7 files changed, 54 insertions, 351 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/Kbuild
index 553d540f2736..fa7a2862dd1f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/Kbuild
@@ -4,5 +4,4 @@ nvkm-y += nvkm/subdev/instmem/nv04.o
 nvkm-y += nvkm/subdev/instmem/nv40.o
 nvkm-y += nvkm/subdev/instmem/nv50.o
 nvkm-y += nvkm/subdev/instmem/gk20a.o
-
-nvkm-y += nvkm/subdev/instmem/r535.o
+nvkm-y += nvkm/subdev/instmem/gh100.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
index a2cd3330efc6..2f55bab8e132 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
@@ -182,9 +182,11 @@ nvkm_instmem_fini(struct nvkm_subdev *subdev, bool suspend)
 	int ret;
 
 	if (suspend) {
-		ret = imem->func->suspend(imem);
-		if (ret)
-			return ret;
+		if (imem->func->suspend) {
+			ret = imem->func->suspend(imem);
+			if (ret)
+				return ret;
+		}
 
 		imem->suspend = true;
 	}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gh100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gh100.c
new file mode 100644
index 000000000000..8d8dd5f8a6c7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/gh100.c
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
+ */
+#include "priv.h"
+
+#include <nvhw/ref/gh100/pri_nv_xal_ep.h>
+
+static void
+gh100_instmem_set_bar0_window_addr(struct nvkm_device *device, u64 addr)
+{
+	nvkm_wr32(device, NV_XAL_EP_BAR0_WINDOW, addr >> NV_XAL_EP_BAR0_WINDOW_BASE_SHIFT);
+}
+
+static const struct nvkm_instmem_func
+gh100_instmem = {
+	.fini = nv50_instmem_fini,
+	.memory_new = nv50_instobj_new,
+	.memory_wrap = nv50_instobj_wrap,
+	.set_bar0_window_addr = gh100_instmem_set_bar0_window_addr,
+};
+
+int
+gh100_instmem_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
+		  struct nvkm_instmem **pimem)
+{
+	return r535_instmem_new(&gh100_instmem, device, type, inst, pimem);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c
index 6b462f960922..2544b9f0ec85 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv40.c
@@ -239,7 +239,6 @@ nv40_instmem_new(struct nvkm_device *device, enum nvkm_subdev_type type, int ins
 		 struct nvkm_instmem **pimem)
 {
 	struct nv40_instmem *imem;
-	int bar;
 
 	if (!(imem = kzalloc(sizeof(*imem), GFP_KERNEL)))
 		return -ENOMEM;
@@ -247,13 +246,8 @@ nv40_instmem_new(struct nvkm_device *device, enum nvkm_subdev_type type, int ins
 	*pimem = &imem->base;
 
 	/* map bar */
-	if (device->func->resource_size(device, 2))
-		bar = 2;
-	else
-		bar = 3;
-
-	imem->iomem = ioremap_wc(device->func->resource_addr(device, bar),
-				 device->func->resource_size(device, bar));
+	imem->iomem = ioremap_wc(device->func->resource_addr(device, NVKM_BAR2_INST),
+				 device->func->resource_size(device, NVKM_BAR2_INST));
 	if (!imem->iomem) {
 		nvkm_error(&imem->base.subdev, "unable to map PRAMIN BAR\n");
 		return -EFAULT;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c
index dd5b5a17ece0..4ca6fb30743d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c
@@ -65,7 +65,7 @@ nv50_instobj_wr32_slow(struct nvkm_memory *memory, u64 offset, u32 data)
 
 	spin_lock_irqsave(&imem->base.lock, flags);
 	if (unlikely(imem->addr != base)) {
-		nvkm_wr32(device, 0x001700, base >> 16);
+		imem->base.func->set_bar0_window_addr(device, base);
 		imem->addr = base;
 	}
 	nvkm_wr32(device, 0x700000 + addr, data);
@@ -85,7 +85,7 @@ nv50_instobj_rd32_slow(struct nvkm_memory *memory, u64 offset)
 
 	spin_lock_irqsave(&imem->base.lock, flags);
 	if (unlikely(imem->addr != base)) {
-		nvkm_wr32(device, 0x001700, base >> 16);
+		imem->base.func->set_bar0_window_addr(device, base);
 		imem->addr = base;
 	}
 	data = nvkm_rd32(device, 0x700000 + addr);
@@ -172,7 +172,7 @@ nv50_instobj_kmap(struct nv50_instobj *iobj, struct nvkm_vmm *vmm)
 
 	/* Make the mapping visible to the host. */
 	iobj->bar = bar;
-	iobj->map = ioremap_wc(device->func->resource_addr(device, 3) +
+	iobj->map = ioremap_wc(device->func->resource_addr(device, NVKM_BAR2_INST) +
 			       (u32)iobj->bar->addr, size);
 	if (!iobj->map) {
 		nvkm_warn(subdev, "PRAMIN ioremap failed\n");
@@ -353,7 +353,7 @@ nv50_instobj_func = {
 	.map = nv50_instobj_map,
 };
 
-static int
+int
 nv50_instobj_wrap(struct nvkm_instmem *base,
 		  struct nvkm_memory *memory, struct nvkm_memory **pmemory)
 {
@@ -373,7 +373,7 @@ nv50_instobj_wrap(struct nvkm_instmem *base,
 	return 0;
 }
 
-static int
+int
 nv50_instobj_new(struct nvkm_instmem *imem, u32 size, u32 align, bool zero,
 		 struct nvkm_memory **pmemory)
 {
@@ -395,6 +395,12 @@ nv50_instobj_new(struct nvkm_instmem *imem, u32 size, u32 align, bool zero,
  *****************************************************************************/
 
 static void
+nv50_instmem_set_bar0_window_addr(struct nvkm_device *device, u64 addr)
+{
+	nvkm_wr32(device, 0x001700, addr >> 16);
+}
+
+void
 nv50_instmem_fini(struct nvkm_instmem *base)
 {
 	nv50_instmem(base)->addr = ~0ULL;
@@ -415,6 +421,7 @@ nv50_instmem = {
 	.memory_new = nv50_instobj_new,
 	.memory_wrap = nv50_instobj_wrap,
 	.zero = false,
+	.set_bar0_window_addr = nv50_instmem_set_bar0_window_addr,
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h
index 4c14c96fb60a..87bbdd786eaa 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h
@@ -16,10 +16,16 @@ struct nvkm_instmem_func {
 			  bool zero, struct nvkm_memory **);
 	int (*memory_wrap)(struct nvkm_instmem *, struct nvkm_memory *, struct nvkm_memory **);
 	bool zero;
+	void (*set_bar0_window_addr)(struct nvkm_device *, u64 addr);
 };
 
 int nv50_instmem_new_(const struct nvkm_instmem_func *, struct nvkm_device *,
 		      enum nvkm_subdev_type, int, struct nvkm_instmem **);
+void nv50_instmem_fini(struct nvkm_instmem *);
+int nv50_instobj_new(struct nvkm_instmem *, u32 size, u32 align, bool zero,
+		     struct nvkm_memory **);
+int nv50_instobj_wrap(struct nvkm_instmem *, struct nvkm_memory *vram,
+		      struct nvkm_memory **bar2);
 
 void nvkm_instmem_ctor(const struct nvkm_instmem_func *, struct nvkm_device *,
 		       enum nvkm_subdev_type, int, struct nvkm_instmem *);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/r535.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/r535.c
deleted file mode 100644
index 5f3c9c02a4c0..000000000000
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/r535.c
+++ /dev/null
@@ -1,333 +0,0 @@
-/*
- * 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 "priv.h"
-
-#include <subdev/gsp.h>
-
-#include <nvhw/drf.h>
-
-#include <nvrm/nvtypes.h>
-#include <nvrm/535.113.01/common/sdk/nvidia/inc/class/cl84a0.h>
-#include <nvrm/535.113.01/common/sdk/nvidia/inc/ctrl/ctrl2080/ctrl2080internal.h>
-#include <nvrm/535.113.01/common/sdk/nvidia/inc/nvos.h>
-#include <nvrm/535.113.01/nvidia/generated/g_fbsr_nvoc.h>
-#include <nvrm/535.113.01/nvidia/generated/g_rpc-structures.h>
-#include <nvrm/535.113.01/nvidia/kernel/inc/vgpu/rpc_global_enums.h>
-
-struct fbsr_item {
-	const char *type;
-	u64 addr;
-	u64 size;
-
-	struct list_head head;
-};
-
-struct fbsr {
-	struct list_head items;
-
-	u64 size;
-	int regions;
-
-	struct nvkm_gsp_client client;
-	struct nvkm_gsp_device device;
-
-	u64 hmemory;
-	u64 sys_offset;
-};
-
-static int
-fbsr_memlist(struct nvkm_gsp_device *device, u32 handle, enum nvkm_memory_target aper,
-	     u64 phys, u64 size, struct sg_table *sgt, struct nvkm_gsp_object *object)
-{
-	struct nvkm_gsp_client *client = device->object.client;
-	struct nvkm_gsp *gsp = client->gsp;
-	const u32 pages = size / GSP_PAGE_SIZE;
-	rpc_alloc_memory_v13_01 *rpc;
-	int ret;
-
-	rpc = nvkm_gsp_rpc_get(gsp, NV_VGPU_MSG_FUNCTION_ALLOC_MEMORY,
-			       sizeof(*rpc) + pages * sizeof(rpc->pteDesc.pte_pde[0]));
-	if (IS_ERR(rpc))
-		return PTR_ERR(rpc);
-
-	rpc->hClient = client->object.handle;
-	rpc->hDevice = device->object.handle;
-	rpc->hMemory = handle;
-	if (aper == NVKM_MEM_TARGET_HOST) {
-		rpc->hClass = NV01_MEMORY_LIST_SYSTEM;
-		rpc->flags = NVDEF(NVOS02, FLAGS, PHYSICALITY, NONCONTIGUOUS) |
-			     NVDEF(NVOS02, FLAGS, LOCATION, PCI) |
-			     NVDEF(NVOS02, FLAGS, MAPPING, NO_MAP);
-	} else {
-		rpc->hClass = NV01_MEMORY_LIST_FBMEM;
-		rpc->flags = NVDEF(NVOS02, FLAGS, PHYSICALITY, CONTIGUOUS) |
-			     NVDEF(NVOS02, FLAGS, LOCATION, VIDMEM) |
-			     NVDEF(NVOS02, FLAGS, MAPPING, NO_MAP);
-		rpc->format = 6; /* NV_MMU_PTE_KIND_GENERIC_MEMORY */
-	}
-	rpc->pteAdjust = 0;
-	rpc->length = size;
-	rpc->pageCount = pages;
-	rpc->pteDesc.idr = 0;
-	rpc->pteDesc.reserved1 = 0;
-	rpc->pteDesc.length = pages;
-
-	if (sgt) {
-		struct scatterlist *sgl;
-		int pte = 0, idx;
-
-		for_each_sgtable_dma_sg(sgt, sgl, idx) {
-			for (int i = 0; i < sg_dma_len(sgl) / GSP_PAGE_SIZE; i++)
-				rpc->pteDesc.pte_pde[pte++].pte = (sg_dma_address(sgl) >> 12) + i;
-
-		}
-	} else {
-		for (int i = 0; i < pages; i++)
-			rpc->pteDesc.pte_pde[i].pte = (phys >> 12) + i;
-	}
-
-	ret = nvkm_gsp_rpc_wr(gsp, rpc, true);
-	if (ret)
-		return ret;
-
-	object->client = device->object.client;
-	object->parent = &device->object;
-	object->handle = handle;
-	return 0;
-}
-
-static int
-fbsr_send(struct fbsr *fbsr, struct fbsr_item *item)
-{
-	NV2080_CTRL_INTERNAL_FBSR_SEND_REGION_INFO_PARAMS *ctrl;
-	struct nvkm_gsp *gsp = fbsr->client.gsp;
-	struct nvkm_gsp_object memlist;
-	int ret;
-
-	ret = fbsr_memlist(&fbsr->device, fbsr->hmemory, NVKM_MEM_TARGET_VRAM,
-			   item->addr, item->size, NULL, &memlist);
-	if (ret)
-		return ret;
-
-	ctrl = nvkm_gsp_rm_ctrl_get(&gsp->internal.device.subdevice,
-				    NV2080_CTRL_CMD_INTERNAL_FBSR_SEND_REGION_INFO,
-				    sizeof(*ctrl));
-	if (IS_ERR(ctrl)) {
-		ret = PTR_ERR(ctrl);
-		goto done;
-	}
-
-	ctrl->fbsrType = FBSR_TYPE_DMA;
-	ctrl->hClient = fbsr->client.object.handle;
-	ctrl->hVidMem = fbsr->hmemory++;
-	ctrl->vidOffset = 0;
-	ctrl->sysOffset = fbsr->sys_offset;
-	ctrl->size = item->size;
-
-	ret = nvkm_gsp_rm_ctrl_wr(&gsp->internal.device.subdevice, ctrl);
-done:
-	nvkm_gsp_rm_free(&memlist);
-	if (ret)
-		return ret;
-
-	fbsr->sys_offset += item->size;
-	return 0;
-}
-
-static int
-fbsr_init(struct fbsr *fbsr, struct sg_table *sgt, u64 items_size)
-{
-	NV2080_CTRL_INTERNAL_FBSR_INIT_PARAMS *ctrl;
-	struct nvkm_gsp *gsp = fbsr->client.gsp;
-	struct nvkm_gsp_object memlist;
-	int ret;
-
-	ret = fbsr_memlist(&fbsr->device, fbsr->hmemory, NVKM_MEM_TARGET_HOST,
-			   0, fbsr->size, sgt, &memlist);
-	if (ret)
-		return ret;
-
-	ctrl = nvkm_gsp_rm_ctrl_get(&gsp->internal.device.subdevice,
-				    NV2080_CTRL_CMD_INTERNAL_FBSR_INIT, sizeof(*ctrl));
-	if (IS_ERR(ctrl))
-		return PTR_ERR(ctrl);
-
-	ctrl->fbsrType = FBSR_TYPE_DMA;
-	ctrl->numRegions = fbsr->regions;
-	ctrl->hClient = fbsr->client.object.handle;
-	ctrl->hSysMem = fbsr->hmemory++;
-	ctrl->gspFbAllocsSysOffset = items_size;
-
-	ret = nvkm_gsp_rm_ctrl_wr(&gsp->internal.device.subdevice, ctrl);
-	if (ret)
-		return ret;
-
-	nvkm_gsp_rm_free(&memlist);
-	return 0;
-}
-
-static bool
-fbsr_vram(struct fbsr *fbsr, const char *type, u64 addr, u64 size)
-{
-	struct fbsr_item *item;
-
-	if (!(item = kzalloc(sizeof(*item), GFP_KERNEL)))
-		return false;
-
-	item->type = type;
-	item->addr = addr;
-	item->size = size;
-	list_add_tail(&item->head, &fbsr->items);
-	return true;
-}
-
-static bool
-fbsr_inst(struct fbsr *fbsr, const char *type, struct nvkm_memory *memory)
-{
-	return fbsr_vram(fbsr, type, nvkm_memory_addr(memory), nvkm_memory_size(memory));
-}
-
-static void
-r535_instmem_resume(struct nvkm_instmem *imem)
-{
-	/* RM has restored VRAM contents already, so just need to free the sysmem buffer. */
-	if (imem->rm.fbsr_valid) {
-		nvkm_gsp_sg_free(imem->subdev.device, &imem->rm.fbsr);
-		imem->rm.fbsr_valid = false;
-	}
-}
-
-static int
-r535_instmem_suspend(struct nvkm_instmem *imem)
-{
-	struct nvkm_subdev *subdev = &imem->subdev;
-	struct nvkm_device *device = subdev->device;
-	struct nvkm_gsp *gsp = device->gsp;
-	struct nvkm_instobj *iobj;
-	struct fbsr fbsr = {};
-	struct fbsr_item *item, *temp;
-	u64 items_size;
-	int ret;
-
-	INIT_LIST_HEAD(&fbsr.items);
-	fbsr.hmemory = 0xcaf00003;
-
-	/* Create a list of all regions we need RM to save during suspend. */
-	list_for_each_entry(iobj, &imem->list, head) {
-		if (iobj->preserve) {
-			if (!fbsr_inst(&fbsr, "inst", &iobj->memory))
-				return -ENOMEM;
-		}
-	}
-
-	list_for_each_entry(iobj, &imem->boot, head) {
-		if (!fbsr_inst(&fbsr, "boot", &iobj->memory))
-			return -ENOMEM;
-	}
-
-	if (!fbsr_vram(&fbsr, "gsp-non-wpr", gsp->fb.heap.addr, gsp->fb.heap.size))
-		return -ENOMEM;
-
-	/* Determine memory requirements. */
-	list_for_each_entry(item, &fbsr.items, head) {
-		nvkm_debug(subdev, "fbsr: %016llx %016llx %s\n",
-			   item->addr, item->size, item->type);
-		fbsr.size += item->size;
-		fbsr.regions++;
-	}
-
-	items_size = fbsr.size;
-	nvkm_debug(subdev, "fbsr: %d regions (0x%llx bytes)\n", fbsr.regions, items_size);
-
-	fbsr.size += gsp->fb.rsvd_size;
-	fbsr.size += gsp->fb.bios.vga_workspace.size;
-	nvkm_debug(subdev, "fbsr: size: 0x%llx bytes\n", fbsr.size);
-
-	ret = nvkm_gsp_sg(gsp->subdev.device, fbsr.size, &imem->rm.fbsr);
-	if (ret)
-		goto done;
-
-	/* Tell RM about the sysmem which will hold VRAM contents across suspend. */
-	ret = nvkm_gsp_client_device_ctor(gsp, &fbsr.client, &fbsr.device);
-	if (ret)
-		goto done_sgt;
-
-	ret = fbsr_init(&fbsr, &imem->rm.fbsr, items_size);
-	if (WARN_ON(ret))
-		goto done_sgt;
-
-	/* Send VRAM regions that need saving. */
-	list_for_each_entry(item, &fbsr.items, head) {
-		ret = fbsr_send(&fbsr, item);
-		if (WARN_ON(ret))
-			goto done_sgt;
-	}
-
-	imem->rm.fbsr_valid = true;
-
-	/* Cleanup everything except the sysmem backup, which will be removed after resume. */
-done_sgt:
-	if (ret) /* ... unless we failed already. */
-		nvkm_gsp_sg_free(device, &imem->rm.fbsr);
-done:
-	list_for_each_entry_safe(item, temp, &fbsr.items, head) {
-		list_del(&item->head);
-		kfree(item);
-	}
-
-	nvkm_gsp_device_dtor(&fbsr.device);
-	nvkm_gsp_client_dtor(&fbsr.client);
-	return ret;
-}
-
-static void *
-r535_instmem_dtor(struct nvkm_instmem *imem)
-{
-	kfree(imem->func);
-	return imem;
-}
-
-int
-r535_instmem_new(const struct nvkm_instmem_func *hw,
-		 struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
-		 struct nvkm_instmem **pinstmem)
-{
-	struct nvkm_instmem_func *rm;
-	int ret;
-
-	if (!(rm = kzalloc(sizeof(*rm), GFP_KERNEL)))
-		return -ENOMEM;
-
-	rm->dtor = r535_instmem_dtor;
-	rm->fini = hw->fini;
-	rm->suspend = r535_instmem_suspend;
-	rm->resume  = r535_instmem_resume;
-	rm->memory_new = hw->memory_new;
-	rm->memory_wrap = hw->memory_wrap;
-	rm->zero = false;
-
-	ret = nv50_instmem_new_(rm, device, type, inst, pinstmem);
-	if (ret)
-		kfree(rm);
-
-	return ret;
-}