drm/nouveau/gr/gp100: initial support

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

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

@@ -108,6 +108,8 @@
 #define MAXWELL_A                                     /* cl9097.h */ 0x0000b097
 #define MAXWELL_B                                     /* cl9097.h */ 0x0000b197
 
+#define PASCAL_A                                      /* cl9097.h */ 0x0000c097
+
 #define NV74_BSP                                                     0x000074b0
 
 #define GT212_MSVLD                                                  0x000085b1
@@ -141,6 +143,7 @@
 #define KEPLER_COMPUTE_B                                             0x0000a1c0
 #define MAXWELL_COMPUTE_A                                            0x0000b0c0
 #define MAXWELL_COMPUTE_B                                            0x0000b1c0
+#define PASCAL_COMPUTE_A                                             0x0000c0c0
 
 #define NV74_CIPHER                                                  0x000074c1
 #endif

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

@@ -42,4 +42,5 @@ int gk20a_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 int gm107_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 int gm200_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 int gm20b_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
+int gp100_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 #endif

drivers/gpu/drm/nouveau/nvkm/engine/device/base.c

@@ -2177,6 +2177,7 @@ nv130_chipset = {
 	.dma = gf119_dma_new,
 	.disp = gp100_disp_new,
 	.fifo = gp100_fifo_new,
+	.gr = gp100_gr_new,
 };
 
 static int

drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild

@@ -31,6 +31,7 @@ nvkm-y += nvkm/engine/gr/gk20a.o
 nvkm-y += nvkm/engine/gr/gm107.o
 nvkm-y += nvkm/engine/gr/gm200.o
 nvkm-y += nvkm/engine/gr/gm20b.o
+nvkm-y += nvkm/engine/gr/gp100.o
 
 nvkm-y += nvkm/engine/gr/ctxnv40.o
 nvkm-y += nvkm/engine/gr/ctxnv50.o
@@ -48,3 +49,4 @@ nvkm-y += nvkm/engine/gr/ctxgk20a.o
 nvkm-y += nvkm/engine/gr/ctxgm107.o
 nvkm-y += nvkm/engine/gr/ctxgm200.o
 nvkm-y += nvkm/engine/gr/ctxgm20b.o
+nvkm-y += nvkm/engine/gr/ctxgp100.o

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

@@ -101,6 +101,8 @@ void gm200_grctx_generate_405b60(struct gf100_gr *);
 
 extern const struct gf100_grctx_func gm20b_grctx;
 
+extern const struct gf100_grctx_func gp100_grctx;
+
 /* context init value lists */
 
 extern const struct gf100_gr_pack gf100_grctx_pack_icmd[];

drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgp100.c

+/*
+ * Copyright 2016 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 <bskeggs@redhat.com>
+ */
+#include "ctxgf100.h"
+
+#include <subdev/fb.h>
+
+/*******************************************************************************
+ * PGRAPH context implementation
+ ******************************************************************************/
+
+static void
+gp100_grctx_generate_pagepool(struct gf100_grctx *info)
+{
+	const struct gf100_grctx_func *grctx = info->gr->func->grctx;
+	const u32 access = NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS;
+	const int s = 8;
+	const int b = mmio_vram(info, grctx->pagepool_size, (1 << s), access);
+	mmio_refn(info, 0x40800c, 0x00000000, s, b);
+	mmio_wr32(info, 0x408010, 0x80000000);
+	mmio_refn(info, 0x419004, 0x00000000, s, b);
+	mmio_wr32(info, 0x419008, 0x00000000);
+}
+
+static void
+gp100_grctx_generate_attrib(struct gf100_grctx *info)
+{
+	struct gf100_gr *gr = info->gr;
+	const struct gf100_grctx_func *grctx = gr->func->grctx;
+	const u32  alpha = grctx->alpha_nr;
+	const u32 attrib = grctx->attrib_nr;
+	const u32 pertpc = 0x20 * (grctx->attrib_nr_max + grctx->alpha_nr_max);
+	const u32   size = roundup(gr->tpc_total * pertpc, 0x80);
+	const u32 access = NV_MEM_ACCESS_RW;
+	const int s = 12;
+	const int b = mmio_vram(info, size, (1 << s), access);
+	const int max_batches = 0xffff;
+	u32 ao = 0;
+	u32 bo = ao + grctx->alpha_nr_max * gr->tpc_total;
+	int gpc, ppc, n = 0;
+
+	mmio_refn(info, 0x418810, 0x80000000, s, b);
+	mmio_refn(info, 0x419848, 0x10000000, s, b);
+	mmio_refn(info, 0x419c2c, 0x10000000, s, b);
+	mmio_refn(info, 0x419b00, 0x00000000, s, b);
+	mmio_wr32(info, 0x419b04, 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 u = 0x418ea0 + (n * 0x04);
+			const u32 o = PPC_UNIT(gpc, ppc, 0);
+			if (!(gr->ppc_mask[gpc] & (1 << ppc)))
+				continue;
+			mmio_wr32(info, o + 0xc0, bs);
+			mmio_wr32(info, o + 0xf4, bo);
+			mmio_wr32(info, o + 0xf0, bs);
+			bo += grctx->attrib_nr_max * gr->ppc_tpc_nr[gpc][ppc];
+			mmio_wr32(info, o + 0xe4, as);
+			mmio_wr32(info, o + 0xf8, ao);
+			ao += grctx->alpha_nr_max * gr->ppc_tpc_nr[gpc][ppc];
+			mmio_wr32(info, u, bs);
+		}
+	}
+
+	mmio_wr32(info, 0x418eec, 0x00000000);
+	mmio_wr32(info, 0x41befc, 0x00000000);
+}
+
+static void
+gp100_grctx_generate_405b60(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	const u32 dist_nr = DIV_ROUND_UP(gr->tpc_total, 4);
+	u32 dist[TPC_MAX / 4] = {};
+	u32 gpcs[GPC_MAX * 2] = {};
+	u8  tpcnr[GPC_MAX];
+	int tpc, gpc, i;
+
+	memcpy(tpcnr, gr->tpc_nr, sizeof(gr->tpc_nr));
+
+	/* won't result in the same distribution as the binary driver where
+	 * some of the gpcs have more tpcs than others, but this shall do
+	 * for the moment.  the code for earlier gpus has this issue too.
+	 */
+	for (gpc = -1, i = 0; i < gr->tpc_total; i++) {
+		do {
+			gpc = (gpc + 1) % gr->gpc_nr;
+		} while(!tpcnr[gpc]);
+		tpc = gr->tpc_nr[gpc] - tpcnr[gpc]--;
+
+		dist[i / 4] |= ((gpc << 4) | tpc) << ((i % 4) * 8);
+		gpcs[gpc + (gr->gpc_nr * (tpc / 4))] |= i << (tpc * 8);
+	}
+
+	for (i = 0; i < dist_nr; i++)
+		nvkm_wr32(device, 0x405b60 + (i * 4), dist[i]);
+	for (i = 0; i < gr->gpc_nr * 2; i++)
+		nvkm_wr32(device, 0x405ba0 + (i * 4), gpcs[i]);
+}
+
+static void
+gp100_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	const struct gf100_grctx_func *grctx = gr->func->grctx;
+	u32 idle_timeout, tmp;
+	int i;
+
+	gf100_gr_mmio(gr, gr->fuc_sw_ctx);
+
+	idle_timeout = nvkm_mask(device, 0x404154, 0xffffffff, 0x00000000);
+
+	grctx->pagepool(info);
+	grctx->bundle(info);
+	grctx->attrib(info);
+	grctx->unkn(gr);
+
+	gm200_grctx_generate_tpcid(gr);
+	gf100_grctx_generate_r406028(gr);
+	gk104_grctx_generate_r418bb8(gr);
+
+	for (i = 0; i < 8; i++)
+		nvkm_wr32(device, 0x4064d0 + (i * 0x04), 0x00000000);
+	nvkm_wr32(device, 0x406500, 0x00000000);
+
+	nvkm_wr32(device, 0x405b00, (gr->tpc_total << 8) | gr->gpc_nr);
+
+	for (tmp = 0, i = 0; i < gr->gpc_nr; i++)
+		tmp |= ((1 << gr->tpc_nr[i]) - 1) << (i * 5);
+	nvkm_wr32(device, 0x4041c4, tmp);
+
+	gp100_grctx_generate_405b60(gr);
+
+	gf100_gr_icmd(gr, gr->fuc_bundle);
+	nvkm_wr32(device, 0x404154, idle_timeout);
+	gf100_gr_mthd(gr, gr->fuc_method);
+}
+
+const struct gf100_grctx_func
+gp100_grctx = {
+	.main  = gp100_grctx_generate_main,
+	.unkn  = gk104_grctx_generate_unkn,
+	.bundle = gm107_grctx_generate_bundle,
+	.bundle_size = 0x3000,
+	.bundle_min_gpm_fifo_depth = 0x180,
+	.bundle_token_limit = 0x1080,
+	.pagepool = gp100_grctx_generate_pagepool,
+	.pagepool_size = 0x20000,
+	.attrib = gp100_grctx_generate_attrib,
+	.attrib_nr_max = 0x660,
+	.attrib_nr = 0x440,
+	.alpha_nr_max = 0xc00,
+	.alpha_nr = 0x800,
+};

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

@@ -292,4 +292,6 @@ extern const struct gf100_gr_init gm107_gr_init_l1c_0[];
 extern const struct gf100_gr_init gm107_gr_init_wwdx_0[];
 extern const struct gf100_gr_init gm107_gr_init_cbm_0[];
 void gm107_gr_init_bios(struct gf100_gr *);
+
+void gm200_gr_init_gpc_mmu(struct gf100_gr *);
 #endif

drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c

@@ -38,7 +38,7 @@ gm200_gr_rops(struct gf100_gr *gr)
 	return nvkm_rd32(gr->base.engine.subdev.device, 0x12006c);
 }
 
-static void
+void
 gm200_gr_init_gpc_mmu(struct gf100_gr *gr)
 {
 	struct nvkm_device *device = gr->base.engine.subdev.device;

drivers/gpu/drm/nouveau/nvkm/engine/gr/gp100.c

+/*
+ * Copyright 2016 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 <bskeggs@redhat.com>
+ */
+#include "gf100.h"
+#include "ctxgf100.h"
+
+#include <nvif/class.h>
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static void
+gp100_gr_init_rop_active_fbps(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	/*XXX: otherwise identical to gm200 aside from mask.. do everywhere? */
+	const u32 fbp_count = nvkm_rd32(device, 0x12006c) & 0x0000000f;
+	nvkm_mask(device, 0x408850, 0x0000000f, fbp_count); /* zrop */
+	nvkm_mask(device, 0x408958, 0x0000000f, fbp_count); /* crop */
+}
+
+static int
+gp100_gr_init(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, gr->tpc_total);
+	u32 data[TPC_MAX / 8] = {};
+	u8  tpcnr[GPC_MAX];
+	int gpc, tpc, rop;
+	int i;
+
+	gr->func->init_gpc_mmu(gr);
+
+	gf100_gr_mmio(gr, gr->fuc_sw_nonctx);
+
+	nvkm_wr32(device, GPC_UNIT(0, 0x3018), 0x00000001);
+
+	memset(data, 0x00, sizeof(data));
+	memcpy(tpcnr, gr->tpc_nr, sizeof(gr->tpc_nr));
+	for (i = 0, gpc = -1; i < gr->tpc_total; i++) {
+		do {
+			gpc = (gpc + 1) % gr->gpc_nr;
+		} while (!tpcnr[gpc]);
+		tpc = gr->tpc_nr[gpc] - tpcnr[gpc]--;
+
+		data[i / 8] |= tpc << ((i % 8) * 4);
+	}
+
+	nvkm_wr32(device, GPC_BCAST(0x0980), data[0]);
+	nvkm_wr32(device, GPC_BCAST(0x0984), data[1]);
+	nvkm_wr32(device, GPC_BCAST(0x0988), data[2]);
+	nvkm_wr32(device, GPC_BCAST(0x098c), data[3]);
+
+	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0914),
+			  gr->screen_tile_row_offset << 8 | gr->tpc_nr[gpc]);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0910), 0x00040000 |
+							 gr->tpc_total);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0918), magicgpc918);
+	}
+
+	nvkm_wr32(device, GPC_BCAST(0x3fd4), magicgpc918);
+	nvkm_wr32(device, GPC_BCAST(0x08ac), nvkm_rd32(device, 0x100800));
+	nvkm_wr32(device, GPC_BCAST(0x033c), nvkm_rd32(device, 0x100804));
+
+	gr->func->init_rop_active_fbps(gr);
+
+	nvkm_wr32(device, 0x400500, 0x00010001);
+	nvkm_wr32(device, 0x400100, 0xffffffff);
+	nvkm_wr32(device, 0x40013c, 0xffffffff);
+	nvkm_wr32(device, 0x400124, 0x00000002);
+	nvkm_wr32(device, 0x409c24, 0x000f0002);
+	nvkm_wr32(device, 0x405848, 0xc0000000);
+	nvkm_mask(device, 0x40584c, 0x00000000, 0x00000001);
+	nvkm_wr32(device, 0x404000, 0xc0000000);
+	nvkm_wr32(device, 0x404600, 0xc0000000);
+	nvkm_wr32(device, 0x408030, 0xc0000000);
+	nvkm_wr32(device, 0x404490, 0xc0000000);
+	nvkm_wr32(device, 0x406018, 0xc0000000);
+	nvkm_wr32(device, 0x407020, 0x40000000);
+	nvkm_wr32(device, 0x405840, 0xc0000000);
+	nvkm_wr32(device, 0x405844, 0x00ffffff);
+	nvkm_mask(device, 0x419cc0, 0x00000008, 0x00000008);
+
+	nvkm_mask(device, 0x419c9c, 0x00010000, 0x00010000);
+	nvkm_mask(device, 0x419c9c, 0x00020000, 0x00020000);
+
+	gr->func->init_ppc_exceptions(gr);
+
+	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
+		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x430), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x644), 0x00dffffe);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x64c), 0x00000105);
+		}
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
+	}
+
+	for (rop = 0; rop < gr->rop_nr; rop++) {
+		nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0x40000000);
+		nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0x40000000);
+		nvkm_wr32(device, ROP_UNIT(rop, 0x204), 0xffffffff);
+		nvkm_wr32(device, ROP_UNIT(rop, 0x208), 0xffffffff);
+	}
+
+	nvkm_wr32(device, 0x400108, 0xffffffff);
+	nvkm_wr32(device, 0x400138, 0xffffffff);
+	nvkm_wr32(device, 0x400118, 0xffffffff);
+	nvkm_wr32(device, 0x400130, 0xffffffff);
+	nvkm_wr32(device, 0x40011c, 0xffffffff);
+	nvkm_wr32(device, 0x400134, 0xffffffff);
+
+	gf100_gr_zbc_init(gr);
+
+	return gf100_gr_init_ctxctl(gr);
+}
+
+static const struct gf100_gr_func
+gp100_gr = {
+	.init = gp100_gr_init,
+	.init_gpc_mmu = gm200_gr_init_gpc_mmu,
+	.init_rop_active_fbps = gp100_gr_init_rop_active_fbps,
+	.init_ppc_exceptions = gk104_gr_init_ppc_exceptions,
+	.rops = gm200_gr_rops,
+	.ppc_nr = 2,
+	.grctx = &gp100_grctx,
+	.sclass = {
+		{ -1, -1, FERMI_TWOD_A },
+		{ -1, -1, KEPLER_INLINE_TO_MEMORY_B },
+		{ -1, -1, PASCAL_A, &gf100_fermi },
+		{ -1, -1, PASCAL_COMPUTE_A },
+		{}
+	}
+};
+
+int
+gp100_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
+{
+	return gm200_gr_new_(&gp100_gr, device, index, pgr);
+}