drm/nouveau/gr/r535: initial support

Adds support for allocating GR classes from RM.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230918202149.4343-41-skeggsb@gmail.com

9 files changed, 611 insertions, 2 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
index bd4b8cf40d4c..8cd57e0e7333 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
@@ -2762,6 +2762,7 @@ nv192_chipset = {
 	.disp     = { 0x00000001, ad102_disp_new },
 	.dma      = { 0x00000001, gv100_dma_new },
 	.fifo     = { 0x00000001, ga102_fifo_new },
+	.gr       = { 0x00000001, ad102_gr_new },
 	.sec2     = { 0x00000001, ga102_sec2_new },
 };
 
@@ -2783,6 +2784,7 @@ nv193_chipset = {
 	.disp     = { 0x00000001, ad102_disp_new },
 	.dma      = { 0x00000001, gv100_dma_new },
 	.fifo     = { 0x00000001, ga102_fifo_new },
+	.gr       = { 0x00000001, ad102_gr_new },
 	.sec2     = { 0x00000001, ga102_sec2_new },
 };
 
@@ -2804,6 +2806,7 @@ nv194_chipset = {
 	.disp     = { 0x00000001, ad102_disp_new },
 	.dma      = { 0x00000001, gv100_dma_new },
 	.fifo     = { 0x00000001, ga102_fifo_new },
+	.gr       = { 0x00000001, ad102_gr_new },
 	.sec2     = { 0x00000001, ga102_sec2_new },
 };
 
@@ -2825,6 +2828,7 @@ nv196_chipset = {
 	.disp     = { 0x00000001, ad102_disp_new },
 	.dma      = { 0x00000001, gv100_dma_new },
 	.fifo     = { 0x00000001, ga102_fifo_new },
+	.gr       = { 0x00000001, ad102_gr_new },
 	.sec2     = { 0x00000001, ga102_sec2_new },
 };
 
@@ -2846,6 +2850,7 @@ nv197_chipset = {
 	.disp     = { 0x00000001, ad102_disp_new },
 	.dma      = { 0x00000001, gv100_dma_new },
 	.fifo     = { 0x00000001, ga102_fifo_new },
+	.gr       = { 0x00000001, ad102_gr_new },
 	.sec2     = { 0x00000001, ga102_sec2_new },
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/r535.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/r535.c
index 102e5a586e01..82303e4c5c2e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/r535.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/r535.c
@@ -29,6 +29,7 @@
 #include <subdev/gsp.h>
 #include <subdev/mmu.h>
 #include <subdev/vfn.h>
+#include <engine/gr.h>
 
 #include <nvhw/drf.h>
 
@@ -67,6 +68,8 @@ r535_chan_ramfc_clear(struct nvkm_chan *chan)
 
 	dma_free_coherent(fifo->engine.subdev.device->dev, fifo->rm.mthdbuf_size,
 			  chan->rm.mthdbuf.ptr, chan->rm.mthdbuf.addr);
+
+	nvkm_cgrp_vctx_put(chan->cgrp, &chan->rm.grctx);
 }
 
 #define CHID_PER_USERD 8
@@ -76,12 +79,19 @@ r535_chan_ramfc_write(struct nvkm_chan *chan, u64 offset, u64 length, u32 devm,
 {
 	struct nvkm_fifo *fifo = chan->cgrp->runl->fifo;
 	struct nvkm_engn *engn;
+	struct nvkm_device *device = fifo->engine.subdev.device;
 	NV_CHANNELGPFIFO_ALLOCATION_PARAMETERS *args;
 	const int userd_p = chan->id / CHID_PER_USERD;
 	const int userd_i = chan->id % CHID_PER_USERD;
 	u32 eT = ~0;
 	int ret;
 
+	if (unlikely(device->gr && !device->gr->engine.subdev.oneinit)) {
+		ret = nvkm_subdev_oneinit(&device->gr->engine.subdev);
+		if (ret)
+			return ret;
+	}
+
 	nvkm_runl_foreach_engn(engn, chan->cgrp->runl) {
 		eT = engn->id;
 		break;
@@ -347,6 +357,27 @@ r535_ce = {
 	.nonstall = r535_engn_nonstall,
 };
 
+static int
+r535_gr_ctor(struct nvkm_engn *engn, struct nvkm_vctx *vctx, struct nvkm_chan *chan)
+{
+	/* RM requires GR context buffers to remain mapped until after the
+	 * channel has been destroyed (as opposed to after the last gr obj
+	 * has been deleted).
+	 *
+	 * Take an extra ref here, which will be released once the channel
+	 * object has been deleted.
+	 */
+	refcount_inc(&vctx->refs);
+	chan->rm.grctx = vctx;
+	return 0;
+}
+
+static const struct nvkm_engn_func
+r535_gr = {
+	.nonstall = r535_engn_nonstall,
+	.ctor2 = r535_gr_ctor,
+};
+
 static void
 r535_runl_allow(struct nvkm_runl *runl, u32 engm)
 {
@@ -477,6 +508,9 @@ r535_fifo_runl_ctor(struct nvkm_fifo *fifo)
 		case NVKM_ENGINE_CE:
 			engn = nvkm_runl_add(runl, nv2080, &r535_ce, type, inst);
 			break;
+		case NVKM_ENGINE_GR:
+			engn = nvkm_runl_add(runl, nv2080, &r535_gr, type, inst);
+			break;
 		case NVKM_ENGINE_SW:
 			continue;
 		default:
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild
index b5418f05ccd8..1555f8c40b4f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild
@@ -41,6 +41,9 @@ nvkm-y += nvkm/engine/gr/gp10b.o
 nvkm-y += nvkm/engine/gr/gv100.o
 nvkm-y += nvkm/engine/gr/tu102.o
 nvkm-y += nvkm/engine/gr/ga102.o
+nvkm-y += nvkm/engine/gr/ad102.o
+
+nvkm-y += nvkm/engine/gr/r535.o
 
 nvkm-y += nvkm/engine/gr/ctxnv40.o
 nvkm-y += nvkm/engine/gr/ctxnv50.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ad102.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ad102.c
new file mode 100644
index 000000000000..7bfa6240d283
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ad102.c
@@ -0,0 +1,46 @@
+/*
+ * Copyright 2023 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.
+ */
+#include "gf100.h"
+
+#include <subdev/gsp.h>
+
+#include <nvif/class.h>
+
+static const struct gf100_gr_func
+ad102_gr = {
+	.sclass = {
+		{ -1, -1, FERMI_TWOD_A },
+		{ -1, -1, KEPLER_INLINE_TO_MEMORY_B },
+		{ -1, -1, ADA_A },
+		{ -1, -1, ADA_COMPUTE_A },
+		{}
+	}
+};
+
+int
+ad102_gr_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
+{
+	if (nvkm_gsp_rm(device->gsp))
+		return r535_gr_new(&ad102_gr, device, type, inst, pgr);
+
+	return -ENODEV;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ga102.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ga102.c
index 2b51f1d0c281..d285c597aff9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ga102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ga102.c
@@ -352,7 +352,7 @@ int
 ga102_gr_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
 {
 	if (nvkm_gsp_rm(device->gsp))
-		return -ENODEV;
+		return r535_gr_new(&ga102_gr, device, type, inst, pgr);
 
 	return gf100_gr_new_(ga102_gr_fwif, device, type, inst, pgr);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
index 54f686ba39ac..b0e0c9305034 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
@@ -445,4 +445,6 @@ void gp108_gr_acr_bld_patch(struct nvkm_acr *, u32, s64);
 
 int gf100_gr_new_(const struct gf100_gr_fwif *, struct nvkm_device *, enum nvkm_subdev_type, int,
 		  struct nvkm_gr **);
+int r535_gr_new(const struct gf100_gr_func *, struct nvkm_device *, enum nvkm_subdev_type, int,
+		struct nvkm_gr **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/r535.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/r535.c
new file mode 100644
index 000000000000..a7501b52137c
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/r535.c
@@ -0,0 +1,508 @@
+/*
+ * Copyright 2023 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.
+ */
+#include "gf100.h"
+
+#include <core/memory.h>
+#include <subdev/gsp.h>
+#include <subdev/mmu/vmm.h>
+#include <engine/fifo/priv.h>
+
+#include <nvif/if900d.h>
+
+#include <nvhw/drf.h>
+
+#include <nvrm/nvtypes.h>
+#include <nvrm/535.54.03/common/sdk/nvidia/inc/alloc/alloc_channel.h>
+#include <nvrm/535.54.03/common/sdk/nvidia/inc/ctrl/ctrl0080/ctrl0080fifo.h>
+#include <nvrm/535.54.03/common/sdk/nvidia/inc/ctrl/ctrl2080/ctrl2080gpu.h>
+#include <nvrm/535.54.03/common/sdk/nvidia/inc/ctrl/ctrl2080/ctrl2080internal.h>
+#include <nvrm/535.54.03/nvidia/generated/g_kernel_channel_nvoc.h>
+
+#define r535_gr(p) container_of((p), struct r535_gr, base)
+
+#define R515_GR_MAX_CTXBUFS 9
+
+struct r535_gr {
+	struct nvkm_gr base;
+
+	struct {
+		u16 bufferId;
+		u32 size;
+		u8  page;
+		u8  align;
+		bool global;
+		bool init;
+		bool ro;
+	} ctxbuf[R515_GR_MAX_CTXBUFS];
+	int ctxbuf_nr;
+
+	struct nvkm_memory *ctxbuf_mem[R515_GR_MAX_CTXBUFS];
+};
+
+struct r535_gr_chan {
+	struct nvkm_object object;
+	struct r535_gr *gr;
+
+	struct nvkm_vmm *vmm;
+	struct nvkm_chan *chan;
+
+	struct nvkm_memory *mem[R515_GR_MAX_CTXBUFS];
+	struct nvkm_vma    *vma[R515_GR_MAX_CTXBUFS];
+};
+
+struct r535_gr_obj {
+	struct nvkm_object object;
+	struct nvkm_gsp_object rm;
+};
+
+static void *
+r535_gr_obj_dtor(struct nvkm_object *object)
+{
+	struct r535_gr_obj *obj = container_of(object, typeof(*obj), object);
+
+	nvkm_gsp_rm_free(&obj->rm);
+	return obj;
+}
+
+static const struct nvkm_object_func
+r535_gr_obj = {
+	.dtor = r535_gr_obj_dtor,
+};
+
+static int
+r535_gr_obj_ctor(const struct nvkm_oclass *oclass, void *argv, u32 argc,
+		 struct nvkm_object **pobject)
+{
+	struct r535_gr_chan *chan = container_of(oclass->parent, typeof(*chan), object);
+	struct r535_gr_obj *obj;
+
+	if (!(obj = kzalloc(sizeof(*obj), GFP_KERNEL)))
+		return -ENOMEM;
+
+	nvkm_object_ctor(&r535_gr_obj, oclass, &obj->object);
+	*pobject = &obj->object;
+
+	return nvkm_gsp_rm_alloc(&chan->chan->rm.object, oclass->handle, oclass->base.oclass, 0,
+				 &obj->rm);
+}
+
+static void *
+r535_gr_chan_dtor(struct nvkm_object *object)
+{
+	struct r535_gr_chan *grc = container_of(object, typeof(*grc), object);
+	struct r535_gr *gr = grc->gr;
+
+	for (int i = 0; i < gr->ctxbuf_nr; i++) {
+		nvkm_vmm_put(grc->vmm, &grc->vma[i]);
+		nvkm_memory_unref(&grc->mem[i]);
+	}
+
+	nvkm_vmm_unref(&grc->vmm);
+	return grc;
+}
+
+static const struct nvkm_object_func
+r535_gr_chan = {
+	.dtor = r535_gr_chan_dtor,
+};
+
+static int
+r535_gr_promote_ctx(struct r535_gr *gr, bool golden, struct nvkm_vmm *vmm,
+		    struct nvkm_memory **pmem, struct nvkm_vma **pvma,
+		    struct nvkm_gsp_object *chan)
+{
+	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	NV2080_CTRL_GPU_PROMOTE_CTX_PARAMS *ctrl;
+
+	ctrl = nvkm_gsp_rm_ctrl_get(&vmm->rm.device.subdevice,
+				    NV2080_CTRL_CMD_GPU_PROMOTE_CTX, sizeof(*ctrl));
+	if (WARN_ON(IS_ERR(ctrl)))
+		return PTR_ERR(ctrl);
+
+	ctrl->engineType = 1;
+	ctrl->hChanClient = vmm->rm.client.object.handle;
+	ctrl->hObject = chan->handle;
+
+	for (int i = 0; i < gr->ctxbuf_nr; i++) {
+		NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ENTRY *entry =
+			&ctrl->promoteEntry[ctrl->entryCount];
+		const bool alloc = golden || !gr->ctxbuf[i].global;
+		int ret;
+
+		entry->bufferId = gr->ctxbuf[i].bufferId;
+		entry->bInitialize = gr->ctxbuf[i].init && alloc;
+
+		if (alloc) {
+			ret = nvkm_memory_new(device, gr->ctxbuf[i].init ?
+					      NVKM_MEM_TARGET_INST : NVKM_MEM_TARGET_INST_SR_LOST,
+					      gr->ctxbuf[i].size, 1 << gr->ctxbuf[i].page,
+					      gr->ctxbuf[i].init, &pmem[i]);
+			if (WARN_ON(ret))
+				return ret;
+
+			if (gr->ctxbuf[i].bufferId ==
+					NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_PRIV_ACCESS_MAP)
+				entry->bNonmapped = 1;
+		} else {
+			if (gr->ctxbuf[i].bufferId ==
+				NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_UNRESTRICTED_PRIV_ACCESS_MAP)
+				continue;
+
+			pmem[i] = nvkm_memory_ref(gr->ctxbuf_mem[i]);
+		}
+
+		if (!entry->bNonmapped) {
+			struct gf100_vmm_map_v0 args = {
+				.priv = 1,
+				.ro   = gr->ctxbuf[i].ro,
+			};
+
+			mutex_lock(&vmm->mutex.vmm);
+			ret = nvkm_vmm_get_locked(vmm, false, true, false, 0, gr->ctxbuf[i].align,
+						  nvkm_memory_size(pmem[i]), &pvma[i]);
+			mutex_unlock(&vmm->mutex.vmm);
+			if (ret)
+				return ret;
+
+			ret = nvkm_memory_map(pmem[i], 0, vmm, pvma[i], &args, sizeof(args));
+			if (ret)
+				return ret;
+
+			entry->gpuVirtAddr = pvma[i]->addr;
+		}
+
+		if (entry->bInitialize) {
+			entry->gpuPhysAddr = nvkm_memory_addr(pmem[i]);
+			entry->size = gr->ctxbuf[i].size;
+			entry->physAttr = 4;
+		}
+
+		nvkm_debug(subdev,
+			   "promote %02d: pa %016llx/%08x sz %016llx va %016llx init:%d nm:%d\n",
+			   entry->bufferId, entry->gpuPhysAddr, entry->physAttr, entry->size,
+			   entry->gpuVirtAddr, entry->bInitialize, entry->bNonmapped);
+
+		ctrl->entryCount++;
+	}
+
+	return nvkm_gsp_rm_ctrl_wr(&vmm->rm.device.subdevice, ctrl);
+}
+
+static int
+r535_gr_chan_new(struct nvkm_gr *base, struct nvkm_chan *chan, const struct nvkm_oclass *oclass,
+		 struct nvkm_object **pobject)
+{
+	struct r535_gr *gr = r535_gr(base);
+	struct r535_gr_chan *grc;
+	int ret;
+
+	if (!(grc = kzalloc(sizeof(*grc), GFP_KERNEL)))
+		return -ENOMEM;
+
+	nvkm_object_ctor(&r535_gr_chan, oclass, &grc->object);
+	grc->gr = gr;
+	grc->vmm = nvkm_vmm_ref(chan->vmm);
+	grc->chan = chan;
+	*pobject = &grc->object;
+
+	ret = r535_gr_promote_ctx(gr, false, grc->vmm, grc->mem, grc->vma, &chan->rm.object);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static u64
+r535_gr_units(struct nvkm_gr *gr)
+{
+	struct nvkm_gsp *gsp = gr->engine.subdev.device->gsp;
+
+	return (gsp->gr.tpcs << 8) | gsp->gr.gpcs;
+}
+
+static int
+r535_gr_oneinit(struct nvkm_gr *base)
+{
+	NV2080_CTRL_INTERNAL_STATIC_GR_GET_CONTEXT_BUFFERS_INFO_PARAMS *info;
+	struct r535_gr *gr = container_of(base, typeof(*gr), base);
+	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	struct nvkm_gsp *gsp = device->gsp;
+	struct nvkm_mmu *mmu = device->mmu;
+	struct {
+		struct nvkm_memory *inst;
+		struct nvkm_vmm *vmm;
+		struct nvkm_gsp_object chan;
+		struct nvkm_vma *vma[R515_GR_MAX_CTXBUFS];
+	} golden = {};
+	int ret;
+
+	/* Allocate a channel to use for golden context init. */
+	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST, 0x12000, 0, true, &golden.inst);
+	if (ret)
+		goto done;
+
+	ret = nvkm_vmm_new(device, 0x1000, 0, NULL, 0, NULL, "grGoldenVmm", &golden.vmm);
+	if (ret)
+		goto done;
+
+	ret = mmu->func->promote_vmm(golden.vmm);
+	if (ret)
+		goto done;
+
+	{
+		NV_CHANNELGPFIFO_ALLOCATION_PARAMETERS *args;
+
+		args = nvkm_gsp_rm_alloc_get(&golden.vmm->rm.device.object, 0xf1f00000,
+					     device->fifo->func->chan.user.oclass,
+					     sizeof(*args), &golden.chan);
+		if (IS_ERR(args)) {
+			ret = PTR_ERR(args);
+			goto done;
+		}
+
+		args->gpFifoOffset = 0;
+		args->gpFifoEntries = 0x1000 / 8;
+		args->flags =
+			NVDEF(NVOS04, FLAGS, CHANNEL_TYPE, PHYSICAL) |
+			NVDEF(NVOS04, FLAGS, VPR, FALSE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_SKIP_MAP_REFCOUNTING, FALSE) |
+			NVVAL(NVOS04, FLAGS, GROUP_CHANNEL_RUNQUEUE, 0) |
+			NVDEF(NVOS04, FLAGS, PRIVILEGED_CHANNEL, TRUE) |
+			NVDEF(NVOS04, FLAGS, DELAY_CHANNEL_SCHEDULING, FALSE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_DENY_PHYSICAL_MODE_CE, FALSE) |
+			NVVAL(NVOS04, FLAGS, CHANNEL_USERD_INDEX_VALUE, 0) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_USERD_INDEX_FIXED, FALSE) |
+			NVVAL(NVOS04, FLAGS, CHANNEL_USERD_INDEX_PAGE_VALUE, 0) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_USERD_INDEX_PAGE_FIXED, TRUE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_DENY_AUTH_LEVEL_PRIV, FALSE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_SKIP_SCRUBBER, FALSE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_CLIENT_MAP_FIFO, FALSE) |
+			NVDEF(NVOS04, FLAGS, SET_EVICT_LAST_CE_PREFETCH_CHANNEL, FALSE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_VGPU_PLUGIN_CONTEXT, FALSE) |
+			NVDEF(NVOS04, FLAGS, CHANNEL_PBDMA_ACQUIRE_TIMEOUT, FALSE) |
+			NVDEF(NVOS04, FLAGS, GROUP_CHANNEL_THREAD, DEFAULT) |
+			NVDEF(NVOS04, FLAGS, MAP_CHANNEL, FALSE) |
+			NVDEF(NVOS04, FLAGS, SKIP_CTXBUFFER_ALLOC, FALSE);
+		args->hVASpace = golden.vmm->rm.object.handle;
+		args->engineType = 1;
+		args->instanceMem.base = nvkm_memory_addr(golden.inst);
+		args->instanceMem.size = 0x1000;
+		args->instanceMem.addressSpace = 2;
+		args->instanceMem.cacheAttrib = 1;
+		args->ramfcMem.base = nvkm_memory_addr(golden.inst);
+		args->ramfcMem.size = 0x200;
+		args->ramfcMem.addressSpace = 2;
+		args->ramfcMem.cacheAttrib = 1;
+		args->userdMem.base = nvkm_memory_addr(golden.inst) + 0x1000;
+		args->userdMem.size = 0x200;
+		args->userdMem.addressSpace = 2;
+		args->userdMem.cacheAttrib = 1;
+		args->mthdbufMem.base = nvkm_memory_addr(golden.inst) + 0x2000;
+		args->mthdbufMem.size = 0x5000;
+		args->mthdbufMem.addressSpace = 2;
+		args->mthdbufMem.cacheAttrib = 1;
+		args->internalFlags =
+			NVDEF(NV_KERNELCHANNEL, ALLOC_INTERNALFLAGS, PRIVILEGE, ADMIN) |
+			NVDEF(NV_KERNELCHANNEL, ALLOC_INTERNALFLAGS, ERROR_NOTIFIER_TYPE, NONE) |
+			NVDEF(NV_KERNELCHANNEL, ALLOC_INTERNALFLAGS, ECC_ERROR_NOTIFIER_TYPE, NONE);
+
+		ret = nvkm_gsp_rm_alloc_wr(&golden.chan, args);
+		if (ret)
+			goto done;
+	}
+
+	/* Fetch context buffer info from RM and allocate each of them here to use
+	 * during golden context init (or later as a global context buffer).
+	 *
+	 * Also build the information that'll be used to create channel contexts.
+	 */
+	info = nvkm_gsp_rm_ctrl_rd(&gsp->internal.device.subdevice,
+				   NV2080_CTRL_CMD_INTERNAL_STATIC_KGR_GET_CONTEXT_BUFFERS_INFO,
+				   sizeof(*info));
+	if (WARN_ON(IS_ERR(info))) {
+		ret = PTR_ERR(info);
+		goto done;
+	}
+
+	for (int i = 0; i < ARRAY_SIZE(info->engineContextBuffersInfo[0].engine); i++) {
+		static const struct {
+			u32     id0; /* NV0080_CTRL_FIFO_GET_ENGINE_CONTEXT_PROPERTIES_ENGINE_ID */
+			u32     id1; /* NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID */
+			bool global;
+			bool   init;
+			bool     ro;
+		} map[] = {
+#define _A(n,N,G,I,R) { .id0 = NV0080_CTRL_FIFO_GET_ENGINE_CONTEXT_PROPERTIES_ENGINE_ID_##n, \
+			.id1 = NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_##N, \
+			.global = (G), .init = (I), .ro = (R) }
+#define _B(N,G,I,R) _A(GRAPHICS_##N, N, (G), (I), (R))
+			/*                                       global   init     ro */
+			_A(           GRAPHICS,             MAIN, false,  true, false),
+			_B(                                PATCH, false,  true, false),
+			_A( GRAPHICS_BUNDLE_CB, BUFFER_BUNDLE_CB,  true, false, false),
+			_B(                             PAGEPOOL,  true, false, false),
+			_B(                         ATTRIBUTE_CB,  true, false, false),
+			_B(                        RTV_CB_GLOBAL,  true, false, false),
+			_B(                           FECS_EVENT,  true,  true, false),
+			_B(                      PRIV_ACCESS_MAP,  true,  true,  true),
+#undef _B
+#undef _A
+		};
+		u32 size = info->engineContextBuffersInfo[0].engine[i].size;
+		u8 align, page;
+		int id;
+
+		for (id = 0; id < ARRAY_SIZE(map); id++) {
+			if (map[id].id0 == i)
+				break;
+		}
+
+		nvkm_debug(subdev, "%02x: size:0x%08x %s\n", i,
+			   size, (id < ARRAY_SIZE(map)) ? "*" : "");
+		if (id >= ARRAY_SIZE(map))
+			continue;
+
+		if (map[id].id1 == NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_MAIN)
+			size = ALIGN(size, 0x1000) + 64 * 0x1000; /* per-subctx headers */
+
+		if      (size >= 1 << 21) page = 21;
+		else if (size >= 1 << 16) page = 16;
+		else			  page = 12;
+
+		if (map[id].id1 == NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_ATTRIBUTE_CB)
+			align = order_base_2(size);
+		else
+			align = page;
+
+		if (WARN_ON(gr->ctxbuf_nr == ARRAY_SIZE(gr->ctxbuf)))
+			continue;
+
+		gr->ctxbuf[gr->ctxbuf_nr].bufferId = map[id].id1;
+		gr->ctxbuf[gr->ctxbuf_nr].size     = size;
+		gr->ctxbuf[gr->ctxbuf_nr].page     = page;
+		gr->ctxbuf[gr->ctxbuf_nr].align    = align;
+		gr->ctxbuf[gr->ctxbuf_nr].global   = map[id].global;
+		gr->ctxbuf[gr->ctxbuf_nr].init     = map[id].init;
+		gr->ctxbuf[gr->ctxbuf_nr].ro       = map[id].ro;
+		gr->ctxbuf_nr++;
+
+		if (map[id].id1 == NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_PRIV_ACCESS_MAP) {
+			if (WARN_ON(gr->ctxbuf_nr == ARRAY_SIZE(gr->ctxbuf)))
+				continue;
+
+			gr->ctxbuf[gr->ctxbuf_nr] = gr->ctxbuf[gr->ctxbuf_nr - 1];
+			gr->ctxbuf[gr->ctxbuf_nr].bufferId =
+				NV2080_CTRL_GPU_PROMOTE_CTX_BUFFER_ID_UNRESTRICTED_PRIV_ACCESS_MAP;
+			gr->ctxbuf_nr++;
+		}
+	}
+
+	nvkm_gsp_rm_ctrl_done(&gsp->internal.device.subdevice, info);
+
+	/* Promote golden context to RM. */
+	ret = r535_gr_promote_ctx(gr, true, golden.vmm, gr->ctxbuf_mem, golden.vma, &golden.chan);
+	if (ret)
+		goto done;
+
+	/* Allocate 3D class on channel to trigger golden context init in RM. */
+	{
+		int i;
+
+		for (i = 0; gr->base.func->sclass[i].ctor; i++) {
+			if ((gr->base.func->sclass[i].oclass & 0xff) == 0x97) {
+				struct nvkm_gsp_object threed;
+
+				ret = nvkm_gsp_rm_alloc(&golden.chan, 0x97000000,
+							gr->base.func->sclass[i].oclass, 0,
+							&threed);
+				if (ret)
+					goto done;
+
+				nvkm_gsp_rm_free(&threed);
+				break;
+			}
+		}
+
+		if (WARN_ON(!gr->base.func->sclass[i].ctor)) {
+			ret = -EINVAL;
+			goto done;
+		}
+	}
+
+done:
+	nvkm_gsp_rm_free(&golden.chan);
+	for (int i = gr->ctxbuf_nr - 1; i >= 0; i--)
+		nvkm_vmm_put(golden.vmm, &golden.vma[i]);
+	nvkm_vmm_unref(&golden.vmm);
+	nvkm_memory_unref(&golden.inst);
+	return ret;
+
+}
+
+static void *
+r535_gr_dtor(struct nvkm_gr *base)
+{
+	struct r535_gr *gr = r535_gr(base);
+
+	while (gr->ctxbuf_nr)
+		nvkm_memory_unref(&gr->ctxbuf_mem[--gr->ctxbuf_nr]);
+
+	kfree(gr->base.func);
+	return gr;
+}
+
+int
+r535_gr_new(const struct gf100_gr_func *hw,
+	    struct nvkm_device *device, enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
+{
+	struct nvkm_gr_func *rm;
+	struct r535_gr *gr;
+	int nclass;
+
+	for (nclass = 0; hw->sclass[nclass].oclass; nclass++);
+
+	if (!(rm = kzalloc(sizeof(*rm) + (nclass + 1) * sizeof(rm->sclass[0]), GFP_KERNEL)))
+		return -ENOMEM;
+
+	rm->dtor = r535_gr_dtor;
+	rm->oneinit = r535_gr_oneinit;
+	rm->units = r535_gr_units;
+	rm->chan_new = r535_gr_chan_new;
+
+	for (int i = 0; i < nclass; i++) {
+		rm->sclass[i].minver = hw->sclass[i].minver;
+		rm->sclass[i].maxver = hw->sclass[i].maxver;
+		rm->sclass[i].oclass = hw->sclass[i].oclass;
+		rm->sclass[i].ctor = r535_gr_obj_ctor;
+	}
+
+	if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL))) {
+		kfree(rm);
+		return -ENOMEM;
+	}
+
+	*pgr = &gr->base;
+
+	return nvkm_gr_ctor(rm, device, type, inst, true, &gr->base);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/tu102.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/tu102.c
index bda8054c6b59..b7a458e9040a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/tu102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/tu102.c
@@ -219,7 +219,7 @@ int
 tu102_gr_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst, struct nvkm_gr **pgr)
 {
 	if (nvkm_gsp_rm(device->gsp))
-		return -ENODEV;
+		return r535_gr_new(&tu102_gr, device, type, inst, pgr);
 
 	return gf100_gr_new_(tu102_gr_fwif, device, type, inst, pgr);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/r535.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/r535.c
index f36eed2fb3b7..71b90adb4382 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/r535.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/r535.c
@@ -855,6 +855,10 @@ r535_gsp_intr_get_table(struct nvkm_gsp *gsp)
 			type = NVKM_ENGINE_CE;
 			inst = ctrl->table[i].engineIdx - MC_ENGINE_IDX_CE0;
 			break;
+		case MC_ENGINE_IDX_GR0:
+			type = NVKM_ENGINE_GR;
+			inst = 0;
+			break;
 		default:
 			continue;
 		}
@@ -930,6 +934,13 @@ r535_gsp_rpc_get_gsp_static_info(struct nvkm_gsp *gsp)
 		gsp->fb.rsvd_size = gsp->fb.heap.addr - rsvd_base;
 	}
 
+	for (int gpc = 0; gpc < ARRAY_SIZE(rpc->tpcInfo); gpc++) {
+		if (rpc->gpcInfo.gpcMask & BIT(gpc)) {
+			gsp->gr.tpcs += hweight32(rpc->tpcInfo[gpc].tpcMask);
+			gsp->gr.gpcs++;
+		}
+	}
+
 	nvkm_gsp_rpc_done(gsp, rpc);
 	return 0;
 }