drm/gf100-/gr: implement the proper SetShaderExceptions method

We have another version of it implemented in SW, however, that version
isn't serialised with normal PGRAPH operation and can possibly clobber
the enables for another context.

This is the same method that's implemented by the NVIDIA binary driver.

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

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

@@ -27,8 +27,8 @@ static struct nouveau_oclass
 gk20a_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa040, &nouveau_object_ofuncs },
-	{ KEPLER_C, &nvc0_fermi_ofuncs },
-	{ 0xa0c0, &nouveau_object_ofuncs },
+	{ KEPLER_C, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ KEPLER_COMPUTE_A, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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

@@ -36,8 +36,8 @@ static struct nouveau_oclass
 gm107_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa140, &nouveau_object_ofuncs },
-	{ MAXWELL_A, &nvc0_fermi_ofuncs },
-	{ 0xb0c0, &nouveau_object_ofuncs },
+	{ MAXWELL_A, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ MAXWELL_COMPUTE_A, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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

@@ -214,12 +214,38 @@ nvc0_fermi_ofuncs = {
 	.mthd = nvc0_fermi_mthd,
 };
 
+static int
+nvc0_graph_set_shader_exceptions(struct nouveau_object *object, u32 mthd,
+				 void *pdata, u32 size)
+{
+	struct nvc0_graph_priv *priv = (void *)nv_engine(object);
+	if (size >= sizeof(u32)) {
+		u32 data = *(u32 *)pdata ? 0xffffffff : 0x00000000;
+		nv_wr32(priv, 0x419e44, data);
+		nv_wr32(priv, 0x419e4c, data);
+		return 0;
+	}
+	return -EINVAL;
+}
+
+struct nouveau_omthds
+nvc0_graph_9097_omthds[] = {
+	{ 0x1528, 0x1528, nvc0_graph_set_shader_exceptions },
+	{}
+};
+
+struct nouveau_omthds
+nvc0_graph_90c0_omthds[] = {
+	{ 0x1528, 0x1528, nvc0_graph_set_shader_exceptions },
+	{}
+};
+
 struct nouveau_oclass
 nvc0_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0x9039, &nouveau_object_ofuncs },
-	{ FERMI_A, &nvc0_fermi_ofuncs },
-	{ 0x90c0, &nouveau_object_ofuncs },
+	{ FERMI_A, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ FERMI_COMPUTE_A, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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

@@ -150,6 +150,8 @@ 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_omthds nvc0_graph_9097_omthds[];
+extern struct nouveau_omthds nvc0_graph_90c0_omthds[];
 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 },
-	{ FERMI_A, &nvc0_fermi_ofuncs },
-	{ FERMI_B, &nvc0_fermi_ofuncs },
-	{ 0x90c0, &nouveau_object_ofuncs },
+	{ FERMI_A, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ FERMI_B, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ FERMI_COMPUTE_A, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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 },
-	{ FERMI_A, &nvc0_fermi_ofuncs },
-	{ FERMI_B, &nvc0_fermi_ofuncs },
-	{ FERMI_C, &nvc0_fermi_ofuncs },
-	{ 0x90c0, &nouveau_object_ofuncs },
+	{ FERMI_A, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ FERMI_B, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ FERMI_C, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ FERMI_COMPUTE_A, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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

@@ -35,8 +35,8 @@ static struct nouveau_oclass
 nve4_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa040, &nouveau_object_ofuncs },
-	{ KEPLER_A, &nvc0_fermi_ofuncs },
-	{ 0xa0c0, &nouveau_object_ofuncs },
+	{ KEPLER_A, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ KEPLER_COMPUTE_A, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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

@@ -33,8 +33,8 @@ struct nouveau_oclass
 nvf0_graph_sclass[] = {
 	{ 0x902d, &nouveau_object_ofuncs },
 	{ 0xa140, &nouveau_object_ofuncs },
-	{ KEPLER_B, &nvc0_fermi_ofuncs },
-	{ 0xa1c0, &nouveau_object_ofuncs },
+	{ KEPLER_B, &nvc0_fermi_ofuncs, nvc0_graph_9097_omthds },
+	{ KEPLER_COMPUTE_B, &nouveau_object_ofuncs, nvc0_graph_90c0_omthds },
 	{}
 };
 

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

@@ -83,6 +83,14 @@
 
 #define MAXWELL_A                                                    0x0000b097
 
+#define FERMI_COMPUTE_A                                              0x000090c0
+#define FERMI_COMPUTE_B                                              0x000091c0
+
+#define KEPLER_COMPUTE_A                                             0x0000a0c0
+#define KEPLER_COMPUTE_B                                             0x0000a1c0
+
+#define MAXWELL_COMPUTE_A                                            0x0000b0c0
+
 
 /*******************************************************************************
  * client