[mirror_ubuntu-bionic-kernel.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;
	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;
	list_for_each_entry(r, &node->regions, rl_entry) {
		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;
		}
	}

	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);
		}

		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, 0, &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;
	33	struct nvkm_mmu *mmu = vm->mmu;
	34	struct nvkm_mm_node *r;
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;
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) {
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 BS	67	}
	68	}
	69
c9582455	70	mmu->func->flush(vm);
a11c3198 BS	71	}
a11c3198 BS	72
2e2cfbe6	73	static void
42594600 BS	74	nvkm_vm_map_sg_table(struct nvkm_vma *vma, u64 delta, u64 length,
42594600 BS	75	struct nvkm_mem *mem)
22b33e8e	76	{
42594600 BS	77	struct nvkm_vm *vm = vma->vm;
42594600 BS	78	struct nvkm_mmu *mmu = vm->mmu;
c9582455	79	int big = vma->node->type != mmu->func->spg_shift;
22b33e8e DA	80	u32 offset = vma->node->offset + (delta >> 12);
	81	u32 bits = vma->node->type - 12;
	82	u32 num = length >> vma->node->type;
c9582455 BS	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);
22b33e8e DA	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) {
d0659d32	92	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
22b33e8e DA	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
c9582455	103	mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
22b33e8e DA	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
c9582455	118	mmu->func->map_sg(vma, pgt, mem, pte, 1, &addr);
22b33e8e DA	119	num--;
	120	pte++;
	121	if (num == 0)
	122	goto finish;
	123	}
	124	}
	125
	126	}
	127	finish:
c9582455	128	mmu->func->flush(vm);
22b33e8e DA	129	}
22b33e8e DA	130
2e2cfbe6	131	static void
42594600 BS	132	nvkm_vm_map_sg(struct nvkm_vma *vma, u64 delta, u64 length,
42594600 BS	133	struct nvkm_mem *mem)
a11c3198	134	{
42594600 BS	135	struct nvkm_vm *vm = vma->vm;
42594600 BS	136	struct nvkm_mmu *mmu = vm->mmu;
f7b24c42	137	dma_addr_t *list = mem->pages;
c9582455	138	int big = vma->node->type != mmu->func->spg_shift;
a11c3198 BS	139	u32 offset = vma->node->offset + (delta >> 12);
	140	u32 bits = vma->node->type - 12;
	141	u32 num = length >> vma->node->type;
c9582455 BS	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);
a11c3198 BS	145	u32 end, len;
	146
	147	while (num) {
d0659d32	148	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
a11c3198 BS	149
	150	end = (pte + num);
	151	if (unlikely(end >= max))
	152	end = max;
	153	len = end - pte;
	154
c9582455	155	mmu->func->map_sg(vma, pgt, mem, pte, len, list);
a11c3198 BS	156
	157	num -= len;
	158	pte += len;
	159	list += len;
	160	if (unlikely(end >= max)) {
	161	pde++;
	162	pte = 0;
	163	}
	164	}
	165
c9582455	166	mmu->func->flush(vm);
a11c3198 BS	167	}
a11c3198 BS	168
2e2cfbe6	169	void
42594600	170	nvkm_vm_map(struct nvkm_vma vma, struct nvkm_mem node)
2e2cfbe6 BS	171	{
2e2cfbe6 BS	172	if (node->sg)
42594600	173	nvkm_vm_map_sg_table(vma, 0, node->size << 12, node);
2e2cfbe6 BS	174	else
2e2cfbe6 BS	175	if (node->pages)
42594600	176	nvkm_vm_map_sg(vma, 0, node->size << 12, node);
2e2cfbe6	177	else
42594600	178	nvkm_vm_map_at(vma, 0, node);
2e2cfbe6 BS	179	}
2e2cfbe6 BS	180
a11c3198	181	void
42594600	182	nvkm_vm_unmap_at(struct nvkm_vma *vma, u64 delta, u64 length)
a11c3198	183	{
42594600 BS	184	struct nvkm_vm *vm = vma->vm;
42594600 BS	185	struct nvkm_mmu *mmu = vm->mmu;
c9582455	186	int big = vma->node->type != mmu->func->spg_shift;
a11c3198 BS	187	u32 offset = vma->node->offset + (delta >> 12);
	188	u32 bits = vma->node->type - 12;
	189	u32 num = length >> vma->node->type;
c9582455 BS	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);
a11c3198 BS	193	u32 end, len;
	194
	195	while (num) {
d0659d32	196	struct nvkm_memory *pgt = vm->pgt[pde].mem[big];
a11c3198 BS	197
	198	end = (pte + num);
	199	if (unlikely(end >= max))
	200	end = max;
	201	len = end - pte;
	202
c9582455	203	mmu->func->unmap(vma, pgt, pte, len);
a11c3198 BS	204
	205	num -= len;
	206	pte += len;
	207	if (unlikely(end >= max)) {
	208	pde++;
	209	pte = 0;
	210	}
	211	}
	212
c9582455	213	mmu->func->flush(vm);
a11c3198 BS	214	}
	215
	216	void
42594600	217	nvkm_vm_unmap(struct nvkm_vma *vma)
a11c3198	218	{
42594600	219	nvkm_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
a11c3198 BS	220	}
	221
	222	static void
42594600	223	nvkm_vm_unmap_pgt(struct nvkm_vm *vm, int big, u32 fpde, u32 lpde)
a11c3198	224	{
42594600 BS	225	struct nvkm_mmu *mmu = vm->mmu;
	226	struct nvkm_vm_pgd *vpgd;
	227	struct nvkm_vm_pgt *vpgt;
d0659d32	228	struct nvkm_memory *pgt;
a11c3198 BS	229	u32 pde;
	230
	231	for (pde = fpde; pde <= lpde; pde++) {
	232	vpgt = &vm->pgt[pde - vm->fpde];
3ee01281	233	if (--vpgt->refcount[big])
a11c3198 BS	234	continue;
a11c3198 BS	235
d0659d32 BS	236	pgt = vpgt->mem[big];
d0659d32 BS	237	vpgt->mem[big] = NULL;
3ee01281	238
a11c3198	239	list_for_each_entry(vpgd, &vm->pgd_list, head) {
c9582455	240	mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
a11c3198 BS	241	}
a11c3198 BS	242
d0659d32	243	nvkm_memory_del(&pgt);
a11c3198 BS	244	}
	245	}
	246
	247	static int
42594600	248	nvkm_vm_map_pgt(struct nvkm_vm *vm, u32 pde, u32 type)
a11c3198	249	{
42594600 BS	250	struct nvkm_mmu *mmu = vm->mmu;
	251	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
	252	struct nvkm_vm_pgd *vpgd;
c9582455	253	int big = (type != mmu->func->spg_shift);
a11c3198 BS	254	u32 pgt_size;
	255	int ret;
	256
c9582455	257	pgt_size = (1 << (mmu->func->pgt_bits + 12)) >> type;
a11c3198 BS	258	pgt_size *= 8;
a11c3198 BS	259
d0659d32 BS	260	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
d0659d32 BS	261	pgt_size, 0x1000, true, &vpgt->mem[big]);
a11c3198 BS	262	if (unlikely(ret))
	263	return ret;
	264
a11c3198	265	list_for_each_entry(vpgd, &vm->pgd_list, head) {
c9582455	266	mmu->func->map_pgt(vpgd->obj, pde, vpgt->mem);
a11c3198 BS	267	}
a11c3198 BS	268
1de68568	269	vpgt->refcount[big]++;
a11c3198 BS	270	return 0;
	271	}
	272
	273	int
42594600 BS	274	nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access,
42594600 BS	275	struct nvkm_vma *vma)
a11c3198	276	{
42594600	277	struct nvkm_mmu *mmu = vm->mmu;
a11c3198 BS	278	u32 align = (1 << page_shift) >> 12;
	279	u32 msize = size >> 12;
	280	u32 fpde, lpde, pde;
	281	int ret;
	282
1de68568	283	mutex_lock(&vm->mutex);
42594600 BS	284	ret = nvkm_mm_head(&vm->mm, 0, page_shift, msize, msize, align,
42594600 BS	285	&vma->node);
a11c3198	286	if (unlikely(ret != 0)) {
1de68568	287	mutex_unlock(&vm->mutex);
a11c3198 BS	288	return ret;
	289	}
	290
c9582455 BS	291	fpde = (vma->node->offset >> mmu->func->pgt_bits);
c9582455 BS	292	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
ebb945a9	293
a11c3198	294	for (pde = fpde; pde <= lpde; pde++) {
42594600	295	struct nvkm_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
c9582455	296	int big = (vma->node->type != mmu->func->spg_shift);
a11c3198	297
3ee01281 BS	298	if (likely(vpgt->refcount[big])) {
3ee01281 BS	299	vpgt->refcount[big]++;
a11c3198 BS	300	continue;
	301	}
	302
42594600	303	ret = nvkm_vm_map_pgt(vm, pde, vma->node->type);
a11c3198 BS	304	if (ret) {
a11c3198 BS	305	if (pde != fpde)
42594600 BS	306	nvkm_vm_unmap_pgt(vm, big, fpde, pde - 1);
42594600 BS	307	nvkm_mm_free(&vm->mm, &vma->node);
1de68568	308	mutex_unlock(&vm->mutex);
a11c3198 BS	309	return ret;
	310	}
	311	}
1de68568	312	mutex_unlock(&vm->mutex);
a11c3198	313
780194b1	314	vma->vm = NULL;
42594600	315	nvkm_vm_ref(vm, &vma->vm, NULL);
a11c3198 BS	316	vma->offset = (u64)vma->node->offset << 12;
	317	vma->access = access;
	318	return 0;
	319	}
	320
	321	void
42594600	322	nvkm_vm_put(struct nvkm_vma *vma)
a11c3198	323	{
227c95d9 BS	324	struct nvkm_mmu *mmu;
227c95d9 BS	325	struct nvkm_vm *vm;
a11c3198 BS	326	u32 fpde, lpde;
	327
	328	if (unlikely(vma->node == NULL))
	329	return;
227c95d9 BS	330	vm = vma->vm;
	331	mmu = vm->mmu;
	332
c9582455 BS	333	fpde = (vma->node->offset >> mmu->func->pgt_bits);
c9582455 BS	334	lpde = (vma->node->offset + vma->node->length - 1) >> mmu->func->pgt_bits;
a11c3198	335
1de68568	336	mutex_lock(&vm->mutex);
c9582455	337	nvkm_vm_unmap_pgt(vm, vma->node->type != mmu->func->spg_shift, fpde, lpde);
42594600	338	nvkm_mm_free(&vm->mm, &vma->node);
1de68568	339	mutex_unlock(&vm->mutex);
780194b1	340
42594600	341	nvkm_vm_ref(NULL, &vma->vm, NULL);
a11c3198 BS	342	}
a11c3198 BS	343
d8e83994 BS	344	int
	345	nvkm_vm_boot(struct nvkm_vm *vm, u64 size)
	346	{
	347	struct nvkm_mmu *mmu = vm->mmu;
d0659d32	348	struct nvkm_memory *pgt;
d8e83994 BS	349	int ret;
d8e83994 BS	350
d0659d32	351	ret = nvkm_memory_new(mmu->subdev.device, NVKM_MEM_TARGET_INST,
c9582455	352	(size >> mmu->func->spg_shift) * 8, 0x1000, true, &pgt);
d8e83994 BS	353	if (ret == 0) {
d8e83994 BS	354	vm->pgt[0].refcount[0] = 1;
d0659d32 BS	355	vm->pgt[0].mem[0] = pgt;
d0659d32 BS	356	nvkm_memory_boot(pgt, vm);
d8e83994 BS	357	}
	358
	359	return ret;
	360	}
	361
a11c3198	362	int
42594600	363	nvkm_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
1de68568	364	u32 block, struct lock_class_key key, struct nvkm_vm *pvm)
a11c3198	365	{
1de68568	366	static struct lock_class_key _key;
42594600	367	struct nvkm_vm *vm;
a11c3198	368	u64 mm_length = (offset + length) - mm_offset;
a11c3198 BS	369	int ret;
a11c3198 BS	370
cfd376b6	371	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
a11c3198 BS	372	if (!vm)
	373	return -ENOMEM;
	374
1de68568	375	__mutex_init(&vm->mutex, "&vm->mutex", key ? key : &_key);
3863c9bc	376	INIT_LIST_HEAD(&vm->pgd_list);
5ce3bf3c	377	vm->mmu = mmu;
e0bacd2f	378	kref_init(&vm->refcount);
c9582455 BS	379	vm->fpde = offset >> (mmu->func->pgt_bits + 12);
c9582455 BS	380	vm->lpde = (offset + length - 1) >> (mmu->func->pgt_bits + 12);
a11c3198	381
d005f51e	382	vm->pgt = vzalloc((vm->lpde - vm->fpde + 1) * sizeof(*vm->pgt));
a11c3198 BS	383	if (!vm->pgt) {
	384	kfree(vm);
	385	return -ENOMEM;
	386	}
	387
42594600 BS	388	ret = nvkm_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
42594600 BS	389	block >> 12);
a11c3198	390	if (ret) {
d005f51e	391	vfree(vm->pgt);
a11c3198 BS	392	kfree(vm);
	393	return ret;
	394	}
	395
cfd376b6 MS	396	*pvm = vm;
cfd376b6 MS	397
a11c3198 BS	398	return 0;
	399	}
	400
3863c9bc	401	int
42594600	402	nvkm_vm_new(struct nvkm_device *device, u64 offset, u64 length, u64 mm_offset,
1de68568	403	struct lock_class_key key, struct nvkm_vm *pvm)
3863c9bc	404	{
c9582455 BS	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);
3863c9bc BS	409	}
3863c9bc BS	410
a11c3198	411	static int
42594600	412	nvkm_vm_link(struct nvkm_vm vm, struct nvkm_gpuobj pgd)
a11c3198	413	{
42594600 BS	414	struct nvkm_mmu *mmu = vm->mmu;
42594600 BS	415	struct nvkm_vm_pgd *vpgd;
a11c3198 BS	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
f027f491	425	vpgd->obj = pgd;
a11c3198	426
1de68568	427	mutex_lock(&vm->mutex);
3ee01281	428	for (i = vm->fpde; i <= vm->lpde; i++)
c9582455	429	mmu->func->map_pgt(pgd, i, vm->pgt[i - vm->fpde].mem);
a11c3198	430	list_add(&vpgd->head, &vm->pgd_list);
1de68568	431	mutex_unlock(&vm->mutex);
a11c3198 BS	432	return 0;
	433	}
	434
	435	static void
42594600	436	nvkm_vm_unlink(struct nvkm_vm vm, struct nvkm_gpuobj mpgd)
a11c3198	437	{
42594600	438	struct nvkm_vm_pgd vpgd, tmp;
a11c3198	439
15ba79ad	440	if (!mpgd)
a11c3198 BS	441	return;
a11c3198 BS	442
1de68568	443	mutex_lock(&vm->mutex);
a11c3198	444	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
15ba79ad	445	if (vpgd->obj == mpgd) {
15ba79ad BS	446	list_del(&vpgd->head);
	447	kfree(vpgd);
	448	break;
	449	}
a11c3198	450	}
1de68568	451	mutex_unlock(&vm->mutex);
a11c3198 BS	452	}
	453
	454	static void
42594600	455	nvkm_vm_del(struct kref *kref)
a11c3198	456	{
42594600 BS	457	struct nvkm_vm vm = container_of(kref, typeof(vm), refcount);
42594600 BS	458	struct nvkm_vm_pgd vpgd, tmp;
a11c3198 BS	459
a11c3198 BS	460	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
42594600	461	nvkm_vm_unlink(vm, vpgd->obj);
a11c3198	462	}
a11c3198	463
42594600	464	nvkm_mm_fini(&vm->mm);
d005f51e	465	vfree(vm->pgt);
a11c3198 BS	466	kfree(vm);
	467	}
	468
	469	int
42594600	470	nvkm_vm_ref(struct nvkm_vm ref, struct nvkm_vm ptr, struct nvkm_gpuobj pgd)
a11c3198	471	{
e0bacd2f	472	if (ref) {
42594600	473	int ret = nvkm_vm_link(ref, pgd);
a11c3198 BS	474	if (ret)
	475	return ret;
	476
e0bacd2f	477	kref_get(&ref->refcount);
a11c3198 BS	478	}
a11c3198 BS	479
e0bacd2f	480	if (*ptr) {
42594600 BS	481	nvkm_vm_unlink(*ptr, pgd);
42594600 BS	482	kref_put(&(*ptr)->refcount, nvkm_vm_del);
a11c3198 BS	483	}
a11c3198 BS	484
e0bacd2f	485	*ptr = ref;
a11c3198 BS	486	return 0;
a11c3198 BS	487	}
c9582455 BS	488
	489	static int
	490	nvkm_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
	498	static int
	499	nvkm_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
	507	static void *
	508	nvkm_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
	516	static const struct nvkm_subdev_func
	517	nvkm_mmu = {
	518	.dtor = nvkm_mmu_dtor,
	519	.oneinit = nvkm_mmu_oneinit,
	520	.init = nvkm_mmu_init,
	521	};
	522
	523	void
	524	nvkm_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
	534	int
	535	nvkm_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	}