 46 | list_for_each_entry(r, &node->regions, rl_entry) { |
a11c3198 BS |
47 | u64 phys = (u64)r->offset << 12; |
48 | u32 num = r->length >> bits; | |
49 | ||
50 | while (num) { | |
3ee01281 | 51 | struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big]; |
a11c3198 BS |
52 | |
53 | end = (pte + num); | |
54 | if (unlikely(end >= max)) | |
55 | end = max; | |
56 | len = end - pte; | |
57 | ||
3863c9bc | 58 | vmm->map(vma, pgt, node, pte, len, phys, delta); |
a11c3198 BS |
59 | |
60 | num -= len; | |
61 | pte += len; | |
62 | if (unlikely(end >= max)) { | |
73c337e7 | 63 | phys += len << (bits + 12); |
a11c3198 BS |
64 | pde++; |
65 | pte = 0; | |
66 | } | |
8f7286f8 BS |
67 | |
68 | delta += (u64)len << vma->node->type; | |
a11c3198 BS |
69 | } |
70 | } | |
71 | ||
3863c9bc | 72 | vmm->flush(vm); |
a11c3198 BS |
73 | } |
74 | ||
75 | void | |
d5f42394 | 76 | nouveau_vm_map(struct nouveau_vma *vma, struct nouveau_mem *node) |
a11c3198 | 77 | { |
d5f42394 | 78 | nouveau_vm_map_at(vma, 0, node); |
a11c3198 BS |
79 | } |
80 | ||
22b33e8e DA |
81 | void |
82 | nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length, | |
83 | struct nouveau_mem *mem) | |
84 | { | |
85 | struct nouveau_vm *vm = vma->vm; | |
3863c9bc BS |
86 | struct nouveau_vmmgr *vmm = vm->vmm; |
87 | int big = vma->node->type != vmm->spg_shift; | |
22b33e8e DA |
88 | u32 offset = vma->node->offset + (delta >> 12); |
89 | u32 bits = vma->node->type - 12; | |
90 | u32 num = length >> vma->node->type; | |
3863c9bc BS |
91 | u32 pde = (offset >> vmm->pgt_bits) - vm->fpde; |
92 | u32 pte = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits; | |
93 | u32 max = 1 << (vmm->pgt_bits - bits); | |
22b33e8e DA |
94 | unsigned m, sglen; |
95 | u32 end, len; | |
96 | int i; | |
97 | struct scatterlist *sg; | |
98 | ||
99 | for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) { | |
100 | struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big]; | |
101 | sglen = sg_dma_len(sg) >> PAGE_SHIFT; | |
102 | ||
103 | end = pte + sglen; | |
104 | if (unlikely(end >= max)) | |
105 | end = max; | |
106 | len = end - pte; | |
107 | ||
108 | for (m = 0; m < len; m++) { | |
109 | dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT); | |
110 | ||
3863c9bc | 111 | vmm->map_sg(vma, pgt, mem, pte, 1, &addr); |
22b33e8e DA |
112 | num--; |
113 | pte++; | |
114 | ||
115 | if (num == 0) | |
116 | goto finish; | |
117 | } | |
118 | if (unlikely(end >= max)) { | |
119 | pde++; | |
120 | pte = 0; | |
121 | } | |
122 | if (m < sglen) { | |
123 | for (; m < sglen; m++) { | |
124 | dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT); | |
125 | ||
3863c9bc | 126 | vmm->map_sg(vma, pgt, mem, pte, 1, &addr); |
22b33e8e DA |
127 | num--; |
128 | pte++; | |
129 | if (num == 0) | |
130 | goto finish; | |
131 | } | |
132 | } | |
133 | ||
134 | } | |
135 | finish: | |
3863c9bc | 136 | vmm->flush(vm); |
22b33e8e DA |
137 | } |
138 | ||
a11c3198 BS |
139 | void |
140 | nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length, | |
f7b24c42 | 141 | struct nouveau_mem *mem) |
a11c3198 BS |
142 | { |
143 | struct nouveau_vm *vm = vma->vm; | |
3863c9bc | 144 | struct nouveau_vmmgr *vmm = vm->vmm; |
f7b24c42 | 145 | dma_addr_t *list = mem->pages; |
3863c9bc | 146 | int big = vma->node->type != vmm->spg_shift; |
a11c3198 BS |
147 | u32 offset = vma->node->offset + (delta >> 12); |
148 | u32 bits = vma->node->type - 12; | |
149 | u32 num = length >> vma->node->type; | |
3863c9bc BS |
150 | u32 pde = (offset >> vmm->pgt_bits) - vm->fpde; |
151 | u32 pte = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits; | |
152 | u32 max = 1 << (vmm->pgt_bits - bits); | |
a11c3198 BS |
153 | u32 end, len; |
154 | ||
155 | while (num) { | |
3ee01281 | 156 | struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big]; |
a11c3198 BS |
157 | |
158 | end = (pte + num); | |
159 | if (unlikely(end >= max)) | |
160 | end = max; | |
161 | len = end - pte; | |
162 | ||
3863c9bc | 163 | vmm->map_sg(vma, pgt, mem, pte, len, list); |
a11c3198 BS |
164 | |
165 | num -= len; | |
166 | pte += len; | |
167 | list += len; | |
168 | if (unlikely(end >= max)) { | |
169 | pde++; | |
170 | pte = 0; | |
171 | } | |
172 | } | |
173 | ||
3863c9bc | 174 | vmm->flush(vm); |
a11c3198 BS |
175 | } |
176 | ||
177 | void | |
178 | nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length) | |
179 | { | |
180 | struct nouveau_vm *vm = vma->vm; | |
3863c9bc BS |
181 | struct nouveau_vmmgr *vmm = vm->vmm; |
182 | int big = vma->node->type != vmm->spg_shift; | |
a11c3198 BS |
183 | u32 offset = vma->node->offset + (delta >> 12); |
184 | u32 bits = vma->node->type - 12; | |
185 | u32 num = length >> vma->node->type; | |
3863c9bc BS |
186 | u32 pde = (offset >> vmm->pgt_bits) - vm->fpde; |
187 | u32 pte = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits; | |
188 | u32 max = 1 << (vmm->pgt_bits - bits); | |
a11c3198 BS |
189 | u32 end, len; |
190 | ||
191 | while (num) { | |
3ee01281 | 192 | struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big]; |
a11c3198 BS |
193 | |
194 | end = (pte + num); | |
195 | if (unlikely(end >= max)) | |
196 | end = max; | |
197 | len = end - pte; | |
198 | ||
3863c9bc | 199 | vmm->unmap(pgt, pte, len); |
a11c3198 BS |
200 | |
201 | num -= len; | |
202 | pte += len; | |
203 | if (unlikely(end >= max)) { | |
204 | pde++; | |
205 | pte = 0; | |
206 | } | |
207 | } | |
208 | ||
3863c9bc | 209 | vmm->flush(vm); |
a11c3198 BS |
210 | } |
211 | ||
212 | void | |
213 | nouveau_vm_unmap(struct nouveau_vma *vma) | |
214 | { | |
215 | nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12); | |
216 | } | |
217 | ||
218 | static void | |
3ee01281 | 219 | nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde) |
a11c3198 | 220 | { |
3863c9bc | 221 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 BS |
222 | struct nouveau_vm_pgd *vpgd; |
223 | struct nouveau_vm_pgt *vpgt; | |
224 | struct nouveau_gpuobj *pgt; | |
225 | u32 pde; | |
226 | ||
227 | for (pde = fpde; pde <= lpde; pde++) { | |
228 | vpgt = &vm->pgt[pde - vm->fpde]; | |
3ee01281 | 229 | if (--vpgt->refcount[big]) |
a11c3198 BS |
230 | continue; |
231 | ||
3ee01281 BS |
232 | pgt = vpgt->obj[big]; |
233 | vpgt->obj[big] = NULL; | |
234 | ||
a11c3198 | 235 | list_for_each_entry(vpgd, &vm->pgd_list, head) { |
3863c9bc | 236 | vmm->map_pgt(vpgd->obj, pde, vpgt->obj); |
a11c3198 BS |
237 | } |
238 | ||
4e67bee8 | 239 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 | 240 | nouveau_gpuobj_ref(NULL, &pgt); |
4e67bee8 | 241 | mutex_lock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
242 | } |
243 | } | |
244 | ||
245 | static int | |
246 | nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type) | |
247 | { | |
3863c9bc | 248 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 BS |
249 | struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde]; |
250 | struct nouveau_vm_pgd *vpgd; | |
251 | struct nouveau_gpuobj *pgt; | |
3863c9bc | 252 | int big = (type != vmm->spg_shift); |
a11c3198 BS |
253 | u32 pgt_size; |
254 | int ret; | |
255 | ||
3863c9bc | 256 | pgt_size = (1 << (vmm->pgt_bits + 12)) >> type; |
a11c3198 BS |
257 | pgt_size *= 8; |
258 | ||
4e67bee8 | 259 | mutex_unlock(&nv_subdev(vmm)->mutex); |
3863c9bc | 260 | ret = nouveau_gpuobj_new(nv_object(vm->vmm), NULL, pgt_size, 0x1000, |
a11c3198 | 261 | NVOBJ_FLAG_ZERO_ALLOC, &pgt); |
4e67bee8 | 262 | mutex_lock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
263 | if (unlikely(ret)) |
264 | return ret; | |
265 | ||
266 | /* someone beat us to filling the PDE while we didn't have the lock */ | |
3ee01281 | 267 | if (unlikely(vpgt->refcount[big]++)) { |
4e67bee8 | 268 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 | 269 | nouveau_gpuobj_ref(NULL, &pgt); |
4e67bee8 | 270 | mutex_lock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
271 | return 0; |
272 | } | |
273 | ||
3ee01281 | 274 | vpgt->obj[big] = pgt; |
a11c3198 | 275 | list_for_each_entry(vpgd, &vm->pgd_list, head) { |
3863c9bc | 276 | vmm->map_pgt(vpgd->obj, pde, vpgt->obj); |
a11c3198 BS |
277 | } |
278 | ||
a11c3198 BS |
279 | return 0; |
280 | } | |
281 | ||
282 | int | |
283 | nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift, | |
284 | u32 access, struct nouveau_vma *vma) | |
285 | { | |
3863c9bc | 286 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 BS |
287 | u32 align = (1 << page_shift) >> 12; |
288 | u32 msize = size >> 12; | |
289 | u32 fpde, lpde, pde; | |
290 | int ret; | |
291 | ||
4e67bee8 | 292 | mutex_lock(&nv_subdev(vmm)->mutex); |
496734bf BS |
293 | ret = nouveau_mm_head(&vm->mm, page_shift, msize, msize, align, |
294 | &vma->node); | |
a11c3198 | 295 | if (unlikely(ret != 0)) { |
4e67bee8 | 296 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
297 | return ret; |
298 | } | |
299 | ||
3863c9bc BS |
300 | fpde = (vma->node->offset >> vmm->pgt_bits); |
301 | lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits; | |
ebb945a9 | 302 | |
a11c3198 BS |
303 | for (pde = fpde; pde <= lpde; pde++) { |
304 | struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde]; | |
3863c9bc | 305 | int big = (vma->node->type != vmm->spg_shift); |
a11c3198 | 306 | |
3ee01281 BS |
307 | if (likely(vpgt->refcount[big])) { |
308 | vpgt->refcount[big]++; | |
a11c3198 BS |
309 | continue; |
310 | } | |
311 | ||
312 | ret = nouveau_vm_map_pgt(vm, pde, vma->node->type); | |
313 | if (ret) { | |
314 | if (pde != fpde) | |
3ee01281 | 315 | nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1); |
496734bf | 316 | nouveau_mm_free(&vm->mm, &vma->node); |
4e67bee8 | 317 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
318 | return ret; |
319 | } | |
320 | } | |
4e67bee8 | 321 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
322 | |
323 | vma->vm = vm; | |
324 | vma->offset = (u64)vma->node->offset << 12; | |
325 | vma->access = access; | |
326 | return 0; | |
327 | } | |
328 | ||
329 | void | |
330 | nouveau_vm_put(struct nouveau_vma *vma) | |
331 | { | |
332 | struct nouveau_vm *vm = vma->vm; | |
3863c9bc | 333 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 BS |
334 | u32 fpde, lpde; |
335 | ||
336 | if (unlikely(vma->node == NULL)) | |
337 | return; | |
3863c9bc BS |
338 | fpde = (vma->node->offset >> vmm->pgt_bits); |
339 | lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits; | |
a11c3198 | 340 | |
4e67bee8 | 341 | mutex_lock(&nv_subdev(vmm)->mutex); |
3863c9bc | 342 | nouveau_vm_unmap_pgt(vm, vma->node->type != vmm->spg_shift, fpde, lpde); |
496734bf | 343 | nouveau_mm_free(&vm->mm, &vma->node); |
4e67bee8 | 344 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
345 | } |
346 | ||
347 | int | |
3863c9bc BS |
348 | nouveau_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length, |
349 | u64 mm_offset, u32 block, struct nouveau_vm **pvm) | |
a11c3198 | 350 | { |
a11c3198 BS |
351 | struct nouveau_vm *vm; |
352 | u64 mm_length = (offset + length) - mm_offset; | |
a11c3198 BS |
353 | int ret; |
354 | ||
cfd376b6 | 355 | vm = kzalloc(sizeof(*vm), GFP_KERNEL); |
a11c3198 BS |
356 | if (!vm) |
357 | return -ENOMEM; | |
358 | ||
3863c9bc BS |
359 | INIT_LIST_HEAD(&vm->pgd_list); |
360 | vm->vmm = vmm; | |
361 | vm->refcount = 1; | |
362 | vm->fpde = offset >> (vmm->pgt_bits + 12); | |
363 | vm->lpde = (offset + length - 1) >> (vmm->pgt_bits + 12); | |
a11c3198 | 364 | |
3863c9bc | 365 | vm->pgt = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt), GFP_KERNEL); |
a11c3198 BS |
366 | if (!vm->pgt) { |
367 | kfree(vm); | |
368 | return -ENOMEM; | |
369 | } | |
370 | ||
a11c3198 BS |
371 | ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12, |
372 | block >> 12); | |
373 | if (ret) { | |
3863c9bc | 374 | kfree(vm->pgt); |
a11c3198 BS |
375 | kfree(vm); |
376 | return ret; | |
377 | } | |
378 | ||
cfd376b6 MS |
379 | *pvm = vm; |
380 | ||
a11c3198 BS |
381 | return 0; |
382 | } | |
383 | ||
3863c9bc BS |
384 | int |
385 | nouveau_vm_new(struct nouveau_device *device, u64 offset, u64 length, | |
386 | u64 mm_offset, struct nouveau_vm **pvm) | |
387 | { | |
388 | struct nouveau_vmmgr *vmm = nouveau_vmmgr(device); | |
389 | return vmm->create(vmm, offset, length, mm_offset, pvm); | |
390 | } | |
391 | ||
a11c3198 BS |
392 | static int |
393 | nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd) | |
394 | { | |
3863c9bc | 395 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 BS |
396 | struct nouveau_vm_pgd *vpgd; |
397 | int i; | |
398 | ||
399 | if (!pgd) | |
400 | return 0; | |
401 | ||
402 | vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL); | |
403 | if (!vpgd) | |
404 | return -ENOMEM; | |
405 | ||
406 | nouveau_gpuobj_ref(pgd, &vpgd->obj); | |
407 | ||
4e67bee8 | 408 | mutex_lock(&nv_subdev(vmm)->mutex); |
3ee01281 | 409 | for (i = vm->fpde; i <= vm->lpde; i++) |
3863c9bc | 410 | vmm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj); |
a11c3198 | 411 | list_add(&vpgd->head, &vm->pgd_list); |
4e67bee8 | 412 | mutex_unlock(&nv_subdev(vmm)->mutex); |
a11c3198 BS |
413 | return 0; |
414 | } | |
415 | ||
416 | static void | |
15ba79ad | 417 | nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd) |
a11c3198 | 418 | { |
4e67bee8 | 419 | struct nouveau_vmmgr *vmm = vm->vmm; |
a11c3198 | 420 | struct nouveau_vm_pgd *vpgd, *tmp; |
15ba79ad | 421 | struct nouveau_gpuobj *pgd = NULL; |
a11c3198 | 422 | |
15ba79ad | 423 | if (!mpgd) |
a11c3198 BS |
424 | return; |
425 | ||
4e67bee8 | 426 | mutex_lock(&nv_subdev(vmm)->mutex); |
a11c3198 | 427 | list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) { |
15ba79ad BS |
428 | if (vpgd->obj == mpgd) { |
429 | pgd = vpgd->obj; | |
430 | list_del(&vpgd->head); | |
431 | kfree(vpgd); | |
432 | break; | |
433 | } | |
a11c3198 | 434 | } |
4e67bee8 | 435 | mutex_unlock(&nv_subdev(vmm)->mutex); |
15ba79ad BS |
436 | |
437 | nouveau_gpuobj_ref(NULL, &pgd); | |
a11c3198 BS |
438 | } |
439 | ||
440 | static void | |
441 | nouveau_vm_del(struct nouveau_vm *vm) | |
442 | { | |
443 | struct nouveau_vm_pgd *vpgd, *tmp; | |
444 | ||
445 | list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) { | |
446 | nouveau_vm_unlink(vm, vpgd->obj); | |
447 | } | |
a11c3198 | 448 | |
ad9ac437 | 449 | nouveau_mm_fini(&vm->mm); |
a11c3198 BS |
450 | kfree(vm->pgt); |
451 | kfree(vm); | |
452 | } | |
453 | ||
454 | int | |
455 | nouveau_vm_ref(struct nouveau_vm *ref, struct nouveau_vm **ptr, | |
456 | struct nouveau_gpuobj *pgd) | |
457 | { | |
458 | struct nouveau_vm *vm; | |
459 | int ret; | |
460 | ||
461 | vm = ref; | |
462 | if (vm) { | |
463 | ret = nouveau_vm_link(vm, pgd); | |
464 | if (ret) | |
465 | return ret; | |
466 | ||
467 | vm->refcount++; | |
468 | } | |
469 | ||
470 | vm = *ptr; | |
471 | *ptr = ref; | |
472 | ||
473 | if (vm) { | |
474 | nouveau_vm_unlink(vm, pgd); | |
475 | ||
476 | if (--vm->refcount == 0) | |
477 | nouveau_vm_del(vm); | |
478 | } | |
479 | ||
480 | return 0; | |
481 | } |