1 files changed, 205 insertions, 140 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c
index 6435f1257572..b5210b31c1b2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c
@@ -24,14 +24,70 @@
 #include "gf100.h"
 #include "ctxgf100.h"
 
-#include <subdev/secboot.h>
+#include <core/firmware.h>
+#include <subdev/acr.h>
+
+#include <nvfw/flcn.h>
 
 #include <nvif/class.h>
 
+int
+gm200_gr_nofw(struct gf100_gr *gr, int ver, const struct gf100_gr_fwif *fwif)
+{
+	nvkm_warn(&gr->base.engine.subdev, "firmware unavailable\n");
+	return -ENODEV;
+}
+
 /*******************************************************************************
  * PGRAPH engine/subdev functions
  ******************************************************************************/
 
+static void
+gm200_gr_acr_bld_patch(struct nvkm_acr *acr, u32 bld, s64 adjust)
+{
+	struct flcn_bl_dmem_desc_v1 hdr;
+	nvkm_robj(acr->wpr, bld, &hdr, sizeof(hdr));
+	hdr.code_dma_base = hdr.code_dma_base + adjust;
+	hdr.data_dma_base = hdr.data_dma_base + adjust;
+	nvkm_wobj(acr->wpr, bld, &hdr, sizeof(hdr));
+	flcn_bl_dmem_desc_v1_dump(&acr->subdev, &hdr);
+}
+
+static void
+gm200_gr_acr_bld_write(struct nvkm_acr *acr, u32 bld,
+		       struct nvkm_acr_lsfw *lsfw)
+{
+	const u64 base = lsfw->offset.img + lsfw->app_start_offset;
+	const u64 code = base + lsfw->app_resident_code_offset;
+	const u64 data = base + lsfw->app_resident_data_offset;
+	const struct flcn_bl_dmem_desc_v1 hdr = {
+		.ctx_dma = FALCON_DMAIDX_UCODE,
+		.code_dma_base = code,
+		.non_sec_code_off = lsfw->app_resident_code_offset,
+		.non_sec_code_size = lsfw->app_resident_code_size,
+		.code_entry_point = lsfw->app_imem_entry,
+		.data_dma_base = data,
+		.data_size = lsfw->app_resident_data_size,
+	};
+
+	nvkm_wobj(acr->wpr, bld, &hdr, sizeof(hdr));
+}
+
+const struct nvkm_acr_lsf_func
+gm200_gr_gpccs_acr = {
+	.flags = NVKM_ACR_LSF_FORCE_PRIV_LOAD,
+	.bld_size = sizeof(struct flcn_bl_dmem_desc_v1),
+	.bld_write = gm200_gr_acr_bld_write,
+	.bld_patch = gm200_gr_acr_bld_patch,
+};
+
+const struct nvkm_acr_lsf_func
+gm200_gr_fecs_acr = {
+	.bld_size = sizeof(struct flcn_bl_dmem_desc_v1),
+	.bld_write = gm200_gr_acr_bld_write,
+	.bld_patch = gm200_gr_acr_bld_patch,
+};
+
 int
 gm200_gr_rops(struct gf100_gr *gr)
 {
@@ -39,6 +95,22 @@ gm200_gr_rops(struct gf100_gr *gr)
 }
 
 void
+gm200_gr_init_ds_hww_esr_2(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	nvkm_wr32(device, 0x405848, 0xc0000000);
+	nvkm_mask(device, 0x40584c, 0x00000001, 0x00000001);
+}
+
+void
+gm200_gr_init_num_active_ltcs(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	nvkm_wr32(device, GPC_BCAST(0x08ac), nvkm_rd32(device, 0x100800));
+	nvkm_wr32(device, GPC_BCAST(0x033c), nvkm_rd32(device, 0x100804));
+}
+
+void
 gm200_gr_init_gpc_mmu(struct gf100_gr *gr)
 {
 	struct nvkm_device *device = gr->base.engine.subdev.device;
@@ -61,160 +133,82 @@ gm200_gr_init_rop_active_fbps(struct gf100_gr *gr)
 	nvkm_mask(device, 0x408958, 0x0000000f, fbp_count); /* crop */
 }
 
-int
-gm200_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);
-
-	gm107_gr_init_bios(gr);
-
-	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, 0x000e0000);
-	nvkm_wr32(device, 0x405848, 0xc0000000);
-	nvkm_wr32(device, 0x40584c, 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);
-
-	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), 0x00000005);
-		}
-		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);
-
-	nvkm_wr32(device, 0x400054, 0x2c350f63);
+static u8
+gm200_gr_tile_map_6_24[] = {
+	0, 1, 2, 3, 4, 5, 3, 4, 5, 0, 1, 2, 0, 1, 2, 3, 4, 5, 3, 4, 5, 0, 1, 2,
+};
 
-	gf100_gr_zbc_init(gr);
+static u8
+gm200_gr_tile_map_4_16[] = {
+	0, 1, 2, 3, 2, 3, 0, 1, 3, 0, 1, 2, 1, 2, 3, 0,
+};
 
-	return gf100_gr_init_ctxctl(gr);
-}
+static u8
+gm200_gr_tile_map_2_8[] = {
+	0, 1, 1, 0, 0, 1, 1, 0,
+};
 
 int
-gm200_gr_new_(const struct gf100_gr_func *func, struct nvkm_device *device,
-	      int index, struct nvkm_gr **pgr)
+gm200_gr_oneinit_sm_id(struct gf100_gr *gr)
 {
-	struct gf100_gr *gr;
-	int ret;
-
-	if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL)))
-		return -ENOMEM;
-	*pgr = &gr->base;
-
-	ret = gf100_gr_ctor(func, device, index, gr);
-	if (ret)
-		return ret;
+	/*XXX: There's a different algorithm here I've not yet figured out. */
+	return gf100_gr_oneinit_sm_id(gr);
+}
 
-	/* Load firmwares for non-secure falcons */
-	if (!nvkm_secboot_is_managed(device->secboot,
-				     NVKM_SECBOOT_FALCON_FECS)) {
-		if ((ret = gf100_gr_ctor_fw(gr, "gr/fecs_inst", &gr->fuc409c)) ||
-		    (ret = gf100_gr_ctor_fw(gr, "gr/fecs_data", &gr->fuc409d)))
-			return ret;
-	}
-	if (!nvkm_secboot_is_managed(device->secboot,
-				     NVKM_SECBOOT_FALCON_GPCCS)) {
-		if ((ret = gf100_gr_ctor_fw(gr, "gr/gpccs_inst", &gr->fuc41ac)) ||
-		    (ret = gf100_gr_ctor_fw(gr, "gr/gpccs_data", &gr->fuc41ad)))
-			return ret;
+void
+gm200_gr_oneinit_tiles(struct gf100_gr *gr)
+{
+	/*XXX: Not sure what this is about.  The algorithm from NVGPU
+	 *     seems to work for all boards I tried from earlier (and
+	 *     later) GPUs except in these specific configurations.
+	 *
+	 *     Let's just hardcode them for now.
+	 */
+	if (gr->gpc_nr == 2 && gr->tpc_total == 8) {
+		memcpy(gr->tile, gm200_gr_tile_map_2_8, gr->tpc_total);
+		gr->screen_tile_row_offset = 1;
+	} else
+	if (gr->gpc_nr == 4 && gr->tpc_total == 16) {
+		memcpy(gr->tile, gm200_gr_tile_map_4_16, gr->tpc_total);
+		gr->screen_tile_row_offset = 4;
+	} else
+	if (gr->gpc_nr == 6 && gr->tpc_total == 24) {
+		memcpy(gr->tile, gm200_gr_tile_map_6_24, gr->tpc_total);
+		gr->screen_tile_row_offset = 5;
+	} else {
+		gf100_gr_oneinit_tiles(gr);
 	}
-
-	if ((ret = gk20a_gr_av_to_init(gr, "gr/sw_nonctx", &gr->fuc_sw_nonctx)) ||
-	    (ret = gk20a_gr_aiv_to_init(gr, "gr/sw_ctx", &gr->fuc_sw_ctx)) ||
-	    (ret = gk20a_gr_av_to_init(gr, "gr/sw_bundle_init", &gr->fuc_bundle)) ||
-	    (ret = gk20a_gr_av_to_method(gr, "gr/sw_method_init", &gr->fuc_method)))
-		return ret;
-
-	return 0;
 }
 
 static const struct gf100_gr_func
 gm200_gr = {
-	.init = gm200_gr_init,
+	.oneinit_tiles = gm200_gr_oneinit_tiles,
+	.oneinit_sm_id = gm200_gr_oneinit_sm_id,
+	.init = gf100_gr_init,
 	.init_gpc_mmu = gm200_gr_init_gpc_mmu,
+	.init_bios = gm107_gr_init_bios,
+	.init_vsc_stream_master = gk104_gr_init_vsc_stream_master,
+	.init_zcull = gf117_gr_init_zcull,
+	.init_num_active_ltcs = gm200_gr_init_num_active_ltcs,
 	.init_rop_active_fbps = gm200_gr_init_rop_active_fbps,
+	.init_fecs_exceptions = gf100_gr_init_fecs_exceptions,
+	.init_ds_hww_esr_2 = gm200_gr_init_ds_hww_esr_2,
+	.init_sked_hww_esr = gk104_gr_init_sked_hww_esr,
+	.init_419cc0 = gf100_gr_init_419cc0,
 	.init_ppc_exceptions = gk104_gr_init_ppc_exceptions,
+	.init_tex_hww_esr = gf100_gr_init_tex_hww_esr,
+	.init_504430 = gm107_gr_init_504430,
+	.init_shader_exceptions = gm107_gr_init_shader_exceptions,
+	.init_rop_exceptions = gf100_gr_init_rop_exceptions,
+	.init_exception2 = gf100_gr_init_exception2,
+	.init_400054 = gm107_gr_init_400054,
+	.trap_mp = gf100_gr_trap_mp,
+	.fecs.reset = gf100_gr_fecs_reset,
 	.rops = gm200_gr_rops,
+	.tpc_nr = 4,
 	.ppc_nr = 2,
 	.grctx = &gm200_grctx,
+	.zbc = &gf100_gr_zbc,
 	.sclass = {
 		{ -1, -1, FERMI_TWOD_A },
 		{ -1, -1, KEPLER_INLINE_TO_MEMORY_B },
@@ -225,7 +219,78 @@ gm200_gr = {
 };
 
 int
-gm200_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
+gm200_gr_load(struct gf100_gr *gr, int ver, const struct gf100_gr_fwif *fwif)
+{
+	int ret;
+
+	ret = nvkm_acr_lsfw_load_bl_inst_data_sig(&gr->base.engine.subdev,
+						  &gr->fecs.falcon,
+						  NVKM_ACR_LSF_FECS,
+						  "gr/fecs_", ver, fwif->fecs);
+	if (ret)
+		return ret;
+
+	ret = nvkm_acr_lsfw_load_bl_inst_data_sig(&gr->base.engine.subdev,
+						  &gr->gpccs.falcon,
+						  NVKM_ACR_LSF_GPCCS,
+						  "gr/gpccs_", ver,
+						  fwif->gpccs);
+	if (ret)
+		return ret;
+
+	gr->firmware = true;
+
+	return gk20a_gr_load_sw(gr, "gr/", ver);
+}
+
+MODULE_FIRMWARE("nvidia/gm200/gr/fecs_bl.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/fecs_inst.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/fecs_data.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/fecs_sig.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/gpccs_bl.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/gpccs_inst.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/gpccs_data.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/gpccs_sig.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/sw_ctx.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/sw_nonctx.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/sw_bundle_init.bin");
+MODULE_FIRMWARE("nvidia/gm200/gr/sw_method_init.bin");
+
+MODULE_FIRMWARE("nvidia/gm204/gr/fecs_bl.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/fecs_inst.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/fecs_data.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/fecs_sig.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/gpccs_bl.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/gpccs_inst.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/gpccs_data.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/gpccs_sig.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/sw_ctx.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/sw_nonctx.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/sw_bundle_init.bin");
+MODULE_FIRMWARE("nvidia/gm204/gr/sw_method_init.bin");
+
+MODULE_FIRMWARE("nvidia/gm206/gr/fecs_bl.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/fecs_inst.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/fecs_data.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/fecs_sig.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/gpccs_bl.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/gpccs_inst.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/gpccs_data.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/gpccs_sig.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/sw_ctx.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/sw_nonctx.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/sw_bundle_init.bin");
+MODULE_FIRMWARE("nvidia/gm206/gr/sw_method_init.bin");
+
+static const struct gf100_gr_fwif
+gm200_gr_fwif[] = {
+	{  0, gm200_gr_load, &gm200_gr, &gm200_gr_fecs_acr, &gm200_gr_gpccs_acr },
+	{ -1, gm200_gr_nofw },
+	{}
+};
+
+int
+gm200_gr_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
 {
-	return gm200_gr_new_(&gm200_gr, device, index, pgr);
+	return gf100_gr_new_(gm200_gr_fwif, device, type, inst, pgr);
 }