drm/nv40: implement ctxprog/state generation

The context programs are *very* simple compared to the ones used by
the binary driver.  There's notes in nv40_grctx.c explaining most of
the things we don't implement.  If we discover if/why any of it is
required further down the track, we'll handle it then.

The PGRAPH state generated for each chipset should match what NVIDIA
do almost exactly (there's a couple of exceptions).  If someone has
a lot of time on their hands, they could figure out the mapping of
object/method to PGRAPH register and demagic the initial state a little,
it's not terribly important however.

At time of commit, confirmed to be working at least well enough for
accelerated X (and where tested, for 3D apps) on NV40, NV43, NV44, NV46,
NV49, NV4A, NV4B and NV4E.

A module option has been added to force the use of external firmware
blobs if it becomes required.

8 files changed, 1031 insertions, 182 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index 1d90d4d0144f..48c290b5da8c 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -8,14 +8,15 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nouveau_sgdma.o nouveau_dma.o \
              nouveau_bo.o nouveau_fence.o nouveau_gem.o nouveau_ttm.o \
              nouveau_hw.o nouveau_calc.o nouveau_bios.o nouveau_i2c.o \
-	     nouveau_display.o nouveau_connector.o nouveau_fbcon.o \
-	     nouveau_dp.o \
+             nouveau_display.o nouveau_connector.o nouveau_fbcon.o \
+             nouveau_dp.o nouveau_grctx.o \
              nv04_timer.o \
              nv04_mc.o nv40_mc.o nv50_mc.o \
              nv04_fb.o nv10_fb.o nv40_fb.o \
              nv04_fifo.o nv10_fifo.o nv40_fifo.o nv50_fifo.o \
              nv04_graph.o nv10_graph.o nv20_graph.o \
              nv40_graph.o nv50_graph.o \
+             nv40_grctx.o \
              nv04_instmem.o nv50_instmem.o \
              nv50_crtc.o nv50_dac.o nv50_sor.o \
              nv50_cursor.o nv50_display.o nv50_fbcon.o \
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.c b/drivers/gpu/drm/nouveau/nouveau_drv.c
index 071738591c1a..2dd3c02742fa 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.c
@@ -35,6 +35,10 @@
 
 #include "drm_pciids.h"
 
+MODULE_PARM_DESC(ctxfw, "Use external firmware blob for grctx init (NV40)");
+int nouveau_ctxfw = 0;
+module_param_named(ctxfw, nouveau_ctxfw, int, 0400);
+
 MODULE_PARM_DESC(noagp, "Disable AGP");
 int nouveau_noagp;
 module_param_named(noagp, nouveau_noagp, int, 0400);
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index bc6ffa5eeac7..5f8cbb79c499 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -54,6 +54,7 @@ struct nouveau_fpriv {
 #include "nouveau_drm.h"
 #include "nouveau_reg.h"
 #include "nouveau_bios.h"
+struct nouveau_grctx;
 
 #define MAX_NUM_DCB_ENTRIES 16
 
@@ -317,6 +318,7 @@ struct nouveau_pgraph_engine {
 	bool accel_blocked;
 	void *ctxprog;
 	void *ctxvals;
+	int grctx_size;
 
 	int  (*init)(struct drm_device *);
 	void (*takedown)(struct drm_device *);
@@ -647,6 +649,7 @@ extern int nouveau_fbpercrtc;
 extern char *nouveau_tv_norm;
 extern int nouveau_reg_debug;
 extern char *nouveau_vbios;
+extern int nouveau_ctxfw;
 
 /* nouveau_state.c */
 extern void nouveau_preclose(struct drm_device *dev, struct drm_file *);
@@ -959,9 +962,7 @@ extern int  nv40_graph_create_context(struct nouveau_channel *);
 extern void nv40_graph_destroy_context(struct nouveau_channel *);
 extern int  nv40_graph_load_context(struct nouveau_channel *);
 extern int  nv40_graph_unload_context(struct drm_device *);
-extern int  nv40_grctx_init(struct drm_device *);
-extern void nv40_grctx_fini(struct drm_device *);
-extern void nv40_grctx_vals_load(struct drm_device *, struct nouveau_gpuobj *);
+extern void nv40_grctx_init(struct nouveau_grctx *);
 
 /* nv50_graph.c */
 extern struct nouveau_pgraph_object_class nv50_graph_grclass[];
@@ -975,6 +976,12 @@ extern int  nv50_graph_load_context(struct nouveau_channel *);
 extern int  nv50_graph_unload_context(struct drm_device *);
 extern void nv50_graph_context_switch(struct drm_device *);
 
+/* nouveau_grctx.c */
+extern int  nouveau_grctx_prog_load(struct drm_device *);
+extern void nouveau_grctx_vals_load(struct drm_device *,
+				    struct nouveau_gpuobj *);
+extern void nouveau_grctx_fini(struct drm_device *);
+
 /* nv04_instmem.c */
 extern int  nv04_instmem_init(struct drm_device *);
 extern void nv04_instmem_takedown(struct drm_device *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_grctx.c b/drivers/gpu/drm/nouveau/nouveau_grctx.c
new file mode 100644
index 000000000000..d5d84c87f757
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_grctx.c
@@ -0,0 +1,158 @@
+/*
+ * Copyright 2009 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 <linux/firmware.h>
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+
+struct nouveau_ctxprog {
+	uint32_t signature;
+	uint8_t  version;
+	uint16_t length;
+	uint32_t data[];
+} __attribute__ ((packed));
+
+struct nouveau_ctxvals {
+	uint32_t signature;
+	uint8_t  version;
+	uint32_t length;
+	struct {
+		uint32_t offset;
+		uint32_t value;
+	} data[];
+} __attribute__ ((packed));
+
+int
+nouveau_grctx_prog_load(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	const int chipset = dev_priv->chipset;
+	const struct firmware *fw;
+	const struct nouveau_ctxprog *cp;
+	const struct nouveau_ctxvals *cv;
+	char name[32];
+	int ret, i;
+
+	if (!pgraph->ctxprog) {
+		sprintf(name, "nouveau/nv%02x.ctxprog", chipset);
+		ret = request_firmware(&fw, name, &dev->pdev->dev);
+		if (ret) {
+			NV_ERROR(dev, "No ctxprog for NV%02x\n", chipset);
+			return ret;
+		}
+
+		pgraph->ctxprog = kmalloc(fw->size, GFP_KERNEL);
+		if (!pgraph->ctxprog) {
+			NV_ERROR(dev, "OOM copying ctxprog\n");
+			release_firmware(fw);
+			return -ENOMEM;
+		}
+		memcpy(pgraph->ctxprog, fw->data, fw->size);
+
+		cp = pgraph->ctxprog;
+		if (le32_to_cpu(cp->signature) != 0x5043564e ||
+		    cp->version != 0 ||
+		    le16_to_cpu(cp->length) != ((fw->size - 7) / 4)) {
+			NV_ERROR(dev, "ctxprog invalid\n");
+			release_firmware(fw);
+			nouveau_grctx_fini(dev);
+			return -EINVAL;
+		}
+		release_firmware(fw);
+	}
+
+	if (!pgraph->ctxvals) {
+		sprintf(name, "nouveau/nv%02x.ctxvals", chipset);
+		ret = request_firmware(&fw, name, &dev->pdev->dev);
+		if (ret) {
+			NV_ERROR(dev, "No ctxvals for NV%02x\n", chipset);
+			nouveau_grctx_fini(dev);
+			return ret;
+		}
+
+		pgraph->ctxvals = kmalloc(fw->size, GFP_KERNEL);
+		if (!pgraph->ctxprog) {
+			NV_ERROR(dev, "OOM copying ctxprog\n");
+			release_firmware(fw);
+			nouveau_grctx_fini(dev);
+			return -ENOMEM;
+		}
+		memcpy(pgraph->ctxvals, fw->data, fw->size);
+
+		cv = (void *)pgraph->ctxvals;
+		if (le32_to_cpu(cv->signature) != 0x5643564e ||
+		    cv->version != 0 ||
+		    le32_to_cpu(cv->length) != ((fw->size - 9) / 8)) {
+			NV_ERROR(dev, "ctxvals invalid\n");
+			release_firmware(fw);
+			nouveau_grctx_fini(dev);
+			return -EINVAL;
+		}
+		release_firmware(fw);
+	}
+
+	cp = pgraph->ctxprog;
+
+	nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
+	for (i = 0; i < le16_to_cpu(cp->length); i++)
+		nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA,
+			le32_to_cpu(cp->data[i]));
+
+	return 0;
+}
+
+void
+nouveau_grctx_fini(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+
+	if (pgraph->ctxprog) {
+		kfree(pgraph->ctxprog);
+		pgraph->ctxprog = NULL;
+	}
+
+	if (pgraph->ctxvals) {
+		kfree(pgraph->ctxprog);
+		pgraph->ctxvals = NULL;
+	}
+}
+
+void
+nouveau_grctx_vals_load(struct drm_device *dev, struct nouveau_gpuobj *ctx)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	struct nouveau_ctxvals *cv = pgraph->ctxvals;
+	int i;
+
+	if (!cv)
+		return;
+
+	for (i = 0; i < le32_to_cpu(cv->length); i++)
+		nv_wo32(dev, ctx, le32_to_cpu(cv->data[i].offset),
+			le32_to_cpu(cv->data[i].value));
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_grctx.h b/drivers/gpu/drm/nouveau/nouveau_grctx.h
new file mode 100644
index 000000000000..5d39c4ce8006
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_grctx.h
@@ -0,0 +1,133 @@
+#ifndef __NOUVEAU_GRCTX_H__
+#define __NOUVEAU_GRCTX_H__
+
+struct nouveau_grctx {
+	struct drm_device *dev;
+
+	enum {
+		NOUVEAU_GRCTX_PROG,
+		NOUVEAU_GRCTX_VALS
+	} mode;
+	void *data;
+
+	uint32_t ctxprog_max;
+	uint32_t ctxprog_len;
+	uint32_t ctxprog_reg;
+	int      ctxprog_label[32];
+	uint32_t ctxvals_pos;
+	uint32_t ctxvals_base;
+};
+
+#ifdef CP_CTX
+static inline void
+cp_out(struct nouveau_grctx *ctx, uint32_t inst)
+{
+	uint32_t *ctxprog = ctx->data;
+
+	if (ctx->mode != NOUVEAU_GRCTX_PROG)
+		return;
+
+	BUG_ON(ctx->ctxprog_len == ctx->ctxprog_max);
+	ctxprog[ctx->ctxprog_len++] = inst;
+}
+
+static inline void
+cp_lsr(struct nouveau_grctx *ctx, uint32_t val)
+{
+	cp_out(ctx, CP_LOAD_SR | val);
+}
+
+static inline void
+cp_ctx(struct nouveau_grctx *ctx, uint32_t reg, uint32_t length)
+{
+	ctx->ctxprog_reg = (reg - 0x00400000) >> 2;
+
+	ctx->ctxvals_base = ctx->ctxvals_pos;
+	ctx->ctxvals_pos = ctx->ctxvals_base + length;
+
+	if (length > (CP_CTX_COUNT >> CP_CTX_COUNT_SHIFT)) {
+		cp_lsr(ctx, length);
+		length = 0;
+	}
+
+	cp_out(ctx, CP_CTX | (length << CP_CTX_COUNT_SHIFT) | ctx->ctxprog_reg);
+}
+
+static inline void
+cp_name(struct nouveau_grctx *ctx, int name)
+{
+	uint32_t *ctxprog = ctx->data;
+	int i;
+
+	if (ctx->mode != NOUVEAU_GRCTX_PROG)
+		return;
+
+	ctx->ctxprog_label[name] = ctx->ctxprog_len;
+	for (i = 0; i < ctx->ctxprog_len; i++) {
+		if ((ctxprog[i] & 0xfff00000) != 0xff400000)
+			continue;
+		if ((ctxprog[i] & CP_BRA_IP) != ((name) << CP_BRA_IP_SHIFT))
+			continue;
+		ctxprog[i] = (ctxprog[i] & 0x00ff00ff) |
+			     (ctx->ctxprog_len << CP_BRA_IP_SHIFT);
+	}
+}
+
+static inline void
+_cp_bra(struct nouveau_grctx *ctx, u32 mod, int flag, int state, int name)
+{
+	int ip = 0;
+
+	if (mod != 2) {
+		ip = ctx->ctxprog_label[name] << CP_BRA_IP_SHIFT;
+		if (ip == 0)
+			ip = 0xff000000 | (name << CP_BRA_IP_SHIFT);
+	}
+
+	cp_out(ctx, CP_BRA | (mod << 18) | ip | flag |
+		    (state ? 0 : CP_BRA_IF_CLEAR));
+}
+#define cp_bra(c,f,s,n) _cp_bra((c), 0, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
+#ifdef CP_BRA_MOD
+#define cp_cal(c,f,s,n) _cp_bra((c), 1, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
+#define cp_ret(c,f,s) _cp_bra((c), 2, CP_FLAG_##f, CP_FLAG_##f##_##s, 0)
+#endif
+
+static inline void
+_cp_wait(struct nouveau_grctx *ctx, int flag, int state)
+{
+	cp_out(ctx, CP_WAIT | flag | (state ? CP_WAIT_SET : 0));
+}
+#define cp_wait(c,f,s) _cp_wait((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
+
+static inline void
+_cp_set(struct nouveau_grctx *ctx, int flag, int state)
+{
+	cp_out(ctx, CP_SET | flag | (state ? CP_SET_1 : 0));
+}
+#define cp_set(c,f,s) _cp_set((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
+
+static inline void
+cp_pos(struct nouveau_grctx *ctx, int offset)
+{
+	ctx->ctxvals_pos = offset;
+	ctx->ctxvals_base = ctx->ctxvals_pos;
+
+	cp_lsr(ctx, ctx->ctxvals_pos);
+	cp_out(ctx, CP_SET_CONTEXT_POINTER);
+}
+
+static inline void
+gr_def(struct nouveau_grctx *ctx, uint32_t reg, uint32_t val)
+{
+	if (ctx->mode != NOUVEAU_GRCTX_VALS)
+		return;
+
+	reg = (reg - 0x00400000) / 4;
+	reg = (reg - ctx->ctxprog_reg) + ctx->ctxvals_base;
+
+	nv_wo32(ctx->dev, ctx->data, reg, val);
+}
+#endif
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index 7e8547cb5833..2b332bb55acf 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -24,36 +24,10 @@
  *
  */
 
-#include <linux/firmware.h>
-
 #include "drmP.h"
 #include "drm.h"
 #include "nouveau_drv.h"
-
-MODULE_FIRMWARE("nouveau/nv40.ctxprog");
-MODULE_FIRMWARE("nouveau/nv40.ctxvals");
-MODULE_FIRMWARE("nouveau/nv41.ctxprog");
-MODULE_FIRMWARE("nouveau/nv41.ctxvals");
-MODULE_FIRMWARE("nouveau/nv42.ctxprog");
-MODULE_FIRMWARE("nouveau/nv42.ctxvals");
-MODULE_FIRMWARE("nouveau/nv43.ctxprog");
-MODULE_FIRMWARE("nouveau/nv43.ctxvals");
-MODULE_FIRMWARE("nouveau/nv44.ctxprog");
-MODULE_FIRMWARE("nouveau/nv44.ctxvals");
-MODULE_FIRMWARE("nouveau/nv46.ctxprog");
-MODULE_FIRMWARE("nouveau/nv46.ctxvals");
-MODULE_FIRMWARE("nouveau/nv47.ctxprog");
-MODULE_FIRMWARE("nouveau/nv47.ctxvals");
-MODULE_FIRMWARE("nouveau/nv49.ctxprog");
-MODULE_FIRMWARE("nouveau/nv49.ctxvals");
-MODULE_FIRMWARE("nouveau/nv4a.ctxprog");
-MODULE_FIRMWARE("nouveau/nv4a.ctxvals");
-MODULE_FIRMWARE("nouveau/nv4b.ctxprog");
-MODULE_FIRMWARE("nouveau/nv4b.ctxvals");
-MODULE_FIRMWARE("nouveau/nv4c.ctxprog");
-MODULE_FIRMWARE("nouveau/nv4c.ctxvals");
-MODULE_FIRMWARE("nouveau/nv4e.ctxprog");
-MODULE_FIRMWARE("nouveau/nv4e.ctxvals");
+#include "nouveau_grctx.h"
 
 struct nouveau_channel *
 nv40_graph_channel(struct drm_device *dev)
@@ -83,27 +57,30 @@ nv40_graph_create_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *ctx;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	int ret;
 
-	/* Allocate a 175KiB block of PRAMIN to store the context.  This
-	 * is massive overkill for a lot of chipsets, but it should be safe
-	 * until we're able to implement this properly (will happen at more
-	 * or less the same time we're able to write our own context programs.
-	 */
-	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, 175*1024, 16,
-					  NVOBJ_FLAG_ZERO_ALLOC,
-					  &chan->ramin_grctx);
+	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pgraph->grctx_size,
+				     16, NVOBJ_FLAG_ZERO_ALLOC,
+				     &chan->ramin_grctx);
 	if (ret)
 		return ret;
-	ctx = chan->ramin_grctx->gpuobj;
 
 	/* Initialise default context values */
 	dev_priv->engine.instmem.prepare_access(dev, true);
-	nv40_grctx_vals_load(dev, ctx);
-	nv_wo32(dev, ctx, 0, ctx->im_pramin->start);
-	dev_priv->engine.instmem.finish_access(dev);
+	if (!pgraph->ctxprog) {
+		struct nouveau_grctx ctx = {};
 
+		ctx.dev = chan->dev;
+		ctx.mode = NOUVEAU_GRCTX_VALS;
+		ctx.data = chan->ramin_grctx->gpuobj;
+		nv40_grctx_init(&ctx);
+	} else {
+		nouveau_grctx_vals_load(dev, chan->ramin_grctx->gpuobj);
+	}
+	nv_wo32(dev, chan->ramin_grctx->gpuobj, 0,
+		     chan->ramin_grctx->gpuobj->im_pramin->start);
+	dev_priv->engine.instmem.finish_access(dev);
 	return 0;
 }
 
@@ -204,139 +181,6 @@ nv40_graph_unload_context(struct drm_device *dev)
 	return ret;
 }
 
-struct nouveau_ctxprog {
-	uint32_t signature;
-	uint8_t  version;
-	uint16_t length;
-	uint32_t data[];
-} __attribute__ ((packed));
-
-struct nouveau_ctxvals {
-	uint32_t signature;
-	uint8_t  version;
-	uint32_t length;
-	struct {
-		uint32_t offset;
-		uint32_t value;
-	} data[];
-} __attribute__ ((packed));
-
-int
-nv40_grctx_init(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	const int chipset = dev_priv->chipset;
-	const struct firmware *fw;
-	const struct nouveau_ctxprog *cp;
-	const struct nouveau_ctxvals *cv;
-	char name[32];
-	int ret, i;
-
-	pgraph->accel_blocked = true;
-
-	if (!pgraph->ctxprog) {
-		sprintf(name, "nouveau/nv%02x.ctxprog", chipset);
-		ret = request_firmware(&fw, name, &dev->pdev->dev);
-		if (ret) {
-			NV_ERROR(dev, "No ctxprog for NV%02x\n", chipset);
-			return ret;
-		}
-
-		pgraph->ctxprog = kmalloc(fw->size, GFP_KERNEL);
-		if (!pgraph->ctxprog) {
-			NV_ERROR(dev, "OOM copying ctxprog\n");
-			release_firmware(fw);
-			return -ENOMEM;
-		}
-		memcpy(pgraph->ctxprog, fw->data, fw->size);
-
-		cp = pgraph->ctxprog;
-		if (le32_to_cpu(cp->signature) != 0x5043564e ||
-		    cp->version != 0 ||
-		    le16_to_cpu(cp->length) != ((fw->size - 7) / 4)) {
-			NV_ERROR(dev, "ctxprog invalid\n");
-			release_firmware(fw);
-			nv40_grctx_fini(dev);
-			return -EINVAL;
-		}
-		release_firmware(fw);
-	}
-
-	if (!pgraph->ctxvals) {
-		sprintf(name, "nouveau/nv%02x.ctxvals", chipset);
-		ret = request_firmware(&fw, name, &dev->pdev->dev);
-		if (ret) {
-			NV_ERROR(dev, "No ctxvals for NV%02x\n", chipset);
-			nv40_grctx_fini(dev);
-			return ret;
-		}
-
-		pgraph->ctxvals = kmalloc(fw->size, GFP_KERNEL);
-		if (!pgraph->ctxprog) {
-			NV_ERROR(dev, "OOM copying ctxprog\n");
-			release_firmware(fw);
-			nv40_grctx_fini(dev);
-			return -ENOMEM;
-		}
-		memcpy(pgraph->ctxvals, fw->data, fw->size);
-
-		cv = (void *)pgraph->ctxvals;
-		if (le32_to_cpu(cv->signature) != 0x5643564e ||
-		    cv->version != 0 ||
-		    le32_to_cpu(cv->length) != ((fw->size - 9) / 8)) {
-			NV_ERROR(dev, "ctxvals invalid\n");
-			release_firmware(fw);
-			nv40_grctx_fini(dev);
-			return -EINVAL;
-		}
-		release_firmware(fw);
-	}
-
-	cp = pgraph->ctxprog;
-
-	nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
-	for (i = 0; i < le16_to_cpu(cp->length); i++)
-		nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA,
-			le32_to_cpu(cp->data[i]));
-
-	pgraph->accel_blocked = false;
-	return 0;
-}
-
-void
-nv40_grctx_fini(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-
-	if (pgraph->ctxprog) {
-		kfree(pgraph->ctxprog);
-		pgraph->ctxprog = NULL;
-	}
-
-	if (pgraph->ctxvals) {
-		kfree(pgraph->ctxprog);
-		pgraph->ctxvals = NULL;
-	}
-}
-
-void
-nv40_grctx_vals_load(struct drm_device *dev, struct nouveau_gpuobj *ctx)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_ctxvals *cv = pgraph->ctxvals;
-	int i;
-
-	if (!cv)
-		return;
-
-	for (i = 0; i < le32_to_cpu(cv->length); i++)
-		nv_wo32(dev, ctx, le32_to_cpu(cv->data[i].offset),
-			le32_to_cpu(cv->data[i].value));
-}
-
 /*
  * G70		0x47
  * G71		0x49
@@ -359,7 +203,26 @@ nv40_graph_init(struct drm_device *dev)
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) |
 			 NV_PMC_ENABLE_PGRAPH);
 
-	nv40_grctx_init(dev);
+	if (nouveau_ctxfw) {
+		nouveau_grctx_prog_load(dev);
+		dev_priv->engine.graph.grctx_size = 175 * 1024;
+	}
+
+	if (!dev_priv->engine.graph.ctxprog) {
+		struct nouveau_grctx ctx = {};
+		uint32_t cp[256];
+
+		ctx.dev = dev;
+		ctx.mode = NOUVEAU_GRCTX_PROG;
+		ctx.data = cp;
+		ctx.ctxprog_max = 256;
+		nv40_grctx_init(&ctx);
+		dev_priv->engine.graph.grctx_size = ctx.ctxvals_pos * 4;
+
+		nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_INDEX, 0);
+		for (i = 0; i < ctx.ctxprog_len; i++)
+			nv_wr32(dev, NV40_PGRAPH_CTXCTL_UCODE_DATA, cp[i]);
+	}
 
 	/* No context present currently */
 	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
@@ -539,6 +402,7 @@ nv40_graph_init(struct drm_device *dev)
 
 void nv40_graph_takedown(struct drm_device *dev)
 {
+	nouveau_grctx_fini(dev);
 }
 
 struct nouveau_pgraph_object_class nv40_graph_grclass[] = {
diff --git a/drivers/gpu/drm/nouveau/nv40_grctx.c b/drivers/gpu/drm/nouveau/nv40_grctx.c
new file mode 100644
index 000000000000..11b11c31f543
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv40_grctx.c
@@ -0,0 +1,678 @@
+/*
+ * Copyright 2009 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
+ */
+
+/* NVIDIA context programs handle a number of other conditions which are
+ * not implemented in our versions.  It's not clear why NVIDIA context
+ * programs have this code, nor whether it's strictly necessary for
+ * correct operation.  We'll implement additional handling if/when we
+ * discover it's necessary.
+ *
+ * - On context save, NVIDIA set 0x400314 bit 0 to 1 if the "3D state"
+ *   flag is set, this gets saved into the context.
+ * - On context save, the context program for all cards load nsource
+ *   into a flag register and check for ILLEGAL_MTHD.  If it's set,
+ *   opcode 0x60000d is called before resuming normal operation.
+ * - Some context programs check more conditions than the above.  NV44
+ *   checks: ((nsource & 0x0857) || (0x400718 & 0x0100) || (intr & 0x0001))
+ *   and calls 0x60000d before resuming normal operation.
+ * - At the very beginning of NVIDIA's context programs, flag 9 is checked
+ *   and if true 0x800001 is called with count=0, pos=0, the flag is cleared
+ *   and then the ctxprog is aborted.  It looks like a complicated NOP,
+ *   its purpose is unknown.
+ * - In the section of code that loads the per-vs state, NVIDIA check
+ *   flag 10.  If it's set, they only transfer the small 0x300 byte block
+ *   of state + the state for a single vs as opposed to the state for
+ *   all vs units.  It doesn't seem likely that it'll occur in normal
+ *   operation, especially seeing as it appears NVIDIA may have screwed
+ *   up the ctxprogs for some cards and have an invalid instruction
+ *   rather than a cp_lsr(ctx, dwords_for_1_vs_unit) instruction.
+ * - There's a number of places where context offset 0 (where we place
+ *   the PRAMIN offset of the context) is loaded into either 0x408000,
+ *   0x408004 or 0x408008.  Not sure what's up there either.
+ * - The ctxprogs for some cards save 0x400a00 again during the cleanup
+ *   path for auto-loadctx.
+ */
+
+#define CP_FLAG_CLEAR                 0
+#define CP_FLAG_SET                   1
+#define CP_FLAG_SWAP_DIRECTION        ((0 * 32) + 0)
+#define CP_FLAG_SWAP_DIRECTION_LOAD   0
+#define CP_FLAG_SWAP_DIRECTION_SAVE   1
+#define CP_FLAG_USER_SAVE             ((0 * 32) + 5)
+#define CP_FLAG_USER_SAVE_NOT_PENDING 0
+#define CP_FLAG_USER_SAVE_PENDING     1
+#define CP_FLAG_USER_LOAD             ((0 * 32) + 6)
+#define CP_FLAG_USER_LOAD_NOT_PENDING 0
+#define CP_FLAG_USER_LOAD_PENDING     1
+#define CP_FLAG_STATUS                ((3 * 32) + 0)
+#define CP_FLAG_STATUS_IDLE           0
+#define CP_FLAG_STATUS_BUSY           1
+#define CP_FLAG_AUTO_SAVE             ((3 * 32) + 4)
+#define CP_FLAG_AUTO_SAVE_NOT_PENDING 0
+#define CP_FLAG_AUTO_SAVE_PENDING     1
+#define CP_FLAG_AUTO_LOAD             ((3 * 32) + 5)
+#define CP_FLAG_AUTO_LOAD_NOT_PENDING 0
+#define CP_FLAG_AUTO_LOAD_PENDING     1
+#define CP_FLAG_UNK54                 ((3 * 32) + 6)
+#define CP_FLAG_UNK54_CLEAR           0
+#define CP_FLAG_UNK54_SET             1
+#define CP_FLAG_ALWAYS                ((3 * 32) + 8)
+#define CP_FLAG_ALWAYS_FALSE          0
+#define CP_FLAG_ALWAYS_TRUE           1
+#define CP_FLAG_UNK57                 ((3 * 32) + 9)
+#define CP_FLAG_UNK57_CLEAR           0
+#define CP_FLAG_UNK57_SET             1
+
+#define CP_CTX                   0x00100000
+#define CP_CTX_COUNT             0x000fc000
+#define CP_CTX_COUNT_SHIFT               14
+#define CP_CTX_REG               0x00003fff
+#define CP_LOAD_SR               0x00200000
+#define CP_LOAD_SR_VALUE         0x000fffff
+#define CP_BRA                   0x00400000
+#define CP_BRA_IP                0x0000ff00
+#define CP_BRA_IP_SHIFT                   8
+#define CP_BRA_IF_CLEAR          0x00000080
+#define CP_BRA_FLAG              0x0000007f
+#define CP_WAIT                  0x00500000
+#define CP_WAIT_SET              0x00000080
+#define CP_WAIT_FLAG             0x0000007f
+#define CP_SET                   0x00700000
+#define CP_SET_1                 0x00000080
+#define CP_SET_FLAG              0x0000007f
+#define CP_NEXT_TO_SWAP          0x00600007
+#define CP_NEXT_TO_CURRENT       0x00600009
+#define CP_SET_CONTEXT_POINTER   0x0060000a
+#define CP_END                   0x0060000e
+#define CP_LOAD_MAGIC_UNK01      0x00800001 /* unknown */
+#define CP_LOAD_MAGIC_NV44TCL    0x00800029 /* per-vs state (0x4497) */
+#define CP_LOAD_MAGIC_NV40TCL    0x00800041 /* per-vs state (0x4097) */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_grctx.h"
+
+/* TODO:
+ *  - get vs count from 0x1540
+ *  - document unimplemented bits compared to nvidia
+ *    - nsource handling
+ *    - R0 & 0x0200 handling
+ *    - single-vs handling
+ *    - 400314 bit 0
+ */
+
+static int
+nv40_graph_4097(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if ((dev_priv->chipset & 0xf0) == 0x60)
+		return 0;
+
+	return !!(0x0baf & (1 << dev_priv->chipset));
+}
+
+static int
+nv40_graph_vs_count(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	switch (dev_priv->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		return 8;
+	case 0x40:
+		return 6;
+	case 0x41:
+	case 0x42:
+		return 5;
+	case 0x43:
+	case 0x44:
+	case 0x46:
+	case 0x4a:
+		return 3;
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+	default:
+		return 1;
+	}
+}
+
+
+enum cp_label {
+	cp_check_load = 1,
+	cp_setup_auto_load,
+	cp_setup_load,
+	cp_setup_save,
+	cp_swap_state,
+	cp_swap_state3d_3_is_save,
+	cp_prepare_exit,
+	cp_exit,
+};
+
+static void
+nv40_graph_construct_general(struct nouveau_grctx *ctx)
+{
+	struct drm_nouveau_private *dev_priv = ctx->dev->dev_private;
+	int i;
+
+	cp_ctx(ctx, 0x4000a4, 1);
+	gr_def(ctx, 0x4000a4, 0x00000008);
+	cp_ctx(ctx, 0x400144, 58);
+	gr_def(ctx, 0x400144, 0x00000001);
+	cp_ctx(ctx, 0x400314, 1);
+	gr_def(ctx, 0x400314, 0x00000000);
+	cp_ctx(ctx, 0x400400, 10);
+	cp_ctx(ctx, 0x400480, 10);
+	cp_ctx(ctx, 0x400500, 19);
+	gr_def(ctx, 0x400514, 0x00040000);
+	gr_def(ctx, 0x400524, 0x55555555);
+	gr_def(ctx, 0x400528, 0x55555555);
+	gr_def(ctx, 0x40052c, 0x55555555);
+	gr_def(ctx, 0x400530, 0x55555555);
+	cp_ctx(ctx, 0x400560, 6);
+	gr_def(ctx, 0x400568, 0x0000ffff);
+	gr_def(ctx, 0x40056c, 0x0000ffff);
+	cp_ctx(ctx, 0x40057c, 5);
+	cp_ctx(ctx, 0x400710, 3);
+	gr_def(ctx, 0x400710, 0x20010001);
+	gr_def(ctx, 0x400714, 0x0f73ef00);
+	cp_ctx(ctx, 0x400724, 1);
+	gr_def(ctx, 0x400724, 0x02008821);
+	cp_ctx(ctx, 0x400770, 3);
+	if (dev_priv->chipset == 0x40) {
+		cp_ctx(ctx, 0x400814, 4);
+		cp_ctx(ctx, 0x400828, 5);
+		cp_ctx(ctx, 0x400840, 5);
+		gr_def(ctx, 0x400850, 0x00000040);
+		cp_ctx(ctx, 0x400858, 4);
+		gr_def(ctx, 0x400858, 0x00000040);
+		gr_def(ctx, 0x40085c, 0x00000040);
+		gr_def(ctx, 0x400864, 0x80000000);
+		cp_ctx(ctx, 0x40086c, 9);
+		gr_def(ctx, 0x40086c, 0x80000000);
+		gr_def(ctx, 0x400870, 0x80000000);
+		gr_def(ctx, 0x400874, 0x80000000);
+		gr_def(ctx, 0x400878, 0x80000000);
+		gr_def(ctx, 0x400888, 0x00000040);
+		gr_def(ctx, 0x40088c, 0x80000000);
+		cp_ctx(ctx, 0x4009c0, 8);
+		gr_def(ctx, 0x4009cc, 0x80000000);
+		gr_def(ctx, 0x4009dc, 0x80000000);
+	} else {
+		cp_ctx(ctx, 0x400840, 20);
+		if (!nv40_graph_4097(ctx->dev)) {
+			for (i = 0; i < 8; i++)
+				gr_def(ctx, 0x400860 + (i * 4), 0x00000001);
+		}
+		gr_def(ctx, 0x400880, 0x00000040);
+		gr_def(ctx, 0x400884, 0x00000040);
+		gr_def(ctx, 0x400888, 0x00000040);
+		cp_ctx(ctx, 0x400894, 11);
+		gr_def(ctx, 0x400894, 0x00000040);
+		if (nv40_graph_4097(ctx->dev)) {
+			for (i = 0; i < 8; i++)
+				gr_def(ctx, 0x4008a0 + (i * 4), 0x80000000);
+		}
+		cp_ctx(ctx, 0x4008e0, 2);
+		cp_ctx(ctx, 0x4008f8, 2);
+		if (dev_priv->chipset == 0x4c ||
+		    (dev_priv->chipset & 0xf0) == 0x60)
+			cp_ctx(ctx, 0x4009f8, 1);
+	}
+	cp_ctx(ctx, 0x400a00, 73);
+	gr_def(ctx, 0x400b0c, 0x0b0b0b0c);
+	cp_ctx(ctx, 0x401000, 4);
+	cp_ctx(ctx, 0x405004, 1);
+	switch (dev_priv->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x403448, 1);
+		gr_def(ctx, 0x403448, 0x00001010);
+		break;
+	default:
+		cp_ctx(ctx, 0x403440, 1);
+		switch (dev_priv->chipset) {
+		case 0x40:
+			gr_def(ctx, 0x403440, 0x00000010);
+			break;
+		case 0x44:
+		case 0x46:
+		case 0x4a:
+			gr_def(ctx, 0x403440, 0x00003010);
+			break;
+		case 0x41:
+		case 0x42:
+		case 0x43:
+		case 0x4c:
+		case 0x4e:
+		case 0x67:
+		default:
+			gr_def(ctx, 0x403440, 0x00001010);
+			break;
+		}
+		break;
+	}
+}
+
+static void
+nv40_graph_construct_state3d(struct nouveau_grctx *ctx)
+{
+	struct drm_nouveau_private *dev_priv = ctx->dev->dev_private;
+	int i;
+
+	if (dev_priv->chipset == 0x40) {
+		cp_ctx(ctx, 0x401880, 51);
+		gr_def(ctx, 0x401940, 0x00000100);
+	} else
+	if (dev_priv->chipset == 0x46 || dev_priv->chipset == 0x47 ||
+	    dev_priv->chipset == 0x49 || dev_priv->chipset == 0x4b) {
+		cp_ctx(ctx, 0x401880, 32);
+		for (i = 0; i < 16; i++)
+			gr_def(ctx, 0x401880 + (i * 4), 0x00000111);
+		if (dev_priv->chipset == 0x46)
+			cp_ctx(ctx, 0x401900, 16);
+		cp_ctx(ctx, 0x401940, 3);
+	}
+	cp_ctx(ctx, 0x40194c, 18);
+	gr_def(ctx, 0x401954, 0x00000111);
+	gr_def(ctx, 0x401958, 0x00080060);
+	gr_def(ctx, 0x401974, 0x00000080);
+	gr_def(ctx, 0x401978, 0xffff0000);
+	gr_def(ctx, 0x40197c, 0x00000001);
+	gr_def(ctx, 0x401990, 0x46400000);
+	if (dev_priv->chipset == 0x40) {
+		cp_ctx(ctx, 0x4019a0, 2);
+		cp_ctx(ctx, 0x4019ac, 5);
+	} else {
+		cp_ctx(ctx, 0x4019a0, 1);
+		cp_ctx(ctx, 0x4019b4, 3);
+	}
+	gr_def(ctx, 0x4019bc, 0xffff0000);
+	switch (dev_priv->chipset) {
+	case 0x46:
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x4019c0, 18);
+		for (i = 0; i < 16; i++)
+			gr_def(ctx, 0x4019c0 + (i * 4), 0x88888888);
+		break;
+	}
+	cp_ctx(ctx, 0x401a08, 8);
+	gr_def(ctx, 0x401a10, 0x0fff0000);
+	gr_def(ctx, 0x401a14, 0x0fff0000);
+	gr_def(ctx, 0x401a1c, 0x00011100);
+	cp_ctx(ctx, 0x401a2c, 4);
+	cp_ctx(ctx, 0x401a44, 26);
+	for (i = 0; i < 16; i++)
+		gr_def(ctx, 0x401a44 + (i * 4), 0x07ff0000);
+	gr_def(ctx, 0x401a8c, 0x4b7fffff);
+	if (dev_priv->chipset == 0x40) {
+		cp_ctx(ctx, 0x401ab8, 3);
+	} else {
+		cp_ctx(ctx, 0x401ab8, 1);
+		cp_ctx(ctx, 0x401ac0, 1);
+	}
+	cp_ctx(ctx, 0x401ad0, 8);
+	gr_def(ctx, 0x401ad0, 0x30201000);
+	gr_def(ctx, 0x401ad4, 0x70605040);
+	gr_def(ctx, 0x401ad8, 0xb8a89888);
+	gr_def(ctx, 0x401adc, 0xf8e8d8c8);
+	cp_ctx(ctx, 0x401b10, dev_priv->chipset == 0x40 ? 2 : 1);
+	gr_def(ctx, 0x401b10, 0x40100000);
+	cp_ctx(ctx, 0x401b18, dev_priv->chipset == 0x40 ? 6 : 5);
+	gr_def(ctx, 0x401b28, dev_priv->chipset == 0x40 ?
+			      0x00000004 : 0x00000000);
+	cp_ctx(ctx, 0x401b30, 25);
+	gr_def(ctx, 0x401b34, 0x0000ffff);
+	gr_def(ctx, 0x401b68, 0x435185d6);
+	gr_def(ctx, 0x401b6c, 0x2155b699);
+	gr_def(ctx, 0x401b70, 0xfedcba98);
+	gr_def(ctx, 0x401b74, 0x00000098);
+	gr_def(ctx, 0x401b84, 0xffffffff);
+	gr_def(ctx, 0x401b88, 0x00ff7000);
+	gr_def(ctx, 0x401b8c, 0x0000ffff);
+	if (dev_priv->chipset != 0x44 && dev_priv->chipset != 0x4a &&
+	    dev_priv->chipset != 0x4e)
+		cp_ctx(ctx, 0x401b94, 1);
+	cp_ctx(ctx, 0x401b98, 8);
+	gr_def(ctx, 0x401b9c, 0x00ff0000);
+	cp_ctx(ctx, 0x401bc0, 9);
+	gr_def(ctx, 0x401be0, 0x00ffff00);
+	cp_ctx(ctx, 0x401c00, 192);
+	for (i = 0; i < 16; i++) { /* fragment texture units */
+		gr_def(ctx, 0x401c40 + (i * 4), 0x00018488);
+		gr_def(ctx, 0x401c80 + (i * 4), 0x00028202);
+		gr_def(ctx, 0x401d00 + (i * 4), 0x0000aae4);
+		gr_def(ctx, 0x401d40 + (i * 4), 0x01012000);
+		gr_def(ctx, 0x401d80 + (i * 4), 0x00080008);
+		gr_def(ctx, 0x401e00 + (i * 4), 0x00100008);
+	}
+	for (i = 0; i < 4; i++) { /* vertex texture units */
+		gr_def(ctx, 0x401e90 + (i * 4), 0x0001bc80);
+		gr_def(ctx, 0x401ea0 + (i * 4), 0x00000202);
+		gr_def(ctx, 0x401ec0 + (i * 4), 0x00000008);
+		gr_def(ctx, 0x401ee0 + (i * 4), 0x00080008);
+	}
+	cp_ctx(ctx, 0x400f5c, 3);
+	gr_def(ctx, 0x400f5c, 0x00000002);
+	cp_ctx(ctx, 0x400f84, 1);
+}
+
+static void
+nv40_graph_construct_state3d_2(struct nouveau_grctx *ctx)
+{
+	struct drm_nouveau_private *dev_priv = ctx->dev->dev_private;
+	int i;
+
+	cp_ctx(ctx, 0x402000, 1);
+	cp_ctx(ctx, 0x402404, dev_priv->chipset == 0x40 ? 1 : 2);
+	switch (dev_priv->chipset) {
+	case 0x40:
+		gr_def(ctx, 0x402404, 0x00000001);
+		break;
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+		gr_def(ctx, 0x402404, 0x00000020);
+		break;
+	case 0x46:
+	case 0x49:
+	case 0x4b:
+		gr_def(ctx, 0x402404, 0x00000421);
+		break;
+	default:
+		gr_def(ctx, 0x402404, 0x00000021);
+	}
+	if (dev_priv->chipset != 0x40)
+		gr_def(ctx, 0x402408, 0x030c30c3);
+	switch (dev_priv->chipset) {
+	case 0x44:
+	case 0x46:
+	case 0x4a:
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+		cp_ctx(ctx, 0x402440, 1);
+		gr_def(ctx, 0x402440, 0x00011001);
+		break;
+	default:
+		break;
+	}
+	cp_ctx(ctx, 0x402480, dev_priv->chipset == 0x40 ? 8 : 9);
+	gr_def(ctx, 0x402488, 0x3e020200);
+	gr_def(ctx, 0x40248c, 0x00ffffff);
+	switch (dev_priv->chipset) {
+	case 0x40:
+		gr_def(ctx, 0x402490, 0x60103f00);
+		break;
+	case 0x47:
+		gr_def(ctx, 0x402490, 0x40103f00);
+		break;
+	case 0x41:
+	case 0x42:
+	case 0x49:
+	case 0x4b:
+		gr_def(ctx, 0x402490, 0x20103f00);
+		break;
+	default:
+		gr_def(ctx, 0x402490, 0x0c103f00);
+		break;
+	}
+	gr_def(ctx, 0x40249c, dev_priv->chipset <= 0x43 ?
+			      0x00020000 : 0x00040000);
+	cp_ctx(ctx, 0x402500, 31);
+	gr_def(ctx, 0x402530, 0x00008100);
+	if (dev_priv->chipset == 0x40)
+		cp_ctx(ctx, 0x40257c, 6);
+	cp_ctx(ctx, 0x402594, 16);
+	cp_ctx(ctx, 0x402800, 17);
+	gr_def(ctx, 0x402800, 0x00000001);
+	switch (dev_priv->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x402864, 1);
+		gr_def(ctx, 0x402864, 0x00001001);
+		cp_ctx(ctx, 0x402870, 3);
+		gr_def(ctx, 0x402878, 0x00000003);
+		if (dev_priv->chipset != 0x47) { /* belong at end!! */
+			cp_ctx(ctx, 0x402900, 1);
+			cp_ctx(ctx, 0x402940, 1);
+			cp_ctx(ctx, 0x402980, 1);
+			cp_ctx(ctx, 0x4029c0, 1);
+			cp_ctx(ctx, 0x402a00, 1);
+			cp_ctx(ctx, 0x402a40, 1);
+			cp_ctx(ctx, 0x402a80, 1);
+			cp_ctx(ctx, 0x402ac0, 1);
+		}
+		break;
+	case 0x40:
+		cp_ctx(ctx, 0x402844, 1);
+		gr_def(ctx, 0x402844, 0x00000001);
+		cp_ctx(ctx, 0x402850, 1);
+		break;
+	default:
+		cp_ctx(ctx, 0x402844, 1);
+		gr_def(ctx, 0x402844, 0x00001001);
+		cp_ctx(ctx, 0x402850, 2);
+		gr_def(ctx, 0x402854, 0x00000003);
+		break;
+	}
+
+	cp_ctx(ctx, 0x402c00, 4);
+	gr_def(ctx, 0x402c00, dev_priv->chipset == 0x40 ?
+			      0x80800001 : 0x00888001);
+	switch (dev_priv->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x402c20, 40);
+		for (i = 0; i < 32; i++)
+			gr_def(ctx, 0x402c40 + (i * 4), 0xffffffff);
+		cp_ctx(ctx, 0x4030b8, 13);
+		gr_def(ctx, 0x4030dc, 0x00000005);
+		gr_def(ctx, 0x4030e8, 0x0000ffff);
+		break;
+	default:
+		cp_ctx(ctx, 0x402c10, 4);
+		if (dev_priv->chipset == 0x40)
+			cp_ctx(ctx, 0x402c20, 36);
+		else
+		if (dev_priv->chipset <= 0x42)
+			cp_ctx(ctx, 0x402c20, 24);
+		else
+		if (dev_priv->chipset <= 0x4a)
+			cp_ctx(ctx, 0x402c20, 16);
+		else
+			cp_ctx(ctx, 0x402c20, 8);
+		cp_ctx(ctx, 0x402cb0, dev_priv->chipset == 0x40 ? 12 : 13);
+		gr_def(ctx, 0x402cd4, 0x00000005);
+		if (dev_priv->chipset != 0x40)
+			gr_def(ctx, 0x402ce0, 0x0000ffff);
+		break;
+	}
+
+	cp_ctx(ctx, 0x403400, dev_priv->chipset == 0x40 ? 4 : 3);
+	cp_ctx(ctx, 0x403410, dev_priv->chipset == 0x40 ? 4 : 3);
+	cp_ctx(ctx, 0x403420, nv40_graph_vs_count(ctx->dev));
+	for (i = 0; i < nv40_graph_vs_count(ctx->dev); i++)
+		gr_def(ctx, 0x403420 + (i * 4), 0x00005555);
+
+	if (dev_priv->chipset != 0x40) {
+		cp_ctx(ctx, 0x403600, 1);
+		gr_def(ctx, 0x403600, 0x00000001);
+	}
+	cp_ctx(ctx, 0x403800, 1);
+
+	cp_ctx(ctx, 0x403c18, 1);
+	gr_def(ctx, 0x403c18, 0x00000001);
+	switch (dev_priv->chipset) {
+	case 0x46:
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x405018, 1);
+		gr_def(ctx, 0x405018, 0x08e00001);
+		cp_ctx(ctx, 0x405c24, 1);
+		gr_def(ctx, 0x405c24, 0x000e3000);
+		break;
+	}
+	if (dev_priv->chipset != 0x4e)
+		cp_ctx(ctx, 0x405800, 11);
+	cp_ctx(ctx, 0x407000, 1);
+}
+
+static void
+nv40_graph_construct_state3d_3(struct nouveau_grctx *ctx)
+{
+	int len = nv40_graph_4097(ctx->dev) ? 0x0684 : 0x0084;
+
+	cp_out (ctx, 0x300000);
+	cp_lsr (ctx, len - 4);
+	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_swap_state3d_3_is_save);
+	cp_lsr (ctx, len);
+	cp_name(ctx, cp_swap_state3d_3_is_save);
+	cp_out (ctx, 0x800001);
+
+	ctx->ctxvals_pos += len;
+}
+
+static void
+nv40_graph_construct_shader(struct nouveau_grctx *ctx)
+{
+	struct drm_device *dev = ctx->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *obj = ctx->data;
+	int vs, vs_nr, vs_len, vs_nr_b0, vs_nr_b1, b0_offset, b1_offset;
+	int offset, i;
+
+	vs_nr    = nv40_graph_vs_count(ctx->dev);
+	vs_nr_b0 = 363;
+	vs_nr_b1 = dev_priv->chipset == 0x40 ? 128 : 64;
+	if (dev_priv->chipset == 0x40) {
+		b0_offset = 0x2200/4; /* 33a0 */
+		b1_offset = 0x55a0/4; /* 1500 */
+		vs_len = 0x6aa0/4;
+	} else
+	if (dev_priv->chipset == 0x41 || dev_priv->chipset == 0x42) {
+		b0_offset = 0x2200/4; /* 2200 */
+		b1_offset = 0x4400/4; /* 0b00 */
+		vs_len = 0x4f00/4;
+	} else {
+		b0_offset = 0x1d40/4; /* 2200 */
+		b1_offset = 0x3f40/4; /* 0b00 : 0a40 */
+		vs_len = nv40_graph_4097(dev) ? 0x4a40/4 : 0x4980/4;
+	}
+
+	cp_lsr(ctx, vs_len * vs_nr + 0x300/4);
+	cp_out(ctx, nv40_graph_4097(dev) ? 0x800041 : 0x800029);
+
+	offset = ctx->ctxvals_pos;
+	ctx->ctxvals_pos += (0x0300/4 + (vs_nr * vs_len));
+
+	if (ctx->mode != NOUVEAU_GRCTX_VALS)
+		return;
+
+	offset += 0x0280/4;
+	for (i = 0; i < 16; i++, offset += 2)
+		nv_wo32(dev, obj, offset, 0x3f800000);
+
+	for (vs = 0; vs < vs_nr; vs++, offset += vs_len) {
+		for (i = 0; i < vs_nr_b0 * 6; i += 6)
+			nv_wo32(dev, obj, offset + b0_offset + i, 0x00000001);
+		for (i = 0; i < vs_nr_b1 * 4; i += 4)
+			nv_wo32(dev, obj, offset + b1_offset + i, 0x3f800000);
+	}
+}
+
+void
+nv40_grctx_init(struct nouveau_grctx *ctx)
+{
+	/* decide whether we're loading/unloading the context */
+	cp_bra (ctx, AUTO_SAVE, PENDING, cp_setup_save);
+	cp_bra (ctx, USER_SAVE, PENDING, cp_setup_save);
+
+	cp_name(ctx, cp_check_load);
+	cp_bra (ctx, AUTO_LOAD, PENDING, cp_setup_auto_load);
+	cp_bra (ctx, USER_LOAD, PENDING, cp_setup_load);
+	cp_bra (ctx, ALWAYS, TRUE, cp_exit);
+
+	/* setup for context load */
+	cp_name(ctx, cp_setup_auto_load);
+	cp_wait(ctx, STATUS, IDLE);
+	cp_out (ctx, CP_NEXT_TO_SWAP);
+	cp_name(ctx, cp_setup_load);
+	cp_wait(ctx, STATUS, IDLE);
+	cp_set (ctx, SWAP_DIRECTION, LOAD);
+	cp_out (ctx, 0x00910880); /* ?? */
+	cp_out (ctx, 0x00901ffe); /* ?? */
+	cp_out (ctx, 0x01940000); /* ?? */
+	cp_lsr (ctx, 0x20);
+	cp_out (ctx, 0x0060000b); /* ?? */
+	cp_wait(ctx, UNK57, CLEAR);
+	cp_out (ctx, 0x0060000c); /* ?? */
+	cp_bra (ctx, ALWAYS, TRUE, cp_swap_state);
+
+	/* setup for context save */
+	cp_name(ctx, cp_setup_save);
+	cp_set (ctx, SWAP_DIRECTION, SAVE);
+
+	/* general PGRAPH state */
+	cp_name(ctx, cp_swap_state);
+	cp_pos (ctx, 0x00020/4);
+	nv40_graph_construct_general(ctx);
+	cp_wait(ctx, STATUS, IDLE);
+
+	/* 3D state, block 1 */
+	cp_bra (ctx, UNK54, CLEAR, cp_prepare_exit);
+	nv40_graph_construct_state3d(ctx);
+	cp_wait(ctx, STATUS, IDLE);
+
+	/* 3D state, block 2 */
+	nv40_graph_construct_state3d_2(ctx);
+
+	/* Some other block of "random" state */
+	nv40_graph_construct_state3d_3(ctx);
+
+	/* Per-vertex shader state */
+	cp_pos (ctx, ctx->ctxvals_pos);
+	nv40_graph_construct_shader(ctx);
+
+	/* pre-exit state updates */
+	cp_name(ctx, cp_prepare_exit);
+	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_check_load);
+	cp_bra (ctx, USER_SAVE, PENDING, cp_exit);
+	cp_out (ctx, CP_NEXT_TO_CURRENT);
+
+	cp_name(ctx, cp_exit);
+	cp_set (ctx, USER_SAVE, NOT_PENDING);
+	cp_set (ctx, USER_LOAD, NOT_PENDING);
+	cp_out (ctx, CP_END);
+}
+
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index 177d8229336f..ca79f32be44c 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -107,9 +107,13 @@ nv50_graph_init_regs(struct drm_device *dev)
 static int
 nv50_graph_init_ctxctl(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
 	NV_DEBUG(dev, "\n");
 
-	nv40_grctx_init(dev);
+	nouveau_grctx_prog_load(dev);
+	if (!dev_priv->engine.graph.ctxprog)
+		dev_priv->engine.graph.accel_blocked = true;
 
 	nv_wr32(dev, 0x400320, 4);
 	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0);
@@ -140,7 +144,7 @@ void
 nv50_graph_takedown(struct drm_device *dev)
 {
 	NV_DEBUG(dev, "\n");
-	nv40_grctx_fini(dev);
+	nouveau_grctx_fini(dev);
 }
 
 void
@@ -207,7 +211,7 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 	dev_priv->engine.instmem.finish_access(dev);
 
 	dev_priv->engine.instmem.prepare_access(dev, true);
-	nv40_grctx_vals_load(dev, ctx);
+	nouveau_grctx_vals_load(dev, ctx);
 	nv_wo32(dev, ctx, 0x00000/4, chan->ramin->instance >> 12);
 	if ((dev_priv->chipset & 0xf0) == 0xa0)
 		nv_wo32(dev, ctx, 0x00004/4, 0x00000000);