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#include "priv.h"
25
26#include <core/gpuobj.h>
27#include <subdev/fb.h>
28
29void
30nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
31{
32	struct nvkm_vm *vm = vma->vm;
33	struct nvkm_mmu *mmu = vm->mmu;
34	struct nvkm_mm_node *r;
35	int big = vma->node->type != mmu->func->spg_shift;
36	u32 offset = vma->node->offset + (delta >> 12);
37	u32 bits = vma->node->type - 12;
38	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
39	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
40	u32 max  = 1 << (mmu->func->pgt_bits - bits);
41	u32 end, len;
42
43	delta = 0;
44	list_for_each_entry(r, &node->regions, rl_entry) {
45		u64 phys = (u64)r->offset << 12;
46		u32 num  = r->length >> bits;
47
48		while (num) {
49			struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
50
51			end = (pte + num);
52			if (unlikely(end >= max))
53				end = max;
54			len = end - pte;
55
56			mmu->func->map(vma, pgt, node, pte, len, phys, delta);
57
58			num -= len;
59			pte += len;
60			if (unlikely(end >= max)) {
61				phys += len << (bits + 12);
62				pde++;
63				pte = 0;
64			}
65
66			delta += (u64)len << vma->node->type;
67		}
68	}
69
70	mmu->func->flush(vm);
71}
72
73static void
74nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
75		     struct nvkm_mem *mem)
76{
77	struct nvkm_vm *vm = vma->vm;
78	struct nvkm_mmu *mmu = vm->mmu;
79	int big = vma->node->type != mmu->func->spg_shift;
80	u32 offset = vma->node->offset + (delta >> 12);
81	u32 bits = vma->node->type - 12;
82	u32 num  = length >> vma->node->type;
83	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
84	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
85	u32 max  = 1 << (mmu->func->pgt_bits - bits);
86	unsigned m, sglen;
87	u32 end, len;
88	int i;
89	struct scatterlist *sg;
90
91	for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) {
92		struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
93		sglen = sg_dma_len(sg) >> PAGE_SHIFT;
94
95		end = pte + sglen;
96		if (unlikely(end >= max))
97			end = max;
98		len = end - pte;
99
100		for (m = 0; m < len; m++) {
101			dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
102
103			mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
104			num--;
105			pte++;
106
107			if (num == 0)
108				goto finish;
109		}
110		if (unlikely(end >= max)) {
111			pde++;
112			pte = 0;
113		}
114		if (m < sglen) {
115			for (; m < sglen; m++) {
116				dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
117
118				mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
119				num--;
120				pte++;
121				if (num == 0)
122					goto finish;
123			}
124		}
125
126	}
127finish:
128	mmu->func->flush(vm);
129}
130
131static void
132nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
133	       struct nvkm_mem *mem)
134{
135	struct nvkm_vm *vm = vma->vm;
136	struct nvkm_mmu *mmu = vm->mmu;
137	dma_addr_t *list = mem->pages;
138	int big = vma->node->type != mmu->func->spg_shift;
139	u32 offset = vma->node->offset + (delta >> 12);
140	u32 bits = vma->node->type - 12;
141	u32 num  = length >> vma->node->type;
142	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
143	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
144	u32 max  = 1 << (mmu->func->pgt_bits - bits);
145	u32 end, len;
146
147	while (num) {
148		struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
149
150		end = (pte + num);
151		if (unlikely(end >= max))
152			end = max;
153		len = end - pte;
154
155		mmu->func->map_sg(vma, pgt, mem, pte, len, list);
156
157		num  -= len;
158		pte  += len;
159		list += len;
160		if (unlikely(end >= max)) {
161			pde++;
162			pte = 0;
163		}
164	}
165
166	mmu->func->flush(vm);
167}
168
169void
170nvkm_vm_map(struct nvkm_vma *vma, struct nvkm_mem *node)
171{
172	if (node->sg)
173		nvkm_vm_map_sg_table(vma, 0, node->size << 12, node);
174	else
175	if (node->pages)
176		nvkm_vm_map_sg(vma, 0, node->size << 12, node);
177	else
178		nvkm_vm_map_at(vma, 0, node);
179}
180
181void
182nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
183{
184	struct nvkm_vm *vm = vma->vm;
185	struct nvkm_mmu *mmu = vm->mmu;
186	int big = vma->node->type != mmu->func->spg_shift;
187	u32 offset = vma->node->offset + (delta >> 12);
188	u32 bits = vma->node->type - 12;
189	u32 num  = length >> vma->node->type;
190	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
191	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
192	u32 max  = 1 << (mmu->func->pgt_bits - bits);
193	u32 end, len;
194
195	while (num) {
196		struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
197
198		end = (pte + num);
199		if (unlikely(end >= max))
200			end = max;
201		len = end - pte;
202
203		mmu->func->unmap(vma, pgt, pte, len);
204
205		num -= len;
206		pte += len;
207		if (unlikely(end >= max)) {
208			pde++;
209			pte = 0;
210		}
211	}
212
213	mmu->func->flush(vm);
214}
215
216void
217nvkm_vm_unmap(struct nvkm_vma *vma)
218{
219	nvkm_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
220}
221
222static void
223nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde)
224{
225	struct nvkm_mmu *mmu = vm->mmu;
226	struct nvkm_vm_pgd *vpgd;
227	struct nvkm_vm_pgt *vpgt;
228	struct nvkm_memory *pgt;
229	u32 pde;
230
231	for (pde = fpde; pde <= lpde; pde++) {
232		vpgt = &vm->pgt[pde - vm->fpde];
233		if (--vpgt->refcount[big])
234			continue;
235
236		pgt = vpgt->mem[big];
237		vpgt->mem[big] = NULL;
238
239		list_for_each_entry(vpgd, &vm->pgd_list, head) {
240			mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
241		}
242
243		nvkm_memory_del(&pgt);
244	}
245}
246
247static int
248nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type)
249{
250	struct nvkm_mmu *mmu = vm->mmu;
251	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
252	struct nvkm_vm_pgd *vpgd;
253	int big = (type != mmu->func->spg_shift);
254	u32 pgt_size;
255	int ret;
256
257	pgt_size  = (1 << (mmu->func->pgt_bits + 12)) >> type;
258	pgt_size *= 8;
259
260	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
261			      pgt_size, 0x1000, true, &vpgt->mem[big]);
262	if (unlikely(ret))
263		return ret;
264
265	list_for_each_entry(vpgd, &vm->pgd_list, head) {
266		mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
267	}
268
269	vpgt->refcount[big]++;
270	return 0;
271}
272
273int
274nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access,
275	    struct nvkm_vma *vma)
276{
277	struct nvkm_mmu *mmu = vm->mmu;
278	u32 align = (1 << page_shift) >> 12;
279	u32 msize = size >> 12;
280	u32 fpde, lpde, pde;
281	int ret;
282
283	mutex_lock(&vm->mutex);
284	ret = nvkm_mm_head(&vm->mm, 0, page_shift, msize, msize, align,
285			   &vma->node);
286	if (unlikely(ret != 0)) {
287		mutex_unlock(&vm->mutex);
288		return ret;
289	}
290
291	fpde = (vma->node->offset >> mmu->func->pgt_bits);
292	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
293
294	for (pde = fpde; pde <= lpde; pde++) {
295		struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
296		int big = (vma->node->type != mmu->func->spg_shift);
297
298		if (likely(vpgt->refcount[big])) {
299			vpgt->refcount[big]++;
300			continue;
301		}
302
303		ret = nvkm_vm_map_pgt(vm, pde, vma->node->type);
304		if (ret) {
305			if (pde != fpde)
306				nvkm_vm_unmap_pgt(vm, big, fpde, pde - 1);
307			nvkm_mm_free(&vm->mm, &vma->node);
308			mutex_unlock(&vm->mutex);
309			return ret;
310		}
311	}
312	mutex_unlock(&vm->mutex);
313
314	vma->vm = NULL;
315	nvkm_vm_ref(vm, &vma->vm, NULL);
316	vma->offset = (u64)vma->node->offset << 12;
317	vma->access = access;
318	return 0;
319}
320
321void
322nvkm_vm_put(struct nvkm_vma *vma)
323{
324	struct nvkm_mmu *mmu;
325	struct nvkm_vm *vm;
326	u32 fpde, lpde;
327
328	if (unlikely(vma->node == NULL))
329		return;
330	vm = vma->vm;
331	mmu = vm->mmu;
332
333	fpde = (vma->node->offset >> mmu->func->pgt_bits);
334	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
335
336	mutex_lock(&vm->mutex);
337	nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde);
338	nvkm_mm_free(&vm->mm, &vma->node);
339	mutex_unlock(&vm->mutex);
340
341	nvkm_vm_ref(NULL, &vma->vm, NULL);
342}
343
344int
345nvkm_vm_boot(struct nvkm_vm *vm, u64 size)
346{
347	struct nvkm_mmu *mmu = vm->mmu;
348	struct nvkm_memory *pgt;
349	int ret;
350
351	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
352			      (size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt);
353	if (ret == 0) {
354		vm->pgt[0].refcount[0] = 1;
355		vm->pgt[0].mem[0] = pgt;
356		nvkm_memory_boot(pgt, vm);
357	}
358
359	return ret;
360}
361
362int
363nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
364	       u32 block, struct lock_class_key *key, struct nvkm_vm **pvm)
365{
366	static struct lock_class_key _key;
367	struct nvkm_vm *vm;
368	u64 mm_length = (offset + length) - mm_offset;
369	int ret;
370
371	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
372	if (!vm)
373		return -ENOMEM;
374
375	__mutex_init(&vm->mutex, "&vm->mutex", key ? key : &_key);
376	INIT_LIST_HEAD(&vm->pgd_list);
377	vm->mmu = mmu;
378	kref_init(&vm->refcount);
379	vm->fpde = offset >> (mmu->func->pgt_bits + 12);
380	vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12);
381
382	vm->pgt  = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt));
383	if (!vm->pgt) {
384		kfree(vm);
385		return -ENOMEM;
386	}
387
388	ret = nvkm_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
389			   block >> 12);
390	if (ret) {
391		vfree(vm->pgt);
392		kfree(vm);
393		return ret;
394	}
395
396	*pvm = vm;
397
398	return 0;
399}
400
401int
402nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset,
403	    struct lock_class_key *key, struct nvkm_vm **pvm)
404{
405	struct nvkm_mmu *mmu = device->mmu;
406	if (!mmu->func->create)
407		return -EINVAL;
408	return mmu->func->create(mmu, offset, length, mm_offset, key, pvm);
409}
410
411static int
412nvkm_vm_link(struct nvkm_vm *vm, struct nvkm_gpuobj *pgd)
413{
414	struct nvkm_mmu *mmu = vm->mmu;
415	struct nvkm_vm_pgd *vpgd;
416	int i;
417
418	if (!pgd)
419		return 0;
420
421	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
422	if (!vpgd)
423		return -ENOMEM;
424
425	vpgd->obj = pgd;
426
427	mutex_lock(&vm->mutex);
428	for (i = vm->fpde; i <= vm->lpde; i++)
429		mmu->func->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem);
430	list_add(&vpgd->head, &vm->pgd_list);
431	mutex_unlock(&vm->mutex);
432	return 0;
433}
434
435static void
436nvkm_vm_unlink(struct nvkm_vm *vm, struct nvkm_gpuobj *mpgd)
437{
438	struct nvkm_vm_pgd *vpgd, *tmp;
439
440	if (!mpgd)
441		return;
442
443	mutex_lock(&vm->mutex);
444	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
445		if (vpgd->obj == mpgd) {
446			list_del(&vpgd->head);
447			kfree(vpgd);
448			break;
449		}
450	}
451	mutex_unlock(&vm->mutex);
452}
453
454static void
455nvkm_vm_del(struct kref *kref)
456{
457	struct nvkm_vm *vm = container_of(kref, typeof(*vm), refcount);
458	struct nvkm_vm_pgd *vpgd, *tmp;
459
460	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
461		nvkm_vm_unlink(vm, vpgd->obj);
462	}
463
464	nvkm_mm_fini(&vm->mm);
465	vfree(vm->pgt);
466	kfree(vm);
467}
468
469int
470nvkm_vm_ref(struct nvkm_vm *ref, struct nvkm_vm **ptr, struct nvkm_gpuobj *pgd)
471{
472	if (ref) {
473		int ret = nvkm_vm_link(ref, pgd);
474		if (ret)
475			return ret;
476
477		kref_get(&ref->refcount);
478	}
479
480	if (*ptr) {
481		nvkm_vm_unlink(*ptr, pgd);
482		kref_put(&(*ptr)->refcount, nvkm_vm_del);
483	}
484
485	*ptr = ref;
486	return 0;
487}
488
489static int
490nvkm_mmu_oneinit(struct nvkm_subdev *subdev)
491{
492	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
493	if (mmu->func->oneinit)
494		return mmu->func->oneinit(mmu);
495	return 0;
496}
497
498static int
499nvkm_mmu_init(struct nvkm_subdev *subdev)
500{
501	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
502	if (mmu->func->init)
503		mmu->func->init(mmu);
504	return 0;
505}
506
507static void *
508nvkm_mmu_dtor(struct nvkm_subdev *subdev)
509{
510	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
511	if (mmu->func->dtor)
512		return mmu->func->dtor(mmu);
513	return mmu;
514}
515
516static const struct nvkm_subdev_func
517nvkm_mmu = {
518	.dtor = nvkm_mmu_dtor,
519	.oneinit = nvkm_mmu_oneinit,
520	.init = nvkm_mmu_init,
521};
522
523void
524nvkm_mmu_ctor(const struct nvkm_mmu_func *func, struct nvkm_device *device,
525	      int index, struct nvkm_mmu *mmu)
526{
527	nvkm_subdev_ctor(&nvkm_mmu, device, index, 0, &mmu->subdev);
528	mmu->func = func;
529	mmu->limit = func->limit;
530	mmu->dma_bits = func->dma_bits;
531	mmu->lpg_shift = func->lpg_shift;
532}
533
534int
535nvkm_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device,
536	      int index, struct nvkm_mmu **pmmu)
537{
538	if (!(*pmmu = kzalloc(sizeof(**pmmu), GFP_KERNEL)))
539		return -ENOMEM;
540	nvkm_mmu_ctor(func, device, index, *pmmu);
541	return 0;
542}
543