[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / gpu / drm / nouveau / nouveau_vm.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 <drm/drmP.h>
#include "nouveau_drv.h"
#include "nouveau_mm.h"
#include "nouveau_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_mm_node *r;
	int big = vma->node->type != vm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vm->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;

			vm->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;
		}
	}

	vm->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;
	int big = vma->node->type != vm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vm->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);

			vm->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);

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

	}
finish:
	vm->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;
	dma_addr_t *list = mem->pages;
	int big = vma->node->type != vm->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (vm->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;

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

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

	vm->flush(vm);
}

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

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

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

	vm->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_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) {
			vm->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_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
	struct nouveau_vm_pgd *vpgd;
	struct nouveau_gpuobj *pgt;
	int big = (type != vm->spg_shift);
	u32 pgt_size;
	int ret;

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

	mutex_unlock(&vm->mm.mutex);
	ret = nouveau_gpuobj_new(vm->dev, 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) {
		vm->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)
{
	u32 align = (1 << page_shift) >> 12;
	u32 msize = size >> 12;
	u32 fpde, lpde, pde;
	int ret;

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

	fpde = (vma->node->offset >> vm->pgt_bits);
	lpde = (vma->node->offset + vma->node->length - 1) >> vm->pgt_bits;
	for (pde = fpde; pde <= lpde; pde++) {
		struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
		int big = (vma->node->type != vm->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_put(&vm->mm, vma->node);
			mutex_unlock(&vm->mm.mutex);
			vma->node = NULL;
			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;
	u32 fpde, lpde;

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

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

int
nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
	       struct nouveau_vm **pvm)
{
	struct drm_nouveau_private *dev_priv = dev->dev_private;
	struct nouveau_vm *vm;
	u64 mm_length = (offset + length) - mm_offset;
	u32 block, pgt_bits;
	int ret;

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

	if (dev_priv->card_type == NV_50) {
		vm->map_pgt = nv50_vm_map_pgt;
		vm->map = nv50_vm_map;
		vm->map_sg = nv50_vm_map_sg;
		vm->unmap = nv50_vm_unmap;
		vm->flush = nv50_vm_flush;
		vm->spg_shift = 12;
		vm->lpg_shift = 16;

		pgt_bits = 29;
		block = (1 << pgt_bits);
		if (length < block)
			block = length;

	} else
	if (dev_priv->card_type >= NV_C0) {
		vm->map_pgt = nvc0_vm_map_pgt;
		vm->map = nvc0_vm_map;
		vm->map_sg = nvc0_vm_map_sg;
		vm->unmap = nvc0_vm_unmap;
		vm->flush = nvc0_vm_flush;
		vm->spg_shift = 12;
		vm->lpg_shift = 17;
		pgt_bits = 27;
		block = 4096;
	} else {
		kfree(vm);
		return -ENOSYS;
	}

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

	INIT_LIST_HEAD(&vm->pgd_list);
	vm->dev = dev;
	vm->refcount = 1;
	vm->pgt_bits = pgt_bits - 12;

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

	*pvm = vm;
	return 0;
}

static int
nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
{
	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++)
		vm->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
760285e7	25	#include <drm/drmP.h>
a11c3198 BS	26	#include "nouveau_drv.h"
	27	#include "nouveau_mm.h"
	28	#include "nouveau_vm.h"
	29
	30	void
d5f42394	31	nouveau_vm_map_at(struct nouveau_vma vma, u64 delta, struct nouveau_mem node)
a11c3198 BS	32	{
	33	struct nouveau_vm *vm = vma->vm;
	34	struct nouveau_mm_node *r;
3ee01281	35	int big = vma->node->type != vm->spg_shift;
a11c3198 BS	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);
	41	u32 end, len;
	42
8f7286f8	43	delta = 0;
d5f42394	44	list_for_each_entry(r, &node->regions, rl_entry) {
a11c3198 BS	45	u64 phys = (u64)r->offset << 12;
	46	u32 num = r->length >> bits;
	47
	48	while (num) {
3ee01281	49	struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
a11c3198 BS	50
	51	end = (pte + num);
	52	if (unlikely(end >= max))
	53	end = max;
	54	len = end - pte;
	55
8f7286f8	56	vm->map(vma, pgt, node, pte, len, phys, delta);
a11c3198 BS	57
	58	num -= len;
	59	pte += len;
	60	if (unlikely(end >= max)) {
73c337e7	61	phys += len << (bits + 12);
a11c3198 BS	62	pde++;
	63	pte = 0;
	64	}
8f7286f8 BS	65
8f7286f8 BS	66	delta += (u64)len << vma->node->type;
a11c3198 BS	67	}
	68	}
	69
	70	vm->flush(vm);
	71	}
	72
	73	void
d5f42394	74	nouveau_vm_map(struct nouveau_vma vma, struct nouveau_mem node)
a11c3198	75	{
d5f42394	76	nouveau_vm_map_at(vma, 0, node);
a11c3198 BS	77	}
a11c3198 BS	78
22b33e8e DA	79	void
	80	nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length,
	81	struct nouveau_mem *mem)
	82	{
	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);
	91	unsigned m, sglen;
	92	u32 end, len;
	93	int i;
	94	struct scatterlist *sg;
	95
	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;
	99
	100	end = pte + sglen;
	101	if (unlikely(end >= max))
	102	end = max;
	103	len = end - pte;
	104
	105	for (m = 0; m < len; m++) {
	106	dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
	107
	108	vm->map_sg(vma, pgt, mem, pte, 1, &addr);
	109	num--;
	110	pte++;
	111
	112	if (num == 0)
	113	goto finish;
	114	}
	115	if (unlikely(end >= max)) {
	116	pde++;
	117	pte = 0;
	118	}
	119	if (m < sglen) {
	120	for (; m < sglen; m++) {
	121	dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
	122
	123	vm->map_sg(vma, pgt, mem, pte, 1, &addr);
	124	num--;
	125	pte++;
	126	if (num == 0)
	127	goto finish;
	128	}
	129	}
	130
	131	}
	132	finish:
	133	vm->flush(vm);
	134	}
	135
a11c3198 BS	136	void
a11c3198 BS	137	nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
f7b24c42	138	struct nouveau_mem *mem)
a11c3198 BS	139	{
a11c3198 BS	140	struct nouveau_vm *vm = vma->vm;
f7b24c42	141	dma_addr_t *list = mem->pages;
3ee01281	142	int big = vma->node->type != vm->spg_shift;
a11c3198 BS	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);
	149	u32 end, len;
	150
	151	while (num) {
3ee01281	152	struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
a11c3198 BS	153
	154	end = (pte + num);
	155	if (unlikely(end >= max))
	156	end = max;
	157	len = end - pte;
	158
26c0c9e3	159	vm->map_sg(vma, pgt, mem, pte, len, list);
a11c3198 BS	160
	161	num -= len;
	162	pte += len;
	163	list += len;
	164	if (unlikely(end >= max)) {
	165	pde++;
	166	pte = 0;
	167	}
	168	}
	169
	170	vm->flush(vm);
	171	}
	172
	173	void
	174	nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
	175	{
	176	struct nouveau_vm *vm = vma->vm;
3ee01281	177	int big = vma->node->type != vm->spg_shift;
a11c3198 BS	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);
	184	u32 end, len;
	185
	186	while (num) {
3ee01281	187	struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
a11c3198 BS	188
	189	end = (pte + num);
	190	if (unlikely(end >= max))
	191	end = max;
	192	len = end - pte;
	193
	194	vm->unmap(pgt, pte, len);
	195
	196	num -= len;
	197	pte += len;
	198	if (unlikely(end >= max)) {
	199	pde++;
	200	pte = 0;
	201	}
	202	}
	203
	204	vm->flush(vm);
	205	}
	206
	207	void
	208	nouveau_vm_unmap(struct nouveau_vma *vma)
	209	{
	210	nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
	211	}
	212
	213	static void
3ee01281	214	nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
a11c3198 BS	215	{
	216	struct nouveau_vm_pgd *vpgd;
	217	struct nouveau_vm_pgt *vpgt;
	218	struct nouveau_gpuobj *pgt;
	219	u32 pde;
	220
	221	for (pde = fpde; pde <= lpde; pde++) {
	222	vpgt = &vm->pgt[pde - vm->fpde];
3ee01281	223	if (--vpgt->refcount[big])
a11c3198 BS	224	continue;
a11c3198 BS	225
3ee01281 BS	226	pgt = vpgt->obj[big];
	227	vpgt->obj[big] = NULL;
	228
a11c3198	229	list_for_each_entry(vpgd, &vm->pgd_list, head) {
3ee01281	230	vm->map_pgt(vpgd->obj, pde, vpgt->obj);
a11c3198 BS	231	}
a11c3198 BS	232
987eec10	233	mutex_unlock(&vm->mm.mutex);
a11c3198	234	nouveau_gpuobj_ref(NULL, &pgt);
987eec10	235	mutex_lock(&vm->mm.mutex);
a11c3198 BS	236	}
	237	}
	238
	239	static int
	240	nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
	241	{
	242	struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
	243	struct nouveau_vm_pgd *vpgd;
	244	struct nouveau_gpuobj *pgt;
3ee01281	245	int big = (type != vm->spg_shift);
a11c3198 BS	246	u32 pgt_size;
	247	int ret;
	248
	249	pgt_size = (1 << (vm->pgt_bits + 12)) >> type;
	250	pgt_size *= 8;
	251
987eec10	252	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	253	ret = nouveau_gpuobj_new(vm->dev, NULL, pgt_size, 0x1000,
a11c3198 BS	254	NVOBJ_FLAG_ZERO_ALLOC, &pgt);
987eec10	255	mutex_lock(&vm->mm.mutex);
a11c3198 BS	256	if (unlikely(ret))
	257	return ret;
	258
	259	/* someone beat us to filling the PDE while we didn't have the lock */
3ee01281	260	if (unlikely(vpgt->refcount[big]++)) {
987eec10	261	mutex_unlock(&vm->mm.mutex);
a11c3198	262	nouveau_gpuobj_ref(NULL, &pgt);
987eec10	263	mutex_lock(&vm->mm.mutex);
a11c3198 BS	264	return 0;
	265	}
	266
3ee01281	267	vpgt->obj[big] = pgt;
a11c3198	268	list_for_each_entry(vpgd, &vm->pgd_list, head) {
3ee01281	269	vm->map_pgt(vpgd->obj, pde, vpgt->obj);
a11c3198 BS	270	}
a11c3198 BS	271
a11c3198 BS	272	return 0;
	273	}
	274
	275	int
	276	nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
	277	u32 access, struct nouveau_vma *vma)
	278	{
	279	u32 align = (1 << page_shift) >> 12;
	280	u32 msize = size >> 12;
	281	u32 fpde, lpde, pde;
	282	int ret;
	283
987eec10 BS	284	mutex_lock(&vm->mm.mutex);
987eec10 BS	285	ret = nouveau_mm_get(&vm->mm, page_shift, msize, 0, align, &vma->node);
a11c3198	286	if (unlikely(ret != 0)) {
987eec10	287	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	288	return ret;
	289	}
	290
	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];
3ee01281	295	int big = (vma->node->type != vm->spg_shift);
a11c3198	296
3ee01281 BS	297	if (likely(vpgt->refcount[big])) {
3ee01281 BS	298	vpgt->refcount[big]++;
a11c3198 BS	299	continue;
	300	}
	301
	302	ret = nouveau_vm_map_pgt(vm, pde, vma->node->type);
	303	if (ret) {
	304	if (pde != fpde)
3ee01281	305	nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
987eec10 BS	306	nouveau_mm_put(&vm->mm, vma->node);
987eec10 BS	307	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	308	vma->node = NULL;
	309	return ret;
	310	}
	311	}
987eec10	312	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	313
	314	vma->vm = vm;
	315	vma->offset = (u64)vma->node->offset << 12;
	316	vma->access = access;
	317	return 0;
	318	}
	319
	320	void
	321	nouveau_vm_put(struct nouveau_vma *vma)
	322	{
	323	struct nouveau_vm *vm = vma->vm;
	324	u32 fpde, lpde;
	325
	326	if (unlikely(vma->node == NULL))
	327	return;
	328	fpde = (vma->node->offset >> vm->pgt_bits);
	329	lpde = (vma->node->offset + vma->node->length - 1) >> vm->pgt_bits;
	330
987eec10	331	mutex_lock(&vm->mm.mutex);
3ee01281	332	nouveau_vm_unmap_pgt(vm, vma->node->type != vm->spg_shift, fpde, lpde);
987eec10	333	nouveau_mm_put(&vm->mm, vma->node);
a11c3198	334	vma->node = NULL;
987eec10	335	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	336	}
	337
	338	int
	339	nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
a11c3198 BS	340	struct nouveau_vm **pvm)
	341	{
	342	struct drm_nouveau_private *dev_priv = dev->dev_private;
	343	struct nouveau_vm *vm;
	344	u64 mm_length = (offset + length) - mm_offset;
3ee01281	345	u32 block, pgt_bits;
a11c3198 BS	346	int ret;
	347
	348	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
	349	if (!vm)
	350	return -ENOMEM;
	351
	352	if (dev_priv->card_type == NV_50) {
	353	vm->map_pgt = nv50_vm_map_pgt;
a11c3198 BS	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;
3ee01281 BS	358	vm->spg_shift = 12;
3ee01281 BS	359	vm->lpg_shift = 16;
4c74eb7f	360
3ee01281	361	pgt_bits = 29;
4c74eb7f BS	362	block = (1 << pgt_bits);
	363	if (length < block)
	364	block = length;
	365
	366	} else
2e9733ff	367	if (dev_priv->card_type >= NV_C0) {
4c74eb7f BS	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;
	373	vm->spg_shift = 12;
	374	vm->lpg_shift = 17;
	375	pgt_bits = 27;
b84f30e6	376	block = 4096;
a11c3198 BS	377	} else {
	378	kfree(vm);
	379	return -ENOSYS;
	380	}
	381
	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);
	385	if (!vm->pgt) {
	386	kfree(vm);
	387	return -ENOMEM;
	388	}
	389
	390	INIT_LIST_HEAD(&vm->pgd_list);
	391	vm->dev = dev;
	392	vm->refcount = 1;
	393	vm->pgt_bits = pgt_bits - 12;
a11c3198	394
a11c3198 BS	395	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
	396	block >> 12);
	397	if (ret) {
	398	kfree(vm);
	399	return ret;
	400	}
	401
	402	*pvm = vm;
	403	return 0;
	404	}
	405
	406	static int
	407	nouveau_vm_link(struct nouveau_vm vm, struct nouveau_gpuobj pgd)
	408	{
	409	struct nouveau_vm_pgd *vpgd;
	410	int i;
	411
	412	if (!pgd)
	413	return 0;
	414
	415	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
	416	if (!vpgd)
	417	return -ENOMEM;
	418
	419	nouveau_gpuobj_ref(pgd, &vpgd->obj);
	420
987eec10	421	mutex_lock(&vm->mm.mutex);
3ee01281 BS	422	for (i = vm->fpde; i <= vm->lpde; i++)
3ee01281 BS	423	vm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
a11c3198	424	list_add(&vpgd->head, &vm->pgd_list);
987eec10	425	mutex_unlock(&vm->mm.mutex);
a11c3198 BS	426	return 0;
	427	}
	428
	429	static void
15ba79ad	430	nouveau_vm_unlink(struct nouveau_vm vm, struct nouveau_gpuobj mpgd)
a11c3198 BS	431	{
a11c3198 BS	432	struct nouveau_vm_pgd vpgd, tmp;
15ba79ad	433	struct nouveau_gpuobj *pgd = NULL;
a11c3198	434
15ba79ad	435	if (!mpgd)
a11c3198 BS	436	return;
a11c3198 BS	437
987eec10	438	mutex_lock(&vm->mm.mutex);
a11c3198	439	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
15ba79ad BS	440	if (vpgd->obj == mpgd) {
	441	pgd = vpgd->obj;
	442	list_del(&vpgd->head);
	443	kfree(vpgd);
	444	break;
	445	}
a11c3198	446	}
987eec10	447	mutex_unlock(&vm->mm.mutex);
15ba79ad BS	448
15ba79ad BS	449	nouveau_gpuobj_ref(NULL, &pgd);
a11c3198 BS	450	}
	451
	452	static void
	453	nouveau_vm_del(struct nouveau_vm *vm)
	454	{
	455	struct nouveau_vm_pgd vpgd, tmp;
	456
	457	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
	458	nouveau_vm_unlink(vm, vpgd->obj);
	459	}
a11c3198	460
ad9ac437	461	nouveau_mm_fini(&vm->mm);
a11c3198 BS	462	kfree(vm->pgt);
	463	kfree(vm);
	464	}
	465
	466	int
	467	nouveau_vm_ref(struct nouveau_vm ref, struct nouveau_vm *ptr,
	468	struct nouveau_gpuobj *pgd)
	469	{
	470	struct nouveau_vm *vm;
	471	int ret;
	472
	473	vm = ref;
	474	if (vm) {
	475	ret = nouveau_vm_link(vm, pgd);
	476	if (ret)
	477	return ret;
	478
	479	vm->refcount++;
	480	}
	481
	482	vm = *ptr;
	483	*ptr = ref;
	484
	485	if (vm) {
	486	nouveau_vm_unlink(vm, pgd);
	487
	488	if (--vm->refcount == 0)
	489	nouveau_vm_del(vm);
	490	}
	491
	492	return 0;
	493	}