[deliverable/linux.git] / drivers / gpu / drm / nouveau / core / subdev / vm / base.c

/*
 * Copyright 2010 Red Hat Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 *
 * Authors: Ben Skeggs
 */

#include <core/gpuobj.h>
#include <core/mm.h>

#include <subdev/fb.h>
#include <subdev/vm.h>

void
nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node)
{
	struct nouveau_vm *vm = vma->vm;
	struct nouveau_vmmgr *vmm = vm->vmm;
	struct nouveau_mm_node *r;
	int big = vma->node->type != vmm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vmm->pgt_bits - bits);
	u32 end, len;

	delta = 0;
	list_for_each_entry(r, &node->regions, rl_entry) {
		u64 phys = (u64)r->offset << 12;
		u32 num  = r->length >> bits;

		while (num) {
			struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];

			end = (pte + num);
			if (unlikely(end >= max))
				end = max;
			len = end - pte;

			vmm->map(vma, pgt, node, pte, len, phys, delta);

			num -= len;
			pte += len;
			if (unlikely(end >= max)) {
				phys += len << (bits + 12);
				pde++;
				pte = 0;
			}

			delta += (u64)len << vma->node->type;
		}
	}

	vmm->flush(vm);
}

void
nouveau_vm_map(struct nouveau_vma *vma, struct nouveau_mem *node)
{
	nouveau_vm_map_at(vma, 0, node);
}

void
nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length,
			struct nouveau_mem *mem)
{
	struct nouveau_vm *vm = vma->vm;
	struct nouveau_vmmgr *vmm = vm->vmm;
	int big = vma->node->type != vmm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vmm->pgt_bits - bits);
	unsigned m, sglen;
	u32 end, len;
	int i;
	struct scatterlist *sg;

	for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) {
		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
		sglen = sg_dma_len(sg) >> PAGE_SHIFT;

		end = pte + sglen;
		if (unlikely(end >= max))
			end = max;
		len = end - pte;

		for (m = 0; m < len; m++) {
			dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);

			vmm->map_sg(vma, pgt, mem, pte, 1, &addr);
			num--;
			pte++;

			if (num == 0)
				goto finish;
		}
		if (unlikely(end >= max)) {
			pde++;
			pte = 0;
		}
		if (m < sglen) {
			for (; m < sglen; m++) {
				dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);

				vmm->map_sg(vma, pgt, mem, pte, 1, &addr);
				num--;
				pte++;
				if (num == 0)
					goto finish;
			}
		}

	}
finish:
	vmm->flush(vm);
}

void
nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
		  struct nouveau_mem *mem)
{
	struct nouveau_vm *vm = vma->vm;
	struct nouveau_vmmgr *vmm = vm->vmm;
	dma_addr_t *list = mem->pages;
	int big = vma->node->type != vmm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vmm->pgt_bits - bits);
	u32 end, len;

	while (num) {
		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];

		end = (pte + num);
		if (unlikely(end >= max))
			end = max;
		len = end - pte;

		vmm->map_sg(vma, pgt, mem, pte, len, list);

		num  -= len;
		pte  += len;
		list += len;
		if (unlikely(end >= max)) {
			pde++;
			pte = 0;
		}
	}

	vmm->flush(vm);
}

void
nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
{
	struct nouveau_vm *vm = vma->vm;
	struct nouveau_vmmgr *vmm = vm->vmm;
	int big = vma->node->type != vmm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vmm->pgt_bits - bits);
	u32 end, len;

	while (num) {
		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];

		end = (pte + num);
		if (unlikely(end >= max))
			end = max;
		len = end - pte;

		vmm->unmap(pgt, pte, len);

		num -= len;
		pte += len;
		if (unlikely(end >= max)) {
			pde++;
			pte = 0;
		}
	}

	vmm->flush(vm);
}

void
nouveau_vm_unmap(struct nouveau_vma *vma)
{
	nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
}

static void
nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
{
	struct nouveau_vmmgr *vmm = vm->vmm;
	struct nouveau_vm_pgd *vpgd;
	struct nouveau_vm_pgt *vpgt;
	struct nouveau_gpuobj *pgt;
	u32 pde;

	for (pde = fpde; pde <= lpde; pde++) {
		vpgt = &vm->pgt[pde - vm->fpde];
		if (--vpgt->refcount[big])
			continue;

		pgt = vpgt->obj[big];
		vpgt->obj[big] = NULL;

		list_for_each_entry(vpgd, &vm->pgd_list, head) {
			vmm->map_pgt(vpgd->obj, pde, vpgt->obj);
		}

		mutex_unlock(&vm->mm.mutex);
		nouveau_gpuobj_ref(NULL, &pgt);
		mutex_lock(&vm->mm.mutex);
	}
}

static int
nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
{
	struct nouveau_vmmgr *vmm = vm->vmm;
	struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
	struct nouveau_vm_pgd *vpgd;
	struct nouveau_gpuobj *pgt;
	int big = (type != vmm->spg_shift);
	u32 pgt_size;
	int ret;

	pgt_size  = (1 << (vmm->pgt_bits + 12)) >> type;
	pgt_size *= 8;

	mutex_unlock(&vm->mm.mutex);
	ret = nouveau_gpuobj_new(nv_object(vm->vmm), NULL, pgt_size, 0x1000,
				 NVOBJ_FLAG_ZERO_ALLOC, &pgt);
	mutex_lock(&vm->mm.mutex);
	if (unlikely(ret))
		return ret;

	/* someone beat us to filling the PDE while we didn't have the lock */
	if (unlikely(vpgt->refcount[big]++)) {
		mutex_unlock(&vm->mm.mutex);
		nouveau_gpuobj_ref(NULL, &pgt);
		mutex_lock(&vm->mm.mutex);
		return 0;
	}

	vpgt->obj[big] = pgt;
	list_for_each_entry(vpgd, &vm->pgd_list, head) {
		vmm->map_pgt(vpgd->obj, pde, vpgt->obj);
	}

	return 0;
}

int
nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
	       u32 access, struct nouveau_vma *vma)
{
	struct nouveau_vmmgr *vmm = vm->vmm;
	u32 align = (1 << page_shift) >> 12;
	u32 msize = size >> 12;
	u32 fpde, lpde, pde;
	int ret;

	mutex_lock(&vm->mm.mutex);
	ret = nouveau_mm_head(&vm->mm, page_shift, msize, msize, align,
			     &vma->node);
	if (unlikely(ret != 0)) {
		mutex_unlock(&vm->mm.mutex);
		return ret;
	}

	fpde = (vma->node->offset >> vmm->pgt_bits);
	lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;

	for (pde = fpde; pde <= lpde; pde++) {
		struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
		int big = (vma->node->type != vmm->spg_shift);

		if (likely(vpgt->refcount[big])) {
			vpgt->refcount[big]++;
			continue;
		}

		ret = nouveau_vm_map_pgt(vm, pde, vma->node->type);
		if (ret) {
			if (pde != fpde)
				nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
			nouveau_mm_free(&vm->mm, &vma->node);
			mutex_unlock(&vm->mm.mutex);
			return ret;
		}
	}
	mutex_unlock(&vm->mm.mutex);

	vma->vm     = vm;
	vma->offset = (u64)vma->node->offset << 12;
	vma->access = access;
	return 0;
}

void
nouveau_vm_put(struct nouveau_vma *vma)
{
	struct nouveau_vm *vm = vma->vm;
	struct nouveau_vmmgr *vmm = vm->vmm;
	u32 fpde, lpde;

	if (unlikely(vma->node == NULL))
		return;
	fpde = (vma->node->offset >> vmm->pgt_bits);
	lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;

	mutex_lock(&vm->mm.mutex);
	nouveau_vm_unmap_pgt(vm, vma->node->type != vmm->spg_shift, fpde, lpde);
	nouveau_mm_free(&vm->mm, &vma->node);
	mutex_unlock(&vm->mm.mutex);
}

int
nouveau_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length,
		  u64 mm_offset, u32 block, struct nouveau_vm **pvm)
{
	struct nouveau_vm *vm;
	u64 mm_length = (offset + length) - mm_offset;
	int ret;

	vm = *pvm = kzalloc(sizeof(*vm), GFP_KERNEL);
	if (!vm)
		return -ENOMEM;

	INIT_LIST_HEAD(&vm->pgd_list);
	vm->vmm = vmm;
	vm->refcount = 1;
	vm->fpde = offset >> (vmm->pgt_bits + 12);
	vm->lpde = (offset + length - 1) >> (vmm->pgt_bits + 12);

	vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt), GFP_KERNEL);
	if (!vm->pgt) {
		kfree(vm);
		return -ENOMEM;
	}

	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
			      block >> 12);
	if (ret) {
		kfree(vm->pgt);
		kfree(vm);
		return ret;
	}

	return 0;
}

int
nouveau_vm_new(struct nouveau_device *device, u64 offset, u64 length,
	       u64 mm_offset, struct nouveau_vm **pvm)
{
	struct nouveau_vmmgr *vmm = nouveau_vmmgr(device);
	return vmm->create(vmm, offset, length, mm_offset, pvm);
}

static int
nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
{
	struct nouveau_vmmgr *vmm = vm->vmm;
	struct nouveau_vm_pgd *vpgd;
	int i;

	if (!pgd)
		return 0;

	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
	if (!vpgd)
		return -ENOMEM;

	nouveau_gpuobj_ref(pgd, &vpgd->obj);

	mutex_lock(&vm->mm.mutex);
	for (i = vm->fpde; i <= vm->lpde; i++)
		vmm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
	list_add(&vpgd->head, &vm->pgd_list);
	mutex_unlock(&vm->mm.mutex);
	return 0;
}

static void
nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
{
	struct nouveau_vm_pgd *vpgd, *tmp;
	struct nouveau_gpuobj *pgd = NULL;

	if (!mpgd)
		return;

	mutex_lock(&vm->mm.mutex);
	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
		if (vpgd->obj == mpgd) {
			pgd = vpgd->obj;
			list_del(&vpgd->head);
			kfree(vpgd);
			break;
		}
	}
	mutex_unlock(&vm->mm.mutex);

	nouveau_gpuobj_ref(NULL, &pgd);
}

static void
nouveau_vm_del(struct nouveau_vm *vm)
{
	struct nouveau_vm_pgd *vpgd, *tmp;

	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
		nouveau_vm_unlink(vm, vpgd->obj);
	}

	nouveau_mm_fini(&vm->mm);
	kfree(vm->pgt);
	kfree(vm);
}

int
nouveau_vm_ref(struct nouveau_vm *ref, struct nouveau_vm **ptr,
	       struct nouveau_gpuobj *pgd)
{
	struct nouveau_vm *vm;
	int ret;

	vm = ref;
	if (vm) {
		ret = nouveau_vm_link(vm, pgd);
		if (ret)
			return ret;

		vm->refcount++;
	}

	vm = *ptr;
	*ptr = ref;

	if (vm) {
		nouveau_vm_unlink(vm, pgd);

		if (--vm->refcount == 0)
			nouveau_vm_del(vm);
	}

	return 0;
}
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
3863c9bc BS	28	#include <subdev/fb.h>
02a841d4	29	#include <subdev/vm.h>
a11c3198 BS	30
a11c3198 BS	31	void
d5f42394	32	nouveau_vm_map_at(struct nouveau_vma vma, u64 delta, struct nouveau_mem node)
a11c3198 BS	33	{
a11c3198 BS	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);
a11c3198 BS	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;
a11c3198 BS	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
8f7286f8 BS	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	}
a11c3198 BS	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;
3863c9bc BS	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	}
22b33e8e DA	138
a11c3198 BS	139	void
a11c3198 BS	140	nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
f7b24c42	141	struct nouveau_mem *mem)
a11c3198 BS	142	{
a11c3198 BS	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;
3863c9bc BS	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;
a11c3198 BS	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	}
a11c3198 BS	238
987eec10	239	mutex_unlock(&vm->mm.mutex);
a11c3198	240	nouveau_gpuobj_ref(NULL, &pgt);
987eec10	241	mutex_lock(&vm->mm.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;
a11c3198 BS	258
987eec10	259	mutex_unlock(&vm->mm.mutex);
3863c9bc	260	ret = nouveau_gpuobj_new(nv_object(vm->vmm), NULL, pgt_size, 0x1000,
a11c3198	261	NVOBJ_FLAG_ZERO_ALLOC, &pgt);
987eec10	262	mutex_lock(&vm->mm.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]++)) {
987eec10	268	mutex_unlock(&vm->mm.mutex);
a11c3198	269	nouveau_gpuobj_ref(NULL, &pgt);
987eec10	270	mutex_lock(&vm->mm.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	}
a11c3198 BS	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
987eec10	292	mutex_lock(&vm->mm.mutex);
496734bf BS	293	ret = nouveau_mm_head(&vm->mm, page_shift, msize, msize, align,
496734bf BS	294	&vma->node);
a11c3198	295	if (unlikely(ret != 0)) {
987eec10	296	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	297	return ret;
	298	}
	299
3863c9bc BS	300	fpde = (vma->node->offset >> vmm->pgt_bits);
3863c9bc BS	301	lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;
ebb945a9	302
a11c3198 BS	303	for (pde = fpde; pde <= lpde; pde++) {
a11c3198 BS	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])) {
3ee01281 BS	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);
987eec10	317	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	318	return ret;
	319	}
	320	}
987eec10	321	mutex_unlock(&vm->mm.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);
3863c9bc BS	339	lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;
a11c3198	340
987eec10	341	mutex_lock(&vm->mm.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);
987eec10	344	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	345	}
	346
	347	int
3863c9bc BS	348	nouveau_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length,
3863c9bc BS	349	u64 mm_offset, u32 block, struct nouveau_vm **pvm)
a11c3198	350	{
a11c3198 BS	351	struct nouveau_vm *vm;
a11c3198 BS	352	u64 mm_length = (offset + length) - mm_offset;
a11c3198 BS	353	int ret;
a11c3198 BS	354
3863c9bc	355	vm = pvm = 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
a11c3198 BS	379	return 0;
	380	}
	381
3863c9bc BS	382	int
	383	nouveau_vm_new(struct nouveau_device *device, u64 offset, u64 length,
	384	u64 mm_offset, struct nouveau_vm **pvm)
	385	{
	386	struct nouveau_vmmgr *vmm = nouveau_vmmgr(device);
	387	return vmm->create(vmm, offset, length, mm_offset, pvm);
	388	}
	389
a11c3198 BS	390	static int
	391	nouveau_vm_link(struct nouveau_vm vm, struct nouveau_gpuobj pgd)
	392	{
3863c9bc	393	struct nouveau_vmmgr *vmm = vm->vmm;
a11c3198 BS	394	struct nouveau_vm_pgd *vpgd;
	395	int i;
	396
	397	if (!pgd)
	398	return 0;
	399
	400	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
	401	if (!vpgd)
	402	return -ENOMEM;
	403
	404	nouveau_gpuobj_ref(pgd, &vpgd->obj);
	405
987eec10	406	mutex_lock(&vm->mm.mutex);
3ee01281	407	for (i = vm->fpde; i <= vm->lpde; i++)
3863c9bc	408	vmm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
a11c3198	409	list_add(&vpgd->head, &vm->pgd_list);
987eec10	410	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	411	return 0;
	412	}
	413
	414	static void
15ba79ad	415	nouveau_vm_unlink(struct nouveau_vm vm, struct nouveau_gpuobj mpgd)
a11c3198 BS	416	{
a11c3198 BS	417	struct nouveau_vm_pgd vpgd, tmp;
15ba79ad	418	struct nouveau_gpuobj *pgd = NULL;
a11c3198	419
15ba79ad	420	if (!mpgd)
a11c3198 BS	421	return;
a11c3198 BS	422
987eec10	423	mutex_lock(&vm->mm.mutex);
a11c3198	424	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
15ba79ad BS	425	if (vpgd->obj == mpgd) {
	426	pgd = vpgd->obj;
	427	list_del(&vpgd->head);
	428	kfree(vpgd);
	429	break;
	430	}
a11c3198	431	}
987eec10	432	mutex_unlock(&vm->mm.mutex);
15ba79ad BS	433
15ba79ad BS	434	nouveau_gpuobj_ref(NULL, &pgd);
a11c3198 BS	435	}
	436
	437	static void
	438	nouveau_vm_del(struct nouveau_vm *vm)
	439	{
	440	struct nouveau_vm_pgd vpgd, tmp;
	441
	442	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
	443	nouveau_vm_unlink(vm, vpgd->obj);
	444	}
a11c3198	445
ad9ac437	446	nouveau_mm_fini(&vm->mm);
a11c3198 BS	447	kfree(vm->pgt);
	448	kfree(vm);
	449	}
	450
	451	int
	452	nouveau_vm_ref(struct nouveau_vm ref, struct nouveau_vm *ptr,
	453	struct nouveau_gpuobj *pgd)
	454	{
	455	struct nouveau_vm *vm;
	456	int ret;
	457
	458	vm = ref;
	459	if (vm) {
	460	ret = nouveau_vm_link(vm, pgd);
	461	if (ret)
	462	return ret;
	463
	464	vm->refcount++;
	465	}
	466
	467	vm = *ptr;
	468	*ptr = ref;
	469
	470	if (vm) {
	471	nouveau_vm_unlink(vm, pgd);
	472
	473	if (--vm->refcount == 0)
	474	nouveau_vm_del(vm);
	475	}
	476
	477	return 0;
	478	}