drm/nouveau/fifo: add new engine context handling

Builds on the context tracking that was added earlier.

- marks engine context PTEs as 'priv' where possible

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>

39 files changed, 382 insertions, 930 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
index 1616a329db9e..da8abcbeb965 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
@@ -8,11 +8,6 @@ struct nvkm_fault_data;
 
 #define NVKM_FIFO_ENGN_NR 16
 
-struct nvkm_fifo_engn {
-	struct nvkm_object *object;
-	int refcount;
-};
-
 struct nvkm_chan {
 	const struct nvkm_chan_func *func;
 	char name[64];
@@ -41,13 +36,10 @@ struct nvkm_chan {
 
 	struct list_head cctxs;
 
-	struct nvkm_fifo *fifo;
 	struct nvkm_object object;
 
 	struct list_head head;
 	struct nvkm_gpuobj *push;
-
-	struct nvkm_fifo_engn engn[NVKM_FIFO_ENGN_NR];
 };
 
 struct nvkm_chan *nvkm_chan_get_chid(struct nvkm_engine *, int id, unsigned long *irqflags);
@@ -78,7 +70,6 @@ struct nvkm_fifo {
 		struct nvkm_vma *bar1;
 	} userd;
 
-	int nr;
 	spinlock_t lock;
 	struct mutex mutex;
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
index c77af1b3ffa0..9362afe37f3e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
@@ -318,7 +318,7 @@ int
 nvkm_fifo_ctor(const struct nvkm_fifo_func *func, struct nvkm_device *device,
 	       enum nvkm_subdev_type type, int inst, struct nvkm_fifo *fifo)
 {
-	int ret, nr;
+	int ret;
 
 	fifo->func = func;
 	INIT_LIST_HEAD(&fifo->runqs);
@@ -335,9 +335,6 @@ nvkm_fifo_ctor(const struct nvkm_fifo_func *func, struct nvkm_device *device,
 	if (ret)
 		return ret;
 
-	nr = func->chid_nr(fifo);
-	fifo->nr = nr;
-
 	if (func->nonstall) {
 		ret = nvkm_event_init(func->nonstall, &fifo->engine.subdev, 1, 1,
 				      &fifo->nonstall.event);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c
index ac42dfc9d4cc..ea53fb3d5d06 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.c
@@ -25,6 +25,7 @@
 #include "runl.h"
 #include "priv.h"
 
+#include <core/gpuobj.h>
 #include <subdev/mmu.h>
 
 static void
@@ -37,6 +38,7 @@ nvkm_cgrp_ectx_put(struct nvkm_cgrp *cgrp, struct nvkm_ectx **pectx)
 
 		if (refcount_dec_and_test(&ectx->refs)) {
 			CGRP_TRACE(cgrp, "dtor ectx %d[%s]", engn->id, engn->engine->subdev.name);
+			nvkm_object_del(&ectx->object);
 			list_del(&ectx->head);
 			kfree(ectx);
 		}
@@ -49,6 +51,11 @@ static int
 nvkm_cgrp_ectx_get(struct nvkm_cgrp *cgrp, struct nvkm_engn *engn, struct nvkm_ectx **pectx,
 		   struct nvkm_chan *chan, struct nvkm_client *client)
 {
+	struct nvkm_engine *engine = engn->engine;
+	struct nvkm_oclass cclass = {
+		.client = client,
+		.engine = engine,
+	};
 	struct nvkm_ectx *ectx;
 	int ret = 0;
 
@@ -67,7 +74,18 @@ nvkm_cgrp_ectx_get(struct nvkm_cgrp *cgrp, struct nvkm_engn *engn, struct nvkm_e
 
 	ectx->engn = engn;
 	refcount_set(&ectx->refs, 1);
+	refcount_set(&ectx->uses, 0);
 	list_add_tail(&ectx->head, &cgrp->ectxs);
+
+	/* Allocate the HW structures. */
+	if (engine->func->fifo.cclass)
+		ret = engine->func->fifo.cclass(chan, &cclass, &ectx->object);
+	else if (engine->func->cclass)
+		ret = nvkm_object_new_(engine->func->cclass, &cclass, NULL, 0, &ectx->object);
+
+	if (ret)
+		nvkm_cgrp_ectx_put(cgrp, pectx);
+
 	return ret;
 }
 
@@ -81,6 +99,8 @@ nvkm_cgrp_vctx_put(struct nvkm_cgrp *cgrp, struct nvkm_vctx **pvctx)
 
 		if (refcount_dec_and_test(&vctx->refs)) {
 			CGRP_TRACE(cgrp, "dtor vctx %d[%s]", engn->id, engn->engine->subdev.name);
+			nvkm_vmm_put(vctx->vmm, &vctx->vma);
+			nvkm_gpuobj_del(&vctx->inst);
 
 			nvkm_cgrp_ectx_put(cgrp, &vctx->ectx);
 			if (vctx->vmm) {
@@ -130,6 +150,21 @@ nvkm_cgrp_vctx_get(struct nvkm_cgrp *cgrp, struct nvkm_engn *engn, struct nvkm_c
 	vctx->vmm = nvkm_vmm_ref(chan->vmm);
 	refcount_set(&vctx->refs, 1);
 	list_add_tail(&vctx->head, &cgrp->vctxs);
+
+	/* MMU on some GPUs needs to know engine usage for TLB invalidation. */
+	if (vctx->vmm)
+		atomic_inc(&vctx->vmm->engref[engn->engine->subdev.type]);
+
+	/* Allocate the HW structures. */
+	if (engn->func->bind) {
+		ret = nvkm_object_bind(vctx->ectx->object, NULL, 0, &vctx->inst);
+		if (ret == 0 && engn->func->ctor)
+			ret = engn->func->ctor(engn, vctx);
+	}
+
+	if (ret)
+		nvkm_cgrp_vctx_put(cgrp, pvctx);
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h
index be9fbe9fab37..5f6abd59a6ff 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/cgrp.h
@@ -10,12 +10,18 @@ struct nvkm_vctx {
 	struct nvkm_vmm *vmm;
 	refcount_t refs;
 
+	struct nvkm_gpuobj *inst;
+	struct nvkm_vma *vma;
+
 	struct list_head head;
 };
 
 struct nvkm_ectx {
 	struct nvkm_engn *engn;
 	refcount_t refs;
+	refcount_t uses;
+
+	struct nvkm_object *object;
 
 	struct list_head head;
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
index 00680b3ca9ab..a35439dda054 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.c
@@ -28,7 +28,6 @@
 #include "runl.h"
 #include "priv.h"
 
-#include <core/client.h>
 #include <core/oproxy.h>
 #include <core/ramht.h>
 #include <subdev/mmu.h>
@@ -46,94 +45,15 @@ struct nvkm_fifo_chan_object {
 	int hash;
 };
 
-static struct nvkm_fifo_engn *
-nvkm_fifo_chan_engn(struct nvkm_fifo_chan *chan, struct nvkm_engine *engine)
-{
-	int engi = chan->fifo->func->engine_id(chan->fifo, engine);
-	if (engi >= 0)
-		return &chan->engn[engi];
-	return NULL;
-}
-
-static int
-nvkm_fifo_chan_child_fini(struct nvkm_oproxy *base, bool suspend)
-{
-	struct nvkm_fifo_chan_object *object =
-		container_of(base, typeof(*object), oproxy);
-	struct nvkm_engine *engine  = object->oproxy.object->engine;
-	struct nvkm_fifo_chan *chan = object->chan;
-	struct nvkm_fifo_engn *engn = nvkm_fifo_chan_engn(chan, engine);
-	const char *name = engine->subdev.name;
-	int ret = 0;
-
-	if (chan->func->engine_fini) {
-		ret = chan->func->engine_fini(chan, engine, suspend);
-		if (ret) {
-			nvif_error(&chan->object,
-				   "detach %s failed, %d\n", name, ret);
-			return ret;
-		}
-	}
-
-	if (engn->object) {
-		ret = nvkm_object_fini(engn->object, suspend);
-		if (ret && suspend)
-			return ret;
-	}
-
-	nvif_trace(&chan->object, "detached %s\n", name);
-	return ret;
-}
-
-static int
-nvkm_fifo_chan_child_init(struct nvkm_oproxy *base)
-{
-	struct nvkm_fifo_chan_object *object =
-		container_of(base, typeof(*object), oproxy);
-	struct nvkm_engine *engine  = object->oproxy.object->engine;
-	struct nvkm_fifo_chan *chan = object->chan;
-	struct nvkm_fifo_engn *engn = nvkm_fifo_chan_engn(chan, engine);
-	const char *name = engine->subdev.name;
-	int ret;
-
-	if (engn->object) {
-		ret = nvkm_object_init(engn->object);
-		if (ret)
-			return ret;
-	}
-
-	if (chan->func->engine_init) {
-		ret = chan->func->engine_init(chan, engine);
-		if (ret) {
-			nvif_error(&chan->object,
-				   "attach %s failed, %d\n", name, ret);
-			return ret;
-		}
-	}
-
-	nvif_trace(&chan->object, "attached %s\n", name);
-	return 0;
-}
-
 static void
 nvkm_fifo_chan_child_del(struct nvkm_oproxy *base)
 {
 	struct nvkm_fifo_chan_object *object =
 		container_of(base, typeof(*object), oproxy);
-	struct nvkm_engine *engine  = object->oproxy.base.engine;
 	struct nvkm_fifo_chan *chan = object->chan;
-	struct nvkm_fifo_engn *engn = nvkm_fifo_chan_engn(chan, engine);
 
 	if (chan->func->object_dtor)
 		chan->func->object_dtor(chan, object->hash);
-
-	if (!--engn->refcount) {
-		if (chan->func->engine_dtor)
-			chan->func->engine_dtor(chan, engine);
-		nvkm_object_del(&engn->object);
-		if (chan->vmm)
-			atomic_dec(&chan->vmm->engref[engine->subdev.type]);
-	}
 }
 
 static const struct nvkm_oproxy_func
@@ -147,7 +67,8 @@ nvkm_fifo_chan_child_new(const struct nvkm_oclass *oclass, void *data, u32 size,
 {
 	struct nvkm_engine *engine = oclass->engine;
 	struct nvkm_fifo_chan *chan = nvkm_fifo_chan(oclass->parent);
-	struct nvkm_fifo_engn *engn = nvkm_fifo_chan_engn(chan, engine);
+	struct nvkm_ectx *engn = nvkm_list_find(engn, &chan->cgrp->ectxs, head,
+						engn->engn->engine == engine);
 	struct nvkm_fifo_chan_object *object;
 	int ret = 0;
 
@@ -157,33 +78,6 @@ nvkm_fifo_chan_child_new(const struct nvkm_oclass *oclass, void *data, u32 size,
 	object->chan = chan;
 	*pobject = &object->oproxy.base;
 
-	if (!engn->refcount++) {
-		struct nvkm_oclass cclass = {
-			.client = oclass->client,
-			.engine = oclass->engine,
-		};
-
-		if (chan->vmm)
-			atomic_inc(&chan->vmm->engref[engine->subdev.type]);
-
-		if (engine->func->fifo.cclass) {
-			ret = engine->func->fifo.cclass(chan, &cclass,
-							&engn->object);
-		} else
-		if (engine->func->cclass) {
-			ret = nvkm_object_new_(engine->func->cclass, &cclass,
-					       NULL, 0, &engn->object);
-		}
-		if (ret)
-			return ret;
-
-		if (chan->func->engine_ctor) {
-			ret = chan->func->engine_ctor(chan, oclass->engine,
-						      engn->object);
-			if (ret)
-				return ret;
-		}
-	}
 
 	ret = oclass->base.ctor(&(const struct nvkm_oclass) {
 					.base = oclass->base,
@@ -210,10 +104,16 @@ nvkm_fifo_chan_child_new(const struct nvkm_oclass *oclass, void *data, u32 size,
 }
 
 void
-nvkm_chan_cctx_bind(struct nvkm_chan *chan, struct nvkm_oproxy *oproxy, struct nvkm_cctx *cctx)
+nvkm_chan_cctx_bind(struct nvkm_chan *chan, struct nvkm_engn *engn, struct nvkm_cctx *cctx)
 {
 	struct nvkm_cgrp *cgrp = chan->cgrp;
 	struct nvkm_runl *runl = cgrp->runl;
+	struct nvkm_engine *engine = engn->engine;
+
+	if (!engn->func->bind)
+		return;
+
+	CHAN_TRACE(chan, "%sbind cctx %d[%s]", cctx ? "" : "un", engn->id, engine->subdev.name);
 
 	/* Prevent any channel in channel group from being rescheduled, kick them
 	 * off host and any engine(s) they're loaded on.
@@ -225,10 +125,7 @@ nvkm_chan_cctx_bind(struct nvkm_chan *chan, struct nvkm_oproxy *oproxy, struct n
 	nvkm_chan_preempt(chan, true);
 
 	/* Update context pointer. */
-	if (cctx)
-		nvkm_fifo_chan_child_init(nvkm_oproxy(oproxy->object));
-	else
-		nvkm_fifo_chan_child_fini(nvkm_oproxy(oproxy->object), false);
+	engn->func->bind(engn, cctx, chan);
 
 	/* Resume normal operation. */
 	if (cgrp->hw)
@@ -558,10 +455,6 @@ nvkm_fifo_chan_ctor(const struct nvkm_fifo_chan_func *fn,
 
 	*func = *fifo->func->chan.func;
 	func->dtor = fn->dtor;
-	func->engine_ctor = fn->engine_ctor;
-	func->engine_dtor = fn->engine_dtor;
-	func->engine_init = fn->engine_init;
-	func->engine_fini = fn->engine_fini;
 	func->object_ctor = fn->object_ctor;
 	func->object_dtor = fn->object_dtor;
 
@@ -572,7 +465,6 @@ nvkm_fifo_chan_ctor(const struct nvkm_fifo_chan_func *fn,
 	atomic_set(&chan->errored, 0);
 
 	nvkm_object_ctor(&nvkm_fifo_chan_func, oclass, &chan->object);
-	chan->fifo = fifo;
 	INIT_LIST_HEAD(&chan->cctxs);
 	INIT_LIST_HEAD(&chan->head);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
index 6e41eba0a7e2..bb142593d8b6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chan.h
@@ -52,12 +52,6 @@ struct nvkm_chan_func {
 	u32 (*doorbell_handle)(struct nvkm_chan *);
 
 	void *(*dtor)(struct nvkm_fifo_chan *);
-	int  (*engine_ctor)(struct nvkm_fifo_chan *, struct nvkm_engine *,
-			    struct nvkm_object *);
-	void (*engine_dtor)(struct nvkm_fifo_chan *, struct nvkm_engine *);
-	int  (*engine_init)(struct nvkm_fifo_chan *, struct nvkm_engine *);
-	int  (*engine_fini)(struct nvkm_fifo_chan *, struct nvkm_engine *,
-			    bool suspend);
 	int  (*object_ctor)(struct nvkm_fifo_chan *, struct nvkm_object *);
 	void (*object_dtor)(struct nvkm_fifo_chan *, int);
 };
@@ -78,8 +72,7 @@ int nvkm_chan_preempt_locked(struct nvkm_chan *, bool wait);
 int nvkm_chan_cctx_get(struct nvkm_chan *, struct nvkm_engn *, struct nvkm_cctx **,
 		       struct nvkm_client * /*TODO: remove need for this */);
 void nvkm_chan_cctx_put(struct nvkm_chan *, struct nvkm_cctx **);
-struct nvkm_oproxy;
-void nvkm_chan_cctx_bind(struct nvkm_chan *, struct nvkm_oproxy *, struct nvkm_cctx *);
+void nvkm_chan_cctx_bind(struct nvkm_chan *, struct nvkm_engn *, struct nvkm_cctx *);
 
 #define CHAN_PRCLI(c,l,p,f,a...) CGRP_PRINT((c)->cgrp, l, p, "%04x:[%s]"f, (c)->id, (c)->name, ##a)
 #define CHAN_PRINT(c,l,p,f,a...) CGRP_PRINT((c)->cgrp, l, p, "%04x:"f, (c)->id, ##a)
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c
index f5e3557b9e50..359d8b5b124c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/chang84.c
@@ -31,111 +31,6 @@
 #include <nvif/cl826e.h>
 
 static int
-g84_fifo_chan_engine_addr(struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_DMAOBJ:
-	case NVKM_ENGINE_SW    : return -1;
-	case NVKM_ENGINE_GR    : return 0x0020;
-	case NVKM_ENGINE_VP    :
-	case NVKM_ENGINE_MSPDEC: return 0x0040;
-	case NVKM_ENGINE_MPEG  :
-	case NVKM_ENGINE_MSPPP : return 0x0060;
-	case NVKM_ENGINE_BSP   :
-	case NVKM_ENGINE_MSVLD : return 0x0080;
-	case NVKM_ENGINE_CIPHER:
-	case NVKM_ENGINE_SEC   : return 0x00a0;
-	case NVKM_ENGINE_CE    : return 0x00c0;
-	default:
-		WARN_ON(1);
-		return -1;
-	}
-}
-
-static int
-g84_fifo_chan_engine_fini(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine, bool suspend)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-	struct nv50_fifo *fifo = chan->fifo;
-	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
-	struct nvkm_device *device = subdev->device;
-	u32 engn, save;
-	int offset;
-	bool done;
-
-	offset = g84_fifo_chan_engine_addr(engine);
-	if (offset < 0)
-		return 0;
-
-	engn = fifo->base.func->engine_id(&fifo->base, engine) - 1;
-	save = nvkm_mask(device, 0x002520, 0x0000003f, 1 << engn);
-	nvkm_wr32(device, 0x0032fc, chan->base.inst->addr >> 12);
-	done = nvkm_msec(device, 2000,
-		if (nvkm_rd32(device, 0x0032fc) != 0xffffffff)
-			break;
-	) >= 0;
-	nvkm_wr32(device, 0x002520, save);
-	if (!done) {
-		nvkm_error(subdev, "channel %d [%s] unload timeout\n",
-			   chan->base.chid, chan->base.object.client->name);
-		if (suspend)
-			return -EBUSY;
-	}
-
-	nvkm_kmap(chan->eng);
-	nvkm_wo32(chan->eng, offset + 0x00, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x04, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x08, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x0c, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x10, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x14, 0x00000000);
-	nvkm_done(chan->eng);
-	return 0;
-}
-
-
-static int
-g84_fifo_chan_engine_init(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-	struct nvkm_gpuobj *engn = *nv50_fifo_chan_engine(chan, engine);
-	u64 limit, start;
-	int offset;
-
-	offset = g84_fifo_chan_engine_addr(engine);
-	if (offset < 0)
-		return 0;
-	limit = engn->addr + engn->size - 1;
-	start = engn->addr;
-
-	nvkm_kmap(chan->eng);
-	nvkm_wo32(chan->eng, offset + 0x00, 0x00190000);
-	nvkm_wo32(chan->eng, offset + 0x04, lower_32_bits(limit));
-	nvkm_wo32(chan->eng, offset + 0x08, lower_32_bits(start));
-	nvkm_wo32(chan->eng, offset + 0x0c, upper_32_bits(limit) << 24 |
-					    upper_32_bits(start));
-	nvkm_wo32(chan->eng, offset + 0x10, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x14, 0x00000000);
-	nvkm_done(chan->eng);
-	return 0;
-}
-
-static int
-g84_fifo_chan_engine_ctor(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine,
-			  struct nvkm_object *object)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-
-	if (g84_fifo_chan_engine_addr(engine) < 0)
-		return 0;
-
-	return nvkm_object_bind(object, NULL, 0, nv50_fifo_chan_engine(chan, engine));
-}
-
-static int
 g84_fifo_chan_object_ctor(struct nvkm_fifo_chan *base,
 			  struct nvkm_object *object)
 {
@@ -169,10 +64,6 @@ g84_fifo_chan_object_ctor(struct nvkm_fifo_chan *base,
 static const struct nvkm_fifo_chan_func
 g84_fifo_chan_func = {
 	.dtor = nv50_fifo_chan_dtor,
-	.engine_ctor = g84_fifo_chan_engine_ctor,
-	.engine_dtor = nv50_fifo_chan_engine_dtor,
-	.engine_init = g84_fifo_chan_engine_init,
-	.engine_fini = g84_fifo_chan_engine_fini,
 	.object_ctor = g84_fifo_chan_object_ctor,
 	.object_dtor = nv50_fifo_chan_object_dtor,
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changf100.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changf100.h
index 3c7b6c0d8cf1..739cd750742e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changf100.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changf100.h
@@ -16,10 +16,6 @@ struct gf100_fifo_chan {
 #define GF100_FIFO_ENGN_CE0    4
 #define GF100_FIFO_ENGN_CE1    5
 #define GF100_FIFO_ENGN_SW     15
-	struct gf100_fifo_engn {
-		struct nvkm_gpuobj *inst;
-		struct nvkm_vma *vma;
-	} engn[NVKM_FIFO_ENGN_NR];
 };
 
 extern const struct nvkm_fifo_chan_oclass gf100_fifo_gpfifo_oclass;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h
index 95e788f31590..2043f06f500b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h
@@ -11,10 +11,6 @@ struct gk104_fifo_chan {
 	int runl;
 
 #define GK104_FIFO_ENGN_SW 15
-	struct gk104_fifo_engn {
-		struct nvkm_gpuobj *inst;
-		struct nvkm_vma *vma;
-	} engn[NVKM_FIFO_ENGN_NR];
 };
 
 extern const struct nvkm_fifo_chan_func gk104_fifo_gpfifo_func;
@@ -22,11 +18,6 @@ extern const struct nvkm_fifo_chan_func gk104_fifo_gpfifo_func;
 int gk104_fifo_gpfifo_new(struct gk104_fifo *, const struct nvkm_oclass *,
 			  void *data, u32 size, struct nvkm_object **);
 void *gk104_fifo_gpfifo_dtor(struct nvkm_fifo_chan *);
-struct gk104_fifo_engn *gk104_fifo_gpfifo_engine(struct gk104_fifo_chan *, struct nvkm_engine *);
-int gk104_fifo_gpfifo_engine_ctor(struct nvkm_fifo_chan *, struct nvkm_engine *,
-				  struct nvkm_object *);
-void gk104_fifo_gpfifo_engine_dtor(struct nvkm_fifo_chan *,
-				   struct nvkm_engine *);
 
 int gv100_fifo_gpfifo_new(struct gk104_fifo *, const struct nvkm_oclass *,
 			  void *data, u32 size, struct nvkm_object **);
@@ -34,10 +25,6 @@ int gv100_fifo_gpfifo_new_(const struct nvkm_fifo_chan_func *,
 			   struct gk104_fifo *, u64 *, u16 *, u64, u64, u64,
 			   u64 *, bool, u32 *, const struct nvkm_oclass *,
 			   struct nvkm_object **);
-int gv100_fifo_gpfifo_engine_init(struct nvkm_fifo_chan *,
-				  struct nvkm_engine *);
-int gv100_fifo_gpfifo_engine_fini(struct nvkm_fifo_chan *,
-				  struct nvkm_engine *, bool);
 
 int tu102_fifo_gpfifo_new(struct gk104_fifo *, const struct nvkm_oclass *,
 			  void *data, u32 size, struct nvkm_object **);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv04.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv04.h
index 837f9f7f69ce..5ebfa2241750 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv04.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv04.h
@@ -8,12 +8,10 @@
 struct nv04_fifo_chan {
 	struct nvkm_fifo_chan base;
 	struct nv04_fifo *fifo;
-	u32 ramfc;
 #define NV04_FIFO_ENGN_SW   0
 #define NV04_FIFO_ENGN_GR   1
 #define NV04_FIFO_ENGN_MPEG 2
 #define NV04_FIFO_ENGN_DMA  3
-	struct nvkm_gpuobj *engn[NVKM_FIFO_ENGN_NR];
 };
 
 extern const struct nvkm_fifo_chan_func nv04_fifo_dma_func;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.c
index 0bf59c4e92b9..cb5606041499 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.c
@@ -28,133 +28,6 @@
 #include <subdev/mmu.h>
 #include <subdev/timer.h>
 
-static int
-nv50_fifo_chan_engine_addr(struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_DMAOBJ:
-	case NVKM_ENGINE_SW    : return -1;
-	case NVKM_ENGINE_GR    : return 0x0000;
-	case NVKM_ENGINE_MPEG  : return 0x0060;
-	default:
-		WARN_ON(1);
-		return -1;
-	}
-}
-
-struct nvkm_gpuobj **
-nv50_fifo_chan_engine(struct nv50_fifo_chan *chan, struct nvkm_engine *engine)
-{
-	int engi = chan->base.fifo->func->engine_id(chan->base.fifo, engine);
-	if (engi >= 0)
-		return &chan->engn[engi];
-	return NULL;
-}
-
-static int
-nv50_fifo_chan_engine_fini(struct nvkm_fifo_chan *base,
-			   struct nvkm_engine *engine, bool suspend)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-	struct nv50_fifo *fifo = chan->fifo;
-	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
-	struct nvkm_device *device = subdev->device;
-	int offset, ret = 0;
-	u32 me;
-
-	offset = nv50_fifo_chan_engine_addr(engine);
-	if (offset < 0)
-		return 0;
-
-	/* HW bug workaround:
-	 *
-	 * PFIFO will hang forever if the connected engines don't report
-	 * that they've processed the context switch request.
-	 *
-	 * In order for the kickoff to work, we need to ensure all the
-	 * connected engines are in a state where they can answer.
-	 *
-	 * Newer chipsets don't seem to suffer from this issue, and well,
-	 * there's also a "ignore these engines" bitmask reg we can use
-	 * if we hit the issue there..
-	 */
-	me = nvkm_mask(device, 0x00b860, 0x00000001, 0x00000001);
-
-	/* do the kickoff... */
-	nvkm_wr32(device, 0x0032fc, chan->base.inst->addr >> 12);
-	if (nvkm_msec(device, 2000,
-		if (nvkm_rd32(device, 0x0032fc) != 0xffffffff)
-			break;
-	) < 0) {
-		nvkm_error(subdev, "channel %d [%s] unload timeout\n",
-			   chan->base.chid, chan->base.object.client->name);
-		if (suspend)
-			ret = -EBUSY;
-	}
-	nvkm_wr32(device, 0x00b860, me);
-
-	if (ret == 0) {
-		nvkm_kmap(chan->eng);
-		nvkm_wo32(chan->eng, offset + 0x00, 0x00000000);
-		nvkm_wo32(chan->eng, offset + 0x04, 0x00000000);
-		nvkm_wo32(chan->eng, offset + 0x08, 0x00000000);
-		nvkm_wo32(chan->eng, offset + 0x0c, 0x00000000);
-		nvkm_wo32(chan->eng, offset + 0x10, 0x00000000);
-		nvkm_wo32(chan->eng, offset + 0x14, 0x00000000);
-		nvkm_done(chan->eng);
-	}
-
-	return ret;
-}
-
-static int
-nv50_fifo_chan_engine_init(struct nvkm_fifo_chan *base,
-			   struct nvkm_engine *engine)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-	struct nvkm_gpuobj *engn = *nv50_fifo_chan_engine(chan, engine);
-	u64 limit, start;
-	int offset;
-
-	offset = nv50_fifo_chan_engine_addr(engine);
-	if (offset < 0)
-		return 0;
-	limit = engn->addr + engn->size - 1;
-	start = engn->addr;
-
-	nvkm_kmap(chan->eng);
-	nvkm_wo32(chan->eng, offset + 0x00, 0x00190000);
-	nvkm_wo32(chan->eng, offset + 0x04, lower_32_bits(limit));
-	nvkm_wo32(chan->eng, offset + 0x08, lower_32_bits(start));
-	nvkm_wo32(chan->eng, offset + 0x0c, upper_32_bits(limit) << 24 |
-					    upper_32_bits(start));
-	nvkm_wo32(chan->eng, offset + 0x10, 0x00000000);
-	nvkm_wo32(chan->eng, offset + 0x14, 0x00000000);
-	nvkm_done(chan->eng);
-	return 0;
-}
-
-void
-nv50_fifo_chan_engine_dtor(struct nvkm_fifo_chan *base,
-			   struct nvkm_engine *engine)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-	nvkm_gpuobj_del(nv50_fifo_chan_engine(chan, engine));
-}
-
-static int
-nv50_fifo_chan_engine_ctor(struct nvkm_fifo_chan *base,
-			   struct nvkm_engine *engine,
-			   struct nvkm_object *object)
-{
-	struct nv50_fifo_chan *chan = nv50_fifo_chan(base);
-
-	if (nv50_fifo_chan_engine_addr(engine) < 0)
-		return 0;
-
-	return nvkm_object_bind(object, NULL, 0, nv50_fifo_chan_engine(chan, engine));
-}
-
 void
 nv50_fifo_chan_object_dtor(struct nvkm_fifo_chan *base, int cookie)
 {
@@ -193,10 +66,6 @@ nv50_fifo_chan_dtor(struct nvkm_fifo_chan *base)
 static const struct nvkm_fifo_chan_func
 nv50_fifo_chan_func = {
 	.dtor = nv50_fifo_chan_dtor,
-	.engine_ctor = nv50_fifo_chan_engine_ctor,
-	.engine_dtor = nv50_fifo_chan_engine_dtor,
-	.engine_init = nv50_fifo_chan_engine_init,
-	.engine_fini = nv50_fifo_chan_engine_fini,
 	.object_ctor = nv50_fifo_chan_object_ctor,
 	.object_dtor = nv50_fifo_chan_object_dtor,
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.h
index 343356a0e5ef..ced832c7f425 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/channv50.h
@@ -9,7 +9,6 @@ struct nv50_fifo_chan {
 	struct nv50_fifo *fifo;
 	struct nvkm_fifo_chan base;
 
-	struct nvkm_gpuobj *eng;
 	struct nvkm_ramht *ramht;
 
 #define NV50_FIFO_ENGN_SW   0
@@ -31,14 +30,11 @@ struct nv50_fifo_chan {
 #define G84_FIFO_ENGN_BSP    6
 #define G84_FIFO_ENGN_MSVLD  6
 #define G84_FIFO_ENGN_DMA    7
-	struct nvkm_gpuobj *engn[NVKM_FIFO_ENGN_NR];
 };
 
 int nv50_fifo_chan_ctor(struct nv50_fifo *, u64 vmm, u64 push,
 			const struct nvkm_oclass *, struct nv50_fifo_chan *);
 void *nv50_fifo_chan_dtor(struct nvkm_fifo_chan *);
-struct nvkm_gpuobj **nv50_fifo_chan_engine(struct nv50_fifo_chan *, struct nvkm_engine *);
-void nv50_fifo_chan_engine_dtor(struct nvkm_fifo_chan *, struct nvkm_engine *);
 void nv50_fifo_chan_object_dtor(struct nvkm_fifo_chan *, int);
 
 int g84_fifo_chan_ctor(struct nv50_fifo *, u64 vmm, u64 push,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/dmanv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/dmanv40.c
index 5e7104457a02..ca6fd8a5af6b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/dmanv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/dmanv40.c
@@ -31,121 +31,6 @@
 #include <nvif/cl006b.h>
 #include <nvif/unpack.h>
 
-static bool
-nv40_fifo_dma_engine(struct nvkm_engine *engine, u32 *reg, u32 *ctx)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_DMAOBJ:
-	case NVKM_ENGINE_SW:
-		return false;
-	case NVKM_ENGINE_GR:
-		*reg = 0x0032e0;
-		*ctx = 0x38;
-		return true;
-	case NVKM_ENGINE_MPEG:
-		if (engine->subdev.device->chipset < 0x44)
-			return false;
-		*reg = 0x00330c;
-		*ctx = 0x54;
-		return true;
-	default:
-		WARN_ON(1);
-		return false;
-	}
-}
-
-static struct nvkm_gpuobj **
-nv40_fifo_dma_engn(struct nv04_fifo_chan *chan, struct nvkm_engine *engine)
-{
-	int engi = chan->base.fifo->func->engine_id(chan->base.fifo, engine);
-	if (engi >= 0)
-		return &chan->engn[engi];
-	return NULL;
-}
-
-static int
-nv40_fifo_dma_engine_fini(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine, bool suspend)
-{
-	struct nv04_fifo_chan *chan = nv04_fifo_chan(base);
-	struct nv04_fifo *fifo = chan->fifo;
-	struct nvkm_device *device = fifo->base.engine.subdev.device;
-	struct nvkm_instmem *imem = device->imem;
-	unsigned long flags;
-	u32 reg, ctx;
-	int chid;
-
-	if (!nv40_fifo_dma_engine(engine, &reg, &ctx))
-		return 0;
-
-	spin_lock_irqsave(&fifo->base.lock, flags);
-	nvkm_mask(device, 0x002500, 0x00000001, 0x00000000);
-
-	chid = nvkm_rd32(device, 0x003204) & (fifo->base.nr - 1);
-	if (chid == chan->base.chid)
-		nvkm_wr32(device, reg, 0x00000000);
-	nvkm_kmap(imem->ramfc);
-	nvkm_wo32(imem->ramfc, chan->ramfc + ctx, 0x00000000);
-	nvkm_done(imem->ramfc);
-
-	nvkm_mask(device, 0x002500, 0x00000001, 0x00000001);
-	spin_unlock_irqrestore(&fifo->base.lock, flags);
-	return 0;
-}
-
-static int
-nv40_fifo_dma_engine_init(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine)
-{
-	struct nv04_fifo_chan *chan = nv04_fifo_chan(base);
-	struct nv04_fifo *fifo = chan->fifo;
-	struct nvkm_device *device = fifo->base.engine.subdev.device;
-	struct nvkm_instmem *imem = device->imem;
-	unsigned long flags;
-	u32 inst, reg, ctx;
-	int chid;
-
-	if (!nv40_fifo_dma_engine(engine, &reg, &ctx))
-		return 0;
-	inst = (*nv40_fifo_dma_engn(chan, engine))->addr >> 4;
-
-	spin_lock_irqsave(&fifo->base.lock, flags);
-	nvkm_mask(device, 0x002500, 0x00000001, 0x00000000);
-
-	chid = nvkm_rd32(device, 0x003204) & (fifo->base.nr - 1);
-	if (chid == chan->base.chid)
-		nvkm_wr32(device, reg, inst);
-	nvkm_kmap(imem->ramfc);
-	nvkm_wo32(imem->ramfc, chan->ramfc + ctx, inst);
-	nvkm_done(imem->ramfc);
-
-	nvkm_mask(device, 0x002500, 0x00000001, 0x00000001);
-	spin_unlock_irqrestore(&fifo->base.lock, flags);
-	return 0;
-}
-
-static void
-nv40_fifo_dma_engine_dtor(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine)
-{
-	struct nv04_fifo_chan *chan = nv04_fifo_chan(base);
-	nvkm_gpuobj_del(nv40_fifo_dma_engn(chan, engine));
-}
-
-static int
-nv40_fifo_dma_engine_ctor(struct nvkm_fifo_chan *base,
-			  struct nvkm_engine *engine,
-			  struct nvkm_object *object)
-{
-	struct nv04_fifo_chan *chan = nv04_fifo_chan(base);
-	u32 reg, ctx;
-
-	if (!nv40_fifo_dma_engine(engine, &reg, &ctx))
-		return 0;
-
-	return nvkm_object_bind(object, NULL, 0, nv40_fifo_dma_engn(chan, engine));
-}
-
 static int
 nv40_fifo_dma_object_ctor(struct nvkm_fifo_chan *base,
 			  struct nvkm_object *object)
@@ -176,10 +61,6 @@ nv40_fifo_dma_object_ctor(struct nvkm_fifo_chan *base,
 static const struct nvkm_fifo_chan_func
 nv40_fifo_dma_func = {
 	.dtor = nv04_fifo_dma_dtor,
-	.engine_ctor = nv40_fifo_dma_engine_ctor,
-	.engine_dtor = nv40_fifo_dma_engine_dtor,
-	.engine_init = nv40_fifo_dma_engine_init,
-	.engine_fini = nv40_fifo_dma_engine_fini,
 	.object_ctor = nv40_fifo_dma_object_ctor,
 	.object_dtor = nv04_fifo_dma_object_dtor,
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
index 6f3f5caaa748..2bc835f2ef74 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g84.c
@@ -26,6 +26,7 @@
 #include "runl.h"
 
 #include <core/ramht.h>
+#include <subdev/timer.h>
 
 #include "nv50.h"
 #include "channv50.h"
@@ -67,7 +68,6 @@ g84_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm, b
 	if (ret)
 		return ret;
 
-	nv50_fifo_chan(chan)->eng = chan->eng;
 	nv50_fifo_chan(chan)->ramht = chan->ramht;
 
 	nvkm_kmap(chan->ramfc);
@@ -106,8 +106,58 @@ g84_chan = {
 	.stop = nv50_chan_stop,
 };
 
+static void
+g84_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	struct nvkm_subdev *subdev = &chan->cgrp->runl->fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u64 start = 0, limit = 0;
+	u32 flags = 0, ptr0, save;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_GR    : ptr0 = 0x0020; break;
+	case NVKM_ENGINE_VP    :
+	case NVKM_ENGINE_MSPDEC: ptr0 = 0x0040; break;
+	case NVKM_ENGINE_MPEG  :
+	case NVKM_ENGINE_MSPPP : ptr0 = 0x0060; break;
+	case NVKM_ENGINE_BSP   :
+	case NVKM_ENGINE_MSVLD : ptr0 = 0x0080; break;
+	case NVKM_ENGINE_CIPHER:
+	case NVKM_ENGINE_SEC   : ptr0 = 0x00a0; break;
+	case NVKM_ENGINE_CE    : ptr0 = 0x00c0; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (!cctx) {
+		save = nvkm_mask(device, 0x002520, 0x0000003f, BIT(engn->id - 1));
+		nvkm_wr32(device, 0x0032fc, chan->inst->addr >> 12);
+		nvkm_msec(device, 2000,
+			if (nvkm_rd32(device, 0x0032fc) != 0xffffffff)
+				break;
+		);
+		nvkm_wr32(device, 0x002520, save);
+	} else {
+		flags = 0x00190000;
+		start = cctx->vctx->inst->addr;
+		limit = start + cctx->vctx->inst->size - 1;
+	}
+
+	nvkm_kmap(chan->eng);
+	nvkm_wo32(chan->eng, ptr0 + 0x00, flags);
+	nvkm_wo32(chan->eng, ptr0 + 0x04, lower_32_bits(limit));
+	nvkm_wo32(chan->eng, ptr0 + 0x08, lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x0c, upper_32_bits(limit) << 24 |
+					  lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x10, 0x00000000);
+	nvkm_wo32(chan->eng, ptr0 + 0x14, 0x00000000);
+	nvkm_done(chan->eng);
+}
+
 const struct nvkm_engn_func
 g84_engn = {
+	.bind = g84_ectx_bind,
 };
 
 static void
@@ -138,28 +188,6 @@ g84_fifo_nonstall = {
 	.fini = g84_fifo_nonstall_block,
 };
 
-int
-g84_fifo_engine_id(struct nvkm_fifo *base, struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_SW    : return G84_FIFO_ENGN_SW;
-	case NVKM_ENGINE_GR    : return G84_FIFO_ENGN_GR;
-	case NVKM_ENGINE_MPEG  :
-	case NVKM_ENGINE_MSPPP : return G84_FIFO_ENGN_MPEG;
-	case NVKM_ENGINE_CE    : return G84_FIFO_ENGN_CE0;
-	case NVKM_ENGINE_VP    :
-	case NVKM_ENGINE_MSPDEC: return G84_FIFO_ENGN_VP;
-	case NVKM_ENGINE_CIPHER:
-	case NVKM_ENGINE_SEC   : return G84_FIFO_ENGN_CIPHER;
-	case NVKM_ENGINE_BSP   :
-	case NVKM_ENGINE_MSVLD : return G84_FIFO_ENGN_BSP;
-	case NVKM_ENGINE_DMAOBJ: return G84_FIFO_ENGN_DMA;
-	default:
-		WARN_ON(1);
-		return -1;
-	}
-}
-
 static int
 g84_fifo_runl_ctor(struct nvkm_fifo *fifo)
 {
@@ -188,7 +216,6 @@ g84_fifo = {
 	.runl_ctor = g84_fifo_runl_ctor,
 	.init = nv50_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = g84_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.nonstall = &g84_fifo_nonstall,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c
index 8016a2228e92..b2634f21d320 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/g98.c
@@ -55,7 +55,6 @@ g98_fifo = {
 	.runl_ctor = g98_fifo_runl_ctor,
 	.init = nv50_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = g84_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.nonstall = &g84_fifo_nonstall,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
index 351b35d580fd..2e2626168bf2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gf100.c
@@ -158,6 +158,47 @@ gf100_chan = {
 	.preempt = gf100_chan_preempt,
 };
 
+static void
+gf100_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	u64 addr = 0ULL;
+	u32 ptr0;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_SW    : return;
+	case NVKM_ENGINE_GR    : ptr0 = 0x0210; break;
+	case NVKM_ENGINE_CE    : ptr0 = 0x0230 + (engn->engine->subdev.inst * 0x10); break;
+	case NVKM_ENGINE_MSPDEC: ptr0 = 0x0250; break;
+	case NVKM_ENGINE_MSPPP : ptr0 = 0x0260; break;
+	case NVKM_ENGINE_MSVLD : ptr0 = 0x0270; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (cctx) {
+		addr  = cctx->vctx->vma->addr;
+		addr |= 4ULL;
+	}
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, ptr0 + 0, lower_32_bits(addr));
+	nvkm_wo32(chan->inst, ptr0 + 4, upper_32_bits(addr));
+	nvkm_done(chan->inst);
+}
+
+static int
+gf100_ectx_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx)
+{
+	int ret;
+
+	ret = nvkm_vmm_get(vctx->vmm, 12, vctx->inst->size, &vctx->vma);
+	if (ret)
+		return ret;
+
+	return nvkm_memory_map(vctx->inst, 0, vctx->vmm, vctx->vma, NULL, 0);
+}
+
 bool
 gf100_engn_mmu_fault_triggered(struct nvkm_engn *engn)
 {
@@ -250,6 +291,8 @@ gf100_engn = {
 	.cxid = gf100_engn_cxid,
 	.mmu_fault_trigger = gf100_engn_mmu_fault_trigger,
 	.mmu_fault_triggered = gf100_engn_mmu_fault_triggered,
+	.ctor = gf100_ectx_ctor,
+	.bind = gf100_ectx_bind,
 };
 
 const struct nvkm_engn_func
@@ -422,22 +465,6 @@ gf100_fifo_nonstall = {
 	.fini = gf100_fifo_nonstall_block,
 };
 
-static int
-gf100_fifo_engine_id(struct nvkm_fifo *base, struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_GR    : return GF100_FIFO_ENGN_GR;
-	case NVKM_ENGINE_MSPDEC: return GF100_FIFO_ENGN_MSPDEC;
-	case NVKM_ENGINE_MSPPP : return GF100_FIFO_ENGN_MSPPP;
-	case NVKM_ENGINE_MSVLD : return GF100_FIFO_ENGN_MSVLD;
-	case NVKM_ENGINE_CE    : return GF100_FIFO_ENGN_CE0 + engine->subdev.inst;
-	case NVKM_ENGINE_SW    : return GF100_FIFO_ENGN_SW;
-	default:
-		WARN_ON(1);
-		return -1;
-	}
-}
-
 static const struct nvkm_enum
 gf100_fifo_mmu_fault_engine[] = {
 	{ 0x00, "PGRAPH", NULL, NVKM_ENGINE_GR },
@@ -935,7 +962,6 @@ gf100_fifo = {
 	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gf100_fifo_mmu_fault,
-	.engine_id = gf100_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gf100_runl,
 	.runq = &gf100_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
index b75e3d089da3..6130f492f48a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -35,6 +35,7 @@
 #include <subdev/top.h>
 
 #include <nvif/class.h>
+#include <nvif/if900d.h>
 
 void
 gk104_chan_stop(struct nvkm_chan *chan)
@@ -130,6 +131,63 @@ gk104_chan = {
 	.preempt = gf100_chan_preempt,
 };
 
+static void
+gk104_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	u32 ptr0, ptr1 = 0;
+	u64 addr = 0ULL;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_SW    : return;
+	case NVKM_ENGINE_GR    : ptr0 = 0x0210; break;
+	case NVKM_ENGINE_SEC   : ptr0 = 0x0220; break;
+	case NVKM_ENGINE_MSPDEC: ptr0 = 0x0250; break;
+	case NVKM_ENGINE_MSPPP : ptr0 = 0x0260; break;
+	case NVKM_ENGINE_MSVLD : ptr0 = 0x0270; break;
+	case NVKM_ENGINE_VIC   : ptr0 = 0x0280; break;
+	case NVKM_ENGINE_MSENC : ptr0 = 0x0290; break;
+	case NVKM_ENGINE_NVDEC :
+		ptr1 = 0x0270;
+		ptr0 = 0x0210;
+		break;
+	case NVKM_ENGINE_NVENC :
+		if (!engn->engine->subdev.inst)
+			ptr1 = 0x0290;
+		ptr0 = 0x0210;
+		break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (cctx) {
+		addr  = cctx->vctx->vma->addr;
+		addr |= 4ULL;
+	}
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, ptr0 + 0, lower_32_bits(addr));
+	nvkm_wo32(chan->inst, ptr0 + 4, upper_32_bits(addr));
+	if (ptr1) {
+		nvkm_wo32(chan->inst, ptr1 + 0, lower_32_bits(addr));
+		nvkm_wo32(chan->inst, ptr1 + 4, upper_32_bits(addr));
+	}
+	nvkm_done(chan->inst);
+}
+
+int
+gk104_ectx_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx)
+{
+	struct gf100_vmm_map_v0 args = { .priv = 1 };
+	int ret;
+
+	ret = nvkm_vmm_get(vctx->vmm, 12, vctx->inst->size, &vctx->vma);
+	if (ret)
+		return ret;
+
+	return nvkm_memory_map(vctx->inst, 0, vctx->vmm, vctx->vma, &args, sizeof(args));
+}
+
 /*TODO: clean this up */
 struct gk104_engn_status {
 	bool busy;
@@ -216,6 +274,8 @@ gk104_engn = {
 	.cxid = gk104_engn_cxid,
 	.mmu_fault_trigger = gf100_engn_mmu_fault_trigger,
 	.mmu_fault_triggered = gf100_engn_mmu_fault_triggered,
+	.ctor = gk104_ectx_ctor,
+	.bind = gk104_ectx_bind,
 };
 
 const struct nvkm_engn_func
@@ -410,24 +470,6 @@ gk104_runl = {
 	.preempt_pending = gf100_runl_preempt_pending,
 };
 
-int
-gk104_fifo_engine_id(struct nvkm_fifo *base, struct nvkm_engine *engine)
-{
-	struct gk104_fifo *fifo = gk104_fifo(base);
-	int engn;
-
-	if (engine->subdev.type == NVKM_ENGINE_SW)
-		return GK104_FIFO_ENGN_SW;
-
-	for (engn = 0; engn < fifo->engine_nr && engine; engn++) {
-		if (fifo->engine[engn].engine == engine)
-			return engn;
-	}
-
-	WARN_ON(1);
-	return -1;
-}
-
 static const struct nvkm_enum
 gk104_fifo_mmu_fault_engine[] = {
 	{ 0x00, "GR", NULL, NVKM_ENGINE_GR },
@@ -778,8 +820,6 @@ gk104_fifo_oneinit(struct nvkm_fifo *base)
 		if (engn < 0)
 			continue;
 
-		fifo->engine[engn].engine = nvkm_device_engine(device, tdev->type, tdev->inst);
-		fifo->engine_nr = max(fifo->engine_nr, engn + 1);
 		fifo->runlist[tdev->runlist].engm |= BIT(engn);
 		fifo->runlist[tdev->runlist].engm_sw |= BIT(engn);
 		if (tdev->type == NVKM_ENGINE_GR)
@@ -825,7 +865,6 @@ gk104_fifo = {
 	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gk104_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gk104_runl,
 	.runq = &gk104_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.h
index 3bcf0720f89d..97fc9ccc2930 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.h
@@ -15,11 +15,6 @@ struct gk104_fifo {
 	struct nvkm_fifo base;
 
 	struct {
-		struct nvkm_engine *engine;
-	} engine[16];
-	int engine_nr;
-
-	struct {
 		u32 engm;
 		u32 engm_sw;
 	} runlist[16];
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c
index 2a617215d924..2afbaf42572d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk110.c
@@ -119,7 +119,6 @@ gk110_fifo = {
 	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gk104_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gk110_runl,
 	.runq = &gk104_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c
index 81d184473f79..3727be30d7f4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk208.c
@@ -63,7 +63,6 @@ gk208_fifo = {
 	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gk104_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gk110_runl,
 	.runq = &gk208_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c
index 47b7d93f5a04..eda1446dce27 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk20a.c
@@ -38,7 +38,6 @@ gk20a_fifo = {
 	.intr_mmu_fault_unit = gf100_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gk104_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gk110_runl,
 	.runq = &gk208_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c
index 3764245fbca1..1089062313cc 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm107.c
@@ -143,7 +143,6 @@ gm107_fifo = {
 	.intr_mmu_fault_unit = gm107_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gm107_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gm107_runl,
 	.runq = &gk208_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c
index 8be59812797a..286c1098b665 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gm200.c
@@ -52,7 +52,6 @@ gm200_fifo = {
 	.intr_mmu_fault_unit = gm107_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gm107_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gm107_runl,
 	.runq = &gk208_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
index a353c6b860ef..45d536d448f0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
@@ -125,7 +125,6 @@ gp100_fifo = {
 	.intr_mmu_fault_unit = gp100_fifo_intr_mmu_fault_unit,
 	.intr_ctxsw_timeout = gf100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gp100_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gp100_runl,
 	.runq = &gk208_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
index c43cd0c24a6d..6ba4335d40bc 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogf100.c
@@ -31,102 +31,6 @@
 #include <nvif/cl906f.h>
 #include <nvif/unpack.h>
 
-static u32
-gf100_fifo_gpfifo_engine_addr(struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_SW    : return 0;
-	case NVKM_ENGINE_GR    : return 0x0210;
-	case NVKM_ENGINE_CE    : return 0x0230 + (engine->subdev.inst * 0x10);
-	case NVKM_ENGINE_MSPDEC: return 0x0250;
-	case NVKM_ENGINE_MSPPP : return 0x0260;
-	case NVKM_ENGINE_MSVLD : return 0x0270;
-	default:
-		WARN_ON(1);
-		return 0;
-	}
-}
-
-static struct gf100_fifo_engn *
-gf100_fifo_gpfifo_engine(struct gf100_fifo_chan *chan, struct nvkm_engine *engine)
-{
-	int engi = chan->base.fifo->func->engine_id(chan->base.fifo, engine);
-	if (engi >= 0)
-		return &chan->engn[engi];
-	return NULL;
-}
-
-static int
-gf100_fifo_gpfifo_engine_fini(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine, bool suspend)
-{
-	const u32 offset = gf100_fifo_gpfifo_engine_addr(engine);
-	struct gf100_fifo_chan *chan = gf100_fifo_chan(base);
-	struct nvkm_gpuobj *inst = chan->base.inst;
-	int ret = 0;
-
-	if (offset) {
-		nvkm_kmap(inst);
-		nvkm_wo32(inst, offset + 0x00, 0x00000000);
-		nvkm_wo32(inst, offset + 0x04, 0x00000000);
-		nvkm_done(inst);
-	}
-
-	return ret;
-}
-
-static int
-gf100_fifo_gpfifo_engine_init(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine)
-{
-	const u32 offset = gf100_fifo_gpfifo_engine_addr(engine);
-	struct gf100_fifo_chan *chan = gf100_fifo_chan(base);
-	struct gf100_fifo_engn *engn = gf100_fifo_gpfifo_engine(chan, engine);
-	struct nvkm_gpuobj *inst = chan->base.inst;
-
-	if (offset) {
-		nvkm_kmap(inst);
-		nvkm_wo32(inst, offset + 0x00, lower_32_bits(engn->vma->addr) | 4);
-		nvkm_wo32(inst, offset + 0x04, upper_32_bits(engn->vma->addr));
-		nvkm_done(inst);
-	}
-
-	return 0;
-}
-
-static void
-gf100_fifo_gpfifo_engine_dtor(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine)
-{
-	struct gf100_fifo_chan *chan = gf100_fifo_chan(base);
-	struct gf100_fifo_engn *engn = gf100_fifo_gpfifo_engine(chan, engine);
-	nvkm_vmm_put(chan->base.vmm, &engn->vma);
-	nvkm_gpuobj_del(&engn->inst);
-}
-
-static int
-gf100_fifo_gpfifo_engine_ctor(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine,
-			      struct nvkm_object *object)
-{
-	struct gf100_fifo_chan *chan = gf100_fifo_chan(base);
-	struct gf100_fifo_engn *engn = gf100_fifo_gpfifo_engine(chan, engine);
-	int ret;
-
-	if (!gf100_fifo_gpfifo_engine_addr(engine))
-		return 0;
-
-	ret = nvkm_object_bind(object, NULL, 0, &engn->inst);
-	if (ret)
-		return ret;
-
-	ret = nvkm_vmm_get(chan->base.vmm, 12, engn->inst->size, &engn->vma);
-	if (ret)
-		return ret;
-
-	return nvkm_memory_map(engn->inst, 0, chan->base.vmm, engn->vma, NULL, 0);
-}
-
 static void *
 gf100_fifo_gpfifo_dtor(struct nvkm_fifo_chan *base)
 {
@@ -136,10 +40,6 @@ gf100_fifo_gpfifo_dtor(struct nvkm_fifo_chan *base)
 static const struct nvkm_fifo_chan_func
 gf100_fifo_gpfifo_func = {
 	.dtor = gf100_fifo_gpfifo_dtor,
-	.engine_ctor = gf100_fifo_gpfifo_engine_ctor,
-	.engine_dtor = gf100_fifo_gpfifo_engine_dtor,
-	.engine_init = gf100_fifo_gpfifo_engine_init,
-	.engine_fini = gf100_fifo_gpfifo_engine_fini,
 };
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
index 77de170c8144..28a7a5eae72c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c
@@ -33,125 +33,6 @@
 #include <nvif/cla06f.h>
 #include <nvif/unpack.h>
 
-static u32
-gk104_fifo_gpfifo_engine_addr(struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_SW    :
-	case NVKM_ENGINE_CE    : return 0;
-	case NVKM_ENGINE_GR    : return 0x0210;
-	case NVKM_ENGINE_SEC   : return 0x0220;
-	case NVKM_ENGINE_MSPDEC: return 0x0250;
-	case NVKM_ENGINE_MSPPP : return 0x0260;
-	case NVKM_ENGINE_MSVLD : return 0x0270;
-	case NVKM_ENGINE_VIC   : return 0x0280;
-	case NVKM_ENGINE_MSENC : return 0x0290;
-	case NVKM_ENGINE_NVDEC : return 0x02100270;
-	case NVKM_ENGINE_NVENC :
-		if (engine->subdev.inst)
-			return 0x0210;
-		return 0x02100290;
-	default:
-		WARN_ON(1);
-		return 0;
-	}
-}
-
-struct gk104_fifo_engn *
-gk104_fifo_gpfifo_engine(struct gk104_fifo_chan *chan, struct nvkm_engine *engine)
-{
-	int engi = chan->base.fifo->func->engine_id(chan->base.fifo, engine);
-	if (engi >= 0)
-		return &chan->engn[engi];
-	return NULL;
-}
-
-static int
-gk104_fifo_gpfifo_engine_fini(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine, bool suspend)
-{
-	struct gk104_fifo_chan *chan = gk104_fifo_chan(base);
-	struct nvkm_gpuobj *inst = chan->base.inst;
-	u32 offset = gk104_fifo_gpfifo_engine_addr(engine);
-
-	if (offset) {
-		nvkm_kmap(inst);
-		nvkm_wo32(inst, (offset & 0xffff) + 0x00, 0x00000000);
-		nvkm_wo32(inst, (offset & 0xffff) + 0x04, 0x00000000);
-		if ((offset >>= 16)) {
-			nvkm_wo32(inst, offset + 0x00, 0x00000000);
-			nvkm_wo32(inst, offset + 0x04, 0x00000000);
-		}
-		nvkm_done(inst);
-	}
-
-	return 0;
-}
-
-static int
-gk104_fifo_gpfifo_engine_init(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine)
-{
-	struct gk104_fifo_chan *chan = gk104_fifo_chan(base);
-	struct gk104_fifo_engn *engn = gk104_fifo_gpfifo_engine(chan, engine);
-	struct nvkm_gpuobj *inst = chan->base.inst;
-	u32 offset = gk104_fifo_gpfifo_engine_addr(engine);
-
-	if (offset) {
-		u32 datalo = lower_32_bits(engn->vma->addr) | 0x00000004;
-		u32 datahi = upper_32_bits(engn->vma->addr);
-		nvkm_kmap(inst);
-		nvkm_wo32(inst, (offset & 0xffff) + 0x00, datalo);
-		nvkm_wo32(inst, (offset & 0xffff) + 0x04, datahi);
-		if ((offset >>= 16)) {
-			nvkm_wo32(inst, offset + 0x00, datalo);
-			nvkm_wo32(inst, offset + 0x04, datahi);
-		}
-		nvkm_done(inst);
-	}
-
-	return 0;
-}
-
-void
-gk104_fifo_gpfifo_engine_dtor(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine)
-{
-	struct gk104_fifo_chan *chan = gk104_fifo_chan(base);
-	struct gk104_fifo_engn *engn = gk104_fifo_gpfifo_engine(chan, engine);
-	nvkm_vmm_put(chan->base.vmm, &engn->vma);
-	nvkm_gpuobj_del(&engn->inst);
-}
-
-int
-gk104_fifo_gpfifo_engine_ctor(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine,
-			      struct nvkm_object *object)
-{
-	struct gk104_fifo_chan *chan = gk104_fifo_chan(base);
-	struct gk104_fifo_engn *engn = gk104_fifo_gpfifo_engine(chan, engine);
-	int ret;
-
-	if (!gk104_fifo_gpfifo_engine_addr(engine)) {
-		if (engine->subdev.type != NVKM_ENGINE_CE ||
-		    engine->subdev.device->card_type < GV100)
-			return 0;
-	}
-
-	ret = nvkm_object_bind(object, NULL, 0, &engn->inst);
-	if (ret)
-		return ret;
-
-	if (!gk104_fifo_gpfifo_engine_addr(engine))
-		return 0;
-
-	ret = nvkm_vmm_get(chan->base.vmm, 12, engn->inst->size, &engn->vma);
-	if (ret)
-		return ret;
-
-	return nvkm_memory_map(engn->inst, 0, chan->base.vmm, engn->vma, NULL, 0);
-}
-
 void *
 gk104_fifo_gpfifo_dtor(struct nvkm_fifo_chan *base)
 {
@@ -162,10 +43,6 @@ gk104_fifo_gpfifo_dtor(struct nvkm_fifo_chan *base)
 const struct nvkm_fifo_chan_func
 gk104_fifo_gpfifo_func = {
 	.dtor = gk104_fifo_gpfifo_dtor,
-	.engine_ctor = gk104_fifo_gpfifo_engine_ctor,
-	.engine_dtor = gk104_fifo_gpfifo_engine_dtor,
-	.engine_init = gk104_fifo_gpfifo_engine_init,
-	.engine_fini = gk104_fifo_gpfifo_engine_fini,
 };
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogv100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogv100.c
index 9a2308be73a2..3c2c76bcbe79 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogv100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogv100.c
@@ -28,87 +28,9 @@
 #include <nvif/clc36f.h>
 #include <nvif/unpack.h>
 
-static int
-gv100_fifo_gpfifo_engine_valid(struct gk104_fifo_chan *chan, bool ce, bool valid)
-{
-	const u32 mask = ce ? 0x00020000 : 0x00010000;
-	const u32 data = valid ? mask : 0x00000000;
-
-	if (1) {
-		/* Update engine context validity. */
-		nvkm_kmap(chan->base.inst);
-		nvkm_mo32(chan->base.inst, 0x0ac, mask, data);
-		nvkm_done(chan->base.inst);
-	}
-
-	return 0;
-}
-
-int
-gv100_fifo_gpfifo_engine_fini(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine, bool suspend)
-{
-	struct gk104_fifo_chan *chan = gk104_fifo_chan(base);
-	struct nvkm_gpuobj *inst = chan->base.inst;
-	int ret;
-
-	if (engine->subdev.type == NVKM_ENGINE_CE) {
-		ret = gv100_fifo_gpfifo_engine_valid(chan, true, false);
-		if (ret && suspend)
-			return ret;
-
-		nvkm_kmap(inst);
-		nvkm_wo32(chan->base.inst, 0x220, 0x00000000);
-		nvkm_wo32(chan->base.inst, 0x224, 0x00000000);
-		nvkm_done(inst);
-		return ret;
-	}
-
-	ret = gv100_fifo_gpfifo_engine_valid(chan, false, false);
-	if (ret && suspend)
-		return ret;
-
-	nvkm_kmap(inst);
-	nvkm_wo32(inst, 0x0210, 0x00000000);
-	nvkm_wo32(inst, 0x0214, 0x00000000);
-	nvkm_done(inst);
-	return ret;
-}
-
-int
-gv100_fifo_gpfifo_engine_init(struct nvkm_fifo_chan *base,
-			      struct nvkm_engine *engine)
-{
-	struct gk104_fifo_chan *chan = gk104_fifo_chan(base);
-	struct gk104_fifo_engn *engn = gk104_fifo_gpfifo_engine(chan, engine);
-	struct nvkm_gpuobj *inst = chan->base.inst;
-
-	if (engine->subdev.type == NVKM_ENGINE_CE) {
-		const u64 bar2 = nvkm_memory_bar2(engn->inst->memory);
-
-		nvkm_kmap(inst);
-		nvkm_wo32(chan->base.inst, 0x220, lower_32_bits(bar2));
-		nvkm_wo32(chan->base.inst, 0x224, upper_32_bits(bar2));
-		nvkm_done(inst);
-
-		return gv100_fifo_gpfifo_engine_valid(chan, true, true);
-	}
-
-	nvkm_kmap(inst);
-	nvkm_wo32(inst, 0x210, lower_32_bits(engn->vma->addr) | 0x00000004);
-	nvkm_wo32(inst, 0x214, upper_32_bits(engn->vma->addr));
-	nvkm_done(inst);
-
-	return gv100_fifo_gpfifo_engine_valid(chan, false, true);
-}
-
 static const struct nvkm_fifo_chan_func
 gv100_fifo_gpfifo = {
 	.dtor = gk104_fifo_gpfifo_dtor,
-	.engine_ctor = gk104_fifo_gpfifo_engine_ctor,
-	.engine_dtor = gk104_fifo_gpfifo_engine_dtor,
-	.engine_init = gv100_fifo_gpfifo_engine_init,
-	.engine_fini = gv100_fifo_gpfifo_engine_fini,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifotu102.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifotu102.c
index 6562bdf39cb0..62a0f9664a9d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifotu102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifotu102.c
@@ -31,10 +31,6 @@
 static const struct nvkm_fifo_chan_func
 tu102_fifo_gpfifo = {
 	.dtor = gk104_fifo_gpfifo_dtor,
-	.engine_ctor = gk104_fifo_gpfifo_engine_ctor,
-	.engine_dtor = gk104_fifo_gpfifo_engine_dtor,
-	.engine_init = gv100_fifo_gpfifo_engine_init,
-	.engine_fini = gv100_fifo_gpfifo_engine_fini,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c
index ffca153a9a28..d55dfc93c61e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gv100.c
@@ -89,16 +89,58 @@ gv100_chan = {
 	.doorbell_handle = gv100_chan_doorbell_handle,
 };
 
+void
+gv100_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	u64 addr = 0ULL;
+
+	if (cctx) {
+		addr  = cctx->vctx->vma->addr;
+		addr |= 4ULL;
+	}
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x210, lower_32_bits(addr));
+	nvkm_wo32(chan->inst, 0x214, upper_32_bits(addr));
+	nvkm_mo32(chan->inst, 0x0ac, 0x00010000, cctx ? 0x00010000 : 0x00000000);
+	nvkm_done(chan->inst);
+}
+
 const struct nvkm_engn_func
 gv100_engn = {
 	.chsw = gk104_engn_chsw,
 	.cxid = gk104_engn_cxid,
+	.ctor = gk104_ectx_ctor,
+	.bind = gv100_ectx_bind,
 };
 
+void
+gv100_ectx_ce_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	const u64 bar2 = cctx ? nvkm_memory_bar2(cctx->vctx->inst->memory) : 0ULL;
+
+	nvkm_kmap(chan->inst);
+	nvkm_wo32(chan->inst, 0x220, lower_32_bits(bar2));
+	nvkm_wo32(chan->inst, 0x224, upper_32_bits(bar2));
+	nvkm_mo32(chan->inst, 0x0ac, 0x00020000, cctx ? 0x00020000 : 0x00000000);
+	nvkm_done(chan->inst);
+}
+
+int
+gv100_ectx_ce_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx)
+{
+	if (nvkm_memory_bar2(vctx->inst->memory) == ~0ULL)
+		return -EFAULT;
+
+	return 0;
+}
+
 const struct nvkm_engn_func
 gv100_engn_ce = {
 	.chsw = gk104_engn_chsw,
 	.cxid = gk104_engn_cxid,
+	.ctor = gv100_ectx_ce_ctor,
+	.bind = gv100_ectx_ce_bind,
 };
 
 static bool
@@ -436,7 +478,6 @@ gv100_fifo = {
 	.intr = gk104_fifo_intr,
 	.intr_ctxsw_timeout = gv100_fifo_intr_ctxsw_timeout,
 	.mmu_fault = &gv100_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &gv100_runl,
 	.runq = &gv100_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
index cf8369b8b881..08de6ee612ce 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv04.c
@@ -234,20 +234,6 @@ const struct nvkm_runl_func
 nv04_runl = {
 };
 
-int
-nv04_fifo_engine_id(struct nvkm_fifo *base, struct nvkm_engine *engine)
-{
-	switch (engine->subdev.type) {
-	case NVKM_ENGINE_SW    : return NV04_FIFO_ENGN_SW;
-	case NVKM_ENGINE_GR    : return NV04_FIFO_ENGN_GR;
-	case NVKM_ENGINE_MPEG  : return NV04_FIFO_ENGN_MPEG;
-	case NVKM_ENGINE_DMAOBJ: return NV04_FIFO_ENGN_DMA;
-	default:
-		WARN_ON(1);
-		return 0;
-	}
-}
-
 static const char *
 nv_dma_state_err(u32 state)
 {
@@ -533,7 +519,6 @@ nv04_fifo = {
 	.runl_ctor = nv04_fifo_runl_ctor,
 	.init = nv04_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = nv04_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.runl = &nv04_runl,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
index c7dbcd236630..9658ac3dac91 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv10.c
@@ -97,7 +97,6 @@ nv10_fifo = {
 	.runl_ctor = nv04_fifo_runl_ctor,
 	.init = nv04_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = nv04_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.runl = &nv04_runl,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
index a829b0b08f9b..fb0b106eb09b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv17.c
@@ -126,7 +126,6 @@ nv17_fifo = {
 	.runl_ctor = nv04_fifo_runl_ctor,
 	.init = nv17_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = nv04_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.runl = &nv04_runl,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
index faf19ec3b323..f27ca3e7bb27 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv40.c
@@ -43,7 +43,6 @@ nv40_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm,
 	const u32 base = chan->id * 128;
 
 	chan->ramfc_offset = base;
-	nv04_fifo_chan(chan)->ramfc = base;
 
 	nvkm_kmap(ramfc);
 	nvkm_wo32(ramfc, base + 0x00, offset);
@@ -109,8 +108,52 @@ nv40_chan = {
 	.stop = nv04_chan_stop,
 };
 
+static void
+nv40_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
+	struct nvkm_device *device = fifo->engine.subdev.device;
+	struct nvkm_memory *ramfc = device->imem->ramfc;
+	u32 inst = 0x00000000, reg, ctx;
+	int chid;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_GR:
+		reg = 0x0032e0;
+		ctx = 0x38;
+		break;
+	case NVKM_ENGINE_MPEG:
+		if (WARN_ON(device->chipset < 0x44))
+			return;
+		reg = 0x00330c;
+		ctx = 0x54;
+		break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (cctx)
+		inst = cctx->vctx->inst->addr >> 4;
+
+	spin_lock_irq(&fifo->lock);
+	nvkm_mask(device, 0x002500, 0x00000001, 0x00000000);
+
+	chid = nvkm_rd32(device, 0x003204) & (fifo->chid->nr - 1);
+	if (chid == chan->id)
+		nvkm_wr32(device, reg, inst);
+
+	nvkm_kmap(ramfc);
+	nvkm_wo32(ramfc, chan->ramfc_offset + ctx, inst);
+	nvkm_done(ramfc);
+
+	nvkm_mask(device, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irq(&fifo->lock);
+}
+
 static const struct nvkm_engn_func
 nv40_engn = {
+	.bind = nv40_ectx_bind,
 };
 
 static const struct nvkm_engn_func
@@ -175,7 +218,6 @@ nv40_fifo = {
 	.runl_ctor = nv04_fifo_runl_ctor,
 	.init = nv40_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = nv04_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.runl = &nv04_runl,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
index 7d5c0d026967..f068a27d9338 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.c
@@ -89,7 +89,6 @@ nv50_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm,
 	if (ret)
 		return ret;
 
-	nv50_fifo_chan(chan)->eng = chan->eng;
 	nv50_fifo_chan(chan)->ramht = chan->ramht;
 
 	nvkm_kmap(chan->ramfc);
@@ -139,8 +138,64 @@ nv50_chan = {
 	.stop = nv50_chan_stop,
 };
 
+static void
+nv50_ectx_bind(struct nvkm_engn *engn, struct nvkm_cctx *cctx, struct nvkm_chan *chan)
+{
+	struct nvkm_subdev *subdev = &chan->cgrp->runl->fifo->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u64 start = 0, limit = 0;
+	u32 flags = 0, ptr0, save;
+
+	switch (engn->engine->subdev.type) {
+	case NVKM_ENGINE_GR    : ptr0 = 0x0000; break;
+	case NVKM_ENGINE_MPEG  : ptr0 = 0x0060; break;
+	default:
+		WARN_ON(1);
+		return;
+	}
+
+	if (!cctx) {
+		/* HW bug workaround:
+		 *
+		 * PFIFO will hang forever if the connected engines don't report
+		 * that they've processed the context switch request.
+		 *
+		 * In order for the kickoff to work, we need to ensure all the
+		 * connected engines are in a state where they can answer.
+		 *
+		 * Newer chipsets don't seem to suffer from this issue, and well,
+		 * there's also a "ignore these engines" bitmask reg we can use
+		 * if we hit the issue there..
+		 */
+		save = nvkm_mask(device, 0x00b860, 0x00000001, 0x00000001);
+
+		/* Tell engines to save out contexts. */
+		nvkm_wr32(device, 0x0032fc, chan->inst->addr >> 12);
+		nvkm_msec(device, 2000,
+			if (nvkm_rd32(device, 0x0032fc) != 0xffffffff)
+				break;
+		);
+		nvkm_wr32(device, 0x00b860, save);
+	} else {
+		flags = 0x00190000;
+		start = cctx->vctx->inst->addr;
+		limit = start + cctx->vctx->inst->size - 1;
+	}
+
+	nvkm_kmap(chan->eng);
+	nvkm_wo32(chan->eng, ptr0 + 0x00, flags);
+	nvkm_wo32(chan->eng, ptr0 + 0x04, lower_32_bits(limit));
+	nvkm_wo32(chan->eng, ptr0 + 0x08, lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x0c, upper_32_bits(limit) << 24 |
+					  lower_32_bits(start));
+	nvkm_wo32(chan->eng, ptr0 + 0x10, 0x00000000);
+	nvkm_wo32(chan->eng, ptr0 + 0x14, 0x00000000);
+	nvkm_done(chan->eng);
+}
+
 static const struct nvkm_engn_func
 nv50_engn = {
+	.bind = nv50_ectx_bind,
 };
 
 const struct nvkm_engn_func
@@ -340,7 +395,6 @@ nv50_fifo = {
 	.runl_ctor = nv04_fifo_runl_ctor,
 	.init = nv50_fifo_init,
 	.intr = nv04_fifo_intr,
-	.engine_id = nv04_fifo_engine_id,
 	.pause = nv04_fifo_pause,
 	.start = nv04_fifo_start,
 	.runl = &nv50_runl,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.h
index 1573ea430cf3..3c6d67906730 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/nv50.h
@@ -12,6 +12,4 @@ int nv50_fifo_new_(const struct nvkm_fifo_func *, struct nvkm_device *, enum nvk
 		   struct nvkm_fifo **);
 
 void *nv50_fifo_dtor(struct nvkm_fifo *);
-
-int g84_fifo_engine_id(struct nvkm_fifo *, struct nvkm_engine *);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
index 43f248a109f6..13b218518984 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/priv.h
@@ -4,11 +4,13 @@
 #define nvkm_fifo(p) container_of((p), struct nvkm_fifo, engine)
 #include <engine/fifo.h>
 #include <core/enum.h>
+struct nvkm_cctx;
 struct nvkm_cgrp;
 struct nvkm_engn;
 struct nvkm_memory;
 struct nvkm_runl;
 struct nvkm_runq;
+struct nvkm_vctx;
 struct gk104_fifo;
 
 struct nvkm_fifo_chan_oclass;
@@ -37,7 +39,6 @@ struct nvkm_fifo_func {
 		const struct nvkm_enum *gpcclient;
 	} *mmu_fault;
 
-	int (*engine_id)(struct nvkm_fifo *, struct nvkm_engine *);
 	void (*pause)(struct nvkm_fifo *, unsigned long *);
 	void (*start)(struct nvkm_fifo *, unsigned long *);
 
@@ -74,7 +75,6 @@ int nv04_fifo_chid_ctor(struct nvkm_fifo *, int);
 int nv04_fifo_runl_ctor(struct nvkm_fifo *);
 void nv04_fifo_init(struct nvkm_fifo *);
 irqreturn_t nv04_fifo_intr(struct nvkm_inth *);
-int nv04_fifo_engine_id(struct nvkm_fifo *, struct nvkm_engine *);
 void nv04_fifo_pause(struct nvkm_fifo *, unsigned long *);
 void nv04_fifo_start(struct nvkm_fifo *, unsigned long *);
 extern const struct nvkm_runl_func nv04_runl;
@@ -138,7 +138,6 @@ extern const struct nvkm_fifo_func_mmu_fault gk104_fifo_mmu_fault;
 extern const struct nvkm_enum gk104_fifo_mmu_fault_reason[];
 extern const struct nvkm_enum gk104_fifo_mmu_fault_hubclient[];
 extern const struct nvkm_enum gk104_fifo_mmu_fault_gpcclient[];
-int gk104_fifo_engine_id(struct nvkm_fifo *, struct nvkm_engine *);
 void gk104_runl_insert_chan(struct nvkm_chan *, struct nvkm_memory *, u64);
 void gk104_runl_commit(struct nvkm_runl *, struct nvkm_memory *, u32, int);
 bool gk104_runl_pending(struct nvkm_runl *);
@@ -153,6 +152,7 @@ bool gk104_runq_idle(struct nvkm_runq *);
 extern const struct nvkm_engn_func gk104_engn;
 bool gk104_engn_chsw(struct nvkm_engn *);
 int gk104_engn_cxid(struct nvkm_engn *, bool *cgid);
+int gk104_ectx_ctor(struct nvkm_engn *, struct nvkm_vctx *);
 extern const struct nvkm_engn_func gk104_engn_ce;
 extern const struct nvkm_chan_func_userd gk104_chan_userd;
 extern const struct nvkm_chan_func_ramfc gk104_chan_ramfc;
@@ -189,7 +189,10 @@ void gv100_runl_insert_chan(struct nvkm_chan *, struct nvkm_memory *, u64);
 void gv100_runl_preempt(struct nvkm_runl *);
 extern const struct nvkm_runq_func gv100_runq;
 extern const struct nvkm_engn_func gv100_engn;
+void gv100_ectx_bind(struct nvkm_engn *, struct nvkm_cctx *, struct nvkm_chan *);
 extern const struct nvkm_engn_func gv100_engn_ce;
+int gv100_ectx_ce_ctor(struct nvkm_engn *, struct nvkm_vctx *);
+void gv100_ectx_ce_bind(struct nvkm_engn *, struct nvkm_cctx *, struct nvkm_chan *);
 extern const struct nvkm_chan_func_userd gv100_chan_userd;
 extern const struct nvkm_chan_func_ramfc gv100_chan_ramfc;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
index c3b469356d5d..86bca72c68bc 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/runl.h
@@ -1,9 +1,11 @@
 #ifndef __NVKM_RUNL_H__
 #define __NVKM_RUNL_H__
 #include <core/os.h>
+struct nvkm_cctx;
 struct nvkm_cgrp;
 struct nvkm_chan;
 struct nvkm_memory;
+struct nvkm_vctx;
 enum nvkm_subdev_type;
 
 struct nvkm_engn {
@@ -12,6 +14,8 @@ struct nvkm_engn {
 		int (*cxid)(struct nvkm_engn *, bool *cgid);
 		void (*mmu_fault_trigger)(struct nvkm_engn *);
 		bool (*mmu_fault_triggered)(struct nvkm_engn *);
+		int (*ctor)(struct nvkm_engn *, struct nvkm_vctx *);
+		void (*bind)(struct nvkm_engn *, struct nvkm_cctx *, struct nvkm_chan *);
 	} *func;
 	struct nvkm_runl *runl;
 	int id;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c
index a8452eee2949..608db9b8dc11 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/tu102.c
@@ -273,7 +273,6 @@ tu102_fifo = {
 	.init_pbdmas = tu102_fifo_init_pbdmas,
 	.intr = tu102_fifo_intr,
 	.mmu_fault = &tu102_fifo_mmu_fault,
-	.engine_id = gk104_fifo_engine_id,
 	.nonstall = &gf100_fifo_nonstall,
 	.runl = &tu102_runl,
 	.runq = &gv100_runq,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
index 74584a29454a..f35bdd245233 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
@@ -25,6 +25,7 @@
 #include "chid.h"
 #include "runl.h"
 
+#include <core/gpuobj.h>
 #include <core/oproxy.h>
 
 #include <nvif/if0020.h>
@@ -74,10 +75,17 @@ nvkm_uchan_object_fini_1(struct nvkm_oproxy *oproxy, bool suspend)
 	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
 	struct nvkm_chan *chan = uobj->chan;
 	struct nvkm_cctx *cctx = uobj->cctx;
+	struct nvkm_ectx *ectx = cctx->vctx->ectx;
+
+	if (!ectx->object)
+		return 0;
 
 	/* Unbind engine context from channel, if no longer required. */
 	if (refcount_dec_and_mutex_lock(&cctx->uses, &chan->cgrp->mutex)) {
-		nvkm_chan_cctx_bind(chan, oproxy, NULL);
+		nvkm_chan_cctx_bind(chan, ectx->engn, NULL);
+
+		if (refcount_dec_and_test(&ectx->uses))
+			nvkm_object_fini(ectx->object, false);
 		mutex_unlock(&chan->cgrp->mutex);
 	}
 
@@ -90,14 +98,24 @@ nvkm_uchan_object_init_0(struct nvkm_oproxy *oproxy)
 	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
 	struct nvkm_chan *chan = uobj->chan;
 	struct nvkm_cctx *cctx = uobj->cctx;
+	struct nvkm_ectx *ectx = cctx->vctx->ectx;
 	int ret = 0;
 
+	if (!ectx->object)
+		return 0;
+
 	/* Bind engine context to channel, if it hasn't been already. */
 	if (!refcount_inc_not_zero(&cctx->uses)) {
 		mutex_lock(&chan->cgrp->mutex);
 		if (!refcount_inc_not_zero(&cctx->uses)) {
+			if (!refcount_inc_not_zero(&ectx->uses)) {
+				ret = nvkm_object_init(ectx->object);
+				if (ret == 0)
+					refcount_set(&ectx->uses, 1);
+			}
+
 			if (ret == 0) {
-				nvkm_chan_cctx_bind(chan, oproxy, cctx);
+				nvkm_chan_cctx_bind(chan, ectx->engn, cctx);
 				refcount_set(&cctx->uses, 1);
 			}
 		}
@@ -112,6 +130,9 @@ nvkm_uchan_object_dtor(struct nvkm_oproxy *oproxy)
 {
 	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
 
+	if (!uobj->cctx)
+		return;
+
 	nvkm_chan_cctx_put(uobj->chan, &uobj->cctx);
 }