drm/nv50: Implement ctxprog/state generation.

This removes dependence on external firmware for NV50 generation cards.
If the generated ctxprogs don't work for you for some reason, please
report it.

Signed-off-by: Marcin Kościelnicki <koriakin@0x04.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

4 files changed, 2411 insertions, 33 deletions

diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index 48c290b5da8..32db806f3b5 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -16,7 +16,7 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.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 \
+             nv40_grctx.o nv50_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.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 6fa9c87083c..2f8ce42f072 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -1029,6 +1029,7 @@ extern void nv50_graph_destroy_context(struct nouveau_channel *);
 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 *);
+extern int  nv50_grctx_init(struct nouveau_grctx *);
 
 /* nouveau_grctx.c */
 extern int  nouveau_grctx_prog_load(struct drm_device *);
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index 6d504801b51..857a09671a3 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -28,30 +28,7 @@
 #include "drm.h"
 #include "nouveau_drv.h"
 
-MODULE_FIRMWARE("nouveau/nv50.ctxprog");
-MODULE_FIRMWARE("nouveau/nv50.ctxvals");
-MODULE_FIRMWARE("nouveau/nv84.ctxprog");
-MODULE_FIRMWARE("nouveau/nv84.ctxvals");
-MODULE_FIRMWARE("nouveau/nv86.ctxprog");
-MODULE_FIRMWARE("nouveau/nv86.ctxvals");
-MODULE_FIRMWARE("nouveau/nv92.ctxprog");
-MODULE_FIRMWARE("nouveau/nv92.ctxvals");
-MODULE_FIRMWARE("nouveau/nv94.ctxprog");
-MODULE_FIRMWARE("nouveau/nv94.ctxvals");
-MODULE_FIRMWARE("nouveau/nv96.ctxprog");
-MODULE_FIRMWARE("nouveau/nv96.ctxvals");
-MODULE_FIRMWARE("nouveau/nv98.ctxprog");
-MODULE_FIRMWARE("nouveau/nv98.ctxvals");
-MODULE_FIRMWARE("nouveau/nva0.ctxprog");
-MODULE_FIRMWARE("nouveau/nva0.ctxvals");
-MODULE_FIRMWARE("nouveau/nva5.ctxprog");
-MODULE_FIRMWARE("nouveau/nva5.ctxvals");
-MODULE_FIRMWARE("nouveau/nva8.ctxprog");
-MODULE_FIRMWARE("nouveau/nva8.ctxvals");
-MODULE_FIRMWARE("nouveau/nvaa.ctxprog");
-MODULE_FIRMWARE("nouveau/nvaa.ctxvals");
-MODULE_FIRMWARE("nouveau/nvac.ctxprog");
-MODULE_FIRMWARE("nouveau/nvac.ctxvals");
+#include "nouveau_grctx.h"
 
 #define IS_G80 ((dev_priv->chipset & 0xf0) == 0x50)
 
@@ -111,9 +88,34 @@ nv50_graph_init_ctxctl(struct drm_device *dev)
 
 	NV_DEBUG(dev, "\n");
 
-	nouveau_grctx_prog_load(dev);
-	if (!dev_priv->engine.graph.ctxprog)
-		dev_priv->engine.graph.accel_blocked = true;
+	if (nouveau_ctxfw) {
+		nouveau_grctx_prog_load(dev);
+		dev_priv->engine.graph.grctx_size = 0x70000;
+	}
+	if (!dev_priv->engine.graph.ctxprog) {
+		struct nouveau_grctx ctx = {};
+		uint32_t *cp = kmalloc(512 * 4, GFP_KERNEL);
+		int i;
+		if (!cp) {
+			NV_ERROR(dev, "Couldn't alloc ctxprog! Disabling acceleration.\n");
+			dev_priv->engine.graph.accel_blocked = true;
+			return 0;
+		}
+		ctx.dev = dev;
+		ctx.mode = NOUVEAU_GRCTX_PROG;
+		ctx.data = cp;
+		ctx.ctxprog_max = 512;
+		if (!nv50_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]);
+		} else {
+			dev_priv->engine.graph.accel_blocked = true;
+		}
+		kfree(cp);
+	}
 
 	nv_wr32(dev, 0x400320, 4);
 	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0);
@@ -193,13 +195,13 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *ramin = chan->ramin->gpuobj;
 	struct nouveau_gpuobj *ctx;
-	uint32_t grctx_size = 0x70000;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	int hdr, ret;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, grctx_size, 0x1000,
-				     NVOBJ_FLAG_ZERO_ALLOC |
+	ret = nouveau_gpuobj_new_ref(dev, chan, NULL, 0, pgraph->grctx_size,
+				     0x1000, NVOBJ_FLAG_ZERO_ALLOC |
 				     NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
 	if (ret)
 		return ret;
@@ -209,7 +211,7 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 	dev_priv->engine.instmem.prepare_access(dev, true);
 	nv_wo32(dev, ramin, (hdr + 0x00)/4, 0x00190002);
 	nv_wo32(dev, ramin, (hdr + 0x04)/4, chan->ramin_grctx->instance +
-					   grctx_size - 1);
+					   pgraph->grctx_size - 1);
 	nv_wo32(dev, ramin, (hdr + 0x08)/4, chan->ramin_grctx->instance);
 	nv_wo32(dev, ramin, (hdr + 0x0c)/4, 0);
 	nv_wo32(dev, ramin, (hdr + 0x10)/4, 0);
@@ -217,7 +219,15 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 	dev_priv->engine.instmem.finish_access(dev);
 
 	dev_priv->engine.instmem.prepare_access(dev, true);
-	nouveau_grctx_vals_load(dev, ctx);
+	if (!pgraph->ctxprog) {
+		struct nouveau_grctx ctx = {};
+		ctx.dev = chan->dev;
+		ctx.mode = NOUVEAU_GRCTX_VALS;
+		ctx.data = chan->ramin_grctx->gpuobj;
+		nv50_grctx_init(&ctx);
+	} else {
+		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);
diff --git a/drivers/gpu/drm/nouveau/nv50_grctx.c b/drivers/gpu/drm/nouveau/nv50_grctx.c
new file mode 100644
index 00000000000..d105fcd42ca
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv50_grctx.c
@@ -0,0 +1,2367 @@
+/*
+ * Copyright 2009 Marcin Kościelnicki
+ *
+ * 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.
+ */
+
+#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_UNK01                 ((0 * 32) + 1)
+#define CP_FLAG_UNK01_CLEAR           0
+#define CP_FLAG_UNK01_SET             1
+#define CP_FLAG_UNK03                 ((0 * 32) + 3)
+#define CP_FLAG_UNK03_CLEAR           0
+#define CP_FLAG_UNK03_SET             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_UNK0B                 ((0 * 32) + 0xb)
+#define CP_FLAG_UNK0B_CLEAR           0
+#define CP_FLAG_UNK0B_SET             1
+#define CP_FLAG_UNK1D                 ((0 * 32) + 0x1d)
+#define CP_FLAG_UNK1D_CLEAR           0
+#define CP_FLAG_UNK1D_SET             1
+#define CP_FLAG_UNK20                 ((1 * 32) + 0)
+#define CP_FLAG_UNK20_CLEAR           0
+#define CP_FLAG_UNK20_SET             1
+#define CP_FLAG_STATUS                ((2 * 32) + 0)
+#define CP_FLAG_STATUS_BUSY           0
+#define CP_FLAG_STATUS_IDLE           1
+#define CP_FLAG_AUTO_SAVE             ((2 * 32) + 4)
+#define CP_FLAG_AUTO_SAVE_NOT_PENDING 0
+#define CP_FLAG_AUTO_SAVE_PENDING     1
+#define CP_FLAG_AUTO_LOAD             ((2 * 32) + 5)
+#define CP_FLAG_AUTO_LOAD_NOT_PENDING 0
+#define CP_FLAG_AUTO_LOAD_PENDING     1
+#define CP_FLAG_XFER                  ((2 * 32) + 11)
+#define CP_FLAG_XFER_IDLE             0
+#define CP_FLAG_XFER_BUSY             1
+#define CP_FLAG_NEWCTX                ((2 * 32) + 12)
+#define CP_FLAG_NEWCTX_BUSY           0
+#define CP_FLAG_NEWCTX_DONE           1
+#define CP_FLAG_ALWAYS                ((2 * 32) + 13)
+#define CP_FLAG_ALWAYS_FALSE          0
+#define CP_FLAG_ALWAYS_TRUE           1
+
+#define CP_CTX                   0x00100000
+#define CP_CTX_COUNT             0x000f0000
+#define CP_CTX_COUNT_SHIFT               16
+#define CP_CTX_REG               0x00003fff
+#define CP_LOAD_SR               0x00200000
+#define CP_LOAD_SR_VALUE         0x000fffff
+#define CP_BRA                   0x00400000
+#define CP_BRA_IP                0x0001ff00
+#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_NEWCTX                0x00600004
+#define CP_NEXT_TO_SWAP          0x00600005
+#define CP_SET_CONTEXT_POINTER   0x00600006
+#define CP_SET_XFER_POINTER      0x00600007
+#define CP_ENABLE                0x00600009
+#define CP_END                   0x0060000c
+#define CP_NEXT_TO_CURRENT       0x0060000d
+#define CP_DISABLE1              0x0090ffff
+#define CP_DISABLE2              0x0091ffff
+#define CP_XFER_1      0x008000ff
+#define CP_XFER_2      0x008800ff
+#define CP_SEEK_1      0x00c000ff
+#define CP_SEEK_2      0x00c800ff
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_grctx.h"
+
+/*
+ * This code deals with PGRAPH contexts on NV50 family cards. Like NV40, it's
+ * the GPU itself that does context-switching, but it needs a special
+ * microcode to do it. And it's the driver's task to supply this microcode,
+ * further known as ctxprog, as well as the initial context values, known
+ * as ctxvals.
+ *
+ * Without ctxprog, you cannot switch contexts. Not even in software, since
+ * the majority of context [xfer strands] isn't accessible directly. You're
+ * stuck with a single channel, and you also suffer all the problems resulting
+ * from missing ctxvals, since you cannot load them.
+ *
+ * Without ctxvals, you're stuck with PGRAPH's default context. It's enough to
+ * run 2d operations, but trying to utilise 3d or CUDA will just lock you up,
+ * since you don't have... some sort of needed setup.
+ *
+ * Nouveau will just disable acceleration if not given ctxprog + ctxvals, since
+ * it's too much hassle to handle no-ctxprog as a special case.
+ */
+
+/*
+ * How ctxprogs work.
+ *
+ * The ctxprog is written in its own kind of microcode, with very small and
+ * crappy set of available commands. You upload it to a small [512 insns]
+ * area of memory on PGRAPH, and it'll be run when PFIFO wants PGRAPH to
+ * switch channel. or when the driver explicitely requests it. Stuff visible
+ * to ctxprog consists of: PGRAPH MMIO registers, PGRAPH context strands,
+ * the per-channel context save area in VRAM [known as ctxvals or grctx],
+ * 4 flags registers, a scratch register, two grctx pointers, plus many
+ * random poorly-understood details.
+ *
+ * When ctxprog runs, it's supposed to check what operations are asked of it,
+ * save old context if requested, optionally reset PGRAPH and switch to the
+ * new channel, and load the new context. Context consists of three major
+ * parts: subset of MMIO registers and two "xfer areas".
+ */
+
+/* TODO:
+ *  - document unimplemented bits compared to nvidia
+ *  - NVAx: make a TP subroutine, use it.
+ *  - use 0x4008fc instead of 0x1540?
+ */
+
+enum cp_label {
+	cp_check_load = 1,
+	cp_setup_auto_load,
+	cp_setup_load,
+	cp_setup_save,
+	cp_swap_state,
+	cp_prepare_exit,
+	cp_exit,
+};
+
+static void nv50_graph_construct_mmio(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_xfer1(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_xfer2(struct nouveau_grctx *ctx);
+
+/* Main function: construct the ctxprog skeleton, call the other functions. */
+
+int
+nv50_grctx_init(struct nouveau_grctx *ctx)
+{
+	struct drm_nouveau_private *dev_priv = ctx->dev->dev_private;
+
+	switch (dev_priv->chipset) {
+	case 0x50:
+	case 0x84:
+	case 0x86:
+	case 0x92:
+	case 0x94:
+	case 0x96:
+	case 0x98:
+	case 0xa0:
+	case 0xa5:
+	case 0xa8:
+	case 0xaa:
+	case 0xac:
+		break;
+	default:
+		NV_ERROR(ctx->dev, "I don't know how to make a ctxprog for "
+				   "your NV%x card.\n", dev_priv->chipset);
+		NV_ERROR(ctx->dev, "Disabling acceleration. Please contact "
+				   "the devs.\n");
+		return -ENOSYS;
+	}
+	/* 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_out (ctx, CP_DISABLE1);
+	cp_out (ctx, CP_DISABLE2);
+	cp_out (ctx, CP_ENABLE);
+	cp_out (ctx, CP_NEXT_TO_SWAP);
+	cp_set (ctx, UNK01, SET);
+	cp_name(ctx, cp_setup_load);
+	cp_out (ctx, CP_NEWCTX);
+	cp_wait(ctx, NEWCTX, BUSY);
+	cp_set (ctx, UNK1D, CLEAR);
+	cp_set (ctx, SWAP_DIRECTION, LOAD);
+	cp_bra (ctx, UNK0B, SET, cp_prepare_exit);
+	cp_bra (ctx, ALWAYS, TRUE, cp_swap_state);
+
+	/* setup for context save */
+	cp_name(ctx, cp_setup_save);
+	cp_set (ctx, UNK1D, SET);
+	cp_wait(ctx, STATUS, BUSY);
+	cp_set (ctx, UNK01, SET);
+	cp_set (ctx, SWAP_DIRECTION, SAVE);
+
+	/* general PGRAPH state */
+	cp_name(ctx, cp_swap_state);
+	cp_set (ctx, UNK03, SET);
+	cp_pos (ctx, 0x00004/4);
+	cp_ctx (ctx, 0x400828, 1); /* needed. otherwise, flickering happens. */
+	cp_pos (ctx, 0x00100/4);
+	nv50_graph_construct_mmio(ctx);
+	nv50_graph_construct_xfer1(ctx);
+	nv50_graph_construct_xfer2(ctx);
+
+	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_check_load);
+
+	cp_set (ctx, UNK20, SET);
+	cp_set (ctx, SWAP_DIRECTION, SAVE); /* no idea why this is needed, but fixes at least one lockup. */
+	cp_lsr (ctx, ctx->ctxvals_base);
+	cp_out (ctx, CP_SET_XFER_POINTER);
+	cp_lsr (ctx, 4);
+	cp_out (ctx, CP_SEEK_1);
+	cp_out (ctx, CP_XFER_1);
+	cp_wait(ctx, XFER, BUSY);
+
+	/* pre-exit state updates */
+	cp_name(ctx, cp_prepare_exit);
+	cp_set (ctx, UNK01, CLEAR);
+	cp_set (ctx, UNK03, CLEAR);
+	cp_set (ctx, UNK1D, CLEAR);
+
+	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);
+	ctx->ctxvals_pos += 0x400; /* padding... no idea why you need it */
+
+	return 0;
+}
+
+/*
+ * Constructs MMIO part of ctxprog and ctxvals. Just a matter of knowing which
+ * registers to save/restore and the default values for them.
+ */
+
+static void
+nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
+{
+	struct drm_nouveau_private *dev_priv = ctx->dev->dev_private;
+	int i, j;
+	int offset, base;
+	uint32_t units = nv_rd32 (ctx->dev, 0x1540);
+
+	/* 0800 */
+	cp_ctx(ctx, 0x400808, 7);
+	gr_def(ctx, 0x400814, 0x00000030);
+	cp_ctx(ctx, 0x400834, 0x32);
+	if (dev_priv->chipset == 0x50) {
+		gr_def(ctx, 0x400834, 0xff400040);
+		gr_def(ctx, 0x400838, 0xfff00080);
+		gr_def(ctx, 0x40083c, 0xfff70090);
+		gr_def(ctx, 0x400840, 0xffe806a8);
+	}
+	gr_def(ctx, 0x400844, 0x00000002);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa)
+		gr_def(ctx, 0x400894, 0x00001000);
+	gr_def(ctx, 0x4008e8, 0x00000003);
+	gr_def(ctx, 0x4008ec, 0x00001000);
+	if (dev_priv->chipset == 0x50)
+		cp_ctx(ctx, 0x400908, 0xb);
+	else if (dev_priv->chipset < 0xa0)
+		cp_ctx(ctx, 0x400908, 0xc);
+	else
+		cp_ctx(ctx, 0x400908, 0xe);
+
+	if (dev_priv->chipset >= 0xa0)
+		cp_ctx(ctx, 0x400b00, 0x1);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa) {
+		cp_ctx(ctx, 0x400b10, 0x1);
+		gr_def(ctx, 0x400b10, 0x0001629d);
+		cp_ctx(ctx, 0x400b20, 0x1);
+		gr_def(ctx, 0x400b20, 0x0001629d);
+	}
+
+	/* 0C00 */
+	cp_ctx(ctx, 0x400c08, 0x2);
+	gr_def(ctx, 0x400c08, 0x0000fe0c);
+
+	/* 1000 */
+	if (dev_priv->chipset < 0xa0) {
+		cp_ctx(ctx, 0x401008, 0x4);
+		gr_def(ctx, 0x401014, 0x00001000);
+	} else if (dev_priv->chipset == 0xa0 || dev_priv->chipset >= 0xaa) {
+		cp_ctx(ctx, 0x401008, 0x5);
+		gr_def(ctx, 0x401018, 0x00001000);
+	} else {
+		cp_ctx(ctx, 0x401008, 0x5);
+		gr_def(ctx, 0x401018, 0x00004000);
+	}
+
+	/* 1400 */
+	cp_ctx(ctx, 0x401400, 0x8);
+	cp_ctx(ctx, 0x401424, 0x3);
+	if (dev_priv->chipset == 0x50)
+		gr_def(ctx, 0x40142c, 0x0001fd87);
+	else
+		gr_def(ctx, 0x40142c, 0x00000187);
+	cp_ctx(ctx, 0x401540, 0x5);
+	gr_def(ctx, 0x401550, 0x00001018);
+
+	/* 1800 */
+	cp_ctx(ctx, 0x401814, 0x1);
+	gr_def(ctx, 0x401814, 0x000000ff);
+	if (dev_priv->chipset == 0x50) {
+		cp_ctx(ctx, 0x40181c, 0xe);
+		gr_def(ctx, 0x401850, 0x00000004);
+	} else if (dev_priv->chipset < 0xa0) {
+		cp_ctx(ctx, 0x40181c, 0xf);
+		gr_def(ctx, 0x401854, 0x00000004);
+	} else {
+		cp_ctx(ctx, 0x40181c, 0x13);
+		gr_def(ctx, 0x401864, 0x00000004);
+	}
+
+	/* 1C00 */
+	cp_ctx(ctx, 0x401c00, 0x1);
+	switch (dev_priv->chipset) {
+	case 0x50:
+		gr_def(ctx, 0x401c00, 0x0001005f);
+		break;
+	case 0x84:
+	case 0x86:
+	case 0x94:
+		gr_def(ctx, 0x401c00, 0x044d00df);
+		break;
+	case 0x92:
+	case 0x96:
+	case 0x98:
+	case 0xa0:
+	case 0xaa:
+	case 0xac:
+		gr_def(ctx, 0x401c00, 0x042500df);
+		break;
+	case 0xa5:
+	case 0xa8:
+		gr_def(ctx, 0x401c00, 0x142500df);
+		break;
+	}
+
+	/* 2400 */
+	cp_ctx(ctx, 0x402400, 0x1);
+	if (dev_priv->chipset == 0x50)
+		cp_ctx(ctx, 0x402408, 0x1);
+	else
+		cp_ctx(ctx, 0x402408, 0x2);
+	gr_def(ctx, 0x402408, 0x00000600);
+
+	/* 2800 */
+	cp_ctx(ctx, 0x402800, 0x1);
+	if (dev_priv->chipset == 0x50)
+		gr_def(ctx, 0x402800, 0x00000006);
+
+	/* 2C00 */
+	cp_ctx(ctx, 0x402c08, 0x6);
+	if (dev_priv->chipset != 0x50)
+		gr_def(ctx, 0x402c14, 0x01000000);
+	gr_def(ctx, 0x402c18, 0x000000ff);
+	if (dev_priv->chipset == 0x50)
+		cp_ctx(ctx, 0x402ca0, 0x1);
+	else
+		cp_ctx(ctx, 0x402ca0, 0x2);
+	if (dev_priv->chipset < 0xa0)
+		gr_def(ctx, 0x402ca0, 0x00000400);
+	else if (dev_priv->chipset == 0xa0 || dev_priv->chipset >= 0xaa)
+		gr_def(ctx, 0x402ca0, 0x00000800);
+	else
+		gr_def(ctx, 0x402ca0, 0x00000400);
+	cp_ctx(ctx, 0x402cac, 0x4);
+
+	/* 3000 */
+	cp_ctx(ctx, 0x403004, 0x1);
+	gr_def(ctx, 0x403004, 0x00000001);
+
+	/* 3404 */
+	if (dev_priv->chipset >= 0xa0) {
+		cp_ctx(ctx, 0x403404, 0x1);
+		gr_def(ctx, 0x403404, 0x00000001);
+	}
+
+	/* 5000 */
+	cp_ctx(ctx, 0x405000, 0x1);
+	switch (dev_priv->chipset) {
+	case 0x50:
+		gr_def(ctx, 0x405000, 0x00300080);
+		break;
+	case 0x84:
+	case 0xa0:
+	case 0xa5:
+	case 0xa8:
+	case 0xaa:
+	case 0xac:
+		gr_def(ctx, 0x405000, 0x000e0080);
+		break;
+	case 0x86:
+	case 0x92:
+	case 0x94:
+	case 0x96:
+	case 0x98:
+		gr_def(ctx, 0x405000, 0x00000080);
+		break;
+	}
+	cp_ctx(ctx, 0x405014, 0x1);
+	gr_def(ctx, 0x405014, 0x00000004);
+	cp_ctx(ctx, 0x40501c, 0x1);
+	cp_ctx(ctx, 0x405024, 0x1);
+	cp_ctx(ctx, 0x40502c, 0x1);
+
+	/* 5400 or maybe 4800 */
+	if (dev_priv->chipset == 0x50) {
+		offset = 0x405400;
+		cp_ctx(ctx, 0x405400, 0xea);
+	} else if (dev_priv->chipset < 0x94) {
+		offset = 0x405400;
+		cp_ctx(ctx, 0x405400, 0xcb);
+	} else if (dev_priv->chipset < 0xa0) {
+		offset = 0x405400;
+		cp_ctx(ctx, 0x405400, 0xcc);
+	} else if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa) {
+		offset = 0x404800;
+		cp_ctx(ctx, 0x404800, 0xda);
+	} else {
+		offset = 0x405400;
+		cp_ctx(ctx, 0x405400, 0xd4);
+	}
+	gr_def(ctx, offset + 0x0c, 0x00000002);
+	gr_def(ctx, offset + 0x10, 0x00000001);
+	if (dev_priv->chipset >= 0x94)
+		offset += 4;
+	gr_def(ctx, offset + 0x1c, 0x00000001);
+	gr_def(ctx, offset + 0x20, 0x00000100);
+	gr_def(ctx, offset + 0x38, 0x00000002);
+	gr_def(ctx, offset + 0x3c, 0x00000001);
+	gr_def(ctx, offset + 0x40, 0x00000001);
+	gr_def(ctx, offset + 0x50, 0x00000001);
+	gr_def(ctx, offset + 0x54, 0x003fffff);
+	gr_def(ctx, offset + 0x58, 0x00001fff);
+	gr_def(ctx, offset + 0x60, 0x00000001);
+	gr_def(ctx, offset + 0x64, 0x00000001);
+	gr_def(ctx, offset + 0x6c, 0x00000001);
+	gr_def(ctx, offset + 0x70, 0x00000001);
+	gr_def(ctx, offset + 0x74, 0x00000001);
+	gr_def(ctx, offset + 0x78, 0x00000004);
+	gr_def(ctx, offset + 0x7c, 0x00000001);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa)
+		offset += 4;
+	gr_def(ctx, offset + 0x80, 0x00000001);
+	gr_def(ctx, offset + 0x84, 0x00000001);
+	gr_def(ctx, offset + 0x88, 0x00000007);
+	gr_def(ctx, offset + 0x8c, 0x00000001);
+	gr_def(ctx, offset + 0x90, 0x00000007);
+	gr_def(ctx, offset + 0x94, 0x00000001);
+	gr_def(ctx, offset + 0x98, 0x00000001);
+	gr_def(ctx, offset + 0x9c, 0x00000001);
+	if (dev_priv->chipset == 0x50) {
+		 gr_def(ctx, offset + 0xb0, 0x00000001);
+		 gr_def(ctx, offset + 0xb4, 0x00000001);
+		 gr_def(ctx, offset + 0xbc, 0x00000001);
+		 gr_def(ctx, offset + 0xc0, 0x0000000a);
+		 gr_def(ctx, offset + 0xd0, 0x00000040);
+		 gr_def(ctx, offset + 0xd8, 0x00000002);
+		 gr_def(ctx, offset + 0xdc, 0x00000100);
+		 gr_def(ctx, offset + 0xe0, 0x00000001);
+		 gr_def(ctx, offset + 0xe4, 0x00000100);
+		 gr_def(ctx, offset + 0x100, 0x00000001);
+		 gr_def(ctx, offset + 0x124, 0x00000004);
+		 gr_def(ctx, offset + 0x13c, 0x00000001);
+		 gr_def(ctx, offset + 0x140, 0x00000100);
+		 gr_def(ctx, offset + 0x148, 0x00000001);
+		 gr_def(ctx, offset + 0x154, 0x00000100);
+		 gr_def(ctx, offset + 0x158, 0x00000001);
+		 gr_def(ctx, offset + 0x15c, 0x00000100);
+		 gr_def(ctx, offset + 0x164, 0x00000001);
+		 gr_def(ctx, offset + 0x170, 0x00000100);
+		 gr_def(ctx, offset + 0x174, 0x00000001);
+		 gr_def(ctx, offset + 0x17c, 0x00000001);
+		 gr_def(ctx, offset + 0x188, 0x00000002);
+		 gr_def(ctx, offset + 0x190, 0x00000001);
+		 gr_def(ctx, offset + 0x198, 0x00000001);
+		 gr_def(ctx, offset + 0x1ac, 0x00000003);
+		 offset += 0xd0;
+	} else {
+		gr_def(ctx, offset + 0xb0, 0x00000001);
+		gr_def(ctx, offset + 0xb4, 0x00000100);
+		gr_def(ctx, offset + 0xbc, 0x00000001);
+		gr_def(ctx, offset + 0xc8, 0x00000100);
+		gr_def(ctx, offset + 0xcc, 0x00000001);
+		gr_def(ctx, offset + 0xd0, 0x00000100);
+		gr_def(ctx, offset + 0xd8, 0x00000001);
+		gr_def(ctx, offset + 0xe4, 0x00000100);
+	}
+	gr_def(ctx, offset + 0xf8, 0x00000004);
+	gr_def(ctx, offset + 0xfc, 0x00000070);
+	gr_def(ctx, offset + 0x100, 0x00000080);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa)
+		offset += 4;
+	gr_def(ctx, offset + 0x114, 0x0000000c);
+	if (dev_priv->chipset == 0x50)
+		offset -= 4;
+	gr_def(ctx, offset + 0x11c, 0x00000008);
+	gr_def(ctx, offset + 0x120, 0x00000014);
+	if (dev_priv->chipset == 0x50) {
+		gr_def(ctx, offset + 0x124, 0x00000026);
+		offset -= 0x18;
+	} else {
+		gr_def(ctx, offset + 0x128, 0x00000029);
+		gr_def(ctx, offset + 0x12c, 0x00000027);
+		gr_def(ctx, offset + 0x130, 0x00000026);
+		gr_def(ctx, offset + 0x134, 0x00000008);
+		gr_def(ctx, offset + 0x138, 0x00000004);
+		gr_def(ctx, offset + 0x13c, 0x00000027);
+	}
+	gr_def(ctx, offset + 0x148, 0x00000001);
+	gr_def(ctx, offset + 0x14c, 0x00000002);
+	gr_def(ctx, offset + 0x150, 0x00000003);
+	gr_def(ctx, offset + 0x154, 0x00000004);
+	gr_def(ctx, offset + 0x158, 0x00000005);
+	gr_def(ctx, offset + 0x15c, 0x00000006);
+	gr_def(ctx, offset + 0x160, 0x00000007);
+	gr_def(ctx, offset + 0x164, 0x00000001);
+	gr_def(ctx, offset + 0x1a8, 0x000000cf);
+	if (dev_priv->chipset == 0x50)
+		offset -= 4;
+	gr_def(ctx, offset + 0x1d8, 0x00000080);
+	gr_def(ctx, offset + 0x1dc, 0x00000004);
+	gr_def(ctx, offset + 0x1e0, 0x00000004);
+	if (dev_priv->chipset == 0x50)
+		offset -= 4;
+	else
+		gr_def(ctx, offset + 0x1e4, 0x00000003);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa) {
+		gr_def(ctx, offset + 0x1ec, 0x00000003);
+		offset += 8;
+	}
+	gr_def(ctx, offset + 0x1e8, 0x00000001);
+	if (dev_priv->chipset == 0x50)
+		offset -= 4;
+	gr_def(ctx, offset + 0x1f4, 0x00000012);
+	gr_def(ctx, offset + 0x1f8, 0x00000010);
+	gr_def(ctx, offset + 0x1fc, 0x0000000c);
+	gr_def(ctx, offset + 0x200, 0x00000001);
+	gr_def(ctx, offset + 0x210, 0x00000004);
+	gr_def(ctx, offset + 0x214, 0x00000002);
+	gr_def(ctx, offset + 0x218, 0x00000004);
+	if (dev_priv->chipset >= 0xa0)
+		offset += 4;
+	gr_def(ctx, offset + 0x224, 0x003fffff);
+	gr_def(ctx, offset + 0x228, 0x00001fff);
+	if (dev_priv->chipset == 0x50)
+		offset -= 0x20;
+	else if (dev_priv->chipset >= 0xa0) {
+		gr_def(ctx, offset + 0x250, 0x00000001);
+		gr_def(ctx, offset + 0x254, 0x00000001);
+		gr_def(ctx, offset + 0x258, 0x00000002);
+		offset += 0x10;
+	}
+	gr_def(ctx, offset + 0x250, 0x00000004);
+	gr_def(ctx, offset + 0x254, 0x00000014);
+	gr_def(ctx, offset + 0x258, 0x00000001);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa)
+		offset += 4;
+	gr_def(ctx, offset + 0x264, 0x00000002);
+	if (dev_priv->chipset >= 0xa0)
+		offset += 8;
+	gr_def(ctx, offset + 0x270, 0x00000001);
+	gr_def(ctx, offset + 0x278, 0x00000002);
+	gr_def(ctx, offset + 0x27c, 0x00001000);
+	if (dev_priv->chipset == 0x50)
+		offset -= 0xc;
+	else {
+		gr_def(ctx, offset + 0x280, 0x00000e00);
+		gr_def(ctx, offset + 0x284, 0x00001000);
+		gr_def(ctx, offset + 0x288, 0x00001e00);
+	}
+	gr_def(ctx, offset + 0x290, 0x00000001);
+	gr_def(ctx, offset + 0x294, 0x00000001);
+	gr_def(ctx, offset + 0x298, 0x00000001);
+	gr_def(ctx, offset + 0x29c, 0x00000001);
+	gr_def(ctx, offset + 0x2a0, 0x00000001);
+	gr_def(ctx, offset + 0x2b0, 0x00000200);
+	if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa) {
+		gr_def(ctx, offset + 0x2b4, 0x00000200);
+		offset += 4;
+	}
+	if (dev_priv->chipset < 0xa0) {
+		gr_def(ctx, offset + 0x2b8, 0x00000001);
+		gr_def(ctx, offset + 0x2bc, 0x00000070);
+		gr_def(ctx, offset + 0x2c0, 0x00000080);
+		gr_def(ctx, offset + 0x2cc, 0x00000001);
+		gr_def(ctx, offset + 0x2d0, 0x00000070);
+		gr_def(ctx, offset + 0x2d4, 0x00000080);
+	} else {
+		gr_def(ctx, offset + 0x2b8, 0x00000001);
+		gr_def(ctx, offset + 0x2bc, 0x000000f0);
+		gr_def(ctx, offset + 0x2c0, 0x000000ff);
+		gr_def(ctx, offset + 0x2cc, 0x00000001);
+		gr_def(ctx, offset + 0x2d0, 0x000000f0);
+		gr_def(ctx, offset + 0x2d4, 0x000000ff);
+		gr_def(ctx, offset + 0x2dc, 0x00000009);
+		offset += 4;
+	}
+	gr_def(ctx, offset + 0x2e4, 0x00000001);
+	gr_def(ctx, offset + 0x2e8, 0x000000cf);
+	gr_def(ctx, offset + 0x2f0, 0x00000001);
+	gr_def(ctx, offset + 0x300, 0x000000cf);
+	gr_def(ctx, offset + 0x308, 0x00000002);
+	gr_def(ctx, offset + 0x310, 0x00000001);
+	gr_def(ctx, offset + 0x318, 0x00000001);
+	gr_def(ctx, offset + 0x320, 0x000000cf);
+	gr_def(ctx, offset + 0x324, 0x000000cf);
+	gr_def(ctx, offset + 0x328, 0x00000001);
+
+	/* 6000? */
+	if (dev_priv->chipset == 0x50)
+		cp_ctx(ctx, 0x4063e0, 0x1);
+
+	/* 6800 */
+	if (dev_priv->chipset < 0x90) {
+		cp_ctx(ctx, 0x406814, 0x2b);
+		gr_def(ctx, 0x406818, 0x00000f80);
+		gr_def(ctx, 0x406860, 0x007f0080);
+		gr_def(ctx, 0x40689c, 0x007f0080);
+	} else {
+		cp_ctx(ctx, 0x406814, 0x4);
+		if (dev_priv->chipset == 0x98)
+			gr_def(ctx, 0x406818, 0x00000f80);
+		else
+			gr_def(ctx, 0x406818, 0x00001f80);
+		if (dev_priv->chipset > 0xa0 && dev_priv->chipset < 0xaa)
+			gr_def(ctx, 0x40681c, 0x00000030);
+		cp_ctx(ctx, 0x406830, 0x3);
+	}
+
+	/* 7000: per-ROP group state */
+	for (i = 0; i < 8; i++) {
+		if (units & (1<<(i+16))) {
+			cp_ctx(ctx, 0x407000 + (i<<8), 3);
+			if (dev_priv->chipset == 0x50)
+				gr_def(ctx, 0x407000 + (i<<8), 0x1b74f820);
+			else if (dev_priv->chipset != 0xa5)
+				gr_def(ctx, 0x407000 + (i<<8), 0x3b74f821);
+			else
+				gr_def(ctx, 0x407000 + (i<<8), 0x7b74f821);
+			gr_def(ctx, 0x407004 + (i<<8), 0x89058001);
+
+			if (dev_priv->chipset == 0x50) {
+				cp_ctx(ctx, 0x407010 + (i<<8), 1);
+			} else if (dev_priv->chipset < 0xa0) {
+				cp_ctx(ctx, 0x407010 + (i<<8), 2);
+				gr_def(ctx, 0x407010 + (i<<8), 0x00001000);
+				gr_def(ctx, 0x407014 + (i<<8), 0x0000001f);
+			} else {
+				cp_ctx(ctx, 0x407010 + (i<<8), 3);
+				gr_def(ctx, 0x407010 + (i<<8), 0x00001000);
+				if (dev_priv->chipset != 0xa5)
+					gr_def(ctx, 0x407014 + (i<<8), 0x000000ff);
+				else
+					gr_def(ctx, 0x407014 + (i<<8), 0x000001ff);
+			}
+
+			cp_ctx(ctx, 0x407080 + (i<<8), 4);
+			if (dev_priv->chipset != 0xa5)
+				gr_def(ctx, 0x407080 + (i<<8), 0x027c10fa);
+			else
+				gr_def(ctx, 0x407080 + (i<<8), 0x827c10fa);
+			if (dev_priv->chipset == 0x50)
+				gr_def(ctx, 0x407084 + (i<<8), 0x000000c0);
+			else
+				gr_def(ctx, 0x407084 + (i<<8), 0x400000c0);
+			gr_def(ctx, 0x407088 + (i<<8), 0xb7892080);
+
+			if (dev_priv->chipset < 0xa0)
+				cp_ctx(ctx, 0x407094 + (i<<8), 1);
+			else if (dev_priv->chipset <= 0xa0 || dev_priv->chipset >= 0xaa)
+				cp_ctx(ctx, 0x407094 + (i<<8), 3);
+			else {
+				cp_ctx(ctx, 0x407094 + (i<<8), 4);
+				gr_def(ctx, 0x4070a0 + (i<<8), 1);
+			}
+		}
+	}
+
+	cp_ctx(ctx, 0x407c00, 0x3);
+	if (dev_priv->chipset < 0x90)
+		gr_def(ctx, 0x407c00, 0x00010040);
+	else if (dev_priv->chipset < 0xa0)
+		gr_def(ctx, 0x407c00, 0x00390040);
+	else
+		gr_def(ctx, 0x407c00, 0x003d0040);
+	gr_def(ctx, 0x407c08, 0x00000022);
+	if (dev_priv->chipset >= 0xa0) {
+		cp_ctx(ctx, 0x407c10, 0x3);
+		cp_ctx(ctx, 0x407c20, 0x1);
+		cp_ctx(ctx, 0x407c2c, 0x1);
+	}
+
+	if (dev_priv->chipset < 0xa0) {
+		cp_ctx(ctx, 0x407d00, 0x9);
+	} else {
+		cp_ctx(ctx, 0x407d00, 0x15);
+	}
+	if (dev_priv->chipset == 0x98)
+		gr_def(ctx, 0x407d08, 0x00380040);
+	else {
+		if (dev_priv->chipset < 0x90)
+			gr_def(ctx, 0x407d08, 0x00010040);
+		else if (dev_priv->chipset < 0xa0)
+			gr_def(ctx, 0x407d08, 0x00390040);
+		else
+			gr_def(ctx, 0x407d08, 0x003d0040);
+		gr_def(ctx, 0x407d0c, 0x00000022);
+	}
+
+	/* 8000+: per-TP state */
+	for (i = 0; i < 10; i++) {
+		if (units & (1<<i)) {
+			if (dev_priv->chipset < 0xa0)
+				base = 0x408000 + (i<<12);
+			else
+				base = 0x408000 + (i<<11);
+			if (dev_priv->chipset < 0xa0)
+				offset = base + 0xc00;
+			else
+				offset = base + 0x80;
+			cp_ctx(ctx, offset + 0x00, 1);
+			gr_def(ctx, offset + 0x00, 0x0000ff0a);
+			cp_ctx(ctx, offset + 0x08, 1);
+
+			/* per-MP state */
+			for (j = 0; j < (dev_priv->chipset < 0xa0 ? 2 : 4); j++) {
+				if (!(units & (1 << (j+24)))) continue;
+				if (dev_priv->chipset < 0xa0)
+					offset = base + 0x200 + (j<<7);
+				else
+					offset = base + 0x100 + (j<<7);
+				cp_ctx(ctx, offset, 0x20);
+				gr_def(ctx, offset + 0x00, 0x01800000);
+				gr_def(ctx, offset + 0x04, 0x00160000);
+				gr_def(ctx, offset + 0x08, 0x01800000);
+				gr_def(ctx, offset + 0x18, 0x0003ffff);
+				switch (dev_priv->chipset) {
+				case 0x50:
+					gr_def(ctx, offset + 0x1c, 0x00080000);
+					break;
+				case 0x84:
+					gr_def(ctx, offset + 0x1c, 0x00880000);
+					break;
+				case 0x86:
+					gr_def(ctx, offset + 0x1c, 0x008c0000);
+					break;
+				case 0x92:
+				case 0x96:
+				case 0x98:
+					gr_def(ctx, offset + 0x1c, 0x118c0000);
+					break;
+				case 0x94:
+					gr_def(ctx, offset + 0x1c, 0x10880000);
+					break;
+				case 0xa0:
+				case 0xa5:
+					gr_def(ctx, offset + 0x1c, 0x310c0000);
+					break;
+				case 0xa8:
+				case 0xaa:
+				case 0xac:
+					gr_def(ctx, offset + 0x1c, 0x300c0000);
+					break;
+				}
+				gr_def(ctx, offset + 0x40, 0x00010401);
+				if (dev_priv->chipset == 0x50)
+					gr_def(ctx, offset + 0x48, 0x00000040);
+				else
+					gr_def(ctx, offset + 0x48, 0x00000078);
+				gr_def(ctx, offset + 0x50, 0x000000bf);
+				gr_def(ctx, offset + 0x58, 0x00001210);
+				if (dev_priv->chipset == 0x50)
+					gr_def(ctx, offset + 0x5c, 0x00000080);
+				else
+					gr_def(ctx, offset + 0x5c, 0x08000080);
+				if (dev_priv->chipset >= 0xa0)
+					gr_def(ctx, offset + 0x68, 0x0000003e);
+			}
+
+			if (dev_priv->chipset < 0xa0)
+				cp_ctx(ctx, base + 0x300, 0x4);
+			else
+				cp_ctx(ctx, base + 0x300, 0x5);
+			if (dev_priv->chipset == 0x50)
+				gr_def(ctx, base + 0x304, 0x00007070);
+			else if (dev_priv->chipset < 0xa0)
+				gr_def(ctx, base + 0x304, 0x00027070);
+			else if (dev_priv->chipset <= 0xa0 || dev_priv->chipset >= 0xaa)
+				gr_def(ctx, base + 0x304, 0x01127070);
+			else
+				gr_def(ctx, base + 0x304, 0x05127070);
+
+			if (dev_priv->chipset < 0xa0)
+				cp_ctx(ctx, base + 0x318, 1);
+			else
+				cp_ctx(ctx, base + 0x320, 1);
+			if (dev_priv->chipset == 0x50)
+				gr_def(ctx, base + 0x318, 0x0003ffff);
+			else if (dev_priv->chipset < 0xa0)
+				gr_def(ctx, base + 0x318, 0x03ffffff);
+			else
+				gr_def(ctx, base + 0x320, 0x07ffffff);
+
+			if (dev_priv->chipset < 0xa0)
+				cp_ctx(ctx, base + 0x324, 5);
+			else
+				cp_ctx(ctx, base + 0x328, 4);
+
+			if (dev_priv->chipset < 0xa0) {
+				cp_ctx(ctx, base + 0x340, 9);
+				offset = base + 0x340;
+			} else if (dev_priv->chipset <= 0xa0 || dev_priv->chipset >= 0xaa) {
+				cp_ctx(ctx, base + 0x33c, 0xb);
+				offset = base + 0x344;
+			} else {
+				cp_ctx(ctx, base + 0x33c, 0xd);
+				offset = base + 0x344;
+			}
+			gr_def(ctx, offset + 0x0, 0x00120407);
+			gr_def(ctx, offset + 0x4, 0x05091507);
+			if (dev_priv->chipset == 0x84)
+				gr_def(ctx, offset + 0x8, 0x05100202);
+			else
+				gr_def(ctx, offset + 0x8, 0x05010202);
+			gr_def(ctx, offset + 0xc, 0x00030201);
+
+			cp_ctx(ctx, base + 0x400, 2);
+			gr_def