drm/gf100-/gr: add support for zero bandwidth clear

Default ZBC table is compatible with binary driver defaults.

Userspace will need to be updated to take full advantage of this
feature, however, some applications will see a performance boost
without updated drivers.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/core/engine/graph/gk20a.c

@@ -27,7 +27,7 @@ static struct nouveau_oclass
 gk20a_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa040, &nouveau_object_ofuncs },
-	{ 0xa297, &nouveau_object_ofuncs },
+	{ KEPLER_C, &nvc0_fermi_ofuncs },
 	{ 0xa0c0, &nouveau_object_ofuncs },
 	{}
 };

drivers/gpu/drm/nouveau/core/engine/graph/gm107.c

@@ -36,7 +36,7 @@ static struct nouveau_oclass
 gm107_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa140, &nouveau_object_ofuncs },
-	{ 0xb097, &nouveau_object_ofuncs },
+	{ MAXWELL_A, &nvc0_fermi_ofuncs },
 	{ 0xb0c0, &nouveau_object_ofuncs },
 	{}
 };
@@ -425,6 +425,9 @@ gm107_graph_init(struct nouveau_object *object)
 	nv_wr32(priv, 0x400134, 0xffffffff);
 
 	nv_wr32(priv, 0x400054, 0x2c350f63);
+
+	nvc0_graph_zbc_init(priv);
+
 	return nvc0_graph_init_ctxctl(priv);
 }
 

drivers/gpu/drm/nouveau/core/engine/graph/nv108.c

@@ -33,7 +33,7 @@ static struct nouveau_oclass
 nv108_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa140, &nouveau_object_ofuncs },
-	{ 0xa197, &nouveau_object_ofuncs },
+	{ KEPLER_B, &nvc0_fermi_ofuncs },
 	{ 0xa1c0, &nouveau_object_ofuncs },
 	{}
 };

drivers/gpu/drm/nouveau/core/engine/graph/nvc0.c

@@ -25,15 +25,200 @@
 #include "nvc0.h"
 #include "ctxnvc0.h"
 
+/*******************************************************************************
+ * Zero Bandwidth Clear
+ ******************************************************************************/
+
+static void
+nvc0_graph_zbc_clear_color(struct nvc0_graph_priv *priv, int zbc)
+{
+	if (priv->zbc_color[zbc].format) {
+		nv_wr32(priv, 0x405804, priv->zbc_color[zbc].ds[0]);
+		nv_wr32(priv, 0x405808, priv->zbc_color[zbc].ds[1]);
+		nv_wr32(priv, 0x40580c, priv->zbc_color[zbc].ds[2]);
+		nv_wr32(priv, 0x405810, priv->zbc_color[zbc].ds[3]);
+	}
+	nv_wr32(priv, 0x405814, priv->zbc_color[zbc].format);
+	nv_wr32(priv, 0x405820, zbc);
+	nv_wr32(priv, 0x405824, 0x00000004); /* TRIGGER | WRITE | COLOR */
+}
+
+static int
+nvc0_graph_zbc_color_get(struct nvc0_graph_priv *priv, int format,
+			 const u32 ds[4], const u32 l2[4])
+{
+	struct nouveau_ltc *ltc = nouveau_ltc(priv);
+	int zbc = -ENOSPC, i;
+
+	for (i = ltc->zbc_min; i <= ltc->zbc_max; i++) {
+		if (priv->zbc_color[i].format) {
+			if (priv->zbc_color[i].format != format)
+				continue;
+			if (memcmp(priv->zbc_color[i].ds, ds, sizeof(
+				   priv->zbc_color[i].ds)))
+				continue;
+			if (memcmp(priv->zbc_color[i].l2, l2, sizeof(
+				   priv->zbc_color[i].l2))) {
+				WARN_ON(1);
+				return -EINVAL;
+			}
+			return i;
+		} else {
+			zbc = (zbc < 0) ? i : zbc;
+		}
+	}
+
+	memcpy(priv->zbc_color[zbc].ds, ds, sizeof(priv->zbc_color[zbc].ds));
+	memcpy(priv->zbc_color[zbc].l2, l2, sizeof(priv->zbc_color[zbc].l2));
+	priv->zbc_color[zbc].format = format;
+	ltc->zbc_color_get(ltc, zbc, l2);
+	nvc0_graph_zbc_clear_color(priv, zbc);
+	return zbc;
+}
+
+static void
+nvc0_graph_zbc_clear_depth(struct nvc0_graph_priv *priv, int zbc)
+{
+	if (priv->zbc_depth[zbc].format)
+		nv_wr32(priv, 0x405818, priv->zbc_depth[zbc].ds);
+	nv_wr32(priv, 0x40581c, priv->zbc_depth[zbc].format);
+	nv_wr32(priv, 0x405820, zbc);
+	nv_wr32(priv, 0x405824, 0x00000005); /* TRIGGER | WRITE | DEPTH */
+}
+
+static int
+nvc0_graph_zbc_depth_get(struct nvc0_graph_priv *priv, int format,
+			 const u32 ds, const u32 l2)
+{
+	struct nouveau_ltc *ltc = nouveau_ltc(priv);
+	int zbc = -ENOSPC, i;
+
+	for (i = ltc->zbc_min; i <= ltc->zbc_max; i++) {
+		if (priv->zbc_depth[i].format) {
+			if (priv->zbc_depth[i].format != format)
+				continue;
+			if (priv->zbc_depth[i].ds != ds)
+				continue;
+			if (priv->zbc_depth[i].l2 != l2) {
+				WARN_ON(1);
+				return -EINVAL;
+			}
+			return i;
+		} else {
+			zbc = (zbc < 0) ? i : zbc;
+		}
+	}
+
+	priv->zbc_depth[zbc].format = format;
+	priv->zbc_depth[zbc].ds = ds;
+	priv->zbc_depth[zbc].l2 = l2;
+	ltc->zbc_depth_get(ltc, zbc, l2);
+	nvc0_graph_zbc_clear_depth(priv, zbc);
+	return zbc;
+}
+
 /*******************************************************************************
  * Graphics object classes
  ******************************************************************************/
 
+static int
+nvc0_fermi_mthd_zbc_color(struct nouveau_object *object, void *data, u32 size)
+{
+	struct nvc0_graph_priv *priv = (void *)object->engine;
+	union {
+		struct fermi_a_zbc_color_v0 v0;
+	} *args = data;
+	int ret;
+
+	if (nvif_unpack(args->v0, 0, 0, false)) {
+		switch (args->v0.format) {
+		case FERMI_A_ZBC_COLOR_V0_FMT_ZERO:
+		case FERMI_A_ZBC_COLOR_V0_FMT_UNORM_ONE:
+		case FERMI_A_ZBC_COLOR_V0_FMT_RF32_GF32_BF32_AF32:
+		case FERMI_A_ZBC_COLOR_V0_FMT_R16_G16_B16_A16:
+		case FERMI_A_ZBC_COLOR_V0_FMT_RN16_GN16_BN16_AN16:
+		case FERMI_A_ZBC_COLOR_V0_FMT_RS16_GS16_BS16_AS16:
+		case FERMI_A_ZBC_COLOR_V0_FMT_RU16_GU16_BU16_AU16:
+		case FERMI_A_ZBC_COLOR_V0_FMT_RF16_GF16_BF16_AF16:
+		case FERMI_A_ZBC_COLOR_V0_FMT_A8R8G8B8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_A8RL8GL8BL8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_A2B10G10R10:
+		case FERMI_A_ZBC_COLOR_V0_FMT_AU2BU10GU10RU10:
+		case FERMI_A_ZBC_COLOR_V0_FMT_A8B8G8R8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_A8BL8GL8RL8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_AN8BN8GN8RN8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_AS8BS8GS8RS8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_AU8BU8GU8RU8:
+		case FERMI_A_ZBC_COLOR_V0_FMT_A2R10G10B10:
+		case FERMI_A_ZBC_COLOR_V0_FMT_BF10GF11RF11:
+			ret = nvc0_graph_zbc_color_get(priv, args->v0.format,
+							     args->v0.ds,
+							     args->v0.l2);
+			if (ret >= 0) {
+				args->v0.index = ret;
+				return 0;
+			}
+			break;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	return ret;
+}
+
+static int
+nvc0_fermi_mthd_zbc_depth(struct nouveau_object *object, void *data, u32 size)
+{
+	struct nvc0_graph_priv *priv = (void *)object->engine;
+	union {
+		struct fermi_a_zbc_depth_v0 v0;
+	} *args = data;
+	int ret;
+
+	if (nvif_unpack(args->v0, 0, 0, false)) {
+		switch (args->v0.format) {
+		case FERMI_A_ZBC_DEPTH_V0_FMT_FP32:
+			ret = nvc0_graph_zbc_depth_get(priv, args->v0.format,
+							     args->v0.ds,
+							     args->v0.l2);
+			return (ret >= 0) ? 0 : -ENOSPC;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	return ret;
+}
+
+static int
+nvc0_fermi_mthd(struct nouveau_object *object, u32 mthd, void *data, u32 size)
+{
+	switch (mthd) {
+	case FERMI_A_ZBC_COLOR:
+		return nvc0_fermi_mthd_zbc_color(object, data, size);
+	case FERMI_A_ZBC_DEPTH:
+		return nvc0_fermi_mthd_zbc_depth(object, data, size);
+	default:
+		break;
+	}
+	return -EINVAL;
+}
+
+struct nouveau_ofuncs
+nvc0_fermi_ofuncs = {
+	.ctor = _nouveau_object_ctor,
+	.dtor = nouveau_object_destroy,
+	.init = nouveau_object_init,
+	.fini = nouveau_object_fini,
+	.mthd = nvc0_fermi_mthd,
+};
+
 struct nouveau_oclass
 nvc0_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0x9039, &nouveau_object_ofuncs },
-	{ 0x9097, &nouveau_object_ofuncs },
+	{ FERMI_A, &nvc0_fermi_ofuncs },
 	{ 0x90c0, &nouveau_object_ofuncs },
 	{}
 };
@@ -406,6 +591,35 @@ nvc0_graph_pack_mmio[] = {
  * PGRAPH engine/subdev functions
  ******************************************************************************/
 
+void
+nvc0_graph_zbc_init(struct nvc0_graph_priv *priv)
+{
+	const u32  zero[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
+	const u32   one[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
+			      0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+	const u32 f32_0[] = { 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+			      0x00000000, 0x00000000, 0x00000000, 0x00000000 };
+	const u32 f32_1[] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000,
+			      0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
+	struct nouveau_ltc *ltc = nouveau_ltc(priv);
+	int index;
+
+	if (!priv->zbc_color[0].format) {
+		nvc0_graph_zbc_color_get(priv, 1,  & zero[0],   &zero[4]);
+		nvc0_graph_zbc_color_get(priv, 2,  &  one[0],    &one[4]);
+		nvc0_graph_zbc_color_get(priv, 4,  &f32_0[0],  &f32_0[4]);
+		nvc0_graph_zbc_color_get(priv, 4,  &f32_1[0],  &f32_1[4]);
+		nvc0_graph_zbc_depth_get(priv, 1, 0x00000000, 0x00000000);
+		nvc0_graph_zbc_depth_get(priv, 1, 0x3f800000, 0x3f800000);
+	}
+
+	for (index = ltc->zbc_min; index <= ltc->zbc_max; index++)
+		nvc0_graph_zbc_clear_color(priv, index);
+	for (index = ltc->zbc_min; index <= ltc->zbc_max; index++)
+		nvc0_graph_zbc_clear_depth(priv, index);
+}
+
 void
 nvc0_graph_mmio(struct nvc0_graph_priv *priv, const struct nvc0_graph_pack *p)
 {
@@ -1223,6 +1437,9 @@ nvc0_graph_init(struct nouveau_object *object)
 	nv_wr32(priv, 0x400134, 0xffffffff);
 
 	nv_wr32(priv, 0x400054, 0x34ce3464);
+
+	nvc0_graph_zbc_init(priv);
+
 	return nvc0_graph_init_ctxctl(priv);
 }
 

drivers/gpu/drm/nouveau/core/engine/graph/nvc0.h

@@ -30,11 +30,15 @@
 #include <core/gpuobj.h>
 #include <core/option.h>
 
+#include <nvif/unpack.h>
+#include <nvif/class.h>
+
 #include <subdev/fb.h>
 #include <subdev/vm.h>
 #include <subdev/bar.h>
 #include <subdev/timer.h>
 #include <subdev/mc.h>
+#include <subdev/ltc.h>
 
 #include <engine/fifo.h>
 #include <engine/graph.h>
@@ -69,6 +73,18 @@ struct nvc0_graph_fuc {
 	u32  size;
 };
 
+struct nvc0_graph_zbc_color {
+	u32 format;
+	u32 ds[4];
+	u32 l2[4];
+};
+
+struct nvc0_graph_zbc_depth {
+	u32 format;
+	u32 ds;
+	u32 l2;
+};
+
 struct nvc0_graph_priv {
 	struct nouveau_graph base;
 
@@ -78,6 +94,9 @@ struct nvc0_graph_priv {
 	struct nvc0_graph_fuc fuc41ad;
 	bool firmware;
 
+	struct nvc0_graph_zbc_color zbc_color[NOUVEAU_LTC_MAX_ZBC_CNT];
+	struct nvc0_graph_zbc_depth zbc_depth[NOUVEAU_LTC_MAX_ZBC_CNT];
+
 	u8 rop_nr;
 	u8 gpc_nr;
 	u8 tpc_nr[GPC_MAX];
@@ -119,11 +138,15 @@ int  nvc0_graph_ctor(struct nouveau_object *, struct nouveau_object *,
 		     struct nouveau_object **);
 void nvc0_graph_dtor(struct nouveau_object *);
 int  nvc0_graph_init(struct nouveau_object *);
+void nvc0_graph_zbc_init(struct nvc0_graph_priv *);
+
 int  nve4_graph_fini(struct nouveau_object *, bool);
 int  nve4_graph_init(struct nouveau_object *);
 
 int  nvf0_graph_fini(struct nouveau_object *, bool);
 
+extern struct nouveau_ofuncs nvc0_fermi_ofuncs;
+
 extern struct nouveau_oclass nvc0_graph_sclass[];
 extern struct nouveau_oclass nvc8_graph_sclass[];
 extern struct nouveau_oclass nvf0_graph_sclass[];

drivers/gpu/drm/nouveau/core/engine/graph/nvc1.c

@@ -33,9 +33,9 @@ static struct nouveau_oclass
 nvc1_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0x9039, &nouveau_object_ofuncs },
-	{ 0x9097, &nouveau_object_ofuncs },
+	{ FERMI_A, &nvc0_fermi_ofuncs },
+	{ FERMI_B, &nvc0_fermi_ofuncs },
 	{ 0x90c0, &nouveau_object_ofuncs },
-	{ 0x9197, &nouveau_object_ofuncs },
 	{}
 };
 

drivers/gpu/drm/nouveau/core/engine/graph/nvc8.c

@@ -33,10 +33,10 @@ struct nouveau_oclass
 nvc8_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0x9039, &nouveau_object_ofuncs },
-	{ 0x9097, &nouveau_object_ofuncs },
+	{ FERMI_A, &nvc0_fermi_ofuncs },
+	{ FERMI_B, &nvc0_fermi_ofuncs },
+	{ FERMI_C, &nvc0_fermi_ofuncs },
 	{ 0x90c0, &nouveau_object_ofuncs },
-	{ 0x9197, &nouveau_object_ofuncs },
-	{ 0x9297, &nouveau_object_ofuncs },
 	{}
 };
 

drivers/gpu/drm/nouveau/core/engine/graph/nve4.c

@@ -35,7 +35,7 @@ static struct nouveau_oclass
 nve4_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa040, &nouveau_object_ofuncs },
-	{ 0xa097, &nouveau_object_ofuncs },
+	{ KEPLER_A, &nvc0_fermi_ofuncs },
 	{ 0xa0c0, &nouveau_object_ofuncs },
 	{}
 };
@@ -303,6 +303,9 @@ nve4_graph_init(struct nouveau_object *object)
 	nv_wr32(priv, 0x400134, 0xffffffff);
 
 	nv_wr32(priv, 0x400054, 0x34ce3464);
+
+	nvc0_graph_zbc_init(priv);
+
 	return nvc0_graph_init_ctxctl(priv);
 }
 

drivers/gpu/drm/nouveau/core/engine/graph/nvf0.c

@@ -33,7 +33,7 @@ struct nouveau_oclass
 nvf0_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa140, &nouveau_object_ofuncs },
-	{ 0xa197, &nouveau_object_ofuncs },
+	{ KEPLER_B, &nvc0_fermi_ofuncs },
 	{ 0xa1c0, &nouveau_object_ofuncs },
 	{}
 };

drivers/gpu/drm/nouveau/nvif/class.h

@@ -73,6 +73,16 @@
 #define GF110_DISP_OVERLAY_CONTROL_DMA                               0x0000907e
 #define GK104_DISP_OVERLAY_CONTROL_DMA                               0x0000917e
 
+#define FERMI_A                                                      0x00009097
+#define FERMI_B                                                      0x00009197
+#define FERMI_C                                                      0x00009297
+
+#define KEPLER_A                                                     0x0000a097
+#define KEPLER_B                                                     0x0000a197
+#define KEPLER_C                                                     0x0000a297
+
+#define MAXWELL_A                                                    0x0000b097
+
 
 /*******************************************************************************
  * client
@@ -491,4 +501,50 @@ struct nv50_disp_overlay_v0 {
 	__u8  pad02[6];
 };
 
+
+/*******************************************************************************
+ * fermi
+ ******************************************************************************/
+
+#define FERMI_A_ZBC_COLOR                                                  0x00
+#define FERMI_A_ZBC_DEPTH                                                  0x01
+
+struct fermi_a_zbc_color_v0 {
+	__u8  version;
+#define FERMI_A_ZBC_COLOR_V0_FMT_ZERO                                      0x01
+#define FERMI_A_ZBC_COLOR_V0_FMT_UNORM_ONE                                 0x02
+#define FERMI_A_ZBC_COLOR_V0_FMT_RF32_GF32_BF32_AF32                       0x04
+#define FERMI_A_ZBC_COLOR_V0_FMT_R16_G16_B16_A16                           0x08
+#define FERMI_A_ZBC_COLOR_V0_FMT_RN16_GN16_BN16_AN16                       0x0c
+#define FERMI_A_ZBC_COLOR_V0_FMT_RS16_GS16_BS16_AS16                       0x10
+#define FERMI_A_ZBC_COLOR_V0_FMT_RU16_GU16_BU16_AU16                       0x14
+#define FERMI_A_ZBC_COLOR_V0_FMT_RF16_GF16_BF16_AF16                       0x16
+#define FERMI_A_ZBC_COLOR_V0_FMT_A8R8G8B8                                  0x18
+#define FERMI_A_ZBC_COLOR_V0_FMT_A8RL8GL8BL8                               0x1c
+#define FERMI_A_ZBC_COLOR_V0_FMT_A2B10G10R10                               0x20
+#define FERMI_A_ZBC_COLOR_V0_FMT_AU2BU10GU10RU10                           0x24
+#define FERMI_A_ZBC_COLOR_V0_FMT_A8B8G8R8                                  0x28
+#define FERMI_A_ZBC_COLOR_V0_FMT_A8BL8GL8RL8                               0x2c
+#define FERMI_A_ZBC_COLOR_V0_FMT_AN8BN8GN8RN8                              0x30
+#define FERMI_A_ZBC_COLOR_V0_FMT_AS8BS8GS8RS8                              0x34
+#define FERMI_A_ZBC_COLOR_V0_FMT_AU8BU8GU8RU8                              0x38
+#define FERMI_A_ZBC_COLOR_V0_FMT_A2R10G10B10                               0x3c
+#define FERMI_A_ZBC_COLOR_V0_FMT_BF10GF11RF11                              0x40
+	__u8  format;
+	__u8  index;
+	__u8  pad03[5];
+	__u32 ds[4];
+	__u32 l2[4];
+};
+
+struct fermi_a_zbc_depth_v0 {
+	__u8  version;
+#define FERMI_A_ZBC_DEPTH_V0_FMT_FP32                                      0x01
+	__u8  format;
+	__u8  index;
+	__u8  pad03[5];
+	__u32 ds;
+	__u32 l2;
+};
+
 #endif