2 * Copyright 2010 Red Hat Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
26 #include "nouveau_drv.h"
27 #include "nouveau_mm.h"
28 #include "nouveau_vm.h"
31 nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node)
33 struct nouveau_vm *vm = vma->vm;
34 struct nouveau_mm_node *r;
35 int big = vma->node->type != vm->spg_shift;
36 u32 offset = vma->node->offset + (delta >> 12);
37 u32 bits = vma->node->type - 12;
38 u32 pde = (offset >> vm->pgt_bits) - vm->fpde;
39 u32 pte = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
40 u32 max = 1 << (vm->pgt_bits - bits);
44 list_for_each_entry(r, &node->regions, rl_entry) {
45 u64 phys = (u64)r->offset << 12;
46 u32 num = r->length >> bits;
49 struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
52 if (unlikely(end >= max))
56 vm->map(vma, pgt, node, pte, len, phys, delta);
60 if (unlikely(end >= max)) {
61 phys += len << (bits + 12);
66 delta += (u64)len << vma->node->type;
74 nouveau_vm_map(struct nouveau_vma *vma, struct nouveau_mem *node)
76 nouveau_vm_map_at(vma, 0, node);
80 nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length,
81 struct nouveau_mem *mem)
83 struct nouveau_vm *vm = vma->vm;
84 int big = vma->node->type != vm->spg_shift;
85 u32 offset = vma->node->offset + (delta >> 12);
86 u32 bits = vma->node->type - 12;
87 u32 num = length >> vma->node->type;
88 u32 pde = (offset >> vm->pgt_bits) - vm->fpde;
89 u32 pte = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
90 u32 max = 1 << (vm->pgt_bits - bits);
94 struct scatterlist *sg;
96 for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) {
97 struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
98 sglen = sg_dma_len(sg) >> PAGE_SHIFT;
101 if (unlikely(end >= max))
105 for (m = 0; m < len; m++) {
106 dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
108 vm->map_sg(vma, pgt, mem, pte, 1, &addr);
115 if (unlikely(end >= max)) {
120 for (; m < sglen; m++) {
121 dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
123 vm->map_sg(vma, pgt, mem, pte, 1, &addr);
137 nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
138 struct nouveau_mem *mem)
140 struct nouveau_vm *vm = vma->vm;
141 dma_addr_t *list = mem->pages;
142 int big = vma->node->type != vm->spg_shift;
143 u32 offset = vma->node->offset + (delta >> 12);
144 u32 bits = vma->node->type - 12;
145 u32 num = length >> vma->node->type;
146 u32 pde = (offset >> vm->pgt_bits) - vm->fpde;
147 u32 pte = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
148 u32 max = 1 << (vm->pgt_bits - bits);
152 struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
155 if (unlikely(end >= max))
159 vm->map_sg(vma, pgt, mem, pte, len, list);
164 if (unlikely(end >= max)) {
174 nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
176 struct nouveau_vm *vm = vma->vm;
177 int big = vma->node->type != vm->spg_shift;
178 u32 offset = vma->node->offset + (delta >> 12);
179 u32 bits = vma->node->type - 12;
180 u32 num = length >> vma->node->type;
181 u32 pde = (offset >> vm->pgt_bits) - vm->fpde;
182 u32 pte = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
183 u32 max = 1 << (vm->pgt_bits - bits);
187 struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
190 if (unlikely(end >= max))
194 vm->unmap(pgt, pte, len);
198 if (unlikely(end >= max)) {
208 nouveau_vm_unmap(struct nouveau_vma *vma)
210 nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
214 nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
216 struct nouveau_vm_pgd *vpgd;
217 struct nouveau_vm_pgt *vpgt;
218 struct nouveau_gpuobj *pgt;
221 for (pde = fpde; pde <= lpde; pde++) {
222 vpgt = &vm->pgt[pde - vm->fpde];
223 if (--vpgt->refcount[big])
226 pgt = vpgt->obj[big];
227 vpgt->obj[big] = NULL;
229 list_for_each_entry(vpgd, &vm->pgd_list, head) {
230 vm->map_pgt(vpgd->obj, pde, vpgt->obj);
233 mutex_unlock(&vm->mm.mutex);
234 nouveau_gpuobj_ref(NULL, &pgt);
235 mutex_lock(&vm->mm.mutex);
240 nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
242 struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
243 struct nouveau_vm_pgd *vpgd;
244 struct nouveau_gpuobj *pgt;
245 int big = (type != vm->spg_shift);
249 pgt_size = (1 << (vm->pgt_bits + 12)) >> type;
252 mutex_unlock(&vm->mm.mutex);
253 ret = nouveau_gpuobj_new(vm->dev, NULL, pgt_size, 0x1000,
254 NVOBJ_FLAG_ZERO_ALLOC, &pgt);
255 mutex_lock(&vm->mm.mutex);
259 /* someone beat us to filling the PDE while we didn't have the lock */
260 if (unlikely(vpgt->refcount[big]++)) {
261 mutex_unlock(&vm->mm.mutex);
262 nouveau_gpuobj_ref(NULL, &pgt);
263 mutex_lock(&vm->mm.mutex);
267 vpgt->obj[big] = pgt;
268 list_for_each_entry(vpgd, &vm->pgd_list, head) {
269 vm->map_pgt(vpgd->obj, pde, vpgt->obj);
276 nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
277 u32 access, struct nouveau_vma *vma)
279 u32 align = (1 << page_shift) >> 12;
280 u32 msize = size >> 12;
284 mutex_lock(&vm->mm.mutex);
285 ret = nouveau_mm_get(&vm->mm, page_shift, msize, 0, align, &vma->node);
286 if (unlikely(ret != 0)) {
287 mutex_unlock(&vm->mm.mutex);
291 fpde = (vma->node->offset >> vm->pgt_bits);
292 lpde = (vma->node->offset + vma->node->length - 1) >> vm->pgt_bits;
293 for (pde = fpde; pde <= lpde; pde++) {
294 struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
295 int big = (vma->node->type != vm->spg_shift);
297 if (likely(vpgt->refcount[big])) {
298 vpgt->refcount[big]++;
302 ret = nouveau_vm_map_pgt(vm, pde, vma->node->type);
305 nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
306 nouveau_mm_put(&vm->mm, vma->node);
307 mutex_unlock(&vm->mm.mutex);
312 mutex_unlock(&vm->mm.mutex);
315 vma->offset = (u64)vma->node->offset << 12;
316 vma->access = access;
321 nouveau_vm_put(struct nouveau_vma *vma)
323 struct nouveau_vm *vm = vma->vm;
326 if (unlikely(vma->node == NULL))
328 fpde = (vma->node->offset >> vm->pgt_bits);
329 lpde = (vma->node->offset + vma->node->length - 1) >> vm->pgt_bits;
331 mutex_lock(&vm->mm.mutex);
332 nouveau_vm_unmap_pgt(vm, vma->node->type != vm->spg_shift, fpde, lpde);
333 nouveau_mm_put(&vm->mm, vma->node);
335 mutex_unlock(&vm->mm.mutex);
339 nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
340 struct nouveau_vm **pvm)
342 struct drm_nouveau_private *dev_priv = dev->dev_private;
343 struct nouveau_vm *vm;
344 u64 mm_length = (offset + length) - mm_offset;
348 vm = kzalloc(sizeof(*vm), GFP_KERNEL);
352 if (dev_priv->card_type == NV_50) {
353 vm->map_pgt = nv50_vm_map_pgt;
354 vm->map = nv50_vm_map;
355 vm->map_sg = nv50_vm_map_sg;
356 vm->unmap = nv50_vm_unmap;
357 vm->flush = nv50_vm_flush;
362 block = (1 << pgt_bits);
367 if (dev_priv->card_type >= NV_C0) {
368 vm->map_pgt = nvc0_vm_map_pgt;
369 vm->map = nvc0_vm_map;
370 vm->map_sg = nvc0_vm_map_sg;
371 vm->unmap = nvc0_vm_unmap;
372 vm->flush = nvc0_vm_flush;
382 vm->fpde = offset >> pgt_bits;
383 vm->lpde = (offset + length - 1) >> pgt_bits;
384 vm->pgt = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt), GFP_KERNEL);
390 INIT_LIST_HEAD(&vm->pgd_list);
393 vm->pgt_bits = pgt_bits - 12;
395 ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
407 nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
409 struct nouveau_vm_pgd *vpgd;
415 vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
419 nouveau_gpuobj_ref(pgd, &vpgd->obj);
421 mutex_lock(&vm->mm.mutex);
422 for (i = vm->fpde; i <= vm->lpde; i++)
423 vm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
424 list_add(&vpgd->head, &vm->pgd_list);
425 mutex_unlock(&vm->mm.mutex);
430 nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
432 struct nouveau_vm_pgd *vpgd, *tmp;
433 struct nouveau_gpuobj *pgd = NULL;
438 mutex_lock(&vm->mm.mutex);
439 list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
440 if (vpgd->obj == mpgd) {
442 list_del(&vpgd->head);
447 mutex_unlock(&vm->mm.mutex);
449 nouveau_gpuobj_ref(NULL, &pgd);
453 nouveau_vm_del(struct nouveau_vm *vm)
455 struct nouveau_vm_pgd *vpgd, *tmp;
457 list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
458 nouveau_vm_unlink(vm, vpgd->obj);
461 nouveau_mm_fini(&vm->mm);
467 nouveau_vm_ref(struct nouveau_vm *ref, struct nouveau_vm **ptr,
468 struct nouveau_gpuobj *pgd)
470 struct nouveau_vm *vm;
475 ret = nouveau_vm_link(vm, pgd);
486 nouveau_vm_unlink(vm, pgd);
488 if (--vm->refcount == 0)