drm/nouveau/core/mm: allow to create bo vma to fix offset inside vm

This allow to create a bo vma (nvkm_vma) at a fix offset inside a vm.
Usefull when we want to force same virtual address on CPU and GPU.

Signed-off-by: Jérôme Glisse <jglisse@redhat.com>

6 files changed, 129 insertions, 0 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h b/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h
index a1f57ffb4163..edaad5265d52 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/mm.h
@@ -34,6 +34,9 @@ int  nvkm_mm_init(struct nvkm_mm *, u64 offset, u64 length, u64 block);
 int  nvkm_mm_fini(struct nvkm_mm *);
 int  nvkm_mm_head(struct nvkm_mm *, u8 heap, u8 type, u64 size_max,
 		  u64 size_min, u64 align, struct nvkm_mm_node **);
+int nvkm_mm_fix(struct nvkm_mm *mm,
+		u8 type, u64 offset, u64 size,
+		struct nvkm_mm_node **pnode);
 int  nvkm_mm_tail(struct nvkm_mm *, u8 heap, u8 type, u64 size_max,
 		  u64 size_min, u64 align, struct nvkm_mm_node **);
 void nvkm_mm_free(struct nvkm_mm *, struct nvkm_mm_node **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
index 126e2da3fe22..73e7893ea20b 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h
@@ -46,6 +46,8 @@ int  nvkm_vm_ref(struct nvkm_vm *, struct nvkm_vm **, struct nvkm_gpuobj *inst);
 int  nvkm_vm_boot(struct nvkm_vm *, u64 size);
 int  nvkm_vm_get(struct nvkm_vm *, u64 size, u32 page_shift, u32 access,
 		 struct nvkm_vma *);
+int nvkm_vm_get_fix(struct nvkm_vm *vm, u64 offset, u64 size, u32 page_shift,
+		    u32 access, struct nvkm_vma *vma);
 void nvkm_vm_put(struct nvkm_vma *);
 void nvkm_vm_map(struct nvkm_vma *, struct nvkm_mem *);
 void nvkm_vm_map_at(struct nvkm_vma *, u64 offset, struct nvkm_mem *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index e427f80344c4..3009045cdf46 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1611,6 +1611,29 @@ nouveau_bo_vma_add(struct nouveau_bo *nvbo, struct nvkm_vm *vm,
 	return 0;
 }
 
+int
+nouveau_bo_vma_add_fix(struct nouveau_bo *nvbo,
+		       struct nvkm_vm *vm,
+		       struct nvkm_vma *vma,
+		       u64 offset, u64 size)
+{
+	int ret;
+
+	ret = nvkm_vm_get_fix(vm, offset, size, nvbo->page_shift,
+			      NV_MEM_ACCESS_RW, vma);
+	if (ret)
+		return ret;
+
+	if ( nvbo->bo.mem.mem_type != TTM_PL_SYSTEM &&
+	    (nvbo->bo.mem.mem_type == TTM_PL_VRAM ||
+	     nvbo->page_shift != vma->vm->mmu->lpg_shift))
+		nvkm_vm_map(vma, nvbo->bo.mem.mm_node);
+
+	list_add_tail(&vma->head, &nvbo->vma_list);
+	vma->refcount = 1;
+	return 0;
+}
+
 void
 nouveau_bo_vma_del(struct nouveau_bo *nvbo, struct nvkm_vma *vma)
 {
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.h b/drivers/gpu/drm/nouveau/nouveau_bo.h
index b06a5385d6dd..b46f3060b823 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.h
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.h
@@ -94,6 +94,10 @@ nouveau_bo_vma_find(struct nouveau_bo *, struct nvkm_vm *);
 
 int  nouveau_bo_vma_add(struct nouveau_bo *, struct nvkm_vm *,
 			struct nvkm_vma *);
+int nouveau_bo_vma_add_fix(struct nouveau_bo *nvbo,
+			struct nvkm_vm *vm,
+			struct nvkm_vma *vma,
+			u64 offset, u64 size);
 void nouveau_bo_vma_del(struct nouveau_bo *, struct nvkm_vma *);
 
 /* TODO: submit equivalent to TTM generic API upstream? */
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/mm.c b/drivers/gpu/drm/nouveau/nvkm/core/mm.c
index 02c6cff8f665..a38630804139 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/mm.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/mm.c
@@ -157,6 +157,56 @@ nvkm_mm_head(struct nvkm_mm *mm, u8 heap, u8 type, u64 size_max, u64 size_min,
 	return -ENOSPC;
 }
 
+int
+nvkm_mm_fix(struct nvkm_mm *mm,
+	    u8 type, u64 offset, u64 size,
+	    struct nvkm_mm_node **pnode)
+{
+	struct nvkm_mm_node *prev, *this, *next;
+	u64 splitoff;
+	u64 s, e;
+
+	/* Catch overflow */
+	if ((offset + size) < offset)
+		return -ENOSPC;
+
+	list_for_each_entry(this, &mm->free, fl_entry) {
+		s = this->offset;
+		e = this->offset + this->length;
+
+		prev = node(this, prev);
+		if (prev && prev->type != type)
+			s = roundup(s, mm->block_size);
+
+		next = node(this, next);
+		if (next && next->type != type)
+			e = rounddown(e, mm->block_size);
+
+		if (offset > e || (offset + size) < s)
+			continue;
+
+		/* Range partialy covered */
+		if (offset < s || (offset + size) > e)
+			return -ENOSPC;
+
+		splitoff = offset - this->offset;
+		if (splitoff && !region_head(mm, this, splitoff))
+			return -ENOMEM;
+
+		this = region_head(mm, this, size);
+		if (!this)
+			return -ENOMEM;
+
+		this->next = NULL;
+		this->type = type;
+		list_del(&this->fl_entry);
+		*pnode = this;
+		return 0;
+	}
+
+	return -ENOSPC;
+}
+
 static struct nvkm_mm_node *
 region_tail(struct nvkm_mm *mm, struct nvkm_mm_node *a, u64 size)
 {
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
index 4a94f58e7f23..4ee75b9bcd94 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c
@@ -316,6 +316,53 @@ nvkm_vm_get(struct nvkm_vm *vm, u64 size, u32 page_shift, u32 access,
 	return 0;
 }
 
+int
+nvkm_vm_get_fix(struct nvkm_vm *vm, u64 offset, u64 size, u32 page_shift,
+		u32 access, struct nvkm_vma *vma)
+{
+	struct nvkm_mmu *mmu = vm->mmu;
+	u64 moffset = offset >> 12;
+	u64 msize = size >> 12;
+	u32 fpde, lpde, pde;
+	int ret;
+
+	mutex_lock(&vm->mutex);
+	ret = nvkm_mm_fix(&vm->mm, page_shift, moffset, msize, &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)
 {