drm/nouveau/mmu/nv44: implement vmm on top of new base

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/include/nvkm/subdev/mmu.h

@@ -48,6 +48,9 @@ struct nvkm_vm {
 
 	bool bootstrapped;
 	atomic_t engref[NVKM_SUBDEV_NR];
+
+	dma_addr_t null;
+	void *nullp;
 };
 
 int  nvkm_vm_new(struct nvkm_device *, u64 offset, u64 length, u64 mm_offset,

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/Kbuild

@@ -15,3 +15,4 @@ nvkm-y += nvkm/subdev/mmu/gp10b.o
 nvkm-y += nvkm/subdev/mmu/vmm.o
 nvkm-y += nvkm/subdev/mmu/vmmnv04.o
 nvkm-y += nvkm/subdev/mmu/vmmnv41.o
+nvkm-y += nvkm/subdev/mmu/vmmnv44.o

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/base.c

@@ -786,14 +786,11 @@ static void *
 nvkm_mmu_dtor(struct nvkm_subdev *subdev)
 {
 	struct nvkm_mmu *mmu = nvkm_mmu(subdev);
-	void *data = mmu;
 
-	if (mmu->func->dtor)
-		data = mmu->func->dtor(mmu);
 	nvkm_vm_ref(NULL, &mmu->vmm, NULL);
 
 	nvkm_mmu_ptc_fini(mmu);
-	return data;
+	return mmu;
 }
 
 static const struct nvkm_subdev_func

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.c

@@ -21,7 +21,6 @@
  *
  * Authors: Ben Skeggs
  */
-#include "nv04.h"
 #include "vmm.h"
 
 #include <nvif/class.h>
@@ -81,32 +80,6 @@ nv04_mmu_oneinit(struct nvkm_mmu *mmu)
 	return 0;
 }
 
-void *
-nv04_mmu_dtor(struct nvkm_mmu *base)
-{
-	struct nv04_mmu *mmu = nv04_mmu(base);
-	struct nvkm_device *device = mmu->base.subdev.device;
-	if (mmu->base.vmm)
-		nvkm_memory_unref(&mmu->base.vmm->pgt[0].mem[0]);
-	if (mmu->nullp) {
-		dma_free_coherent(device->dev, 16 * 1024,
-				  mmu->nullp, mmu->null);
-	}
-	return mmu;
-}
-
-int
-nv04_mmu_new_(const struct nvkm_mmu_func *func, struct nvkm_device *device,
-	      int index, struct nvkm_mmu **pmmu)
-{
-	struct nv04_mmu *mmu;
-	if (!(mmu = kzalloc(sizeof(*mmu), GFP_KERNEL)))
-		return -ENOMEM;
-	*pmmu = &mmu->base;
-	nvkm_mmu_ctor(func, device, index, &mmu->base);
-	return 0;
-}
-
 const struct nvkm_mmu_func
 nv04_mmu = {
 	.oneinit = nv04_mmu_oneinit,

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv04.h

-#ifndef __NV04_MMU_PRIV__
-#define __NV04_MMU_PRIV__
-#define nv04_mmu(p) container_of((p), struct nv04_mmu, base)
-#include "priv.h"
-
-struct nv04_mmu {
-	struct nvkm_mmu base;
-	dma_addr_t null;
-	void *nullp;
-};
-
-int nv04_mmu_new_(const struct nvkm_mmu_func *, struct nvkm_device *,
-		  int index, struct nvkm_mmu **);
-void *nv04_mmu_dtor(struct nvkm_mmu *);
-
-extern const struct nvkm_mmu_func nv04_mmu;
-#endif

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv44.c

@@ -21,12 +21,13 @@
  *
  * Authors: Ben Skeggs
  */
-#include "nv04.h"
+#include "vmm.h"
 
-#include <core/gpuobj.h>
 #include <core/option.h>
 #include <subdev/timer.h>
 
+#include <nvif/class.h>
+
 #define NV44_GART_SIZE (512 * 1024 * 1024)
 #define NV44_GART_PAGE (  4 * 1024)
 
@@ -84,7 +85,6 @@ static void
 nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt,
 	       struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
 {
-	struct nv04_mmu *mmu = nv04_mmu(vma->vm->mmu);
 	u32 tmp[4];
 	int i;
 
@@ -92,7 +92,7 @@ nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt,
 	if (pte & 3) {
 		u32  max = 4 - (pte & 3);
 		u32 part = (cnt > max) ? max : cnt;
-		nv44_vm_fill(pgt, mmu->null, list, pte, part);
+		nv44_vm_fill(pgt, vma->vm->null, list, pte, part);
 		pte  += part;
 		list += part;
 		cnt  -= part;
@@ -109,20 +109,18 @@ nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt,
 	}
 
 	if (cnt)
-		nv44_vm_fill(pgt, mmu->null, list, pte, cnt);
+		nv44_vm_fill(pgt, vma->vm->null, list, pte, cnt);
 	nvkm_done(pgt);
 }
 
 static void
 nv44_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
 {
-	struct nv04_mmu *mmu = nv04_mmu(vma->vm->mmu);
-
 	nvkm_kmap(pgt);
 	if (pte & 3) {
 		u32  max = 4 - (pte & 3);
 		u32 part = (cnt > max) ? max : cnt;
-		nv44_vm_fill(pgt, mmu->null, NULL, pte, part);
+		nv44_vm_fill(pgt, vma->vm->null, NULL, pte, part);
 		pte  += part;
 		cnt  -= part;
 	}
@@ -136,16 +134,15 @@ nv44_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
 	}
 
 	if (cnt)
-		nv44_vm_fill(pgt, mmu->null, NULL, pte, cnt);
+		nv44_vm_fill(pgt, vma->vm->null, NULL, pte, cnt);
 	nvkm_done(pgt);
 }
 
 static void
 nv44_vm_flush(struct nvkm_vm *vm)
 {
-	struct nv04_mmu *mmu = nv04_mmu(vm->mmu);
-	struct nvkm_device *device = mmu->base.subdev.device;
-	nvkm_wr32(device, 0x100814, mmu->base.limit - NV44_GART_PAGE);
+	struct nvkm_device *device = vm->mmu->subdev.device;
+	nvkm_wr32(device, 0x100814, vm->mmu->limit - NV44_GART_PAGE);
 	nvkm_wr32(device, 0x100808, 0x00000020);
 	nvkm_msec(device, 2000,
 		if (nvkm_rd32(device, 0x100808) & 0x00000001)
@@ -159,38 +156,18 @@ nv44_vm_flush(struct nvkm_vm *vm)
  ******************************************************************************/
 
 static int
-nv44_mmu_oneinit(struct nvkm_mmu *base)
+nv44_mmu_oneinit(struct nvkm_mmu *mmu)
 {
-	struct nv04_mmu *mmu = nv04_mmu(base);
-	struct nvkm_device *device = mmu->base.subdev.device;
-	int ret;
-
-	mmu->nullp = dma_alloc_coherent(device->dev, 16 * 1024,
-					&mmu->null, GFP_KERNEL);
-	if (!mmu->nullp) {
-		nvkm_warn(&mmu->base.subdev, "unable to allocate dummy pages\n");
-		mmu->null = 0;
-	}
-
-	ret = nvkm_vm_create(&mmu->base, 0, NV44_GART_SIZE, 0, 4096, NULL,
-			     &mmu->base.vmm);
-	if (ret)
-		return ret;
-
-	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST,
-			      (NV44_GART_SIZE / NV44_GART_PAGE) * 4,
-			      512 * 1024, true,
-			      &mmu->base.vmm->pgt[0].mem[0]);
-	mmu->base.vmm->pgt[0].refcount[0] = 1;
-	return ret;
+	mmu->vmm->pgt[0].mem[0] = mmu->vmm->pd->pt[0]->memory;
+	mmu->vmm->pgt[0].refcount[0] = 1;
+	return 0;
 }
 
 static void
-nv44_mmu_init(struct nvkm_mmu *base)
+nv44_mmu_init(struct nvkm_mmu *mmu)
 {
-	struct nv04_mmu *mmu = nv04_mmu(base);
-	struct nvkm_device *device = mmu->base.subdev.device;
-	struct nvkm_memory *gart = mmu->base.vmm->pgt[0].mem[0];
+	struct nvkm_device *device = mmu->subdev.device;
+	struct nvkm_memory *gart = mmu->vmm->pgt[0].mem[0];
 	u32 addr;
 
 	/* calculate vram address of this PRAMIN block, object must be
@@ -201,7 +178,7 @@ nv44_mmu_init(struct nvkm_mmu *base)
 	addr -= ((nvkm_memory_addr(gart) >> 19) + 1) << 19;
 
 	nvkm_wr32(device, 0x100850, 0x80000000);
-	nvkm_wr32(device, 0x100818, mmu->null);
+	nvkm_wr32(device, 0x100818, mmu->vmm->null);
 	nvkm_wr32(device, 0x100804, NV44_GART_SIZE);
 	nvkm_wr32(device, 0x100850, 0x00008000);
 	nvkm_mask(device, 0x10008c, 0x00000200, 0x00000200);
@@ -212,7 +189,6 @@ nv44_mmu_init(struct nvkm_mmu *base)
 
 static const struct nvkm_mmu_func
 nv44_mmu = {
-	.dtor = nv04_mmu_dtor,
 	.oneinit = nv44_mmu_oneinit,
 	.init = nv44_mmu_init,
 	.limit = NV44_GART_SIZE,
@@ -223,6 +199,7 @@ nv44_mmu = {
 	.map_sg = nv44_vm_map_sg,
 	.unmap = nv44_vm_unmap,
 	.flush = nv44_vm_flush,
+	.vmm = {{ -1, -1, NVIF_CLASS_VMM_NV04}, nv44_vmm_new, true },
 };
 
 int
@@ -232,5 +209,5 @@ nv44_mmu_new(struct nvkm_device *device, int index, struct nvkm_mmu **pmmu)
 	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true))
 		return nv04_mmu_new(device, index, pmmu);
 
-	return nv04_mmu_new_(&nv44_mmu, device, index, pmmu);
+	return nvkm_mmu_new_(&nv44_mmu, device, index, pmmu);
 }

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/priv.h

@@ -9,7 +9,6 @@ int nvkm_mmu_new_(const struct nvkm_mmu_func *, struct nvkm_device *,
 		  int index, struct nvkm_mmu **);
 
 struct nvkm_mmu_func {
-	void *(*dtor)(struct nvkm_mmu *);
 	int (*oneinit)(struct nvkm_mmu *);
 	void (*init)(struct nvkm_mmu *);
 

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c

@@ -70,6 +70,11 @@ nvkm_vmm_pt_new(const struct nvkm_vmm_desc *desc, bool sparse,
 void
 nvkm_vmm_dtor(struct nvkm_vmm *vmm)
 {
+	if (vmm->nullp) {
+		dma_free_coherent(vmm->mmu->subdev.device->dev, 16 * 1024,
+				  vmm->nullp, vmm->null);
+	}
+
 	if (vmm->pd) {
 		nvkm_mmu_ptc_put(vmm->mmu, true, &vmm->pd->pt[0]);
 		nvkm_vmm_pt_del(&vmm->pd);

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h

@@ -110,4 +110,6 @@ int nv04_vmm_new(struct nvkm_mmu *, u64, u64, void *, u32,
 		 struct lock_class_key *, const char *, struct nvkm_vmm **);
 int nv41_vmm_new(struct nvkm_mmu *, u64, u64, void *, u32,
 		 struct lock_class_key *, const char *, struct nvkm_vmm **);
+int nv44_vmm_new(struct nvkm_mmu *, u64, u64, void *, u32,
+		 struct lock_class_key *, const char *, struct nvkm_vmm **);
 #endif

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmnv44.c

+/*
+ * Copyright 2017 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.
+ */
+#include "vmm.h"
+
+static const struct nvkm_vmm_desc_func
+nv44_vmm_desc_pgt = {
+};
+
+static const struct nvkm_vmm_desc
+nv44_vmm_desc_12[] = {
+	{ PGT, 17, 4, 0x80000, &nv44_vmm_desc_pgt },
+	{}
+};
+
+static const struct nvkm_vmm_func
+nv44_vmm = {
+	.page = {
+		{ 12, &nv44_vmm_desc_12[0], NVKM_VMM_PAGE_HOST },
+		{}
+	}
+};
+
+int
+nv44_vmm_new(struct nvkm_mmu *mmu, u64 addr, u64 size, void *argv, u32 argc,
+	     struct lock_class_key *key, const char *name,
+	     struct nvkm_vmm **pvmm)
+{
+	struct nvkm_subdev *subdev = &mmu->subdev;
+	struct nvkm_vmm *vmm;
+	int ret;
+
+	ret = nv04_vmm_new_(&nv44_vmm, mmu, 0, addr, size,
+			    argv, argc, key, name, &vmm);
+	*pvmm = vmm;
+	if (ret)
+		return ret;
+
+	vmm->nullp = dma_alloc_coherent(subdev->device->dev, 16 * 1024,
+					&vmm->null, GFP_KERNEL);
+	if (!vmm->nullp) {
+		nvkm_warn(subdev, "unable to allocate dummy pages\n");
+		vmm->null = 0;
+	}
+
+	return 0;
+}