drm/nouveau: create real execution engine for software object class

Just a cleanup more or less, and to remove the need for special handling of
software objects.

This removes a heap of documentation on dma/graph object formats.  The info
is very out of date with our current understanding, and is far better
documented in rnndb in envytools git.

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

19 files changed, 567 insertions, 282 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index f7103b26ed43..ba3e57c7892d 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -18,6 +18,7 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nv50_fb.o nvc0_fb.o \
              nv04_fifo.o nv10_fifo.o nv40_fifo.o nv50_fifo.o nvc0_fifo.o \
              nve0_fifo.o \
+             nv04_software.o nv50_software.o nvc0_software.o \
              nv04_graph.o nv10_graph.o nv20_graph.o \
              nv40_graph.o nv50_graph.o nvc0_graph.o nve0_graph.o \
              nv40_grctx.o nv50_grctx.o nvc0_grctx.o nve0_grctx.o \
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 730bbb249b01..26c08e9b4a33 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -28,6 +28,7 @@
 #include "nouveau_drm.h"
 #include "nouveau_dma.h"
 #include "nouveau_ramht.h"
+#include "nouveau_software.h"
 
 static int
 nouveau_channel_pushbuf_init(struct nouveau_channel *chan)
@@ -155,8 +156,6 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	}
 
 	NV_DEBUG(dev, "initialising channel %d\n", chan->id);
-	INIT_LIST_HEAD(&chan->nvsw.vbl_wait);
-	INIT_LIST_HEAD(&chan->nvsw.flip);
 	INIT_LIST_HEAD(&chan->fence.pending);
 	spin_lock_init(&chan->fence.lock);
 
@@ -213,6 +212,12 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 		OUT_RING  (chan, 0x00000000);
 	FIRE_RING(chan);
 
+	ret = nouveau_gpuobj_gr_new(chan, NvSw, nouveau_software_class(dev));
+	if (ret) {
+		nouveau_channel_put(&chan);
+		return ret;
+	}
+
 	ret = nouveau_fence_channel_init(chan);
 	if (ret) {
 		nouveau_channel_put(&chan);
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 980b3e187b72..a13f2516d52f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -33,6 +33,7 @@
 #include "nouveau_crtc.h"
 #include "nouveau_dma.h"
 #include "nouveau_connector.h"
+#include "nouveau_software.h"
 #include "nouveau_gpio.h"
 #include "nv50_display.h"
 
@@ -432,6 +433,7 @@ nouveau_page_flip_emit(struct nouveau_channel *chan,
 		       struct nouveau_page_flip_state *s,
 		       struct nouveau_fence **pfence)
 {
+	struct nouveau_software_chan *swch = chan->engctx[NVOBJ_ENGINE_SW];
 	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_device *dev = chan->dev;
 	unsigned long flags;
@@ -439,7 +441,7 @@ nouveau_page_flip_emit(struct nouveau_channel *chan,
 
 	/* Queue it to the pending list */
 	spin_lock_irqsave(&dev->event_lock, flags);
-	list_add_tail(&s->head, &chan->nvsw.flip);
+	list_add_tail(&s->head, &swch->flip);
 	spin_unlock_irqrestore(&dev->event_lock, flags);
 
 	/* Synchronize with the old framebuffer */
@@ -547,20 +549,20 @@ int
 nouveau_finish_page_flip(struct nouveau_channel *chan,
 			 struct nouveau_page_flip_state *ps)
 {
+	struct nouveau_software_chan *swch = chan->engctx[NVOBJ_ENGINE_SW];
 	struct drm_device *dev = chan->dev;
 	struct nouveau_page_flip_state *s;
 	unsigned long flags;
 
 	spin_lock_irqsave(&dev->event_lock, flags);
 
-	if (list_empty(&chan->nvsw.flip)) {
+	if (list_empty(&swch->flip)) {
 		NV_ERROR(dev, "Unexpected pageflip in channel %d.\n", chan->id);
 		spin_unlock_irqrestore(&dev->event_lock, flags);
 		return -EINVAL;
 	}
 
-	s = list_first_entry(&chan->nvsw.flip,
-			     struct nouveau_page_flip_state, head);
+	s = list_first_entry(&swch->flip, struct nouveau_page_flip_state, head);
 	if (s->event) {
 		struct drm_pending_vblank_event *e = s->event;
 		struct timeval now;
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 87619a94fa42..8b0e07740291 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -296,14 +296,6 @@ struct nouveau_channel {
 	uint32_t sw_subchannel[8];
 
 	struct nouveau_vma dispc_vma[4];
-	struct {
-		struct nouveau_gpuobj *vblsem;
-		uint32_t vblsem_head;
-		uint32_t vblsem_offset;
-		uint32_t vblsem_rval;
-		struct list_head vbl_wait;
-		struct list_head flip;
-	} nvsw;
 
 	struct {
 		bool active;
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index ff5969d057ea..aca4719c287f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -32,6 +32,7 @@
 
 #include "nouveau_drv.h"
 #include "nouveau_ramht.h"
+#include "nouveau_software.h"
 #include "nouveau_dma.h"
 
 #define USE_REFCNT(dev) (nouveau_private(dev)->chipset >= 0x10)
@@ -503,11 +504,6 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 	int ret;
 
 	if (dev_priv->card_type < NV_C0) {
-		/* Create an NV_SW object for various sync purposes */
-		ret = nouveau_gpuobj_gr_new(chan, NvSw, NV_SW);
-		if (ret)
-			return ret;
-
 		ret = RING_SPACE(chan, 2);
 		if (ret)
 			return ret;
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index cc419fae794b..637636b55c73 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -35,6 +35,7 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 #include "nouveau_ramht.h"
+#include "nouveau_software.h"
 #include "nouveau_vm.h"
 #include "nv50_display.h"
 
@@ -133,37 +134,6 @@ nouveau_gpuobj_mthd_call2(struct drm_device *dev, int chid,
 	return ret;
 }
 
-/* NVidia uses context objects to drive drawing operations.
-
-   Context objects can be selected into 8 subchannels in the FIFO,
-   and then used via DMA command buffers.
-
-   A context object is referenced by a user defined handle (CARD32). The HW
-   looks up graphics objects in a hash table in the instance RAM.
-
-   An entry in the hash table consists of 2 CARD32. The first CARD32 contains
-   the handle, the second one a bitfield, that contains the address of the
-   object in instance RAM.
-
-   The format of the second CARD32 seems to be:
-
-   NV4 to NV30:
-
-   15: 0  instance_addr >> 4
-   17:16  engine (here uses 1 = graphics)
-   28:24  channel id (here uses 0)
-   31	  valid (use 1)
-
-   NV40:
-
-   15: 0  instance_addr >> 4   (maybe 19-0)
-   21:20  engine (here uses 1 = graphics)
-   I'm unsure about the other bits, but using 0 seems to work.
-
-   The key into the hash table depends on the object handle and channel id and
-   is given as:
-*/
-
 int
 nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 		   uint32_t size, int align, uint32_t flags,
@@ -361,34 +331,6 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, u32 pinst, u64 vinst,
 	return 0;
 }
 
-/*
-   DMA objects are used to reference a piece of memory in the
-   framebuffer, PCI or AGP address space. Each object is 16 bytes big
-   and looks as follows:
-
-   entry[0]
-   11:0  class (seems like I can always use 0 here)
-   12    page table present?
-   13    page entry linear?
-   15:14 access: 0 rw, 1 ro, 2 wo
-   17:16 target: 0 NV memory, 1 NV memory tiled, 2 PCI, 3 AGP
-   31:20 dma adjust (bits 0-11 of the address)
-   entry[1]
-   dma limit (size of transfer)
-   entry[X]
-   1     0 readonly, 1 readwrite
-   31:12 dma frame address of the page (bits 12-31 of the address)
-   entry[N]
-   page table terminator, same value as the first pte, as does nvidia
-   rivatv uses 0xffffffff
-
-   Non linear page tables need a list of frame addresses afterwards,
-   the rivatv project has some info on this.
-
-   The method below creates a DMA object in instance RAM and returns a handle
-   to it that can be used to set up context objects.
-*/
-
 void
 nv50_gpuobj_dma_init(struct nouveau_gpuobj *obj, u32 offset, int class,
 		     u64 base, u64 size, int target, int access,
@@ -540,82 +482,6 @@ nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base,
 	return 0;
 }
 
-/* Context objects in the instance RAM have the following structure.
- * On NV40 they are 32 byte long, on NV30 and smaller 16 bytes.
-
-   NV4 - NV30:
-
-   entry[0]
-   11:0 class
-   12   chroma key enable
-   13   user clip enable
-   14   swizzle enable
-   17:15 patch config:
-       scrcopy_and, rop_and, blend_and, scrcopy, srccopy_pre, blend_pre
-   18   synchronize enable
-   19   endian: 1 big, 0 little
-   21:20 dither mode
-   23    single step enable
-   24    patch status: 0 invalid, 1 valid
-   25    context_surface 0: 1 valid
-   26    context surface 1: 1 valid
-   27    context pattern: 1 valid
-   28    context rop: 1 valid
-   29,30 context beta, beta4
-   entry[1]
-   7:0   mono format
-   15:8  color format
-   31:16 notify instance address
-   entry[2]
-   15:0  dma 0 instance address
-   31:16 dma 1 instance address
-   entry[3]
-   dma method traps
-
-   NV40:
-   No idea what the exact format is. Here's what can be deducted:
-
-   entry[0]:
-   11:0  class  (maybe uses more bits here?)
-   17    user clip enable
-   21:19 patch config
-   25    patch status valid ?
-   entry[1]:
-   15:0  DMA notifier  (maybe 20:0)
-   entry[2]:
-   15:0  DMA 0 instance (maybe 20:0)
-   24    big endian
-   entry[3]:
-   15:0  DMA 1 instance (maybe 20:0)
-   entry[4]:
-   entry[5]:
-   set to 0?
-*/
-static int
-nouveau_gpuobj_sw_new(struct nouveau_channel *chan, u32 handle, u16 class)
-{
-	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
-	struct nouveau_gpuobj *gpuobj;
-	int ret;
-
-	gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
-	if (!gpuobj)
-		return -ENOMEM;
-	gpuobj->dev = chan->dev;
-	gpuobj->engine = NVOBJ_ENGINE_SW;
-	gpuobj->class = class;
-	kref_init(&gpuobj->refcount);
-	gpuobj->cinst = 0x40;
-
-	spin_lock(&dev_priv->ramin_lock);
-	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
-	spin_unlock(&dev_priv->ramin_lock);
-
-	ret = nouveau_ramht_insert(chan, handle, gpuobj);
-	nouveau_gpuobj_ref(NULL, &gpuobj);
-	return ret;
-}
-
 int
 nouveau_gpuobj_gr_new(struct nouveau_channel *chan, u32 handle, int class)
 {
@@ -632,9 +498,6 @@ nouveau_gpuobj_gr_new(struct nouveau_channel *chan, u32 handle, int class)
 		if (oc->id != class)
 			continue;
 
-		if (oc->engine == NVOBJ_ENGINE_SW)
-			return nouveau_gpuobj_sw_new(chan, handle, class);
-
 		if (!chan->engctx[oc->engine]) {
 			ret = eng->context_new(chan, oc->engine);
 			if (ret)
@@ -956,6 +819,17 @@ int nouveau_ioctl_grobj_alloc(struct drm_device *dev, void *data,
 	if (init->handle == ~0)
 		return -EINVAL;
 
+	/* compatibility with userspace that assumes 506e for all chipsets */
+	if (init->class == 0x506e) {
+		init->class = nouveau_software_class(dev);
+		if (init->class == 0x906e)
+			return 0;
+	} else
+	if (init->class == 0x906e) {
+		NV_ERROR(dev, "906e not supported yet\n");
+		return -EINVAL;
+	}
+
 	chan = nouveau_channel_get(file_priv, init->channel);
 	if (IS_ERR(chan))
 		return PTR_ERR(chan);
diff --git a/drivers/gpu/drm/nouveau/nouveau_software.h b/drivers/gpu/drm/nouveau/nouveau_software.h
new file mode 100644
index 000000000000..fe30a8f47f8c
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_software.h
@@ -0,0 +1,68 @@
+#ifndef __NOUVEAU_SOFTWARE_H__
+#define __NOUVEAU_SOFTWARE_H__
+
+struct nouveau_software_priv {
+	struct nouveau_exec_engine base;
+	struct list_head vblank;
+};
+
+struct nouveau_software_chan {
+	struct list_head flip;
+	struct {
+		struct list_head list;
+		struct nouveau_bo *bo;
+		u32 offset;
+		u32 value;
+		u32 head;
+	} vblank;
+};
+
+static inline void
+nouveau_software_vblank(struct drm_device *dev, int crtc)
+{
+	struct nouveau_software_priv *psw = nv_engine(dev, NVOBJ_ENGINE_SW);
+	struct nouveau_software_chan *pch, *tmp;
+
+	list_for_each_entry_safe(pch, tmp, &psw->vblank, vblank.list) {
+		if (pch->vblank.head != crtc)
+			continue;
+
+		nouveau_bo_wr32(pch->vblank.bo, pch->vblank.offset,
+						pch->vblank.value);
+		list_del(&pch->vblank.list);
+		drm_vblank_put(dev, crtc);
+	}
+}
+
+static inline void
+nouveau_software_context_new(struct nouveau_software_chan *pch)
+{
+	INIT_LIST_HEAD(&pch->flip);
+}
+
+static inline void
+nouveau_software_create(struct nouveau_software_priv *psw)
+{
+	INIT_LIST_HEAD(&psw->vblank);
+}
+
+static inline u16
+nouveau_software_class(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	if (dev_priv->card_type <= NV_04)
+		return 0x006e;
+	if (dev_priv->card_type <= NV_40)
+		return 0x016e;
+	if (dev_priv->card_type <= NV_50)
+		return 0x506e;
+	if (dev_priv->card_type <= NV_E0)
+		return 0x906e;
+	return 0x0000;
+}
+
+int nv04_software_create(struct drm_device *);
+int nv50_software_create(struct drm_device *);
+int nvc0_software_create(struct drm_device *);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index e76edb51373d..ed83c425fcf3 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -39,6 +39,7 @@
 #include "nouveau_gpio.h"
 #include "nouveau_pm.h"
 #include "nv50_display.h"
+#include "nouveau_software.h"
 
 static void nouveau_stub_takedown(struct drm_device *dev) {}
 static int nouveau_stub_init(struct drm_device *dev) { return 0; }
@@ -767,6 +768,26 @@ nouveau_card_init(struct drm_device *dev)
 	if (!dev_priv->noaccel) {
 		switch (dev_priv->card_type) {
 		case NV_04:
+		case NV_10:
+		case NV_20:
+		case NV_30:
+		case NV_40:
+			nv04_software_create(dev);
+			break;
+		case NV_50:
+			nv50_software_create(dev);
+			break;
+		case NV_C0:
+		case NV_D0:
+		case NV_E0:
+			nvc0_software_create(dev);
+			break;
+		default:
+			break;
+		}
+
+		switch (dev_priv->card_type) {
+		case NV_04:
 			nv04_graph_create(dev);
 			break;
 		case NV_10:
diff --git a/drivers/gpu/drm/nouveau/nv04_graph.c b/drivers/gpu/drm/nouveau/nv04_graph.c
index dbdea8ed3925..442b4df44fad 100644
--- a/drivers/gpu/drm/nouveau/nv04_graph.c
+++ b/drivers/gpu/drm/nouveau/nv04_graph.c
@@ -550,28 +550,6 @@ nv04_graph_fini(struct drm_device *dev, int engine, bool suspend)
 	return 0;
 }
 
-static int
-nv04_graph_mthd_set_ref(struct nouveau_channel *chan,
-			u32 class, u32 mthd, u32 data)
-{
-	atomic_set(&chan->fence.last_sequence_irq, data);
-	return 0;
-}
-
-int
-nv04_graph_mthd_page_flip(struct nouveau_channel *chan,
-			  u32 class, u32 mthd, u32 data)
-{
-	struct drm_device *dev = chan->dev;
-	struct nouveau_page_flip_state s;
-
-	if (!nouveau_finish_page_flip(chan, &s))
-		nv_set_crtc_base(dev, s.crtc,
-				 s.offset + s.y * s.pitch + s.x * s.bpp / 8);
-
-	return 0;
-}
-
 /*
  * Software methods, why they are needed, and how they all work:
  *
@@ -1345,9 +1323,5 @@ nv04_graph_create(struct drm_device *dev)
 	NVOBJ_MTHD (dev, 0x005e, 0x0198, nv04_graph_mthd_bind_surf2d);
 	NVOBJ_MTHD (dev, 0x005e, 0x02fc, nv04_graph_mthd_set_operation);
 
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0150, nv04_graph_mthd_set_ref);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv04_software.c b/drivers/gpu/drm/nouveau/nv04_software.c
new file mode 100644
index 000000000000..d37de6e3c3b2
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv04_software.c
@@ -0,0 +1,153 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_ramht.h"
+#include "nouveau_software.h"
+#include "nouveau_hw.h"
+
+struct nv04_software_priv {
+	struct nouveau_software_priv base;
+};
+
+struct nv04_software_chan {
+	struct nouveau_software_chan base;
+};
+
+static int
+mthd_fence(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	atomic_set(&chan->fence.last_sequence_irq, data);
+	return 0;
+}
+
+static int
+mthd_flip(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+
+	struct nouveau_page_flip_state state;
+
+	if (!nouveau_finish_page_flip(chan, &state)) {
+		nv_set_crtc_base(chan->dev, state.crtc, state.offset +
+				 state.y * state.pitch +
+				 state.x * state.bpp / 8);
+	}
+
+	return 0;
+}
+
+static int
+nv04_software_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct nv04_software_chan *pch;
+
+	pch = kzalloc(sizeof(*pch), GFP_KERNEL);
+	if (!pch)
+		return -ENOMEM;
+
+	nouveau_software_context_new(&pch->base);
+	atomic_set(&chan->fence.last_sequence_irq, 0);
+	chan->engctx[engine] = pch;
+	return 0;
+}
+
+static void
+nv04_software_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct nv04_software_chan *pch = chan->engctx[engine];
+	chan->engctx[engine] = NULL;
+	kfree(pch);
+}
+
+static int
+nv04_software_object_new(struct nouveau_channel *chan, int engine,
+			 u32 handle, u16 class)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, 0, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 0;
+	obj->class  = class;
+
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
+}
+
+static int
+nv04_software_init(struct drm_device *dev, int engine)
+{
+	return 0;
+}
+
+static int
+nv04_software_fini(struct drm_device *dev, int engine, bool suspend)
+{
+	return 0;
+}
+
+static void
+nv04_software_destroy(struct drm_device *dev, int engine)
+{
+	struct nv04_software_priv *psw = nv_engine(dev, engine);
+
+	NVOBJ_ENGINE_DEL(dev, SW);
+	kfree(psw);
+}
+
+int
+nv04_software_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv04_software_priv *psw;
+
+	psw = kzalloc(sizeof(*psw), GFP_KERNEL);
+	if (!psw)
+		return -ENOMEM;
+
+	psw->base.base.destroy = nv04_software_destroy;
+	psw->base.base.init = nv04_software_init;
+	psw->base.base.fini = nv04_software_fini;
+	psw->base.base.context_new = nv04_software_context_new;
+	psw->base.base.context_del = nv04_software_context_del;
+	psw->base.base.object_new = nv04_software_object_new;
+	nouveau_software_create(&psw->base);
+
+	NVOBJ_ENGINE_ADD(dev, SW, &psw->base.base);
+	if (dev_priv->card_type <= NV_04) {
+		NVOBJ_CLASS(dev, 0x006e, SW);
+		NVOBJ_MTHD (dev, 0x006e, 0x0150, mthd_fence);
+		NVOBJ_MTHD (dev, 0x006e, 0x0500, mthd_flip);
+	} else {
+		NVOBJ_CLASS(dev, 0x016e, SW);
+		NVOBJ_MTHD (dev, 0x016e, 0x0500, mthd_flip);
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nv10_graph.c b/drivers/gpu/drm/nouveau/nv10_graph.c
index 7255e4a4d3f3..10c0eb5d4233 100644
--- a/drivers/gpu/drm/nouveau/nv10_graph.c
+++ b/drivers/gpu/drm/nouveau/nv10_graph.c
@@ -1153,10 +1153,6 @@ nv10_graph_create(struct drm_device *dev)
 	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
 	nouveau_irq_register(dev, 12, nv10_graph_isr);
 
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
 	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
 	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
 	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
diff --git a/drivers/gpu/drm/nouveau/nv20_graph.c b/drivers/gpu/drm/nouveau/nv20_graph.c
index 183e37512ef9..385e2b49a554 100644
--- a/drivers/gpu/drm/nouveau/nv20_graph.c
+++ b/drivers/gpu/drm/nouveau/nv20_graph.c
@@ -796,10 +796,6 @@ nv20_graph_create(struct drm_device *dev)
 	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
 	nouveau_irq_register(dev, 12, nv20_graph_isr);
 
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
-
 	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
 	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
 	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index 6d8147b07c83..03837200762d 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -439,7 +439,6 @@ nv40_graph_create(struct drm_device *dev)
 	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
 	nouveau_irq_register(dev, 12, nv40_graph_isr);
 
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
 	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
 	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
 	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
@@ -462,8 +461,5 @@ nv40_graph_create(struct drm_device *dev)
 	else
 		NVOBJ_CLASS(dev, 0x4097, GR);
 
-	/* nvsw */
-	NVOBJ_CLASS(dev, 0x506e, SW);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index b526e3f61c17..362a4d7cd566 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -645,20 +645,7 @@ nv50_display_script_select(struct drm_device *dev, struct dcb_entry *dcb,
 static void
 nv50_display_vblank_crtc_handler(struct drm_device *dev, int crtc)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_channel *chan, *tmp;
-
-	list_for_each_entry_safe(chan, tmp, &dev_priv->vbl_waiting,
-				 nvsw.vbl_wait) {
-		if (chan->nvsw.vblsem_head != crtc)
-			continue;
-
-		nouveau_bo_wr32(chan->notifier_bo, chan->nvsw.vblsem_offset,
-						chan->nvsw.vblsem_rval);
-		list_del(&chan->nvsw.vbl_wait);
-		drm_vblank_put(dev, crtc);
-	}
-
+	nouveau_software_vblank(dev, crtc);
 	drm_handle_vblank(dev, crtc);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nv50_display.h b/drivers/gpu/drm/nouveau/nv50_display.h
index 5d3dd14d2837..e9db9b97f041 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.h
+++ b/drivers/gpu/drm/nouveau/nv50_display.h
@@ -33,6 +33,7 @@
 #include "nouveau_dma.h"
 #include "nouveau_reg.h"
 #include "nouveau_crtc.h"
+#include "nouveau_software.h"
 #include "nv50_evo.h"
 
 struct nv50_display_crtc {
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index 2698d80c8eb2..a46e060eb399 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -333,70 +333,6 @@ nv50_graph_context_switch(struct drm_device *dev)
 		NV40_PGRAPH_INTR_EN) | NV_PGRAPH_INTR_CONTEXT_SWITCH);
 }
 
-static int
-nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan,
-			   u32 class, u32 mthd, u32 data)
-{
-	struct nouveau_gpuobj *gpuobj;
-
-	gpuobj = nouveau_ramht_find(chan, data);
-	if (!gpuobj)
-		return -ENOENT;
-
-	if (nouveau_notifier_offset(gpuobj, NULL))
-		return -EINVAL;
-
-	chan->nvsw.vblsem = gpuobj;
-	chan->nvsw.vblsem_offset = ~0;
-	return 0;
-}
-
-static int
-nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan,
-			      u32 class, u32 mthd, u32 data)
-{
-	if (nouveau_notifier_offset(chan->nvsw.vblsem, &data))
-		return -ERANGE;
-
-	chan->nvsw.vblsem_offset = data >> 2;
-	return 0;
-}
-
-static int
-nv50_graph_nvsw_vblsem_release_val(struct nouveau_channel *chan,
-				   u32 class, u32 mthd, u32 data)
-{
-	chan->nvsw.vblsem_rval = data;
-	return 0;
-}
-
-static int
-nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan,
-			       u32 class, u32 mthd, u32 data)
-{
-	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (!chan->nvsw.vblsem || chan->nvsw.vblsem_offset == ~0 || data > 1)
-		return -EINVAL;
-
-	drm_vblank_get(dev, data);
-
-	chan->nvsw.vblsem_head = data;
-	list_add(&chan->nvsw.vbl_wait, &dev_priv->vbl_waiting);
-
-	return 0;
-}
-
-static int
-nv50_graph_nvsw_mthd_page_flip(struct nouveau_channel *chan,
-			       u32 class, u32 mthd, u32 data)
-{
-	nouveau_finish_page_flip(chan, NULL);
-	return 0;
-}
-
-
 static void
 nv50_graph_tlb_flush(struct drm_device *dev, int engine)
 {
@@ -1018,14 +954,6 @@ nv50_graph_create(struct drm_device *dev)
 
 	nouveau_irq_register(dev, 12, nv50_graph_isr);
 
-	/* NVSW really doesn't live here... */
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
-	NVOBJ_MTHD (dev, 0x506e, 0x018c, nv50_graph_nvsw_dma_vblsem);
-	NVOBJ_MTHD (dev, 0x506e, 0x0400, nv50_graph_nvsw_vblsem_offset);
-	NVOBJ_MTHD (dev, 0x506e, 0x0404, nv50_graph_nvsw_vblsem_release_val);
-	NVOBJ_MTHD (dev, 0x506e, 0x0408, nv50_graph_nvsw_vblsem_release);
-	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv50_graph_nvsw_mthd_page_flip);
-
 	NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base);
 	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
 	NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */
diff --git a/drivers/gpu/drm/nouveau/nv50_software.c b/drivers/gpu/drm/nouveau/nv50_software.c
new file mode 100644
index 000000000000..0e64d80d5363
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv50_software.c
@@ -0,0 +1,190 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_ramht.h"
+#include "nouveau_software.h"
+
+struct nv50_software_priv {
+	struct nouveau_software_priv base;
+};
+
+struct nv50_software_chan {
+	struct nouveau_software_chan base;
+	struct {
+		struct nouveau_gpuobj *object;
+	} vblank;
+};
+
+static int
+mthd_dma_vblsem(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	struct nv50_software_chan *pch = chan->engctx[NVOBJ_ENGINE_SW];
+	struct nouveau_gpuobj *gpuobj;
+
+	gpuobj = nouveau_ramht_find(chan, data);
+	if (!gpuobj)
+		return -ENOENT;
+
+	if (nouveau_notifier_offset(gpuobj, NULL))
+		return -EINVAL;
+
+	pch->vblank.object = gpuobj;
+	pch->base.vblank.offset = ~0;
+	return 0;
+}
+
+static int
+mthd_vblsem_offset(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	struct nv50_software_chan *pch = chan->engctx[NVOBJ_ENGINE_SW];
+
+	if (nouveau_notifier_offset(pch->vblank.object, &data))
+		return -ERANGE;
+
+	pch->base.vblank.offset = data >> 2;
+	return 0;
+}
+
+static int
+mthd_vblsem_value(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	struct nv50_software_chan *pch = chan->engctx[NVOBJ_ENGINE_SW];
+	pch->base.vblank.value = data;
+	return 0;
+}
+
+static int
+mthd_vblsem_release(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	struct nv50_software_priv *psw = nv_engine(chan->dev, NVOBJ_ENGINE_SW);
+	struct nv50_software_chan *pch = chan->engctx[NVOBJ_ENGINE_SW];
+	struct drm_device *dev = chan->dev;
+
+	if (!pch->vblank.object || pch->base.vblank.offset == ~0 || data > 1)
+		return -EINVAL;
+
+	drm_vblank_get(dev, data);
+
+	pch->base.vblank.head = data;
+	list_add(&pch->base.vblank.list, &psw->base.vblank);
+	return 0;
+}
+
+static int
+mthd_flip(struct nouveau_channel *chan, u32 class, u32 mthd, u32 data)
+{
+	nouveau_finish_page_flip(chan, NULL);
+	return 0;
+}
+
+static int
+nv50_software_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct nv50_software_chan *pch;
+
+	pch = kzalloc(sizeof(*pch), GFP_KERNEL);
+	if (!pch)
+		return -ENOMEM;
+
+	nouveau_software_context_new(&pch->base);
+	pch->base.vblank.bo = chan->notifier_bo;
+
+	chan->engctx[engine] = pch;
+	return 0;
+}
+
+static void
+nv50_software_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct nv50_software_chan *pch = chan->engctx[engine];
+	chan->engctx[engine] = NULL;
+	kfree(pch);
+}
+
+static int
+nv50_software_object_new(struct nouveau_channel *chan, int engine,
+			 u32 handle, u16 class)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, 0, &obj);
+	if (ret)
+		return ret;
+	obj->engine = 0;
+	obj->class  = class;
+
+	ret = nouveau_ramht_insert(chan, handle, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	return ret;
+}
+
+static int
+nv50_software_init(struct drm_device *dev, int engine)
+{
+	return 0;
+}
+
+static int
+nv50_software_fini(struct drm_device *dev, int engine, bool suspend)
+{
+	return 0;
+}
+
+static void
+nv50_software_destroy(struct drm_device *dev, int engine)
+{
+	struct nv50_software_priv *psw = nv_engine(dev, engine);
+
+	NVOBJ_ENGINE_DEL(dev, SW);
+	kfree(psw);
+}
+
+int
+nv50_software_create(struct drm_device *dev)
+{
+	struct nv50_software_priv *psw = kzalloc(sizeof(*psw), GFP_KERNEL);
+	if (!psw)
+		return -ENOMEM;
+
+	psw->base.base.destroy = nv50_software_destroy;
+	psw->base.base.init = nv50_software_init;
+	psw->base.base.fini = nv50_software_fini;
+	psw->base.base.context_new = nv50_software_context_new;
+	psw->base.base.context_del = nv50_software_context_del;
+	psw->base.base.object_new = nv50_software_object_new;
+	nouveau_software_create(&psw->base);
+
+	NVOBJ_ENGINE_ADD(dev, SW, &psw->base.base);
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x018c, mthd_dma_vblsem);
+	NVOBJ_MTHD (dev, 0x506e, 0x0400, mthd_vblsem_offset);
+	NVOBJ_MTHD (dev, 0x506e, 0x0404, mthd_vblsem_value);
+	NVOBJ_MTHD (dev, 0x506e, 0x0408, mthd_vblsem_release);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, mthd_flip);
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvc0_fifo.c b/drivers/gpu/drm/nouveau/nvc0_fifo.c
index 50d68a7a1379..f47f39645c3e 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fifo.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fifo.c
@@ -278,7 +278,6 @@ nvc0_fifo_create(struct drm_device *dev)
 		goto error;
 
 	nouveau_irq_register(dev, 8, nvc0_fifo_isr);
-	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
 	return 0;
 
 error:
diff --git a/drivers/gpu/drm/nouveau/nvc0_software.c b/drivers/gpu/drm/nouveau/nvc0_software.c
new file mode 100644
index 000000000000..a1b43f580995
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvc0_software.c
@@ -0,0 +1,106 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_ramht.h"
+#include "nouveau_software.h"
+
+struct nvc0_software_priv {
+	struct nouveau_software_priv base;
+};
+
+struct nvc0_software_chan {
+	struct nouveau_software_chan base;
+};
+
+static int
+nvc0_software_context_new(struct nouveau_channel *chan, int engine)
+{
+	struct nvc0_software_chan *pch;
+
+	pch = kzalloc(sizeof(*pch), GFP_KERNEL);
+	if (!pch)
+		return -ENOMEM;
+
+	nouveau_software_context_new(&pch->base);
+	chan->engctx[engine] = pch;
+	return 0;
+}
+
+static void
+nvc0_software_context_del(struct nouveau_channel *chan, int engine)
+{
+	struct nvc0_software_chan *pch = chan->engctx[engine];
+	chan->engctx[engine] = NULL;
+	kfree(pch);
+}
+
+static int
+nvc0_software_object_new(struct nouveau_channel *chan, int engine,
+			 u32 handle, u16 class)
+{
+	return 0;
+}
+
+static int
+nvc0_software_init(struct drm_device *dev, int engine)
+{
+	return 0;
+}
+
+static int
+nvc0_software_fini(struct drm_device *dev, int engine, bool suspend)
+{
+	return 0;
+}
+
+static void
+nvc0_software_destroy(struct drm_device *dev, int engine)
+{
+	struct nvc0_software_priv *psw = nv_engine(dev, engine);
+
+	NVOBJ_ENGINE_DEL(dev, SW);
+	kfree(psw);
+}
+
+int
+nvc0_software_create(struct drm_device *dev)
+{
+	struct nvc0_software_priv *psw = kzalloc(sizeof(*psw), GFP_KERNEL);
+	if (!psw)
+		return -ENOMEM;
+
+	psw->base.base.destroy = nvc0_software_destroy;
+	psw->base.base.init = nvc0_software_init;
+	psw->base.base.fini = nvc0_software_fini;
+	psw->base.base.context_new = nvc0_software_context_new;
+	psw->base.base.context_del = nvc0_software_context_del;
+	psw->base.base.object_new = nvc0_software_object_new;
+	nouveau_software_create(&psw->base);
+
+	NVOBJ_ENGINE_ADD(dev, SW, &psw->base.base);
+	NVOBJ_CLASS(dev, 0x906e, SW);
+	return 0;
+}