drm/nouveau/gr/gv100-: modify gr init to match newer version of RM

Will be used as a basis for implementing changes needed for Turing.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c

@@ -1324,10 +1324,8 @@ gf100_grctx_generate_sm_id(struct gf100_gr *gr, int gpc, int tpc, int sm)
 void
 gf100_grctx_generate_floorsweep(struct gf100_gr *gr)
 {
-	struct nvkm_device *device = gr->base.engine.subdev.device;
 	const struct gf100_grctx_func *func = gr->func->grctx;
-	int gpc, sm, i, j;
-	u32 data;
+	int sm;
 
 	for (sm = 0; sm < gr->sm_nr; sm++) {
 		func->sm_id(gr, gr->sm[sm].gpc, gr->sm[sm].tpc, sm);
@@ -1335,12 +1333,8 @@ gf100_grctx_generate_floorsweep(struct gf100_gr *gr)
 			func->tpc_nr(gr, gr->sm[sm].gpc);
 	}
 
-	for (gpc = 0, i = 0; i < 4; i++) {
-		for (data = 0, j = 0; j < 8 && gpc < gr->gpc_nr; j++, gpc++)
-			data |= gr->tpc_nr[gpc] << (j * 4);
-		nvkm_wr32(device, 0x406028 + (i * 4), data);
-		nvkm_wr32(device, 0x405870 + (i * 4), data);
-	}
+	gf100_gr_init_num_tpc_per_gpc(gr, false, true);
+	gf100_gr_init_num_tpc_per_gpc(gr, true, false);
 
 	if (func->r4060a8)
 		func->r4060a8(gr);

drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgv100.c

@@ -67,14 +67,14 @@ gv100_grctx_generate_attrib(struct gf100_grctx *info)
 	const u32 attrib = grctx->attrib_nr;
 	const u32   gfxp = grctx->gfxp_nr;
 	const int s = 12;
-	const int max_batches = 0xffff;
 	u32 size = grctx->alpha_nr_max * gr->tpc_total;
 	u32 ao = 0;
 	u32 bo = ao + size;
 	int gpc, ppc, b, n = 0;
 
-	size += grctx->gfxp_nr * gr->tpc_total;
-	size = ((size * 0x20) + 128) & ~127;
+	for (gpc = 0; gpc < gr->gpc_nr; gpc++)
+		size += grctx->gfxp_nr * gr->ppc_nr[gpc] * gr->ppc_tpc_max;
+	size = ((size * 0x20) + 127) & ~127;
 	b = mmio_vram(info, size, (1 << s), false);
 
 	mmio_refn(info, 0x418810, 0x80000000, s, b);
@@ -84,13 +84,12 @@ gv100_grctx_generate_attrib(struct gf100_grctx *info)
 	mmio_wr32(info, 0x419e04, 0x80000000 | size >> 7);
 	mmio_wr32(info, 0x405830, attrib);
 	mmio_wr32(info, 0x40585c, alpha);
-	mmio_wr32(info, 0x4064c4, ((alpha / 4) << 16) | max_batches);
 
 	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
 		for (ppc = 0; ppc < gr->ppc_nr[gpc]; ppc++, n++) {
 			const u32 as =  alpha * gr->ppc_tpc_nr[gpc][ppc];
-			const u32 bs = attrib * gr->ppc_tpc_nr[gpc][ppc];
-			const u32 gs =   gfxp * gr->ppc_tpc_nr[gpc][ppc];
+			const u32 bs = attrib * gr->ppc_tpc_max;
+			const u32 gs =   gfxp * gr->ppc_tpc_max;
 			const u32 u = 0x418ea0 + (n * 0x04);
 			const u32 o = PPC_UNIT(gpc, ppc, 0);
 			if (!(gr->ppc_mask[gpc] & (1 << ppc)))

drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c

@@ -2169,6 +2169,23 @@ gf100_gr_new_(const struct gf100_gr_func *func, struct nvkm_device *device,
 	return 0;
 }
 
+void
+gf100_gr_init_num_tpc_per_gpc(struct gf100_gr *gr, bool pd, bool ds)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	int gpc, i, j;
+	u32 data;
+
+	for (gpc = 0, i = 0; i < 4; i++) {
+		for (data = 0, j = 0; j < 8 && gpc < gr->gpc_nr; j++, gpc++)
+			data |= gr->tpc_nr[gpc] << (j * 4);
+		if (pd)
+			nvkm_wr32(device, 0x406028 + (i * 4), data);
+		if (ds)
+			nvkm_wr32(device, 0x405870 + (i * 4), data);
+	}
+}
+
 void
 gf100_gr_init_400054(struct gf100_gr *gr)
 {
@@ -2336,8 +2353,8 @@ gf100_gr_init(struct gf100_gr *gr)
 	if (gr->func->init_40601c)
 		gr->func->init_40601c(gr);
 
-	nvkm_wr32(device, 0x404490, 0xc0000000);
 	nvkm_wr32(device, 0x406018, 0xc0000000);
+	nvkm_wr32(device, 0x404490, 0xc0000000);
 
 	if (gr->func->init_sked_hww_esr)
 		gr->func->init_sked_hww_esr(gr);

drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h

@@ -216,6 +216,7 @@ void gf100_gr_init_419eb4(struct gf100_gr *);
 void gf100_gr_init_tex_hww_esr(struct gf100_gr *, int, int);
 void gf100_gr_init_shader_exceptions(struct gf100_gr *, int, int);
 void gf100_gr_init_400054(struct gf100_gr *);
+void gf100_gr_init_num_tpc_per_gpc(struct gf100_gr *, bool, bool);
 extern const struct gf100_gr_func_zbc gf100_gr_zbc;
 
 void gf117_gr_init_zcull(struct gf100_gr *);