diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index 3621e7f23477da53b7af04edc60af11fe6f5d1ee..6828d81ed7b99daea875ffd06c4dd399d4aa05dd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -298,7 +298,7 @@ nouveau_abi16_ioctl_channel_alloc(ABI16_IOCTL_ARGS)
 	else
 		init->pushbuf_domains = NOUVEAU_GEM_DOMAIN_GART;
 
-	if (device->card_type < NV_C0) {
+	if (device->card_type < NV_10) {
 		init->subchan[0].handle = 0x00000000;
 		init->subchan[0].grclass = 0x0000;
 		init->subchan[1].handle = NvSw;
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index e84f4c32331bf291f7cd3eeb3f253cd8fd543f37..cc5152be2cf121a0901d323bf41b57886b56b27a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -346,22 +346,17 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart)
 	for (i = 0; i < NOUVEAU_DMA_SKIPS; i++)
 		OUT_RING(chan, 0x00000000);
 
-	/* allocate software object class (used for fences on <= nv05, and
-	 * to signal flip completion), bind it to a subchannel.
-	 */
-	if ((device->card_type < NV_E0) || gart /* nve0: want_nvsw */) {
+	/* allocate software object class (used for fences on <= nv05) */
+	if (device->card_type < NV_10) {
 		ret = nouveau_object_new(nv_object(client), chan->handle,
-					 NvSw, nouveau_abi16_swclass(chan->drm),
-					 NULL, 0, &object);
+					 NvSw, 0x006e, NULL, 0, &object);
 		if (ret)
 			return ret;
 
 		swch = (void *)object->parent;
 		swch->flip = nouveau_flip_complete;
 		swch->flip_data = chan;
-	}
 
-	if (device->card_type < NV_C0) {
 		ret = RING_SPACE(chan, 2);
 		if (ret)
 			return ret;
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 2418b0de589eb631b4fc7e3297c78bdce5557e45..7a3759f1c41a67bac6b48cdcfd5b10b5ac30ae39 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -37,6 +37,7 @@
 #include <engine/device.h>
 #include <engine/disp.h>
 #include <engine/fifo.h>
+#include <engine/software.h>
 
 #include <subdev/vm.h>
 
@@ -191,6 +192,32 @@ nouveau_accel_init(struct nouveau_drm *drm)
 		return;
 	}
 
+	ret = nouveau_object_new(nv_object(drm), NVDRM_CHAN, NVDRM_NVSW,
+				 nouveau_abi16_swclass(drm), NULL, 0, &object);
+	if (ret == 0) {
+		struct nouveau_software_chan *swch = (void *)object->parent;
+		ret = RING_SPACE(drm->channel, 2);
+		if (ret == 0) {
+			if (device->card_type < NV_C0) {
+				BEGIN_NV04(drm->channel, NvSubSw, 0, 1);
+				OUT_RING  (drm->channel, NVDRM_NVSW);
+			} else
+			if (device->card_type < NV_E0) {
+				BEGIN_NVC0(drm->channel, FermiSw, 0, 1);
+				OUT_RING  (drm->channel, 0x001f0000);
+			}
+		}
+		swch = (void *)object->parent;
+		swch->flip = nouveau_flip_complete;
+		swch->flip_data = drm->channel;
+	}
+
+	if (ret) {
+		NV_ERROR(drm, "failed to allocate software object, %d\n", ret);
+		nouveau_accel_fini(drm);
+		return;
+	}
+
 	if (device->card_type < NV_C0) {
 		ret = nouveau_gpuobj_new(drm->device, NULL, 32, 0, 0,
 					&drm->notify);
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.h b/drivers/gpu/drm/nouveau/nouveau_drm.h
index 71ed2dadae61cc7f6e1789fbf58b318538694d49..4b0fb6c66be918857bc6529b3a7c5a7c7a49a498 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.h
@@ -56,6 +56,7 @@ enum nouveau_drm_handle {
 	NVDRM_CONTROL = 0xdddddddc,
 	NVDRM_PUSH    = 0xbbbb0000, /* |= client chid */
 	NVDRM_CHAN    = 0xcccc0000, /* |= client chid */
+	NVDRM_NVSW    = 0x55550000,
 };
 
 struct nouveau_cli {