drm/nouveau/gv100-: switch to volta semaphore methods

HOPPER_CHANNEL_GPFIFO_A removes the SEMAPHORE[A-D] methods that are
currently used by nouveau to implement fences on GF100 and newer.

Switch to the newer SEM methods available from VOLTA_CHANNEL_GPFIFO,
which are also available on the Hopper/Blackwell host classes.

Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

6 files changed, 151 insertions, 1 deletions

diff --git a/drivers/gpu/drm/nouveau/Kbuild b/drivers/gpu/drm/nouveau/Kbuild
index 0759ba15954b..385d24530d1e 100644
--- a/drivers/gpu/drm/nouveau/Kbuild
+++ b/drivers/gpu/drm/nouveau/Kbuild
@@ -69,5 +69,6 @@ nouveau-y += nv17_fence.o
 nouveau-y += nv50_fence.o
 nouveau-y += nv84_fence.o
 nouveau-y += nvc0_fence.o
+nouveau-y += gv100_fence.o
 
 obj-$(CONFIG_DRM_NOUVEAU) += nouveau.o
diff --git a/drivers/gpu/drm/nouveau/gv100_fence.c b/drivers/gpu/drm/nouveau/gv100_fence.c
new file mode 100644
index 000000000000..cccdeca72002
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/gv100_fence.c
@@ -0,0 +1,93 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
+ */
+#include "nouveau_drv.h"
+#include "nouveau_dma.h"
+#include "nouveau_fence.h"
+
+#include "nv50_display.h"
+
+#include <nvif/push906f.h>
+
+#include <nvhw/class/clc36f.h>
+
+static int
+gv100_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
+{
+	struct nvif_push *push = &chan->chan.push;
+	int ret;
+
+	ret = PUSH_WAIT(push, 8);
+	if (ret)
+		return ret;
+
+	PUSH_MTHD(push, NVC36F, SEM_ADDR_LO, lower_32_bits(virtual),
+				SEM_ADDR_HI, upper_32_bits(virtual),
+				SEM_PAYLOAD_LO, sequence);
+
+	PUSH_MTHD(push, NVC36F, SEM_EXECUTE,
+		  NVDEF(NVC36F, SEM_EXECUTE, OPERATION, RELEASE) |
+		  NVDEF(NVC36F, SEM_EXECUTE, RELEASE_WFI, EN) |
+		  NVDEF(NVC36F, SEM_EXECUTE, PAYLOAD_SIZE, 32BIT) |
+		  NVDEF(NVC36F, SEM_EXECUTE, RELEASE_TIMESTAMP, DIS));
+
+	PUSH_MTHD(push, NVC36F, NON_STALL_INTERRUPT, 0);
+
+	PUSH_KICK(push);
+	return 0;
+}
+
+static int
+gv100_fence_sync32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
+{
+	struct nvif_push *push = &chan->chan.push;
+	int ret;
+
+	ret = PUSH_WAIT(push, 6);
+	if (ret)
+		return ret;
+
+	PUSH_MTHD(push, NVC36F, SEM_ADDR_LO, lower_32_bits(virtual),
+				SEM_ADDR_HI, upper_32_bits(virtual),
+				SEM_PAYLOAD_LO, sequence);
+
+	PUSH_MTHD(push, NVC36F, SEM_EXECUTE,
+		  NVDEF(NVC36F, SEM_EXECUTE, OPERATION, ACQ_CIRC_GEQ) |
+		  NVDEF(NVC36F, SEM_EXECUTE, ACQUIRE_SWITCH_TSG, EN) |
+		  NVDEF(NVC36F, SEM_EXECUTE, PAYLOAD_SIZE, 32BIT));
+
+	PUSH_KICK(push);
+	return 0;
+}
+
+static int
+gv100_fence_context_new(struct nouveau_channel *chan)
+{
+	struct nv84_fence_chan *fctx;
+	int ret;
+
+	ret = nv84_fence_context_new(chan);
+	if (ret)
+		return ret;
+
+	fctx = chan->fence;
+	fctx->base.emit32 = gv100_fence_emit32;
+	fctx->base.sync32 = gv100_fence_sync32;
+	return 0;
+}
+
+int
+gv100_fence_create(struct nouveau_drm *drm)
+{
+	struct nv84_fence_priv *priv;
+	int ret;
+
+	ret = nv84_fence_create(drm);
+	if (ret)
+		return ret;
+
+	priv = drm->fence;
+	priv->base.context_new = gv100_fence_context_new;
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h b/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h
new file mode 100644
index 000000000000..8735dda4c8a7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/include/nvhw/class/clc36f.h
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: MIT
+ *
+ * Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
+ */
+#ifndef _clc36f_h_
+#define _clc36f_h_
+
+#define NVC36F_NON_STALL_INTERRUPT                                 (0x00000020)
+#define NVC36F_NON_STALL_INTERRUPT_HANDLE                                 31:0
+#define NVC36F_SEM_ADDR_LO                                         (0x0000005c)
+#define NVC36F_SEM_ADDR_LO_OFFSET                                         31:2
+#define NVC36F_SEM_ADDR_HI                                         (0x00000060)
+#define NVC36F_SEM_ADDR_HI_OFFSET                                          7:0
+#define NVC36F_SEM_PAYLOAD_LO                                      (0x00000064)
+#define NVC36F_SEM_PAYLOAD_LO_PAYLOAD                                     31:0
+#define NVC36F_SEM_PAYLOAD_HI                                      (0x00000068)
+#define NVC36F_SEM_PAYLOAD_HI_PAYLOAD                                     31:0
+#define NVC36F_SEM_EXECUTE                                         (0x0000006c)
+#define NVC36F_SEM_EXECUTE_OPERATION                                       2:0
+#define NVC36F_SEM_EXECUTE_OPERATION_ACQUIRE                        0x00000000
+#define NVC36F_SEM_EXECUTE_OPERATION_RELEASE                        0x00000001
+#define NVC36F_SEM_EXECUTE_OPERATION_ACQ_STRICT_GEQ                 0x00000002
+#define NVC36F_SEM_EXECUTE_OPERATION_ACQ_CIRC_GEQ                   0x00000003
+#define NVC36F_SEM_EXECUTE_OPERATION_ACQ_AND                        0x00000004
+#define NVC36F_SEM_EXECUTE_OPERATION_ACQ_NOR                        0x00000005
+#define NVC36F_SEM_EXECUTE_OPERATION_REDUCTION                      0x00000006
+#define NVC36F_SEM_EXECUTE_ACQUIRE_SWITCH_TSG                            12:12
+#define NVC36F_SEM_EXECUTE_ACQUIRE_SWITCH_TSG_DIS                   0x00000000
+#define NVC36F_SEM_EXECUTE_ACQUIRE_SWITCH_TSG_EN                    0x00000001
+#define NVC36F_SEM_EXECUTE_RELEASE_WFI                                   20:20
+#define NVC36F_SEM_EXECUTE_RELEASE_WFI_DIS                          0x00000000
+#define NVC36F_SEM_EXECUTE_RELEASE_WFI_EN                           0x00000001
+#define NVC36F_SEM_EXECUTE_PAYLOAD_SIZE                                  24:24
+#define NVC36F_SEM_EXECUTE_PAYLOAD_SIZE_32BIT                       0x00000000
+#define NVC36F_SEM_EXECUTE_PAYLOAD_SIZE_64BIT                       0x00000001
+#define NVC36F_SEM_EXECUTE_RELEASE_TIMESTAMP                             25:25
+#define NVC36F_SEM_EXECUTE_RELEASE_TIMESTAMP_DIS                    0x00000000
+#define NVC36F_SEM_EXECUTE_RELEASE_TIMESTAMP_EN                     0x00000001
+#define NVC36F_SEM_EXECUTE_REDUCTION                                     30:27
+#define NVC36F_SEM_EXECUTE_REDUCTION_IMIN                           0x00000000
+#define NVC36F_SEM_EXECUTE_REDUCTION_IMAX                           0x00000001
+#define NVC36F_SEM_EXECUTE_REDUCTION_IXOR                           0x00000002
+#define NVC36F_SEM_EXECUTE_REDUCTION_IAND                           0x00000003
+#define NVC36F_SEM_EXECUTE_REDUCTION_IOR                            0x00000004
+#define NVC36F_SEM_EXECUTE_REDUCTION_IADD                           0x00000005
+#define NVC36F_SEM_EXECUTE_REDUCTION_INC                            0x00000006
+#define NVC36F_SEM_EXECUTE_REDUCTION_DEC                            0x00000007
+#define NVC36F_SEM_EXECUTE_REDUCTION_FORMAT                              31:31
+#define NVC36F_SEM_EXECUTE_REDUCTION_FORMAT_SIGNED                  0x00000000
+#define NVC36F_SEM_EXECUTE_REDUCTION_FORMAT_UNSIGNED                0x00000001
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/include/nvif/push906f.h b/drivers/gpu/drm/nouveau/include/nvif/push906f.h
index cc2866bc8b0a..79df71de98d2 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/push906f.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/push906f.h
@@ -7,6 +7,7 @@
 #ifndef PUSH906F_SUBC
 // Host methods
 #define PUSH906F_SUBC_NV906F	0
+#define PUSH906F_SUBC_NVC36F	0
 
 // Twod
 #define PUSH906F_SUBC_NV902D	3
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index c69139701056..e7544942791d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -503,11 +503,13 @@ nouveau_accel_init(struct nouveau_drm *drm)
 		case KEPLER_CHANNEL_GPFIFO_B:
 		case MAXWELL_CHANNEL_GPFIFO_A:
 		case PASCAL_CHANNEL_GPFIFO_A:
+			ret = nvc0_fence_create(drm);
+			break;
 		case VOLTA_CHANNEL_GPFIFO_A:
 		case TURING_CHANNEL_GPFIFO_A:
 		case AMPERE_CHANNEL_GPFIFO_A:
 		case AMPERE_CHANNEL_GPFIFO_B:
-			ret = nvc0_fence_create(drm);
+			ret = gv100_fence_create(drm);
 			break;
 		default:
 			break;
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h
index 8bc065acfe35..6a983dd9f7b9 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.h
@@ -83,6 +83,7 @@ void nv17_fence_resume(struct nouveau_drm *drm);
 int nv50_fence_create(struct nouveau_drm *);
 int nv84_fence_create(struct nouveau_drm *);
 int nvc0_fence_create(struct nouveau_drm *);
+int gv100_fence_create(struct nouveau_drm *);
 
 struct nv84_fence_chan {
 	struct nouveau_fence_chan base;