drm/nouveau/fifo: switch to gpuobj accessor macros

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

8 files changed, 220 insertions, 158 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
index 9024dc893218..c4f48112aa05 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
@@ -65,14 +65,16 @@ g84_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *object)
 	}
 
 	nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
-	nv_wo32(base->eng, addr + 0x00, 0x00190000);
-	nv_wo32(base->eng, addr + 0x04, lower_32_bits(limit));
-	nv_wo32(base->eng, addr + 0x08, lower_32_bits(start));
-	nv_wo32(base->eng, addr + 0x0c, upper_32_bits(limit) << 24 |
-					upper_32_bits(start));
-	nv_wo32(base->eng, addr + 0x10, 0x00000000);
-	nv_wo32(base->eng, addr + 0x14, 0x00000000);
+	nvkm_kmap(base->eng);
+	nvkm_wo32(base->eng, addr + 0x00, 0x00190000);
+	nvkm_wo32(base->eng, addr + 0x04, lower_32_bits(limit));
+	nvkm_wo32(base->eng, addr + 0x08, lower_32_bits(start));
+	nvkm_wo32(base->eng, addr + 0x0c, upper_32_bits(limit) << 24 |
+					  upper_32_bits(start));
+	nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
 	bar->flush(bar);
+	nvkm_done(base->eng);
 	return 0;
 }
 
@@ -119,13 +121,15 @@ g84_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 			return -EBUSY;
 	}
 
-	nv_wo32(base->eng, addr + 0x00, 0x00000000);
-	nv_wo32(base->eng, addr + 0x04, 0x00000000);
-	nv_wo32(base->eng, addr + 0x08, 0x00000000);
-	nv_wo32(base->eng, addr + 0x0c, 0x00000000);
-	nv_wo32(base->eng, addr + 0x10, 0x00000000);
-	nv_wo32(base->eng, addr + 0x14, 0x00000000);
+	nvkm_kmap(base->eng);
+	nvkm_wo32(base->eng, addr + 0x00, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x04, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x08, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x0c, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
 	bar->flush(bar);
+	nvkm_done(base->eng);
 	return 0;
 }
 
@@ -216,23 +220,25 @@ g84_fifo_chan_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 	nv_parent(chan)->object_attach = g84_fifo_object_attach;
 	nv_parent(chan)->object_detach = nv50_fifo_object_detach;
 
-	nv_wo32(base->ramfc, 0x08, lower_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x0c, upper_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x10, lower_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x14, upper_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x3c, 0x003f6078);
-	nv_wo32(base->ramfc, 0x44, 0x01003fff);
-	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
-	nv_wo32(base->ramfc, 0x4c, 0xffffffff);
-	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
-	nv_wo32(base->ramfc, 0x78, 0x00000000);
-	nv_wo32(base->ramfc, 0x7c, 0x30000001);
-	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
-				   (4 << 24) /* SEARCH_FULL */ |
-				   (chan->ramht->gpuobj.node->offset >> 4));
-	nv_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
-	nv_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
+	nvkm_kmap(base->ramfc);
+	nvkm_wo32(base->ramfc, 0x08, lower_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x0c, upper_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x10, lower_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x14, upper_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x3c, 0x003f6078);
+	nvkm_wo32(base->ramfc, 0x44, 0x01003fff);
+	nvkm_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nvkm_wo32(base->ramfc, 0x4c, 0xffffffff);
+	nvkm_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nvkm_wo32(base->ramfc, 0x78, 0x00000000);
+	nvkm_wo32(base->ramfc, 0x7c, 0x30000001);
+	nvkm_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				     (4 << 24) /* SEARCH_FULL */ |
+				     (chan->ramht->gpuobj.node->offset >> 4));
+	nvkm_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
+	nvkm_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
 	bar->flush(bar);
+	nvkm_done(base->ramfc);
 	return 0;
 }
 
@@ -294,20 +300,22 @@ g84_fifo_chan_ctor_ind(struct nvkm_object *parent, struct nvkm_object *engine,
 	ioffset = args->v0.ioffset;
 	ilength = order_base_2(args->v0.ilength / 8);
 
-	nv_wo32(base->ramfc, 0x3c, 0x403f6078);
-	nv_wo32(base->ramfc, 0x44, 0x01003fff);
-	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
-	nv_wo32(base->ramfc, 0x50, lower_32_bits(ioffset));
-	nv_wo32(base->ramfc, 0x54, upper_32_bits(ioffset) | (ilength << 16));
-	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
-	nv_wo32(base->ramfc, 0x78, 0x00000000);
-	nv_wo32(base->ramfc, 0x7c, 0x30000001);
-	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
-				   (4 << 24) /* SEARCH_FULL */ |
-				   (chan->ramht->gpuobj.node->offset >> 4));
-	nv_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
-	nv_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
+	nvkm_kmap(base->ramfc);
+	nvkm_wo32(base->ramfc, 0x3c, 0x403f6078);
+	nvkm_wo32(base->ramfc, 0x44, 0x01003fff);
+	nvkm_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nvkm_wo32(base->ramfc, 0x50, lower_32_bits(ioffset));
+	nvkm_wo32(base->ramfc, 0x54, upper_32_bits(ioffset) | (ilength << 16));
+	nvkm_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nvkm_wo32(base->ramfc, 0x78, 0x00000000);
+	nvkm_wo32(base->ramfc, 0x7c, 0x30000001);
+	nvkm_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				     (4 << 24) /* SEARCH_FULL */ |
+				     (chan->ramht->gpuobj.node->offset >> 4));
+	nvkm_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
+	nvkm_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
 	bar->flush(bar);
+	nvkm_done(base->ramfc);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
index b043d08a35c9..d940d41d1182 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
@@ -86,15 +86,17 @@ gf100_fifo_runlist_update(struct gf100_fifo *fifo)
 	cur = fifo->runlist.mem[fifo->runlist.active];
 	fifo->runlist.active = !fifo->runlist.active;
 
+	nvkm_kmap(cur);
 	for (i = 0, p = 0; i < 128; i++) {
 		struct gf100_fifo_chan *chan = (void *)fifo->base.channel[i];
 		if (chan && chan->state == RUNNING) {
-			nv_wo32(cur, p + 0, i);
-			nv_wo32(cur, p + 4, 0x00000004);
+			nvkm_wo32(cur, p + 0, i);
+			nvkm_wo32(cur, p + 4, 0x00000004);
 			p += 8;
 		}
 	}
 	bar->flush(bar);
+	nvkm_done(cur);
 
 	nvkm_wr32(device, 0x002270, cur->addr >> 12);
 	nvkm_wr32(device, 0x002274, 0x01f00000 | (p >> 3));
@@ -112,6 +114,7 @@ gf100_fifo_context_attach(struct nvkm_object *parent,
 {
 	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gf100_fifo_base *base = (void *)parent->parent;
+	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	struct nvkm_engctx *ectx = (void *)object;
 	u32 addr;
 	int ret;
@@ -137,9 +140,11 @@ gf100_fifo_context_attach(struct nvkm_object *parent,
 		nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
 	}
 
-	nv_wo32(base, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
-	nv_wo32(base, addr + 0x04, upper_32_bits(ectx->vma.offset));
+	nvkm_kmap(engn);
+	nvkm_wo32(engn, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
+	nvkm_wo32(engn, addr + 0x04, upper_32_bits(ectx->vma.offset));
 	bar->flush(bar);
+	nvkm_done(engn);
 	return 0;
 }
 
@@ -150,6 +155,7 @@ gf100_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	struct gf100_fifo *fifo = (void *)parent->engine;
 	struct gf100_fifo_base *base = (void *)parent->parent;
 	struct gf100_fifo_chan *chan = (void *)parent;
+	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
 	struct nvkm_bar *bar = device->bar;
@@ -178,9 +184,11 @@ gf100_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 			return -EBUSY;
 	}
 
-	nv_wo32(base, addr + 0x00, 0x00000000);
-	nv_wo32(base, addr + 0x04, 0x00000000);
+	nvkm_kmap(engn);
+	nvkm_wo32(engn, addr + 0x00, 0x00000000);
+	nvkm_wo32(engn, addr + 0x04, 0x00000000);
 	bar->flush(bar);
+	nvkm_done(engn);
 	return 0;
 }
 
@@ -196,6 +204,7 @@ gf100_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	struct gf100_fifo *fifo = (void *)engine;
 	struct gf100_fifo_base *base = (void *)parent;
 	struct gf100_fifo_chan *chan;
+	struct nvkm_gpuobj *ramfc = &base->base.gpuobj;
 	u64 usermem, ioffset, ilength;
 	int ret, i;
 
@@ -231,26 +240,30 @@ gf100_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	ioffset = args->v0.ioffset;
 	ilength = order_base_2(args->v0.ilength / 8);
 
+	nvkm_kmap(fifo->user.mem);
 	for (i = 0; i < 0x1000; i += 4)
-		nv_wo32(fifo->user.mem, usermem + i, 0x00000000);
-
-	nv_wo32(base, 0x08, lower_32_bits(fifo->user.mem->addr + usermem));
-	nv_wo32(base, 0x0c, upper_32_bits(fifo->user.mem->addr + usermem));
-	nv_wo32(base, 0x10, 0x0000face);
-	nv_wo32(base, 0x30, 0xfffff902);
-	nv_wo32(base, 0x48, lower_32_bits(ioffset));
-	nv_wo32(base, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
-	nv_wo32(base, 0x54, 0x00000002);
-	nv_wo32(base, 0x84, 0x20400000);
-	nv_wo32(base, 0x94, 0x30000001);
-	nv_wo32(base, 0x9c, 0x00000100);
-	nv_wo32(base, 0xa4, 0x1f1f1f1f);
-	nv_wo32(base, 0xa8, 0x1f1f1f1f);
-	nv_wo32(base, 0xac, 0x0000001f);
-	nv_wo32(base, 0xb8, 0xf8000000);
-	nv_wo32(base, 0xf8, 0x10003080); /* 0x002310 */
-	nv_wo32(base, 0xfc, 0x10000010); /* 0x002350 */
+		nvkm_wo32(fifo->user.mem, usermem + i, 0x00000000);
+	nvkm_done(fifo->user.mem);
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, 0x08, lower_32_bits(fifo->user.mem->addr + usermem));
+	nvkm_wo32(ramfc, 0x0c, upper_32_bits(fifo->user.mem->addr + usermem));
+	nvkm_wo32(ramfc, 0x10, 0x0000face);
+	nvkm_wo32(ramfc, 0x30, 0xfffff902);
+	nvkm_wo32(ramfc, 0x48, lower_32_bits(ioffset));
+	nvkm_wo32(ramfc, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
+	nvkm_wo32(ramfc, 0x54, 0x00000002);
+	nvkm_wo32(ramfc, 0x84, 0x20400000);
+	nvkm_wo32(ramfc, 0x94, 0x30000001);
+	nvkm_wo32(ramfc, 0x9c, 0x00000100);
+	nvkm_wo32(ramfc, 0xa4, 0x1f1f1f1f);
+	nvkm_wo32(ramfc, 0xa8, 0x1f1f1f1f);
+	nvkm_wo32(ramfc, 0xac, 0x0000001f);
+	nvkm_wo32(ramfc, 0xb8, 0xf8000000);
+	nvkm_wo32(ramfc, 0xf8, 0x10003080); /* 0x002310 */
+	nvkm_wo32(ramfc, 0xfc, 0x10000010); /* 0x002350 */
 	bar->flush(bar);
+	nvkm_done(ramfc);
 	return 0;
 }
 
@@ -341,10 +354,12 @@ gf100_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	nv_wo32(base, 0x0200, lower_32_bits(base->pgd->addr));
-	nv_wo32(base, 0x0204, upper_32_bits(base->pgd->addr));
-	nv_wo32(base, 0x0208, 0xffffffff);
-	nv_wo32(base, 0x020c, 0x000000ff);
+	nvkm_kmap(&base->base.gpuobj);
+	nvkm_wo32(&base->base.gpuobj, 0x0200, lower_32_bits(base->pgd->addr));
+	nvkm_wo32(&base->base.gpuobj, 0x0204, upper_32_bits(base->pgd->addr));
+	nvkm_wo32(&base->base.gpuobj, 0x0208, 0xffffffff);
+	nvkm_wo32(&base->base.gpuobj, 0x020c, 0x000000ff);
+	nvkm_done(&base->base.gpuobj);
 
 	ret = nvkm_vm_ref(nvkm_client(parent)->vm, &base->vm, base->pgd);
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
index ce12e4e12b22..9a1ecef24813 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -106,15 +106,17 @@ gk104_fifo_runlist_update(struct gk104_fifo *fifo, u32 engine)
 	cur = engn->runlist[engn->cur_runlist];
 	engn->cur_runlist = !engn->cur_runlist;
 
+	nvkm_kmap(cur);
 	for (i = 0, p = 0; i < fifo->base.max; i++) {
 		struct gk104_fifo_chan *chan = (void *)fifo->base.channel[i];
 		if (chan && chan->state == RUNNING && chan->engine == engine) {
-			nv_wo32(cur, p + 0, i);
-			nv_wo32(cur, p + 4, 0x00000000);
+			nvkm_wo32(cur, p + 0, i);
+			nvkm_wo32(cur, p + 4, 0x00000000);
 			p += 8;
 		}
 	}
 	bar->flush(bar);
+	nvkm_done(cur);
 
 	nvkm_wr32(device, 0x002270, cur->addr >> 12);
 	nvkm_wr32(device, 0x002274, (engine << 20) | (p >> 3));
@@ -132,6 +134,7 @@ gk104_fifo_context_attach(struct nvkm_object *parent,
 {
 	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gk104_fifo_base *base = (void *)parent->parent;
+	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	struct nvkm_engctx *ectx = (void *)object;
 	u32 addr;
 	int ret;
@@ -161,9 +164,11 @@ gk104_fifo_context_attach(struct nvkm_object *parent,
 		nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
 	}
 
-	nv_wo32(base, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
-	nv_wo32(base, addr + 0x04, upper_32_bits(ectx->vma.offset));
+	nvkm_kmap(engn);
+	nvkm_wo32(engn, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
+	nvkm_wo32(engn, addr + 0x04, upper_32_bits(ectx->vma.offset));
 	bar->flush(bar);
+	nvkm_done(engn);
 	return 0;
 }
 
@@ -195,6 +200,7 @@ gk104_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	struct nvkm_bar *bar = nvkm_bar(parent);
 	struct gk104_fifo_base *base = (void *)parent->parent;
 	struct gk104_fifo_chan *chan = (void *)parent;
+	struct nvkm_gpuobj *engn = &base->base.gpuobj;
 	u32 addr;
 	int ret;
 
@@ -216,9 +222,11 @@ gk104_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 		return ret;
 
 	if (addr) {
-		nv_wo32(base, addr + 0x00, 0x00000000);
-		nv_wo32(base, addr + 0x04, 0x00000000);
+		nvkm_kmap(engn);
+		nvkm_wo32(engn, addr + 0x00, 0x00000000);
+		nvkm_wo32(engn, addr + 0x04, 0x00000000);
 		bar->flush(bar);
+		nvkm_done(engn);
 	}
 
 	return 0;
@@ -237,6 +245,7 @@ gk104_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	struct gk104_fifo_base *base = (void *)parent;
 	struct gk104_fifo_chan *chan;
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
+	struct nvkm_gpuobj *ramfc = &base->base.gpuobj;
 	u64 usermem, ioffset, ilength;
 	int ret, i;
 
@@ -282,24 +291,28 @@ gk104_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	ioffset = args->v0.ioffset;
 	ilength = order_base_2(args->v0.ilength / 8);
 
+	nvkm_kmap(fifo->user.mem);
 	for (i = 0; i < 0x200; i += 4)
-		nv_wo32(fifo->user.mem, usermem + i, 0x00000000);
-
-	nv_wo32(base, 0x08, lower_32_bits(fifo->user.mem->addr + usermem));
-	nv_wo32(base, 0x0c, upper_32_bits(fifo->user.mem->addr + usermem));
-	nv_wo32(base, 0x10, 0x0000face);
-	nv_wo32(base, 0x30, 0xfffff902);
-	nv_wo32(base, 0x48, lower_32_bits(ioffset));
-	nv_wo32(base, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
-	nv_wo32(base, 0x84, 0x20400000);
-	nv_wo32(base, 0x94, 0x30000001);
-	nv_wo32(base, 0x9c, 0x00000100);
-	nv_wo32(base, 0xac, 0x0000001f);
-	nv_wo32(base, 0xe8, chan->base.chid);
-	nv_wo32(base, 0xb8, 0xf8000000);
-	nv_wo32(base, 0xf8, 0x10003080); /* 0x002310 */
-	nv_wo32(base, 0xfc, 0x10000010); /* 0x002350 */
+		nvkm_wo32(fifo->user.mem, usermem + i, 0x00000000);
+	nvkm_done(fifo->user.mem);
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, 0x08, lower_32_bits(fifo->user.mem->addr + usermem));
+	nvkm_wo32(ramfc, 0x0c, upper_32_bits(fifo->user.mem->addr + usermem));
+	nvkm_wo32(ramfc, 0x10, 0x0000face);
+	nvkm_wo32(ramfc, 0x30, 0xfffff902);
+	nvkm_wo32(ramfc, 0x48, lower_32_bits(ioffset));
+	nvkm_wo32(ramfc, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
+	nvkm_wo32(ramfc, 0x84, 0x20400000);
+	nvkm_wo32(ramfc, 0x94, 0x30000001);
+	nvkm_wo32(ramfc, 0x9c, 0x00000100);
+	nvkm_wo32(ramfc, 0xac, 0x0000001f);
+	nvkm_wo32(ramfc, 0xe8, chan->base.chid);
+	nvkm_wo32(ramfc, 0xb8, 0xf8000000);
+	nvkm_wo32(ramfc, 0xf8, 0x10003080); /* 0x002310 */
+	nvkm_wo32(ramfc, 0xfc, 0x10000010); /* 0x002350 */
 	bar->flush(bar);
+	nvkm_done(ramfc);
 	return 0;
 }
 
@@ -387,10 +400,12 @@ gk104_fifo_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	nv_wo32(base, 0x0200, lower_32_bits(base->pgd->addr));
-	nv_wo32(base, 0x0204, upper_32_bits(base->pgd->addr));
-	nv_wo32(base, 0x0208, 0xffffffff);
-	nv_wo32(base, 0x020c, 0x000000ff);
+	nvkm_kmap(&base->base.gpuobj);
+	nvkm_wo32(&base->base.gpuobj, 0x0200, lower_32_bits(base->pgd->addr));
+	nvkm_wo32(&base->base.gpuobj, 0x0204, upper_32_bits(base->pgd->addr));
+	nvkm_wo32(&base->base.gpuobj, 0x0208, 0xffffffff);
+	nvkm_wo32(&base->base.gpuobj, 0x020c, 0x000000ff);
+	nvkm_done(&base->base.gpuobj);
 
 	ret = nvkm_vm_ref(nvkm_client(parent)->vm, &base->vm, base->pgd);
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
index 5eca5de292f6..d5346636b7e3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
@@ -142,16 +142,18 @@ nv04_fifo_chan_ctor(struct nvkm_object *parent,
 	nv_parent(chan)->context_attach = nv04_fifo_context_attach;
 	chan->ramfc = chan->base.chid * 32;
 
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x08, chan->base.pushgpu->addr >> 4);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x10,
+	nvkm_kmap(fifo->ramfc);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x08, chan->base.pushgpu->addr >> 4);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x10,
 			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
 			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
 #ifdef __BIG_ENDIAN
 			     NV_PFIFO_CACHE1_BIG_ENDIAN |
 #endif
 			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_done(fifo->ramfc);
 	return 0;
 }
 
@@ -162,9 +164,11 @@ nv04_fifo_chan_dtor(struct nvkm_object *object)
 	struct nv04_fifo_chan *chan = (void *)object;
 	struct ramfc_desc *c = fifo->ramfc_desc;
 
+	nvkm_kmap(fifo->ramfc);
 	do {
-		nv_wo32(fifo->ramfc, chan->ramfc + c->ctxp, 0x00000000);
+		nvkm_wo32(fifo->ramfc, chan->ramfc + c->ctxp, 0x00000000);
 	} while ((++c)->bits);
+	nvkm_done(fifo->ramfc);
 
 	nvkm_fifo_channel_destroy(&chan->base);
 }
@@ -217,8 +221,8 @@ nv04_fifo_chan_fini(struct nvkm_object *object, bool suspend)
 			u32 rm = ((1ULL << c->bits) - 1) << c->regs;
 			u32 cm = ((1ULL << c->bits) - 1) << c->ctxs;
 			u32 rv = (nvkm_rd32(device, c->regp) &  rm) >> c->regs;
-			u32 cv = (nv_ro32(fctx, c->ctxp + data) & ~cm);
-			nv_wo32(fctx, c->ctxp + data, cv | (rv << c->ctxs));
+			u32 cv = (nvkm_ro32(fctx, c->ctxp + data) & ~cm);
+			nvkm_wo32(fctx, c->ctxp + data, cv | (rv << c->ctxs));
 		} while ((++c)->bits);
 
 		c = fifo->ramfc_desc;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
index 2c38d6f36fe7..5a7b11f92684 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
@@ -86,16 +86,18 @@ nv10_fifo_chan_ctor(struct nvkm_object *parent,
 	nv_parent(chan)->context_attach = nv04_fifo_context_attach;
 	chan->ramfc = chan->base.chid * 32;
 
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x14,
+	nvkm_kmap(fifo->ramfc);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x14,
 			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
 			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
 #ifdef __BIG_ENDIAN
 			     NV_PFIFO_CACHE1_BIG_ENDIAN |
 #endif
 			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_done(fifo->ramfc);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
index 55857665180c..537bb7229e3d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
@@ -93,16 +93,18 @@ nv17_fifo_chan_ctor(struct nvkm_object *parent,
 	nv_parent(chan)->context_attach = nv04_fifo_context_attach;
 	chan->ramfc = chan->base.chid * 64;
 
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x14,
+	nvkm_kmap(fifo->ramfc);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x14,
 			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
 			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
 #ifdef __BIG_ENDIAN
 			     NV_PFIFO_CACHE1_BIG_ENDIAN |
 #endif
 			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nvkm_done(fifo->ramfc);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
index b6c7af527d6a..27f07d1db85f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
@@ -130,7 +130,9 @@ nv40_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *engctx)
 
 	if ((nvkm_rd32(device, 0x003204) & fifo->base.max) == chan->base.chid)
 		nvkm_wr32(device, reg, nv_engctx(engctx)->addr);
-	nv_wo32(fifo->ramfc, chan->ramfc + ctx, nv_engctx(engctx)->addr);
+	nvkm_kmap(fifo->ramfc);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + ctx, nv_engctx(engctx)->addr);
+	nvkm_done(fifo->ramfc);
 
 	nvkm_mask(device, 0x002500, 0x00000001, 0x00000001);
 	spin_unlock_irqrestore(&fifo->base.lock, flags);
@@ -167,7 +169,9 @@ nv40_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 
 	if ((nvkm_rd32(device, 0x003204) & fifo->base.max) == chan->base.chid)
 		nvkm_wr32(device, reg, 0x00000000);
-	nv_wo32(fifo->ramfc, chan->ramfc + ctx, 0x00000000);
+	nvkm_kmap(fifo->ramfc);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + ctx, 0x00000000);
+	nvkm_done(fifo->ramfc);
 
 	nvkm_mask(device, 0x002500, 0x00000001, 0x00000001);
 	spin_unlock_irqrestore(&fifo->base.lock, flags);
@@ -212,17 +216,19 @@ nv40_fifo_chan_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
 	nv_parent(chan)->object_detach = nv04_fifo_object_detach;
 	chan->ramfc = chan->base.chid * 128;
 
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x18, 0x30000000 |
+	nvkm_kmap(fifo->ramfc);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x00, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x04, args->v0.offset);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x18, 0x30000000 |
 			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
 			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
 #ifdef __BIG_ENDIAN
 			     NV_PFIFO_CACHE1_BIG_ENDIAN |
 #endif
 			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
-	nv_wo32(fifo->ramfc, chan->ramfc + 0x3c, 0x0001ffff);
+	nvkm_wo32(fifo->ramfc, chan->ramfc + 0x3c, 0x0001ffff);
+	nvkm_done(fifo->ramfc);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
index 920b3c91b9ed..24154778c01e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
@@ -49,12 +49,13 @@ nv50_fifo_playlist_update_locked(struct nv50_fifo *fifo)
 	cur = fifo->playlist[fifo->cur_playlist];
 	fifo->cur_playlist = !fifo->cur_playlist;
 
+	nvkm_kmap(cur);
 	for (i = fifo->base.min, p = 0; i < fifo->base.max; i++) {
 		if (nvkm_rd32(device, 0x002600 + (i * 4)) & 0x80000000)
-			nv_wo32(cur, p++ * 4, i);
+			nvkm_wo32(cur, p++ * 4, i);
 	}
-
 	bar->flush(bar);
+	nvkm_done(cur);
 
 	nvkm_wr32(device, 0x0032f4, cur->addr >> 12);
 	nvkm_wr32(device, 0x0032ec, p);
@@ -88,14 +89,17 @@ nv50_fifo_context_attach(struct nvkm_object *parent, struct nvkm_object *object)
 	}
 
 	nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
-	nv_wo32(base->eng, addr + 0x00, 0x00190000);
-	nv_wo32(base->eng, addr + 0x04, lower_32_bits(limit));
-	nv_wo32(base->eng, addr + 0x08, lower_32_bits(start));
-	nv_wo32(base->eng, addr + 0x0c, upper_32_bits(limit) << 24 |
-					upper_32_bits(start));
-	nv_wo32(base->eng, addr + 0x10, 0x00000000);
-	nv_wo32(base->eng, addr + 0x14, 0x00000000);
+
+	nvkm_kmap(base->eng);
+	nvkm_wo32(base->eng, addr + 0x00, 0x00190000);
+	nvkm_wo32(base->eng, addr + 0x04, lower_32_bits(limit));
+	nvkm_wo32(base->eng, addr + 0x08, lower_32_bits(start));
+	nvkm_wo32(base->eng, addr + 0x0c, upper_32_bits(limit) << 24 |
+					  upper_32_bits(start));
+	nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
+	nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
 	bar->flush(bar);
+	nvkm_done(base->eng);
 	return 0;
 }
 
@@ -148,13 +152,15 @@ nv50_fifo_context_detach(struct nvkm_object *parent, bool suspend,
 	nvkm_wr32(device, 0x00b860, me);
 
 	if (ret == 0) {
-		nv_wo32(base->eng, addr + 0x00, 0x00000000);
-		nv_wo32(base->eng, addr + 0x04, 0x00000000);
-		nv_wo32(base->eng, addr + 0x08, 0x00000000);
-		nv_wo32(base->eng, addr + 0x0c, 0x00000000);
-		nv_wo32(base->eng, addr + 0x10, 0x00000000);
-		nv_wo32(base->eng, addr + 0x14, 0x00000000);
+		nvkm_kmap(base->eng);
+		nvkm_wo32(base->eng, addr + 0x00, 0x00000000);
+		nvkm_wo32(base->eng, addr + 0x04, 0x00000000);
+		nvkm_wo32(base->eng, addr + 0x08, 0x00000000);
+		nvkm_wo32(base->eng, addr + 0x0c, 0x00000000);
+		nvkm_wo32(base->eng, addr + 0x10, 0x00000000);
+		nvkm_wo32(base->eng, addr + 0x14, 0x00000000);
 		bar->flush(bar);
+		nvkm_done(base->eng);
 	}
 
 	return ret;
@@ -234,21 +240,23 @@ nv50_fifo_chan_ctor_dma(struct nvkm_object *parent, struct nvkm_object *engine,
 	if (ret)
 		return ret;
 
-	nv_wo32(base->ramfc, 0x08, lower_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x0c, upper_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x10, lower_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x14, upper_32_bits(args->v0.offset));
-	nv_wo32(base->ramfc, 0x3c, 0x003f6078);
-	nv_wo32(base->ramfc, 0x44, 0x01003fff);
-	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
-	nv_wo32(base->ramfc, 0x4c, 0xffffffff);
-	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
-	nv_wo32(base->ramfc, 0x78, 0x00000000);
-	nv_wo32(base->ramfc, 0x7c, 0x30000001);
-	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
-				   (4 << 24) /* SEARCH_FULL */ |
-				   (chan->ramht->gpuobj.node->offset >> 4));
+	nvkm_kmap(base->ramfc);
+	nvkm_wo32(base->ramfc, 0x08, lower_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x0c, upper_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x10, lower_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x14, upper_32_bits(args->v0.offset));
+	nvkm_wo32(base->ramfc, 0x3c, 0x003f6078);
+	nvkm_wo32(base->ramfc, 0x44, 0x01003fff);
+	nvkm_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nvkm_wo32(base->ramfc, 0x4c, 0xffffffff);
+	nvkm_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nvkm_wo32(base->ramfc, 0x78, 0x00000000);
+	nvkm_wo32(base->ramfc, 0x7c, 0x30000001);
+	nvkm_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				     (4 << 24) /* SEARCH_FULL */ |
+				     (chan->ramht->gpuobj.node->offset >> 4));
 	bar->flush(bar);
+	nvkm_done(base->ramfc);
 	return 0;
 }
 
@@ -300,18 +308,20 @@ nv50_fifo_chan_ctor_ind(struct nvkm_object *parent, struct nvkm_object *engine,
 	ioffset = args->v0.ioffset;
 	ilength = order_base_2(args->v0.ilength / 8);
 
-	nv_wo32(base->ramfc, 0x3c, 0x403f6078);
-	nv_wo32(base->ramfc, 0x44, 0x01003fff);
-	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
-	nv_wo32(base->ramfc, 0x50, lower_32_bits(ioffset));
-	nv_wo32(base->ramfc, 0x54, upper_32_bits(ioffset) | (ilength << 16));
-	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
-	nv_wo32(base->ramfc, 0x78, 0x00000000);
-	nv_wo32(base->ramfc, 0x7c, 0x30000001);
-	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
-				   (4 << 24) /* SEARCH_FULL */ |
-				   (chan->ramht->gpuobj.node->offset >> 4));
+	nvkm_kmap(base->ramfc);
+	nvkm_wo32(base->ramfc, 0x3c, 0x403f6078);
+	nvkm_wo32(base->ramfc, 0x44, 0x01003fff);
+	nvkm_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nvkm_wo32(base->ramfc, 0x50, lower_32_bits(ioffset));
+	nvkm_wo32(base->ramfc, 0x54, upper_32_bits(ioffset) | (ilength << 16));
+	nvkm_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nvkm_wo32(base->ramfc, 0x78, 0x00000000);
+	nvkm_wo32(base->ramfc, 0x7c, 0x30000001);
+	nvkm_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				     (4 << 24) /* SEARCH_FULL */ |
+				     (chan->ramht->gpuobj.node->offset >> 4));
 	bar->flush(bar);
+	nvkm_done(base->ramfc);
 	return 0;
 }