[linux-2.6-block.git] / drivers / gpu / drm / nouveau / nvkm / subdev / mmu / 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 "priv.h"

#include <core/gpuobj.h>
#include <subdev/fb.h>

void
nvkm_vm_map_at(struct nvkm_vma *vma, u64 delta, struct nvkm_mem *node)
{
	struct nvkm_vm *vm = vma->vm;
	struct nvkm_mmu *mmu = vm->mmu;
	struct nvkm_mm_node *r = node->mem;
	int big = vma->node->type != mmu->func->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (mmu->func->pgt_bits - bits);
	u32 end, len;

	delta = 0;
	while (r) {
		u64 phys = (u64)r->offset << 12;
		u32 num  = r->length >> bits;

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

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

			mmu->func->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;
		}
		r = r->next;
	};

	mmu->func->flush(vm);
}

static void
nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
		     struct nvkm_mem *mem)
{
	struct nvkm_vm *vm = vma->vm;
	struct nvkm_mmu *mmu = vm->mmu;
	int big = vma->node->type != mmu->func->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (mmu->func->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 nvkm_memory *pgt = vm->pgt[pde].mem[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);

			mmu->func->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);

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

	}
finish:
	mmu->func->flush(vm);
}

static void
nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
	       struct nvkm_mem *mem)
{
	struct nvkm_vm *vm = vma->vm;
	struct nvkm_mmu *mmu = vm->mmu;
	dma_addr_t *list = mem->pages;
	int big = vma->node->type != mmu->func->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (mmu->func->pgt_bits - bits);
	u32 end, len;

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

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

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

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

	mmu->func->flush(vm);
}

void
nvkm_vm_map(struct nvkm_vma *vma, struct nvkm_mem *node)
{
	if (node->sg)
		nvkm_vm_map_sg_table(vma, 0, node->size << 12, node);
	else
	if (node->pages)
		nvkm_vm_map_sg(vma, 0, node->size << 12, node);
	else
		nvkm_vm_map_at(vma, 0, node);
}

void
nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
{
	struct nvkm_vm *vm = vma->vm;
	struct nvkm_mmu *mmu = vm->mmu;
	int big = vma->node->type != mmu->func->spg_shift;
	u32 offset = vma->node->offset + (delta >> 12);
	u32 bits = vma->node->type - 12;
	u32 num  = length >> vma->node->type;
	u32 pde  = (offset >> mmu->func->pgt_bits) - vm->fpde;
	u32 pte  = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	u32 max  = 1 << (mmu->func->pgt_bits - bits);
	u32 end, len;

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

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

		mmu->func->unmap(vma, pgt, pte, len);

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

	mmu->func->flush(vm);
}

void
nvkm_vm_unmap(struct nvkm_vma *vma)
{
	nvkm_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
}

static void
nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde)
{
	struct nvkm_mmu *mmu = vm->mmu;
	struct nvkm_vm_pgd *vpgd;
	struct nvkm_vm_pgt *vpgt;
	struct nvkm_memory *pgt;
	u32 pde;

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

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

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

		mmu->func->flush(vm);

		nvkm_memory_del(&pgt);
	}
}

static int
nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type)
{
	struct nvkm_mmu *mmu = vm->mmu;
	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
	struct nvkm_vm_pgd *vpgd;
	int big = (type != mmu->func->spg_shift);
	u32 pgt_size;
	int ret;

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

	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
			      pgt_size, 0x1000, true, &vpgt->mem[big]);
	if (unlikely(ret))
		return ret;

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

	vpgt->refcount[big]++;
	return 0;
}

int
nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access,
	    struct nvkm_vma *vma)
{
	struct nvkm_mmu *mmu = vm->mmu;
	u32 align = (1 << page_shift) >> 12;
	u32 msize = size >> 12;
	u32 fpde, lpde, pde;
	int ret;

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

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

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

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

		ret = nvkm_vm_map_pgt(vm, pde, vma->node->type);
		if (ret) {
			if (pde != fpde)
				nvkm_vm_unmap_pgt(vm, big, fpde, pde - 1);
			nvkm_mm_free(&vm->mm, &vma->node);
			mutex_unlock(&vm->mutex);
			return ret;
		}
	}
	mutex_unlock(&vm->mutex);

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

void
nvkm_vm_put(struct nvkm_vma *vma)
{
	struct nvkm_mmu *mmu;
	struct nvkm_vm *vm;
	u32 fpde, lpde;

	if (unlikely(vma->node == NULL))
		return;
	vm = vma->vm;
	mmu = vm->mmu;

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

	mutex_lock(&vm->mutex);
	nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde);
	nvkm_mm_free(&vm->mm, &vma->node);
	mutex_unlock(&vm->mutex);

	nvkm_vm_ref(NULL, &vma->vm, NULL);
}

int
nvkm_vm_boot(struct nvkm_vm *vm, u64 size)
{
	struct nvkm_mmu *mmu = vm->mmu;
	struct nvkm_memory *pgt;
	int ret;

	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
			      (size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt);
	if (ret == 0) {
		vm->pgt[0].refcount[0] = 1;
		vm->pgt[0].mem[0] = pgt;
		nvkm_memory_boot(pgt, vm);
	}

	return ret;
}

int
nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
	       u32 block, struct lock_class_key *key, struct nvkm_vm **pvm)
{
	static struct lock_class_key _key;
	struct nvkm_vm *vm;
	u64 mm_length = (offset + length) - mm_offset;
	int ret;

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

	__mutex_init(&vm->mutex, "&vm->mutex", key ? key : &_key);
	INIT_LIST_HEAD(&vm->pgd_list);
	vm->mmu = mmu;
	kref_init(&vm->refcount);
	vm->fpde = offset >> (mmu->func->pgt_bits + 12);
	vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12);

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

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

	*pvm = vm;

	return 0;
}

int
nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset,
	    struct lock_class_key *key, struct nvkm_vm **pvm)
{
	struct nvkm_mmu *mmu = device->mmu;
	if (!mmu->func->create)
		return -EINVAL;
	return mmu->func->create(mmu, offset, length, mm_offset, key, pvm);
}

static int
nvkm_vm_link(struct nvkm_vm *vm, struct nvkm_gpuobj *pgd)
{
	struct nvkm_mmu *mmu = vm->mmu;
	struct nvkm_vm_pgd *vpgd;
	int i;

	if (!pgd)
		return 0;

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

	vpgd->obj = pgd;

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

static void
nvkm_vm_unlink(struct nvkm_vm *vm, struct nvkm_gpuobj *mpgd)
{
	struct nvkm_vm_pgd *vpgd, *tmp;

	if (!mpgd)
		return;

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

static void
nvkm_vm_del(struct kref *kref)
{
	struct nvkm_vm *vm = container_of(kref, typeof(*vm), refcount);
	struct nvkm_vm_pgd *vpgd, *tmp;

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

	nvkm_mm_fini(&vm->mm);
	vfree(vm->pgt);
	kfree(vm);
}

int
nvkm_vm_ref(struct nvkm_vm *ref, struct nvkm_vm **ptr, struct nvkm_gpuobj *pgd)
{
	if (ref) {
		int ret = nvkm_vm_link(ref, pgd);
		if (ret)
			return ret;

		kref_get(&ref->refcount);
	}

	if (*ptr) {
		nvkm_vm_unlink(*ptr, pgd);
		kref_put(&(*ptr)->refcount, nvkm_vm_del);
	}

	*ptr = ref;
	return 0;
}

static int
nvkm_mmu_oneinit(struct nvkm_subdev *subdev)
{
	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
	if (mmu->func->oneinit)
		return mmu->func->oneinit(mmu);
	return 0;
}

static int
nvkm_mmu_init(struct nvkm_subdev *subdev)
{
	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
	if (mmu->func->init)
		mmu->func->init(mmu);
	return 0;
}

static void *
nvkm_mmu_dtor(struct nvkm_subdev *subdev)
{
	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
	if (mmu->func->dtor)
		return mmu->func->dtor(mmu);
	return mmu;
}

static const struct nvkm_subdev_func
nvkm_mmu = {
	.dtor = nvkm_mmu_dtor,
	.oneinit = nvkm_mmu_oneinit,
	.init = nvkm_mmu_init,
};

void
nvkm_mmu_ctor(const struct nvkm_mmu_func *func, struct nvkm_device *device,
	      int index, struct nvkm_mmu *mmu)
{
	nvkm_subdev_ctor(&nvkm_mmu, device, index, &mmu->subdev);
	mmu->func = func;
	mmu->limit = func->limit;
	mmu->dma_bits = func->dma_bits;
	mmu->lpg_shift = func->lpg_shift;
}

int
nvkm_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device,
	      int index, struct nvkm_mmu **pmmu)
{
	if (!(*pmmu = kzalloc(sizeof(**pmmu), GFP_KERNEL)))
		return -ENOMEM;
	nvkm_mmu_ctor(func, device, index, *pmmu);
	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	*/
c9582455	24	#include "priv.h"
a11c3198	25
3863c9bc	26	#include <core/gpuobj.h>
c9582455	27	#include <subdev/fb.h>
a11c3198 BS	28
a11c3198 BS	29	void
42594600	30	nvkm_vm_map_at(struct nvkm_vma vma, u64 delta, struct nvkm_mem node)
a11c3198	31	{
42594600 BS	32	struct nvkm_vm *vm = vma->vm;
42594600 BS	33	struct nvkm_mmu *mmu = vm->mmu;
134fdc1a	34	struct nvkm_mm_node *r = node->mem;
c9582455	35	int big = vma->node->type != mmu->func->spg_shift;
a11c3198 BS	36	u32 offset = vma->node->offset + (delta >> 12);
a11c3198 BS	37	u32 bits = vma->node->type - 12;
c9582455 BS	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);
a11c3198 BS	41	u32 end, len;
a11c3198 BS	42
8f7286f8	43	delta = 0;
134fdc1a	44	while (r) {
a11c3198 BS	45	u64 phys = (u64)r->offset << 12;
	46	u32 num = r->length >> bits;
	47
	48	while (num) {
d0659d32	49	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
a11c3198 BS	50
	51	end = (pte + num);
	52	if (unlikely(end >= max))
	53	end = max;
	54	len = end - pte;
	55
c9582455	56	mmu->func->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	67	}
134fdc1a BS	68	r = r->next;
134fdc1a BS	69	};
a11c3198	70
c9582455	71	mmu->func->flush(vm);
a11c3198 BS	72	}
a11c3198 BS	73
2e2cfbe6	74	static void
42594600 BS	75	nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
42594600 BS	76	struct nvkm_mem *mem)
22b33e8e	77	{
42594600 BS	78	struct nvkm_vm *vm = vma->vm;
42594600 BS	79	struct nvkm_mmu *mmu = vm->mmu;
c9582455	80	int big = vma->node->type != mmu->func->spg_shift;
22b33e8e DA	81	u32 offset = vma->node->offset + (delta >> 12);
	82	u32 bits = vma->node->type - 12;
	83	u32 num = length >> vma->node->type;
c9582455 BS	84	u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde;
	85	u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	86	u32 max = 1 << (mmu->func->pgt_bits - bits);
22b33e8e DA	87	unsigned m, sglen;
	88	u32 end, len;
	89	int i;
	90	struct scatterlist *sg;
	91
	92	for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) {
d0659d32	93	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
22b33e8e DA	94	sglen = sg_dma_len(sg) >> PAGE_SHIFT;
	95
	96	end = pte + sglen;
	97	if (unlikely(end >= max))
	98	end = max;
	99	len = end - pte;
	100
	101	for (m = 0; m < len; m++) {
	102	dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
	103
c9582455	104	mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
22b33e8e DA	105	num--;
	106	pte++;
	107
	108	if (num == 0)
	109	goto finish;
	110	}
	111	if (unlikely(end >= max)) {
	112	pde++;
	113	pte = 0;
	114	}
	115	if (m < sglen) {
	116	for (; m < sglen; m++) {
	117	dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
	118
c9582455	119	mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
22b33e8e DA	120	num--;
	121	pte++;
	122	if (num == 0)
	123	goto finish;
	124	}
	125	}
	126
	127	}
	128	finish:
c9582455	129	mmu->func->flush(vm);
22b33e8e DA	130	}
22b33e8e DA	131
2e2cfbe6	132	static void
42594600 BS	133	nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
42594600 BS	134	struct nvkm_mem *mem)
a11c3198	135	{
42594600 BS	136	struct nvkm_vm *vm = vma->vm;
42594600 BS	137	struct nvkm_mmu *mmu = vm->mmu;
f7b24c42	138	dma_addr_t *list = mem->pages;
c9582455	139	int big = vma->node->type != mmu->func->spg_shift;
a11c3198 BS	140	u32 offset = vma->node->offset + (delta >> 12);
	141	u32 bits = vma->node->type - 12;
	142	u32 num = length >> vma->node->type;
c9582455 BS	143	u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde;
	144	u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	145	u32 max = 1 << (mmu->func->pgt_bits - bits);
a11c3198 BS	146	u32 end, len;
	147
	148	while (num) {
d0659d32	149	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
a11c3198 BS	150
	151	end = (pte + num);
	152	if (unlikely(end >= max))
	153	end = max;
	154	len = end - pte;
	155
c9582455	156	mmu->func->map_sg(vma, pgt, mem, pte, len, list);
a11c3198 BS	157
	158	num -= len;
	159	pte += len;
	160	list += len;
	161	if (unlikely(end >= max)) {
	162	pde++;
	163	pte = 0;
	164	}
	165	}
	166
c9582455	167	mmu->func->flush(vm);
a11c3198 BS	168	}
a11c3198 BS	169
2e2cfbe6	170	void
42594600	171	nvkm_vm_map(struct nvkm_vma vma, struct nvkm_mem node)
2e2cfbe6 BS	172	{
2e2cfbe6 BS	173	if (node->sg)
42594600	174	nvkm_vm_map_sg_table(vma, 0, node->size << 12, node);
2e2cfbe6 BS	175	else
2e2cfbe6 BS	176	if (node->pages)
42594600	177	nvkm_vm_map_sg(vma, 0, node->size << 12, node);
2e2cfbe6	178	else
42594600	179	nvkm_vm_map_at(vma, 0, node);
2e2cfbe6 BS	180	}
2e2cfbe6 BS	181
a11c3198	182	void
42594600	183	nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
a11c3198	184	{
42594600 BS	185	struct nvkm_vm *vm = vma->vm;
42594600 BS	186	struct nvkm_mmu *mmu = vm->mmu;
c9582455	187	int big = vma->node->type != mmu->func->spg_shift;
a11c3198 BS	188	u32 offset = vma->node->offset + (delta >> 12);
	189	u32 bits = vma->node->type - 12;
	190	u32 num = length >> vma->node->type;
c9582455 BS	191	u32 pde = (offset >> mmu->func->pgt_bits) - vm->fpde;
	192	u32 pte = (offset & ((1 << mmu->func->pgt_bits) - 1)) >> bits;
	193	u32 max = 1 << (mmu->func->pgt_bits - bits);
a11c3198 BS	194	u32 end, len;
	195
	196	while (num) {
d0659d32	197	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
a11c3198 BS	198
	199	end = (pte + num);
	200	if (unlikely(end >= max))
	201	end = max;
	202	len = end - pte;
	203
c9582455	204	mmu->func->unmap(vma, pgt, pte, len);
a11c3198 BS	205
	206	num -= len;
	207	pte += len;
	208	if (unlikely(end >= max)) {
	209	pde++;
	210	pte = 0;
	211	}
	212	}
	213
c9582455	214	mmu->func->flush(vm);
a11c3198 BS	215	}
	216
	217	void
42594600	218	nvkm_vm_unmap(struct nvkm_vma *vma)
a11c3198	219	{
42594600	220	nvkm_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
a11c3198 BS	221	}
	222
	223	static void
42594600	224	nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde)
a11c3198	225	{
42594600 BS	226	struct nvkm_mmu *mmu = vm->mmu;
	227	struct nvkm_vm_pgd *vpgd;
	228	struct nvkm_vm_pgt *vpgt;
d0659d32	229	struct nvkm_memory *pgt;
a11c3198 BS	230	u32 pde;
	231
	232	for (pde = fpde; pde <= lpde; pde++) {
	233	vpgt = &vm->pgt[pde - vm->fpde];
3ee01281	234	if (--vpgt->refcount[big])
a11c3198 BS	235	continue;
a11c3198 BS	236
d0659d32 BS	237	pgt = vpgt->mem[big];
d0659d32 BS	238	vpgt->mem[big] = NULL;
3ee01281	239
a11c3198	240	list_for_each_entry(vpgd, &vm->pgd_list, head) {
c9582455	241	mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
a11c3198 BS	242	}
a11c3198 BS	243
77913bbc BS	244	mmu->func->flush(vm);
77913bbc BS	245
d0659d32	246	nvkm_memory_del(&pgt);
a11c3198 BS	247	}
	248	}
	249
	250	static int
42594600	251	nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type)
a11c3198	252	{
42594600 BS	253	struct nvkm_mmu *mmu = vm->mmu;
	254	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
	255	struct nvkm_vm_pgd *vpgd;
c9582455	256	int big = (type != mmu->func->spg_shift);
a11c3198 BS	257	u32 pgt_size;
	258	int ret;
	259
c9582455	260	pgt_size = (1 << (mmu->func->pgt_bits + 12)) >> type;
a11c3198 BS	261	pgt_size *= 8;
a11c3198 BS	262
d0659d32 BS	263	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
d0659d32 BS	264	pgt_size, 0x1000, true, &vpgt->mem[big]);
a11c3198 BS	265	if (unlikely(ret))
	266	return ret;
	267
a11c3198	268	list_for_each_entry(vpgd, &vm->pgd_list, head) {
c9582455	269	mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
a11c3198 BS	270	}
a11c3198 BS	271
1de68568	272	vpgt->refcount[big]++;
a11c3198 BS	273	return 0;
	274	}
	275
	276	int
42594600 BS	277	nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access,
42594600 BS	278	struct nvkm_vma *vma)
a11c3198	279	{
42594600	280	struct nvkm_mmu *mmu = vm->mmu;
a11c3198 BS	281	u32 align = (1 << page_shift) >> 12;
	282	u32 msize = size >> 12;
	283	u32 fpde, lpde, pde;
	284	int ret;
	285
1de68568	286	mutex_lock(&vm->mutex);
42594600 BS	287	ret = nvkm_mm_head(&vm->mm, 0, page_shift, msize, msize, align,
42594600 BS	288	&vma->node);
a11c3198	289	if (unlikely(ret != 0)) {
1de68568	290	mutex_unlock(&vm->mutex);
a11c3198 BS	291	return ret;
	292	}
	293
c9582455 BS	294	fpde = (vma->node->offset >> mmu->func->pgt_bits);
c9582455 BS	295	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
ebb945a9	296
a11c3198	297	for (pde = fpde; pde <= lpde; pde++) {
42594600	298	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
c9582455	299	int big = (vma->node->type != mmu->func->spg_shift);
a11c3198	300
3ee01281 BS	301	if (likely(vpgt->refcount[big])) {
3ee01281 BS	302	vpgt->refcount[big]++;
a11c3198 BS	303	continue;
	304	}
	305
42594600	306	ret = nvkm_vm_map_pgt(vm, pde, vma->node->type);
a11c3198 BS	307	if (ret) {
a11c3198 BS	308	if (pde != fpde)
42594600 BS	309	nvkm_vm_unmap_pgt(vm, big, fpde, pde - 1);
42594600 BS	310	nvkm_mm_free(&vm->mm, &vma->node);
1de68568	311	mutex_unlock(&vm->mutex);
a11c3198 BS	312	return ret;
	313	}
	314	}
1de68568	315	mutex_unlock(&vm->mutex);
a11c3198	316
780194b1	317	vma->vm = NULL;
42594600	318	nvkm_vm_ref(vm, &vma->vm, NULL);
a11c3198 BS	319	vma->offset = (u64)vma->node->offset << 12;
	320	vma->access = access;
	321	return 0;
	322	}
	323
	324	void
42594600	325	nvkm_vm_put(struct nvkm_vma *vma)
a11c3198	326	{
227c95d9 BS	327	struct nvkm_mmu *mmu;
227c95d9 BS	328	struct nvkm_vm *vm;
a11c3198 BS	329	u32 fpde, lpde;
	330
	331	if (unlikely(vma->node == NULL))
	332	return;
227c95d9 BS	333	vm = vma->vm;
	334	mmu = vm->mmu;
	335
c9582455 BS	336	fpde = (vma->node->offset >> mmu->func->pgt_bits);
c9582455 BS	337	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
a11c3198	338
1de68568	339	mutex_lock(&vm->mutex);
c9582455	340	nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde);
42594600	341	nvkm_mm_free(&vm->mm, &vma->node);
1de68568	342	mutex_unlock(&vm->mutex);
780194b1	343
42594600	344	nvkm_vm_ref(NULL, &vma->vm, NULL);
a11c3198 BS	345	}
a11c3198 BS	346
d8e83994 BS	347	int
	348	nvkm_vm_boot(struct nvkm_vm *vm, u64 size)
	349	{
	350	struct nvkm_mmu *mmu = vm->mmu;
d0659d32	351	struct nvkm_memory *pgt;
d8e83994 BS	352	int ret;
d8e83994 BS	353
d0659d32	354	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
c9582455	355	(size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt);
d8e83994 BS	356	if (ret == 0) {
d8e83994 BS	357	vm->pgt[0].refcount[0] = 1;
d0659d32 BS	358	vm->pgt[0].mem[0] = pgt;
d0659d32 BS	359	nvkm_memory_boot(pgt, vm);
d8e83994 BS	360	}
	361
	362	return ret;
	363	}
	364
a11c3198	365	int
42594600	366	nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
1de68568	367	u32 block, struct lock_class_key key, struct nvkm_vm *pvm)
a11c3198	368	{
1de68568	369	static struct lock_class_key _key;
42594600	370	struct nvkm_vm *vm;
a11c3198	371	u64 mm_length = (offset + length) - mm_offset;
a11c3198 BS	372	int ret;
a11c3198 BS	373
cfd376b6	374	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
a11c3198 BS	375	if (!vm)
	376	return -ENOMEM;
	377
1de68568	378	__mutex_init(&vm->mutex, "&vm->mutex", key ? key : &_key);
3863c9bc	379	INIT_LIST_HEAD(&vm->pgd_list);
5ce3bf3c	380	vm->mmu = mmu;
e0bacd2f	381	kref_init(&vm->refcount);
c9582455 BS	382	vm->fpde = offset >> (mmu->func->pgt_bits + 12);
c9582455 BS	383	vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12);
a11c3198	384
d005f51e	385	vm->pgt = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt));
a11c3198 BS	386	if (!vm->pgt) {
	387	kfree(vm);
	388	return -ENOMEM;
	389	}
	390
42594600 BS	391	ret = nvkm_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
42594600 BS	392	block >> 12);
a11c3198	393	if (ret) {
d005f51e	394	vfree(vm->pgt);
a11c3198 BS	395	kfree(vm);
	396	return ret;
	397	}
	398
cfd376b6 MS	399	*pvm = vm;
cfd376b6 MS	400
a11c3198 BS	401	return 0;
	402	}
	403
3863c9bc	404	int
42594600	405	nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset,
1de68568	406	struct lock_class_key key, struct nvkm_vm *pvm)
3863c9bc	407	{
c9582455 BS	408	struct nvkm_mmu *mmu = device->mmu;
	409	if (!mmu->func->create)
	410	return -EINVAL;
	411	return mmu->func->create(mmu, offset, length, mm_offset, key, pvm);
3863c9bc BS	412	}
3863c9bc BS	413
a11c3198	414	static int
42594600	415	nvkm_vm_link(struct nvkm_vm vm, struct nvkm_gpuobj pgd)
a11c3198	416	{
42594600 BS	417	struct nvkm_mmu *mmu = vm->mmu;
42594600 BS	418	struct nvkm_vm_pgd *vpgd;
a11c3198 BS	419	int i;
	420
	421	if (!pgd)
	422	return 0;
	423
	424	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
	425	if (!vpgd)
	426	return -ENOMEM;
	427
f027f491	428	vpgd->obj = pgd;
a11c3198	429
1de68568	430	mutex_lock(&vm->mutex);
3ee01281	431	for (i = vm->fpde; i <= vm->lpde; i++)
c9582455	432	mmu->func->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem);
a11c3198	433	list_add(&vpgd->head, &vm->pgd_list);
1de68568	434	mutex_unlock(&vm->mutex);
a11c3198 BS	435	return 0;
	436	}
	437
	438	static void
42594600	439	nvkm_vm_unlink(struct nvkm_vm vm, struct nvkm_gpuobj mpgd)
a11c3198	440	{
42594600	441	struct nvkm_vm_pgd vpgd, tmp;
a11c3198	442
15ba79ad	443	if (!mpgd)
a11c3198 BS	444	return;
a11c3198 BS	445
1de68568	446	mutex_lock(&vm->mutex);
a11c3198	447	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
15ba79ad	448	if (vpgd->obj == mpgd) {
15ba79ad BS	449	list_del(&vpgd->head);
	450	kfree(vpgd);
	451	break;
	452	}
a11c3198	453	}
1de68568	454	mutex_unlock(&vm->mutex);
a11c3198 BS	455	}
	456
	457	static void
42594600	458	nvkm_vm_del(struct kref *kref)
a11c3198	459	{
42594600 BS	460	struct nvkm_vm vm = container_of(kref, typeof(vm), refcount);
42594600 BS	461	struct nvkm_vm_pgd vpgd, tmp;
a11c3198 BS	462
a11c3198 BS	463	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
42594600	464	nvkm_vm_unlink(vm, vpgd->obj);
a11c3198	465	}
a11c3198	466
42594600	467	nvkm_mm_fini(&vm->mm);
d005f51e	468	vfree(vm->pgt);
a11c3198 BS	469	kfree(vm);
	470	}
	471
	472	int
42594600	473	nvkm_vm_ref(struct nvkm_vm ref, struct nvkm_vm ptr, struct nvkm_gpuobj pgd)
a11c3198	474	{
e0bacd2f	475	if (ref) {
42594600	476	int ret = nvkm_vm_link(ref, pgd);
a11c3198 BS	477	if (ret)
	478	return ret;
	479
e0bacd2f	480	kref_get(&ref->refcount);
a11c3198 BS	481	}
a11c3198 BS	482
e0bacd2f	483	if (*ptr) {
42594600 BS	484	nvkm_vm_unlink(*ptr, pgd);
42594600 BS	485	kref_put(&(*ptr)->refcount, nvkm_vm_del);
a11c3198 BS	486	}
a11c3198 BS	487
e0bacd2f	488	*ptr = ref;
a11c3198 BS	489	return 0;
a11c3198 BS	490	}
c9582455 BS	491
	492	static int
	493	nvkm_mmu_oneinit(struct nvkm_subdev *subdev)
	494	{
	495	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
	496	if (mmu->func->oneinit)
	497	return mmu->func->oneinit(mmu);
	498	return 0;
	499	}
	500
	501	static int
	502	nvkm_mmu_init(struct nvkm_subdev *subdev)
	503	{
	504	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
	505	if (mmu->func->init)
	506	mmu->func->init(mmu);
	507	return 0;
	508	}
	509
	510	static void *
	511	nvkm_mmu_dtor(struct nvkm_subdev *subdev)
	512	{
	513	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
	514	if (mmu->func->dtor)
	515	return mmu->func->dtor(mmu);
	516	return mmu;
	517	}
	518
	519	static const struct nvkm_subdev_func
	520	nvkm_mmu = {
	521	.dtor = nvkm_mmu_dtor,
	522	.oneinit = nvkm_mmu_oneinit,
	523	.init = nvkm_mmu_init,
	524	};
	525
	526	void
	527	nvkm_mmu_ctor(const struct nvkm_mmu_func func, struct nvkm_device device,
	528	int index, struct nvkm_mmu *mmu)
	529	{
56d06fa2	530	nvkm_subdev_ctor(&nvkm_mmu, device, index, &mmu->subdev);
c9582455 BS	531	mmu->func = func;
	532	mmu->limit = func->limit;
	533	mmu->dma_bits = func->dma_bits;
	534	mmu->lpg_shift = func->lpg_shift;
	535	}
	536
	537	int
	538	nvkm_mmu_new_(const struct nvkm_mmu_func func, struct nvkm_device device,
	539	int index, struct nvkm_mmu **pmmu)
	540	{
	541	if (!(pmmu = kzalloc(sizeof(*pmmu), GFP_KERNEL)))
	542	return -ENOMEM;
	543	nvkm_mmu_ctor(func, device, index, *pmmu);
	544	return 0;
	545	}