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

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 \

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);

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 *);

drivers/gpu/drm/nouveau/nouveau_grctx.c

+/*
+ * 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));
+}

drivers/gpu/drm/nouveau/nouveau_grctx.h

+#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

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[] = {

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);