nouveau : fix r570 suspend/resume regression

nouveau : fix r570 suspend/resume regression

[Dave Airlie]

Since R570 firmware got enabled, a number of GPUs have not being
successfully suspend/resuming. Lyude showed it on RTX6000, which
I reproduced and wasted a lot of time down various rabbit holes.

There are two required fixes, the first one adds proper sequence
numbers to the rpc messages, which fixes a bunch of NOCAT asserts.

Then we have to pass the runtime vs non-runtime state down to the
GSP fbsr code. This however requires replacing a bool with an enum
which refactors quite a bunch of interfaces unfortunately, but it
was the cleanest way to do it.

The final patch hooks up the interface so normal suspend doesn't
set the GCOFF flags.

Dave.

[PATCH 1/3] nouveau/gsp: use rpc sequence numbers properly.

[Dave Airlie]

From: Dave Airlie <airlied@redhat.com>

There are two layers of sequence numbers, one at the msg level
and one at the rpc level.

570 firmware started asserting on the sequence numbers being
in the right order, and we would see nocat records with asserts
in them.

Add the rpc level sequence number support.

Fixes: 53dac0623853 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
---
 drivers/gpu/drm/nouveau/include/nvkm/subdev/gsp.h     | 6 ++++++
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c | 4 ++--
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/rpc.c | 6 ++++++
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/gsp.c | 2 +-
 4 files changed, 15 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/gsp.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/gsp.h
index b8b97e10ae83..64fed208e4cf 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/gsp.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/gsp.h
@@ -44,6 +44,9 @@ typedef void (*nvkm_gsp_event_func)(struct nvkm_gsp_event *, void *repv, u32 rep
  * NVKM_GSP_RPC_REPLY_NOWAIT - If specified, immediately return to the
  * caller after the GSP RPC command is issued.
  *
+ * NVKM_GSP_RPC_REPLY_NOSEQ - If specified, exactly like NOWAIT
+ * but don't emit RPC sequence number.
+ *
  * NVKM_GSP_RPC_REPLY_RECV - If specified, wait and receive the entire GSP
  * RPC message after the GSP RPC command is issued.
  *
@@ -53,6 +56,7 @@ typedef void (*nvkm_gsp_event_func)(struct nvkm_gsp_event *, void *repv, u32 rep
  */
 enum nvkm_gsp_rpc_reply_policy {
 	NVKM_GSP_RPC_REPLY_NOWAIT = 0,
+	NVKM_GSP_RPC_REPLY_NOSEQ,
 	NVKM_GSP_RPC_REPLY_RECV,
 	NVKM_GSP_RPC_REPLY_POLL,
 };
@@ -242,6 +246,8 @@ struct nvkm_gsp {
 	/* The size of the registry RPC */
 	size_t registry_rpc_size;
 
+	u32 rpc_seq;
+
 #ifdef CONFIG_DEBUG_FS
 	/*
 	 * Logging buffers in debugfs. The wrapper objects need to remain
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
index 2a7e80c6d70f..6e7af2f737b7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
@@ -704,7 +704,7 @@ r535_gsp_rpc_set_registry(struct nvkm_gsp *gsp)
 
 	build_registry(gsp, rpc);
 
-	return nvkm_gsp_rpc_wr(gsp, rpc, NVKM_GSP_RPC_REPLY_NOWAIT);
+	return nvkm_gsp_rpc_wr(gsp, rpc, NVKM_GSP_RPC_REPLY_NOSEQ);
 
 fail:
 	clean_registry(gsp);
@@ -921,7 +921,7 @@ r535_gsp_set_system_info(struct nvkm_gsp *gsp)
 	info->pciConfigMirrorSize = device->pci->func->cfg.size;
 	r535_gsp_acpi_info(gsp, &info->acpiMethodData);
 
-	return nvkm_gsp_rpc_wr(gsp, info, NVKM_GSP_RPC_REPLY_NOWAIT);
+	return nvkm_gsp_rpc_wr(gsp, info, NVKM_GSP_RPC_REPLY_NOSEQ);
 }
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/rpc.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/rpc.c
index 0dc4782df8c0..3ca3de8f4340 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/rpc.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/rpc.c
@@ -557,6 +557,7 @@ r535_gsp_rpc_handle_reply(struct nvkm_gsp *gsp, u32 fn,
 
 	switch (policy) {
 	case NVKM_GSP_RPC_REPLY_NOWAIT:
+	case NVKM_GSP_RPC_REPLY_NOSEQ:
 		break;
 	case NVKM_GSP_RPC_REPLY_RECV:
 		reply = r535_gsp_msg_recv(gsp, fn, gsp_rpc_len);
@@ -588,6 +589,11 @@ r535_gsp_rpc_send(struct nvkm_gsp *gsp, void *payload,
 			       rpc->data, rpc->length - sizeof(*rpc), true);
 	}
 
+	if (policy == NVKM_GSP_RPC_REPLY_NOSEQ)
+		rpc->sequence = 0;
+	else
+		rpc->sequence = gsp->rpc_seq++;
+
 	ret = r535_gsp_cmdq_push(gsp, rpc);
 	if (ret)
 		return ERR_PTR(ret);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/gsp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/gsp.c
index 9d2fa4e66d59..996941c668ba 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/gsp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/gsp.c
@@ -176,7 +176,7 @@ r570_gsp_set_system_info(struct nvkm_gsp *gsp)
 	info->bIsPrimary = video_is_primary_device(device->dev);
 	info->bPreserveVideoMemoryAllocations = false;
 
-	return nvkm_gsp_rpc_wr(gsp, info, NVKM_GSP_RPC_REPLY_NOWAIT);
+	return nvkm_gsp_rpc_wr(gsp, info, NVKM_GSP_RPC_REPLY_NOSEQ);
 }
 
 static void
-- 
2.52.0

[PATCH 2/3] nouveau: add a third state to the fini handler.

[Dave Airlie]

From: Dave Airlie <airlied@redhat.com>

This is just refactoring to allow the lower layers to distinguish
between suspend and runtime suspend.

GSP 570 needs to set a flag with the GPU is going into GCOFF,
this flag taken from the opengpu driver is set whenever runtime
suspend is enterning GCOFF but not for normal suspend paths.

This just refactors the code, a subsequent patch use the information.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Fixes: 53dac0623853 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org>
---
 drivers/gpu/drm/nouveau/include/nvif/client.h |  2 +-
 drivers/gpu/drm/nouveau/include/nvif/driver.h |  2 +-
 .../drm/nouveau/include/nvkm/core/device.h    |  3 ++-
 .../drm/nouveau/include/nvkm/core/engine.h    |  2 +-
 .../drm/nouveau/include/nvkm/core/object.h    |  5 +++--
 .../drm/nouveau/include/nvkm/core/oproxy.h    |  2 +-
 .../drm/nouveau/include/nvkm/core/subdev.h    |  4 ++--
 .../nouveau/include/nvkm/core/suspend_state.h | 11 ++++++++++
 drivers/gpu/drm/nouveau/nouveau_drm.c         |  2 +-
 drivers/gpu/drm/nouveau/nouveau_nvif.c        | 10 +++++++--
 drivers/gpu/drm/nouveau/nvif/client.c         |  4 ++--
 drivers/gpu/drm/nouveau/nvkm/core/engine.c    |  4 ++--
 drivers/gpu/drm/nouveau/nvkm/core/ioctl.c     |  4 ++--
 drivers/gpu/drm/nouveau/nvkm/core/object.c    | 20 +++++++++++++----
 drivers/gpu/drm/nouveau/nvkm/core/oproxy.c    |  2 +-
 drivers/gpu/drm/nouveau/nvkm/core/subdev.c    | 18 ++++++++++++---
 drivers/gpu/drm/nouveau/nvkm/core/uevent.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/engine/ce/ga100.c    |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h |  2 +-
 .../gpu/drm/nouveau/nvkm/engine/device/base.c | 22 ++++++++++++++-----
 .../gpu/drm/nouveau/nvkm/engine/device/pci.c  |  4 ++--
 .../gpu/drm/nouveau/nvkm/engine/device/priv.h |  2 +-
 .../gpu/drm/nouveau/nvkm/engine/device/user.c |  2 +-
 .../gpu/drm/nouveau/nvkm/engine/disp/base.c   |  4 ++--
 .../gpu/drm/nouveau/nvkm/engine/disp/chan.c   |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/falcon.c  |  4 ++--
 .../gpu/drm/nouveau/nvkm/engine/fifo/base.c   |  2 +-
 .../gpu/drm/nouveau/nvkm/engine/fifo/uchan.c  |  6 ++---
 drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c |  4 ++--
 .../gpu/drm/nouveau/nvkm/engine/gr/gf100.c    |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/gr/nv04.c |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/gr/nv10.c |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.c |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.h |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/gr/nv40.c |  4 ++--
 .../gpu/drm/nouveau/nvkm/engine/mpeg/nv44.c   |  2 +-
 .../gpu/drm/nouveau/nvkm/engine/sec2/base.c   |  2 +-
 drivers/gpu/drm/nouveau/nvkm/engine/xtensa.c  |  4 ++--
 .../gpu/drm/nouveau/nvkm/subdev/acr/base.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/bar/base.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/clk/base.c    |  2 +-
 .../drm/nouveau/nvkm/subdev/devinit/base.c    |  4 ++--
 .../gpu/drm/nouveau/nvkm/subdev/fault/base.c  |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/fault/user.c  |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/gpio/base.c   |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/gsp/base.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c   |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/gsp/priv.h    |  8 +++----
 .../drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c   |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/i2c/base.c    |  2 +-
 .../drm/nouveau/nvkm/subdev/instmem/base.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/pci/base.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/pmu/base.c    |  2 +-
 .../gpu/drm/nouveau/nvkm/subdev/therm/base.c  |  6 ++---
 .../gpu/drm/nouveau/nvkm/subdev/timer/base.c  |  2 +-
 56 files changed, 139 insertions(+), 84 deletions(-)
 create mode 100644 drivers/gpu/drm/nouveau/include/nvkm/core/suspend_state.h

diff --git a/drivers/gpu/drm/nouveau/include/nvif/client.h b/drivers/gpu/drm/nouveau/include/nvif/client.h
index 03f1d564eb12..b698c74306f8 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/client.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/client.h
@@ -11,7 +11,7 @@ struct nvif_client {
 
 int  nvif_client_ctor(struct nvif_client *parent, const char *name, struct nvif_client *);
 void nvif_client_dtor(struct nvif_client *);
-int  nvif_client_suspend(struct nvif_client *);
+int  nvif_client_suspend(struct nvif_client *, bool);
 int  nvif_client_resume(struct nvif_client *);
 
 /*XXX*/
diff --git a/drivers/gpu/drm/nouveau/include/nvif/driver.h b/drivers/gpu/drm/nouveau/include/nvif/driver.h
index 7b08ff769039..61c8a177b28f 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/driver.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/driver.h
@@ -8,7 +8,7 @@ struct nvif_driver {
 	const char *name;
 	int (*init)(const char *name, u64 device, const char *cfg,
 		    const char *dbg, void **priv);
-	int (*suspend)(void *priv);
+	int (*suspend)(void *priv, bool runtime);
 	int (*resume)(void *priv);
 	int (*ioctl)(void *priv, void *data, u32 size, void **hack);
 	void __iomem *(*map)(void *priv, u64 handle, u32 size);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/device.h b/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
index 99579e7b9376..954a89d43bad 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
@@ -2,6 +2,7 @@
 #ifndef __NVKM_DEVICE_H__
 #define __NVKM_DEVICE_H__
 #include <core/oclass.h>
+#include <core/suspend_state.h>
 #include <core/intr.h>
 enum nvkm_subdev_type;
 
@@ -93,7 +94,7 @@ struct nvkm_device_func {
 	void *(*dtor)(struct nvkm_device *);
 	int (*preinit)(struct nvkm_device *);
 	int (*init)(struct nvkm_device *);
-	void (*fini)(struct nvkm_device *, bool suspend);
+	void (*fini)(struct nvkm_device *, enum nvkm_suspend_state suspend);
 	int (*irq)(struct nvkm_device *);
 	resource_size_t (*resource_addr)(struct nvkm_device *, enum nvkm_bar_id);
 	resource_size_t (*resource_size)(struct nvkm_device *, enum nvkm_bar_id);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h b/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h
index 738899fcf30b..1e97be6c6564 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/engine.h
@@ -20,7 +20,7 @@ struct nvkm_engine_func {
 	int (*oneinit)(struct nvkm_engine *);
 	int (*info)(struct nvkm_engine *, u64 mthd, u64 *data);
 	int (*init)(struct nvkm_engine *);
-	int (*fini)(struct nvkm_engine *, bool suspend);
+	int (*fini)(struct nvkm_engine *, enum nvkm_suspend_state suspend);
 	int (*reset)(struct nvkm_engine *);
 	int (*nonstall)(struct nvkm_engine *);
 	void (*intr)(struct nvkm_engine *);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/object.h b/drivers/gpu/drm/nouveau/include/nvkm/core/object.h
index 10107ef3ca49..54d356154274 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/object.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/object.h
@@ -2,6 +2,7 @@
 #ifndef __NVKM_OBJECT_H__
 #define __NVKM_OBJECT_H__
 #include <core/oclass.h>
+#include <core/suspend_state.h>
 struct nvkm_event;
 struct nvkm_gpuobj;
 struct nvkm_uevent;
@@ -27,7 +28,7 @@ enum nvkm_object_map {
 struct nvkm_object_func {
 	void *(*dtor)(struct nvkm_object *);
 	int (*init)(struct nvkm_object *);
-	int (*fini)(struct nvkm_object *, bool suspend);
+	int (*fini)(struct nvkm_object *, enum nvkm_suspend_state suspend);
 	int (*mthd)(struct nvkm_object *, u32 mthd, void *data, u32 size);
 	int (*ntfy)(struct nvkm_object *, u32 mthd, struct nvkm_event **);
 	int (*map)(struct nvkm_object *, void *argv, u32 argc,
@@ -49,7 +50,7 @@ int nvkm_object_new(const struct nvkm_oclass *, void *data, u32 size,
 void nvkm_object_del(struct nvkm_object **);
 void *nvkm_object_dtor(struct nvkm_object *);
 int nvkm_object_init(struct nvkm_object *);
-int nvkm_object_fini(struct nvkm_object *, bool suspend);
+int nvkm_object_fini(struct nvkm_object *, enum nvkm_suspend_state);
 int nvkm_object_mthd(struct nvkm_object *, u32 mthd, void *data, u32 size);
 int nvkm_object_ntfy(struct nvkm_object *, u32 mthd, struct nvkm_event **);
 int nvkm_object_map(struct nvkm_object *, void *argv, u32 argc,
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/oproxy.h b/drivers/gpu/drm/nouveau/include/nvkm/core/oproxy.h
index 0e70a9afba33..cf66aee4d111 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/oproxy.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/oproxy.h
@@ -13,7 +13,7 @@ struct nvkm_oproxy {
 struct nvkm_oproxy_func {
 	void (*dtor[2])(struct nvkm_oproxy *);
 	int  (*init[2])(struct nvkm_oproxy *);
-	int  (*fini[2])(struct nvkm_oproxy *, bool suspend);
+	int  (*fini[2])(struct nvkm_oproxy *, enum nvkm_suspend_state suspend);
 };
 
 void nvkm_oproxy_ctor(const struct nvkm_oproxy_func *,
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h b/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h
index bce6e1ba09ea..bd6b1b658e40 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/subdev.h
@@ -40,7 +40,7 @@ struct nvkm_subdev_func {
 	int (*oneinit)(struct nvkm_subdev *);
 	int (*info)(struct nvkm_subdev *, u64 mthd, u64 *data);
 	int (*init)(struct nvkm_subdev *);
-	int (*fini)(struct nvkm_subdev *, bool suspend);
+	int (*fini)(struct nvkm_subdev *, enum nvkm_suspend_state suspend);
 	void (*intr)(struct nvkm_subdev *);
 };
 
@@ -65,7 +65,7 @@ void nvkm_subdev_unref(struct nvkm_subdev *);
 int  nvkm_subdev_preinit(struct nvkm_subdev *);
 int  nvkm_subdev_oneinit(struct nvkm_subdev *);
 int  nvkm_subdev_init(struct nvkm_subdev *);
-int  nvkm_subdev_fini(struct nvkm_subdev *, bool suspend);
+int  nvkm_subdev_fini(struct nvkm_subdev *, enum nvkm_suspend_state suspend);
 int  nvkm_subdev_info(struct nvkm_subdev *, u64, u64 *);
 void nvkm_subdev_intr(struct nvkm_subdev *);
 
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/suspend_state.h b/drivers/gpu/drm/nouveau/include/nvkm/core/suspend_state.h
new file mode 100644
index 000000000000..134120fb71f4
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/suspend_state.h
@@ -0,0 +1,11 @@
+/* SPDX-License-Identifier: MIT */
+#ifndef __NVKM_SUSPEND_STATE_H__
+#define __NVKM_SUSPEND_STATE_H__
+
+enum nvkm_suspend_state {
+	NVKM_POWEROFF,
+	NVKM_SUSPEND,
+	NVKM_RUNTIME_SUSPEND,
+};
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 1527b801f013..dc469e571c0a 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -983,7 +983,7 @@ nouveau_do_suspend(struct nouveau_drm *drm, bool runtime)
 	}
 
 	NV_DEBUG(drm, "suspending object tree...\n");
-	ret = nvif_client_suspend(&drm->_client);
+	ret = nvif_client_suspend(&drm->_client, runtime);
 	if (ret)
 		goto fail_client;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_nvif.c b/drivers/gpu/drm/nouveau/nouveau_nvif.c
index adb802421fda..eeb4ebbc16bf 100644
--- a/drivers/gpu/drm/nouveau/nouveau_nvif.c
+++ b/drivers/gpu/drm/nouveau/nouveau_nvif.c
@@ -62,10 +62,16 @@ nvkm_client_resume(void *priv)
 }
 
 static int
-nvkm_client_suspend(void *priv)
+nvkm_client_suspend(void *priv, bool runtime)
 {
 	struct nvkm_client *client = priv;
-	return nvkm_object_fini(&client->object, true);
+	enum nvkm_suspend_state state;
+
+	if (runtime)
+		state = NVKM_RUNTIME_SUSPEND;
+	else
+		state = NVKM_SUSPEND;
+	return nvkm_object_fini(&client->object, state);
 }
 
 static int
diff --git a/drivers/gpu/drm/nouveau/nvif/client.c b/drivers/gpu/drm/nouveau/nvif/client.c
index fdf5054ed7d8..36d3c99786bd 100644
--- a/drivers/gpu/drm/nouveau/nvif/client.c
+++ b/drivers/gpu/drm/nouveau/nvif/client.c
@@ -30,9 +30,9 @@
 #include <nvif/if0000.h>
 
 int
-nvif_client_suspend(struct nvif_client *client)
+nvif_client_suspend(struct nvif_client *client, bool runtime)
 {
-	return client->driver->suspend(client->object.priv);
+	return client->driver->suspend(client->object.priv, runtime);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/engine.c b/drivers/gpu/drm/nouveau/nvkm/core/engine.c
index 36a31e9eea22..5bf62940d7be 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/engine.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/engine.c
@@ -41,7 +41,7 @@ nvkm_engine_reset(struct nvkm_engine *engine)
 	if (engine->func->reset)
 		return engine->func->reset(engine);
 
-	nvkm_subdev_fini(&engine->subdev, false);
+	nvkm_subdev_fini(&engine->subdev, NVKM_POWEROFF);
 	return nvkm_subdev_init(&engine->subdev);
 }
 
@@ -98,7 +98,7 @@ nvkm_engine_info(struct nvkm_subdev *subdev, u64 mthd, u64 *data)
 }
 
 static int
-nvkm_engine_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_engine_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_engine *engine = nvkm_engine(subdev);
 	if (engine->func->fini)
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c b/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c
index 45051a1249da..b8fc9be67851 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/ioctl.c
@@ -141,7 +141,7 @@ nvkm_ioctl_new(struct nvkm_client *client,
 			}
 			ret = -EEXIST;
 		}
-		nvkm_object_fini(object, false);
+		nvkm_object_fini(object, NVKM_POWEROFF);
 	}
 
 	nvkm_object_del(&object);
@@ -160,7 +160,7 @@ nvkm_ioctl_del(struct nvkm_client *client,
 	nvif_ioctl(object, "delete size %d\n", size);
 	if (!(ret = nvif_unvers(ret, &data, &size, args->none))) {
 		nvif_ioctl(object, "delete\n");
-		nvkm_object_fini(object, false);
+		nvkm_object_fini(object, NVKM_POWEROFF);
 		nvkm_object_del(&object);
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/object.c b/drivers/gpu/drm/nouveau/nvkm/core/object.c
index 390c265cf8af..af9f00f74c28 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/object.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/object.c
@@ -142,13 +142,25 @@ nvkm_object_bind(struct nvkm_object *object, struct nvkm_gpuobj *gpuobj,
 }
 
 int
-nvkm_object_fini(struct nvkm_object *object, bool suspend)
+nvkm_object_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
-	const char *action = suspend ? "suspend" : "fini";
+	const char *action;
 	struct nvkm_object *child;
 	s64 time;
 	int ret;
 
+	switch (suspend) {
+	case NVKM_POWEROFF:
+	default:
+		action = "fini";
+		break;
+	case NVKM_SUSPEND:
+		action = "suspend";
+		break;
+	case NVKM_RUNTIME_SUSPEND:
+		action = "runtime";
+		break;
+	}
 	nvif_debug(object, "%s children...\n", action);
 	time = ktime_to_us(ktime_get());
 	list_for_each_entry_reverse(child, &object->tree, head) {
@@ -212,11 +224,11 @@ nvkm_object_init(struct nvkm_object *object)
 
 fail_child:
 	list_for_each_entry_continue_reverse(child, &object->tree, head)
-		nvkm_object_fini(child, false);
+		nvkm_object_fini(child, NVKM_POWEROFF);
 fail:
 	nvif_error(object, "init failed with %d\n", ret);
 	if (object->func->fini)
-		object->func->fini(object, false);
+		object->func->fini(object, NVKM_POWEROFF);
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/oproxy.c b/drivers/gpu/drm/nouveau/nvkm/core/oproxy.c
index 5db80d1780f0..7c9edf752768 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/oproxy.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/oproxy.c
@@ -87,7 +87,7 @@ nvkm_oproxy_uevent(struct nvkm_object *object, void *argv, u32 argc,
 }
 
 static int
-nvkm_oproxy_fini(struct nvkm_object *object, bool suspend)
+nvkm_oproxy_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_oproxy *oproxy = nvkm_oproxy(object);
 	int ret;
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/subdev.c b/drivers/gpu/drm/nouveau/nvkm/core/subdev.c
index 6c20e827a069..d184655757ee 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/subdev.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/subdev.c
@@ -51,12 +51,24 @@ nvkm_subdev_info(struct nvkm_subdev *subdev, u64 mthd, u64 *data)
 }
 
 int
-nvkm_subdev_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_subdev_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_device *device = subdev->device;
-	const char *action = suspend ? "suspend" : subdev->use.enabled ? "fini" : "reset";
+	const char *action;
 	s64 time;
 
+	switch (suspend) {
+	case NVKM_POWEROFF:
+	default:
+		action = subdev->use.enabled ? "fini": "reset";
+		break;
+	case NVKM_SUSPEND:
+		action = "suspend";
+		break;
+	case NVKM_RUNTIME_SUSPEND:
+		action = "runtime";
+		break;
+	}
 	nvkm_trace(subdev, "%s running...\n", action);
 	time = ktime_to_us(ktime_get());
 
@@ -186,7 +198,7 @@ void
 nvkm_subdev_unref(struct nvkm_subdev *subdev)
 {
 	if (refcount_dec_and_mutex_lock(&subdev->use.refcount, &subdev->use.mutex)) {
-		nvkm_subdev_fini(subdev, false);
+		nvkm_subdev_fini(subdev, NVKM_POWEROFF);
 		mutex_unlock(&subdev->use.mutex);
 	}
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/uevent.c b/drivers/gpu/drm/nouveau/nvkm/core/uevent.c
index cc254c390a57..46beb6e470ee 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/uevent.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/uevent.c
@@ -73,7 +73,7 @@ nvkm_uevent_mthd(struct nvkm_object *object, u32 mthd, void *argv, u32 argc)
 }
 
 static int
-nvkm_uevent_fini(struct nvkm_object *object, bool suspend)
+nvkm_uevent_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_uevent *uevent = nvkm_uevent(object);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/ce/ga100.c b/drivers/gpu/drm/nouveau/nvkm/engine/ce/ga100.c
index 1c0c60138706..1a3caf697608 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/ce/ga100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/ce/ga100.c
@@ -46,7 +46,7 @@ ga100_ce_nonstall(struct nvkm_engine *engine)
 }
 
 int
-ga100_ce_fini(struct nvkm_engine *engine, bool suspend)
+ga100_ce_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	nvkm_inth_block(&engine->subdev.inth);
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h
index 34fd2657134b..f07b45853310 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h
@@ -14,7 +14,7 @@ extern const struct nvkm_object_func gv100_ce_cclass;
 
 int ga100_ce_oneinit(struct nvkm_engine *);
 int ga100_ce_init(struct nvkm_engine *);
-int ga100_ce_fini(struct nvkm_engine *, bool);
+int ga100_ce_fini(struct nvkm_engine *, enum nvkm_suspend_state);
 int ga100_ce_nonstall(struct nvkm_engine *);
 
 u32 gb202_ce_grce_mask(struct nvkm_device *);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
index 2517b65d8faa..b101e14f841e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
@@ -2936,13 +2936,25 @@ nvkm_device_engine(struct nvkm_device *device, int type, int inst)
 }
 
 int
-nvkm_device_fini(struct nvkm_device *device, bool suspend)
+nvkm_device_fini(struct nvkm_device *device, enum nvkm_suspend_state suspend)
 {
-	const char *action = suspend ? "suspend" : "fini";
+	const char *action;
 	struct nvkm_subdev *subdev;
 	int ret;
 	s64 time;
 
+	switch (suspend) {
+	case NVKM_POWEROFF:
+	default:
+		action = "fini";
+		break;
+	case NVKM_SUSPEND:
+		action = "suspend";
+		break;
+	case NVKM_RUNTIME_SUSPEND:
+		action = "runtime";
+		break;
+	}
 	nvdev_trace(device, "%s running...\n", action);
 	time = ktime_to_us(ktime_get());
 
@@ -3032,7 +3044,7 @@ nvkm_device_init(struct nvkm_device *device)
 	if (ret)
 		return ret;
 
-	nvkm_device_fini(device, false);
+	nvkm_device_fini(device, NVKM_POWEROFF);
 
 	nvdev_trace(device, "init running...\n");
 	time = ktime_to_us(ktime_get());
@@ -3060,9 +3072,9 @@ nvkm_device_init(struct nvkm_device *device)
 
 fail_subdev:
 	list_for_each_entry_from(subdev, &device->subdev, head)
-		nvkm_subdev_fini(subdev, false);
+		nvkm_subdev_fini(subdev, NVKM_POWEROFF);
 fail:
-	nvkm_device_fini(device, false);
+	nvkm_device_fini(device, NVKM_POWEROFF);
 
 	nvdev_error(device, "init failed with %d\n", ret);
 	return ret;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c
index 8f0261a0d618..4c29b60460d4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c
@@ -1605,10 +1605,10 @@ nvkm_device_pci_irq(struct nvkm_device *device)
 }
 
 static void
-nvkm_device_pci_fini(struct nvkm_device *device, bool suspend)
+nvkm_device_pci_fini(struct nvkm_device *device, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_device_pci *pdev = nvkm_device_pci(device);
-	if (suspend) {
+	if (suspend != NVKM_POWEROFF) {
 		pci_disable_device(pdev->pdev);
 		pdev->suspend = true;
 	}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/device/priv.h
index 75ee7506d443..d0c40f034244 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/priv.h
@@ -56,5 +56,5 @@ int  nvkm_device_ctor(const struct nvkm_device_func *,
 		      const char *name, const char *cfg, const char *dbg,
 		      struct nvkm_device *);
 int  nvkm_device_init(struct nvkm_device *);
-int  nvkm_device_fini(struct nvkm_device *, bool suspend);
+int  nvkm_device_fini(struct nvkm_device *, enum nvkm_suspend_state suspend);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c
index 58191b7a0494..32ff3181f47b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c
@@ -218,7 +218,7 @@ nvkm_udevice_map(struct nvkm_object *object, void *argv, u32 argc,
 }
 
 static int
-nvkm_udevice_fini(struct nvkm_object *object, bool suspend)
+nvkm_udevice_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_udevice *udev = nvkm_udevice(object);
 	struct nvkm_device *device = udev->device;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/base.c
index b24eb1e560bc..84745f60912e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/base.c
@@ -99,13 +99,13 @@ nvkm_disp_intr(struct nvkm_engine *engine)
 }
 
 static int
-nvkm_disp_fini(struct nvkm_engine *engine, bool suspend)
+nvkm_disp_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_disp *disp = nvkm_disp(engine);
 	struct nvkm_outp *outp;
 
 	if (disp->func->fini)
-		disp->func->fini(disp, suspend);
+		disp->func->fini(disp, suspend != NVKM_POWEROFF);
 
 	list_for_each_entry(outp, &disp->outps, head) {
 		if (outp->func->fini)
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/chan.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/chan.c
index 9b84e357d354..57a62a2de7c7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/chan.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/chan.c
@@ -128,7 +128,7 @@ nvkm_disp_chan_child_get(struct nvkm_object *object, int index, struct nvkm_ocla
 }
 
 static int
-nvkm_disp_chan_fini(struct nvkm_object *object, bool suspend)
+nvkm_disp_chan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_disp_chan *chan = nvkm_disp_chan(object);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/falcon.c b/drivers/gpu/drm/nouveau/nvkm/engine/falcon.c
index fd5ee9f0af36..cf8e356867b4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/falcon.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/falcon.c
@@ -93,13 +93,13 @@ nvkm_falcon_intr(struct nvkm_engine *engine)
 }
 
 static int
-nvkm_falcon_fini(struct nvkm_engine *engine, bool suspend)
+nvkm_falcon_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_falcon *falcon = nvkm_falcon(engine);
 	struct nvkm_device *device = falcon->engine.subdev.device;
 	const u32 base = falcon->addr;
 
-	if (!suspend) {
+	if (suspend == NVKM_POWEROFF) {
 		nvkm_memory_unref(&falcon->core);
 		if (falcon->external) {
 			vfree(falcon->data.data);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
index 6fd4e60634fb..1561287a32f2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/base.c
@@ -122,7 +122,7 @@ nvkm_fifo_class_get(struct nvkm_oclass *oclass, int index, const struct nvkm_dev
 }
 
 static int
-nvkm_fifo_fini(struct nvkm_engine *engine, bool suspend)
+nvkm_fifo_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_fifo *fifo = nvkm_fifo(engine);
 	struct nvkm_runl *runl;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
index 52420a1edca5..c978b97e10c6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/uchan.c
@@ -72,7 +72,7 @@ struct nvkm_uobj {
 };
 
 static int
-nvkm_uchan_object_fini_1(struct nvkm_oproxy *oproxy, bool suspend)
+nvkm_uchan_object_fini_1(struct nvkm_oproxy *oproxy, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_uobj *uobj = container_of(oproxy, typeof(*uobj), oproxy);
 	struct nvkm_chan *chan = uobj->chan;
@@ -87,7 +87,7 @@ nvkm_uchan_object_fini_1(struct nvkm_oproxy *oproxy, bool suspend)
 		nvkm_chan_cctx_bind(chan, ectx->engn, NULL);
 
 		if (refcount_dec_and_test(&ectx->uses))
-			nvkm_object_fini(ectx->object, false);
+			nvkm_object_fini(ectx->object, NVKM_POWEROFF);
 		mutex_unlock(&chan->cgrp->mutex);
 	}
 
@@ -269,7 +269,7 @@ nvkm_uchan_map(struct nvkm_object *object, void *argv, u32 argc,
 }
 
 static int
-nvkm_uchan_fini(struct nvkm_object *object, bool suspend)
+nvkm_uchan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_chan *chan = nvkm_uchan(object)->chan;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c
index f5e68f09df76..cd4908b1b4df 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/base.c
@@ -168,11 +168,11 @@ nvkm_gr_init(struct nvkm_engine *engine)
 }
 
 static int
-nvkm_gr_fini(struct nvkm_engine *engine, bool suspend)
+nvkm_gr_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_gr *gr = nvkm_gr(engine);
 	if (gr->func->fini)
-		return gr->func->fini(gr, suspend);
+		return gr->func->fini(gr, suspend != NVKM_POWEROFF);
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
index 3ea447f6a45b..3608215f0f11 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
@@ -2330,7 +2330,7 @@ gf100_gr_reset(struct nvkm_gr *base)
 
 	WARN_ON(gf100_gr_fecs_halt_pipeline(gr));
 
-	subdev->func->fini(subdev, false);
+	subdev->func->fini(subdev, NVKM_POWEROFF);
 	nvkm_mc_disable(device, subdev->type, subdev->inst);
 	if (gr->func->gpccs.reset)
 		gr->func->gpccs.reset(gr);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv04.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv04.c
index ca822f07b63e..82937df8b8c0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv04.c
@@ -1158,7 +1158,7 @@ nv04_gr_chan_dtor(struct nvkm_object *object)
 }
 
 static int
-nv04_gr_chan_fini(struct nvkm_object *object, bool suspend)
+nv04_gr_chan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nv04_gr_chan *chan = nv04_gr_chan(object);
 	struct nv04_gr *gr = chan->gr;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv10.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv10.c
index 92ef7c9b2910..fcb4e4fce83f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv10.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv10.c
@@ -951,7 +951,7 @@ nv10_gr_context_switch(struct nv10_gr *gr)
 }
 
 static int
-nv10_gr_chan_fini(struct nvkm_object *object, bool suspend)
+nv10_gr_chan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nv10_gr_chan *chan = nv10_gr_chan(object);
 	struct nv10_gr *gr = chan->gr;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.c
index 13407fafe947..ab57b3b40228 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.c
@@ -27,7 +27,7 @@ nv20_gr_chan_init(struct nvkm_object *object)
 }
 
 int
-nv20_gr_chan_fini(struct nvkm_object *object, bool suspend)
+nv20_gr_chan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nv20_gr_chan *chan = nv20_gr_chan(object);
 	struct nv20_gr *gr = chan->gr;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.h b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.h
index c0d2be53413e..786c7832f7ac 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv20.h
@@ -31,5 +31,5 @@ struct nv20_gr_chan {
 
 void *nv20_gr_chan_dtor(struct nvkm_object *);
 int nv20_gr_chan_init(struct nvkm_object *);
-int nv20_gr_chan_fini(struct nvkm_object *, bool);
+int nv20_gr_chan_fini(struct nvkm_object *, enum nvkm_suspend_state);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv40.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv40.c
index b609b0150ba1..e3e797cf3034 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv40.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv40.c
@@ -89,7 +89,7 @@ nv40_gr_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
 }
 
 static int
-nv40_gr_chan_fini(struct nvkm_object *object, bool suspend)
+nv40_gr_chan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nv40_gr_chan *chan = nv40_gr_chan(object);
 	struct nv40_gr *gr = chan->gr;
@@ -101,7 +101,7 @@ nv40_gr_chan_fini(struct nvkm_object *object, bool suspend)
 	nvkm_mask(device, 0x400720, 0x00000001, 0x00000000);
 
 	if (nvkm_rd32(device, 0x40032c) == inst) {
-		if (suspend) {
+		if (suspend != NVKM_POWEROFF) {
 			nvkm_wr32(device, 0x400720, 0x00000000);
 			nvkm_wr32(device, 0x400784, inst);
 			nvkm_mask(device, 0x400310, 0x00000020, 0x00000020);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv44.c b/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv44.c
index 4b1374adbda3..38146f9cc81c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv44.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/mpeg/nv44.c
@@ -65,7 +65,7 @@ nv44_mpeg_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
 }
 
 static int
-nv44_mpeg_chan_fini(struct nvkm_object *object, bool suspend)
+nv44_mpeg_chan_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 
 	struct nv44_mpeg_chan *chan = nv44_mpeg_chan(object);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/sec2/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/sec2/base.c
index f2c60da5d1e8..3e4d6a680ee9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/sec2/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/sec2/base.c
@@ -37,7 +37,7 @@ nvkm_sec2_finimsg(void *priv, struct nvfw_falcon_msg *hdr)
 }
 
 static int
-nvkm_sec2_fini(struct nvkm_engine *engine, bool suspend)
+nvkm_sec2_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_sec2 *sec2 = nvkm_sec2(engine);
 	struct nvkm_subdev *subdev = &sec2->engine.subdev;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/xtensa.c b/drivers/gpu/drm/nouveau/nvkm/engine/xtensa.c
index f7d3ba0afb55..910a5bb2d191 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/xtensa.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/xtensa.c
@@ -76,7 +76,7 @@ nvkm_xtensa_intr(struct nvkm_engine *engine)
 }
 
 static int
-nvkm_xtensa_fini(struct nvkm_engine *engine, bool suspend)
+nvkm_xtensa_fini(struct nvkm_engine *engine, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_xtensa *xtensa = nvkm_xtensa(engine);
 	struct nvkm_device *device = xtensa->engine.subdev.device;
@@ -85,7 +85,7 @@ nvkm_xtensa_fini(struct nvkm_engine *engine, bool suspend)
 	nvkm_wr32(device, base + 0xd84, 0); /* INTR_EN */
 	nvkm_wr32(device, base + 0xd94, 0); /* FIFO_CTRL */
 
-	if (!suspend)
+	if (suspend == NVKM_POWEROFF)
 		nvkm_memory_unref(&xtensa->gpu_fw);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c
index 9b8ca4e898f9..13d829593180 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c
@@ -182,7 +182,7 @@ nvkm_acr_managed_falcon(struct nvkm_device *device, enum nvkm_acr_lsf_id id)
 }
 
 static int
-nvkm_acr_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_acr_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	if (!subdev->use.enabled)
 		return 0;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c
index 91bc53be97ff..7dee55bf9ada 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bar/base.c
@@ -90,7 +90,7 @@ nvkm_bar_bar2_init(struct nvkm_device *device)
 }
 
 static int
-nvkm_bar_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_bar_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_bar *bar = nvkm_bar(subdev);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/base.c
index 178dc56909c2..71420f81714b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/base.c
@@ -577,7 +577,7 @@ nvkm_clk_read(struct nvkm_clk *clk, enum nv_clk_src src)
 }
 
 static int
-nvkm_clk_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_clk_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_clk *clk = nvkm_clk(subdev);
 	flush_work(&clk->work);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/base.c
index 3d9319c319c6..ad5ec9ee1294 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/devinit/base.c
@@ -67,11 +67,11 @@ nvkm_devinit_post(struct nvkm_devinit *init)
 }
 
 static int
-nvkm_devinit_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_devinit_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_devinit *init = nvkm_devinit(subdev);
 	/* force full reinit on resume */
-	if (suspend)
+	if (suspend != NVKM_POWEROFF)
 		init->post = true;
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c
index b53ac9a2552f..d8d32bb5bcd9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/base.c
@@ -51,7 +51,7 @@ nvkm_fault_intr(struct nvkm_subdev *subdev)
 }
 
 static int
-nvkm_fault_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_fault_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_fault *fault = nvkm_fault(subdev);
 	if (fault->func->fini)
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/user.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/user.c
index cd2fbc0472d8..8ab052d18e5d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fault/user.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fault/user.c
@@ -56,7 +56,7 @@ nvkm_ufault_map(struct nvkm_object *object, void *argv, u32 argc,
 }
 
 static int
-nvkm_ufault_fini(struct nvkm_object *object, bool suspend)
+nvkm_ufault_fini(struct nvkm_object *object, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_fault_buffer *buffer = nvkm_fault_buffer(object);
 	buffer->fault->func->buffer.fini(buffer);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gpio/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gpio/base.c
index b196baa376dc..b2c34878a68f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gpio/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gpio/base.c
@@ -144,7 +144,7 @@ nvkm_gpio_intr(struct nvkm_subdev *subdev)
 }
 
 static int
-nvkm_gpio_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_gpio_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_gpio *gpio = nvkm_gpio(subdev);
 	u32 mask = (1ULL << gpio->func->lines) - 1;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/base.c
index 7ccb41761066..30cb843ba35c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/base.c
@@ -48,7 +48,7 @@ nvkm_gsp_intr_stall(struct nvkm_gsp *gsp, enum nvkm_subdev_type type, int inst)
 }
 
 static int
-nvkm_gsp_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_gsp_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_gsp *gsp = nvkm_gsp(subdev);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c
index b0dd5fce7bad..88436a264177 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c
@@ -17,7 +17,7 @@
 #include <nvhw/ref/gh100/dev_riscv_pri.h>
 
 int
-gh100_gsp_fini(struct nvkm_gsp *gsp, bool suspend)
+gh100_gsp_fini(struct nvkm_gsp *gsp, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_falcon *falcon = &gsp->falcon;
 	int ret, time = 4000;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h
index 9dd66a2e3801..71b7203bef50 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h
@@ -59,7 +59,7 @@ struct nvkm_gsp_func {
 	void (*dtor)(struct nvkm_gsp *);
 	int (*oneinit)(struct nvkm_gsp *);
 	int (*init)(struct nvkm_gsp *);
-	int (*fini)(struct nvkm_gsp *, bool suspend);
+	int (*fini)(struct nvkm_gsp *, enum nvkm_suspend_state suspend);
 	int (*reset)(struct nvkm_gsp *);
 
 	struct {
@@ -75,7 +75,7 @@ int tu102_gsp_fwsec_sb_ctor(struct nvkm_gsp *);
 void tu102_gsp_fwsec_sb_dtor(struct nvkm_gsp *);
 int tu102_gsp_oneinit(struct nvkm_gsp *);
 int tu102_gsp_init(struct nvkm_gsp *);
-int tu102_gsp_fini(struct nvkm_gsp *, bool suspend);
+int tu102_gsp_fini(struct nvkm_gsp *, enum nvkm_suspend_state suspend);
 int tu102_gsp_reset(struct nvkm_gsp *);
 u64 tu102_gsp_wpr_heap_size(struct nvkm_gsp *);
 
@@ -87,12 +87,12 @@ int ga102_gsp_reset(struct nvkm_gsp *);
 
 int gh100_gsp_oneinit(struct nvkm_gsp *);
 int gh100_gsp_init(struct nvkm_gsp *);
-int gh100_gsp_fini(struct nvkm_gsp *, bool suspend);
+int gh100_gsp_fini(struct nvkm_gsp *, enum nvkm_suspend_state suspend);
 
 void r535_gsp_dtor(struct nvkm_gsp *);
 int r535_gsp_oneinit(struct nvkm_gsp *);
 int r535_gsp_init(struct nvkm_gsp *);
-int r535_gsp_fini(struct nvkm_gsp *, bool suspend);
+int r535_gsp_fini(struct nvkm_gsp *, enum nvkm_suspend_state suspend);
 
 int nvkm_gsp_new_(const struct nvkm_gsp_fwif *, struct nvkm_device *, enum nvkm_subdev_type, int,
 		  struct nvkm_gsp **);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
index 6e7af2f737b7..2f028a30e07d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
@@ -1721,7 +1721,7 @@ r535_gsp_sr_data_size(struct nvkm_gsp *gsp)
 }
 
 int
-r535_gsp_fini(struct nvkm_gsp *gsp, bool suspend)
+r535_gsp_fini(struct nvkm_gsp *gsp, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_rm *rm = gsp->rm;
 	int ret;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c
index 04b642a1f730..19cb269e7a26 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c
@@ -161,7 +161,7 @@ tu102_gsp_reset(struct nvkm_gsp *gsp)
 }
 
 int
-tu102_gsp_fini(struct nvkm_gsp *gsp, bool suspend)
+tu102_gsp_fini(struct nvkm_gsp *gsp, enum nvkm_suspend_state suspend)
 {
 	u32 mbox0 = 0xff, mbox1 = 0xff;
 	int ret;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/i2c/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/i2c/base.c
index 7ec17e8435a1..454bb21815a2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/i2c/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/i2c/base.c
@@ -135,7 +135,7 @@ nvkm_i2c_intr(struct nvkm_subdev *subdev)
 }
 
 static int
-nvkm_i2c_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_i2c_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_i2c *i2c = nvkm_i2c(subdev);
 	struct nvkm_i2c_pad *pad;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
index 2f55bab8e132..6b9ed61684a0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/instmem/base.c
@@ -176,7 +176,7 @@ nvkm_instmem_boot(struct nvkm_instmem *imem)
 }
 
 static int
-nvkm_instmem_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_instmem_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_instmem *imem = nvkm_instmem(subdev);
 	int ret;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
index 6867934256a7..0f3e0d324a52 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
@@ -74,7 +74,7 @@ nvkm_pci_rom_shadow(struct nvkm_pci *pci, bool shadow)
 }
 
 static int
-nvkm_pci_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_pci_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_pci *pci = nvkm_pci(subdev);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
index 8f2f50ad4ded..9e9004ec4588 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
@@ -77,7 +77,7 @@ nvkm_pmu_intr(struct nvkm_subdev *subdev)
 }
 
 static int
-nvkm_pmu_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_pmu_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_pmu *pmu = nvkm_pmu(subdev);
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c
index fc5ee118e910..1510aba33956 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/base.c
@@ -341,15 +341,15 @@ nvkm_therm_intr(struct nvkm_subdev *subdev)
 }
 
 static int
-nvkm_therm_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_therm_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_therm *therm = nvkm_therm(subdev);
 
 	if (therm->func->fini)
 		therm->func->fini(therm);
 
-	nvkm_therm_fan_fini(therm, suspend);
-	nvkm_therm_sensor_fini(therm, suspend);
+	nvkm_therm_fan_fini(therm, suspend != NVKM_POWEROFF);
+	nvkm_therm_sensor_fini(therm, suspend != NVKM_POWEROFF);
 
 	if (suspend) {
 		therm->suspend = therm->mode;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c
index 8b0da0c06268..a5c3c282b5d0 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/timer/base.c
@@ -149,7 +149,7 @@ nvkm_timer_intr(struct nvkm_subdev *subdev)
 }
 
 static int
-nvkm_timer_fini(struct nvkm_subdev *subdev, bool suspend)
+nvkm_timer_fini(struct nvkm_subdev *subdev, enum nvkm_suspend_state suspend)
 {
 	struct nvkm_timer *tmr = nvkm_timer(subdev);
 	tmr->func->alarm_fini(tmr);
-- 
2.52.0

[PATCH 3/3] nouveau/gsp: fix suspend/resume regression on r570 firmware

[Dave Airlie]

From: Dave Airlie <airlied@redhat.com>

The r570 firmware with certain GPUs (at least RTX6000) needs this
flag to reflect the suspend vs runtime PM state of the driver.

This uses that info to set the correct flags to the firmware.

This fixes a regression on RTX6000 and other GPUs since r570 firmware
was enabled.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Fixes: 53dac0623853 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org>
---
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c  | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/fbsr.c | 8 ++++----
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/rm.h        | 2 +-
 4 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c
index 150e22fde2ac..e962d0e8f837 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c
@@ -208,7 +208,7 @@ r535_fbsr_resume(struct nvkm_gsp *gsp)
 }
 
 static int
-r535_fbsr_suspend(struct nvkm_gsp *gsp)
+r535_fbsr_suspend(struct nvkm_gsp *gsp, bool runtime)
 {
 	struct nvkm_subdev *subdev = &gsp->subdev;
 	struct nvkm_device *device = subdev->device;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
index 2f028a30e07d..7fb13434c051 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
@@ -1748,7 +1748,7 @@ r535_gsp_fini(struct nvkm_gsp *gsp, enum nvkm_suspend_state suspend)
 		sr->sysmemAddrOfSuspendResumeData = gsp->sr.radix3.lvl0.addr;
 		sr->sizeOfSuspendResumeData = len;
 
-		ret = rm->api->fbsr->suspend(gsp);
+		ret = rm->api->fbsr->suspend(gsp, suspend == NVKM_RUNTIME_SUSPEND);
 		if (ret) {
 			nvkm_gsp_mem_dtor(&gsp->sr.meta);
 			nvkm_gsp_radix3_dtor(gsp, &gsp->sr.radix3);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/fbsr.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/fbsr.c
index 2945d5b4e570..8ef8b4f65588 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/fbsr.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r570/fbsr.c
@@ -62,7 +62,7 @@ r570_fbsr_resume(struct nvkm_gsp *gsp)
 }
 
 static int
-r570_fbsr_init(struct nvkm_gsp *gsp, struct sg_table *sgt, u64 size)
+r570_fbsr_init(struct nvkm_gsp *gsp, struct sg_table *sgt, u64 size, bool runtime)
 {
 	NV2080_CTRL_INTERNAL_FBSR_INIT_PARAMS *ctrl;
 	struct nvkm_gsp_object memlist;
@@ -81,7 +81,7 @@ r570_fbsr_init(struct nvkm_gsp *gsp, struct sg_table *sgt, u64 size)
 	ctrl->hClient = gsp->internal.client.object.handle;
 	ctrl->hSysMem = memlist.handle;
 	ctrl->sysmemAddrOfSuspendResumeData = gsp->sr.meta.addr;
-	ctrl->bEnteringGcoffState = 1;
+	ctrl->bEnteringGcoffState = runtime ? 1 : 0;
 
 	ret = nvkm_gsp_rm_ctrl_wr(&gsp->internal.device.subdevice, ctrl);
 	if (ret)
@@ -92,7 +92,7 @@ r570_fbsr_init(struct nvkm_gsp *gsp, struct sg_table *sgt, u64 size)
 }
 
 static int
-r570_fbsr_suspend(struct nvkm_gsp *gsp)
+r570_fbsr_suspend(struct nvkm_gsp *gsp, bool runtime)
 {
 	struct nvkm_subdev *subdev = &gsp->subdev;
 	struct nvkm_device *device = subdev->device;
@@ -133,7 +133,7 @@ r570_fbsr_suspend(struct nvkm_gsp *gsp)
 		return ret;
 
 	/* Initialise FBSR on RM. */
-	ret = r570_fbsr_init(gsp, &gsp->sr.fbsr, size);
+	ret = r570_fbsr_init(gsp, &gsp->sr.fbsr, size, runtime);
 	if (ret) {
 		nvkm_gsp_sg_free(device, &gsp->sr.fbsr);
 		return ret;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/rm.h b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/rm.h
index 393ea775941f..4f0ae6cc085c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/rm.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/rm.h
@@ -78,7 +78,7 @@ struct nvkm_rm_api {
 	} *device;
 
 	const struct nvkm_rm_api_fbsr {
-		int (*suspend)(struct nvkm_gsp *);
+		int (*suspend)(struct nvkm_gsp *, bool runtime);
 		void (*resume)(struct nvkm_gsp *);
 	} *fbsr;
 
-- 
2.52.0

Re: nouveau : fix r570 suspend/resume regression

[lyude]

For the whole series:

Reviewed-by: Lyude Paul <lyude@redhat.com>
Tested-by: Lyude Paul <lyude@redhat.com>

My power bill thanks you :)

On Tue, 2026-02-03 at 15:21 +1000, Dave Airlie wrote:
> Since R570 firmware got enabled, a number of GPUs have not being
> successfully suspend/resuming. Lyude showed it on RTX6000, which
> I reproduced and wasted a lot of time down various rabbit holes.
> 
> There are two required fixes, the first one adds proper sequence
> numbers to the rpc messages, which fixes a bunch of NOCAT asserts.
> 
> Then we have to pass the runtime vs non-runtime state down to the
> GSP fbsr code. This however requires replacing a bool with an enum
> which refactors quite a bunch of interfaces unfortunately, but it
> was the cleanest way to do it.
> 
> The final patch hooks up the interface so normal suspend doesn't
> set the GCOFF flags.
> 
> Dave.

Re: nouveau : fix r570 suspend/resume regression

[Dave Airlie]

This is now pushed to drm-misc-fixes.

Dave.

On Wed, 4 Feb 2026 at 06:08, <lyude@redhat.com> wrote:
>
> For the whole series:
>
> Reviewed-by: Lyude Paul <lyude@redhat.com>
> Tested-by: Lyude Paul <lyude@redhat.com>
>
> My power bill thanks you :)
>
> On Tue, 2026-02-03 at 15:21 +1000, Dave Airlie wrote:
> > Since R570 firmware got enabled, a number of GPUs have not being
> > successfully suspend/resuming. Lyude showed it on RTX6000, which
> > I reproduced and wasted a lot of time down various rabbit holes.
> >
> > There are two required fixes, the first one adds proper sequence
> > numbers to the rpc messages, which fixes a bunch of NOCAT asserts.
> >
> > Then we have to pass the runtime vs non-runtime state down to the
> > GSP fbsr code. This however requires replacing a bool with an enum
> > which refactors quite a bunch of interfaces unfortunately, but it
> > was the cleanest way to do it.
> >
> > The final patch hooks up the interface so normal suspend doesn't
> > set the GCOFF flags.
> >
> > Dave.
>

Re: [PATCH 1/3] nouveau/gsp: use rpc sequence numbers properly.

[Timur Tabi]

On Tue, 2026-02-03 at 15:21 +1000, Dave Airlie wrote:
> diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> index 2a7e80c6d70f..6e7af2f737b7 100644
> --- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> @@ -704,7 +704,7 @@ r535_gsp_rpc_set_registry(struct nvkm_gsp *gsp)
>  
>  	build_registry(gsp, rpc);
>  
> -	return nvkm_gsp_rpc_wr(gsp, rpc, NVKM_GSP_RPC_REPLY_NOWAIT);
> +	return nvkm_gsp_rpc_wr(gsp, rpc, NVKM_GSP_RPC_REPLY_NOSEQ);
>  
>  fail:
>  	clean_registry(gsp);

So are you saying that some RPC commands need to have a sequence number set, and some do not?

Re: [PATCH 1/3] nouveau/gsp: use rpc sequence numbers properly.

[Dave Airlie]

On Thu, 5 Feb 2026 at 07:36, Timur Tabi <ttabi@nvidia.com> wrote:
>
> On Tue, 2026-02-03 at 15:21 +1000, Dave Airlie wrote:
> > diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> > b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> > index 2a7e80c6d70f..6e7af2f737b7 100644
> > --- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> > +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
> > @@ -704,7 +704,7 @@ r535_gsp_rpc_set_registry(struct nvkm_gsp *gsp)
> >
> >       build_registry(gsp, rpc);
> >
> > -     return nvkm_gsp_rpc_wr(gsp, rpc, NVKM_GSP_RPC_REPLY_NOWAIT);
> > +     return nvkm_gsp_rpc_wr(gsp, rpc, NVKM_GSP_RPC_REPLY_NOSEQ);
> >
> >  fail:
> >       clean_registry(gsp);
>
> So are you saying that some RPC commands need to have a sequence number set, and some do not?

I'm copying the behaviour of opengpu here,
src/kernel/gpu/gsp/kernel_gsp.c
if (pSequence)
        vgpu_rpc_message_header_v->sequence = *pSequence = pRpc->sequence++;
    else
        vgpu_rpc_message_header_v->sequence = 0;

src/kernel/vgpu/rpc.c:_issueRpcAsync
doesn't pass pSequence
_issueRpcAndWait does pass it.

The SetSystemInfo and SetRegistry are the two async calls in nouveau.

So I'm not saying some RPC commands need to have a sequence number and
some don't, that would be up to someone who can access GSP source
code, I'm saying that opengpu does this and I'm copying it :-)

Dave.

Re: [PATCH 1/3] nouveau/gsp: use rpc sequence numbers properly.

[John Hubbard]

On 2/4/26 1:56 PM, Dave Airlie wrote:
> On Thu, 5 Feb 2026 at 07:36, Timur Tabi <ttabi@nvidia.com> wrote:
...
>> So are you saying that some RPC commands need to have a sequence number set, and some do not?
> 
> I'm copying the behaviour of opengpu here,
> src/kernel/gpu/gsp/kernel_gsp.c
> if (pSequence)
>         vgpu_rpc_message_header_v->sequence = *pSequence = pRpc->sequence++;
>     else
>         vgpu_rpc_message_header_v->sequence = 0;
> 
> src/kernel/vgpu/rpc.c:_issueRpcAsync
> doesn't pass pSequence
> _issueRpcAndWait does pass it.
> 
> The SetSystemInfo and SetRegistry are the two async calls in nouveau.
> 
> So I'm not saying some RPC commands need to have a sequence number and
> some don't, that would be up to someone who can access GSP source
> code, I'm saying that opengpu does this and I'm copying it :-)
> 

lol I love it. If only someone here had access to the GSP code and
the GSP team! :)

OK, then, I just had a very enlightening call with one of our GSP
RPC experts, in order to learn what the sequence number story really
is. The notes below are sort of Nova-centric, but it should apply
equally to Nouveau.

Let me Cc Eliot, because he was also fixing up other aspects
of GSP RPC calls on Nova, in case this overlaps.

==========================================================
Background
==========

Today there are some loose ends and inconsistencies even in the
Open RM + GSP scenario, for how sequence numbers are used. And these
are being cleaned up and fixed. In fact, I was even able to request,
and receive some nice clean behavior, which will be implemented in
GSP soon (we'll get it when we upgrade, likely sometime this year).

Today, there are 2+ sequence number spaces, one for send-receive pairs
(command + response) RPC calls, and another for GSP-initiated ("async")
messages to the CPU.

The "2+" is because there is an inconsistency (it will be fixed in
GSP), leading to the first two very early RPC calls being in yet another
unaccounted for number space. These:

NovaCore 0000:01:00.0: GSP RPC: send: seq# 0, function=GSP_SET_SYSTEM_INFO, length=0x3f0
NovaCore 0000:01:00.0: GSP RPC: send: seq# 1, function=SET_REGISTRY, length=0xc5

...are not included in the counting, by GSP.

The GSP finally starts counting up when it gets the first "non-async"
(command/response) message, here:

NovaCore 0000:01:00.0: GSP RPC: send: seq# 2, function=GET_GSP_STATIC_INFO, length=0x6c8
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GetGspStaticInfo), length=0x6c8


But even here, it's not what I think we want, because we want the CPU to
get back the same seq num that it sent, for command/response pairs. But
that's not what actually happens (at least not directly).

So for now, seq numbers on Nova and Nouveau are generally "do what
you want, it will work ok". But actually, we will soon be able to
use them for

    a) debugging aids,
    b) detecting missing messages, and
    c) recovering from "CPU sent a message, timed out waiting for
       a response" cases.

==========================================================
Next steps for Nova (and maybe Nouveau, if not already done)
============================================================

a) Change debug output to print the seq number numeric space,
which is either "async message from GSP" or "command response
from GSP".

b) Put a comment in the code to indicate that GSP_SET_SYSTEM_INFO
and SET_REGISTRY do not yet participate in the incrementing seq
number system, but will in future GSP versions.

I'll send patches for Nova soon, to do the above.

==========================================================
Examples to illustrate:
=======================

Today's Nova debug logs on Blackwell:

NovaCore 0000:01:00.0: GSP RPC: send: seq# 0, function=GSP_SET_SYSTEM_INFO, length=0x3f0
NovaCore 0000:01:00.0: GSP RPC: send: seq# 1, function=SET_REGISTRY, length=0xc5
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GspLockdownNotice), length=0x51
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GspLockdownNotice), length=0x51
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(UcodeLibOsPrint), length=0x68
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(UcodeLibOsPrint), length=0x70
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GspPostNoCat), length=0x50c
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GspLockdownNotice), length=0x51
...more of these...
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GspInitDone), length=0x50
NovaCore 0000:01:00.0: GSP RPC: send: seq# 2, function=GET_GSP_STATIC_INFO, length=0x6c8
NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GetGspStaticInfo), length=0x6c8

Future (after changing Nova first, and eventually upgrading GSP):

NovaCore 0000:01:00.0: GSP RPC: send: seq# 0, function=GSP_SET_SYSTEM_INFO, length=0x3f0
NovaCore 0000:01:00.0: GSP RPC: send: seq# 1, function=SET_REGISTRY, length=0xc5
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 0, function=Ok(GspLockdownNotice), length=0x51
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 1, function=Ok(GspLockdownNotice), length=0x51
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 2, function=Ok(UcodeLibOsPrint), length=0x68
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 3, function=Ok(UcodeLibOsPrint), length=0x70
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 4, function=Ok(GspPostNoCat), length=0x50c
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 5 function=Ok(GspLockdownNotice), length=0x51
...more of these...
NovaCore 0000:01:00.0: GSP RPC: async received: seq# 12, function=Ok(GspInitDone), length=0x50
NovaCore 0000:01:00.0: GSP RPC: send: seq# 2, function=GET_GSP_STATIC_INFO, length=0x6c8
NovaCore 0000:01:00.0: GSP RPC: response received: seq# 2, function=Ok(GetGspStaticInfo), length=0x6c8

As you can see, the CPU side can then track the two types of messages
clearly.

thanks,
-- 
John Hubbard

Re: [PATCH 1/3] nouveau/gsp: use rpc sequence numbers properly.

[Dave Airlie]

On Thu, 5 Feb 2026 at 13:01, John Hubbard <jhubbard@nvidia.com> wrote:
>
> On 2/4/26 1:56 PM, Dave Airlie wrote:
> > On Thu, 5 Feb 2026 at 07:36, Timur Tabi <ttabi@nvidia.com> wrote:
> ...
> >> So are you saying that some RPC commands need to have a sequence number set, and some do not?
> >
> > I'm copying the behaviour of opengpu here,
> > src/kernel/gpu/gsp/kernel_gsp.c
> > if (pSequence)
> >         vgpu_rpc_message_header_v->sequence = *pSequence = pRpc->sequence++;
> >     else
> >         vgpu_rpc_message_header_v->sequence = 0;
> >
> > src/kernel/vgpu/rpc.c:_issueRpcAsync
> > doesn't pass pSequence
> > _issueRpcAndWait does pass it.
> >
> > The SetSystemInfo and SetRegistry are the two async calls in nouveau.
> >
> > So I'm not saying some RPC commands need to have a sequence number and
> > some don't, that would be up to someone who can access GSP source
> > code, I'm saying that opengpu does this and I'm copying it :-)
> >
>
> lol I love it. If only someone here had access to the GSP code and
> the GSP team! :)
>
> OK, then, I just had a very enlightening call with one of our GSP
> RPC experts, in order to learn what the sequence number story really
> is. The notes below are sort of Nova-centric, but it should apply
> equally to Nouveau.
>
> Let me Cc Eliot, because he was also fixing up other aspects
> of GSP RPC calls on Nova, in case this overlaps.
>
> ==========================================================
> Background
> ==========
>
> Today there are some loose ends and inconsistencies even in the
> Open RM + GSP scenario, for how sequence numbers are used. And these
> are being cleaned up and fixed. In fact, I was even able to request,
> and receive some nice clean behavior, which will be implemented in
> GSP soon (we'll get it when we upgrade, likely sometime this year).
>
> Today, there are 2+ sequence number spaces, one for send-receive pairs
> (command + response) RPC calls, and another for GSP-initiated ("async")
> messages to the CPU.
>
> The "2+" is because there is an inconsistency (it will be fixed in
> GSP), leading to the first two very early RPC calls being in yet another
> unaccounted for number space. These:
>
> NovaCore 0000:01:00.0: GSP RPC: send: seq# 0, function=GSP_SET_SYSTEM_INFO, length=0x3f0
> NovaCore 0000:01:00.0: GSP RPC: send: seq# 1, function=SET_REGISTRY, length=0xc5
>
> ...are not included in the counting, by GSP.
>
> The GSP finally starts counting up when it gets the first "non-async"
> (command/response) message, here:
>
> NovaCore 0000:01:00.0: GSP RPC: send: seq# 2, function=GET_GSP_STATIC_INFO, length=0x6c8
> NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GetGspStaticInfo), length=0x6c8
>
>
> But even here, it's not what I think we want, because we want the CPU to
> get back the same seq num that it sent, for command/response pairs. But
> that's not what actually happens (at least not directly).
>
> So for now, seq numbers on Nova and Nouveau are generally "do what
> you want, it will work ok". But actually, we will soon be able to
> use them for
>
>     a) debugging aids,
>     b) detecting missing messages, and
>     c) recovering from "CPU sent a message, timed out waiting for
>        a response" cases.
>
> ==========================================================
> Next steps for Nova (and maybe Nouveau, if not already done)
> ============================================================
>
> a) Change debug output to print the seq number numeric space,
> which is either "async message from GSP" or "command response
> from GSP".
>
> b) Put a comment in the code to indicate that GSP_SET_SYSTEM_INFO
> and SET_REGISTRY do not yet participate in the incrementing seq
> number system, but will in future GSP versions.
>
> I'll send patches for Nova soon, to do the above.

So just FYI nova gets this wrong as well.

There are two sequence numbers:

the one in GSP_MSG_QUEUE_ELEMENT and the one in rpc_message_header_v

The rules for the first seem to be just increment, the rules for the
2nd seems to be along the lines above, two msgs get 0, then start from
0.

Currently we never init the rpc message header sequence to anything but 0.

Dave.

Re: [PATCH 1/3] nouveau/gsp: use rpc sequence numbers properly.

[John Hubbard]

On 2/4/26 8:37 PM, Dave Airlie wrote:
> On Thu, 5 Feb 2026 at 13:01, John Hubbard <jhubbard@nvidia.com> wrote:
>> On 2/4/26 1:56 PM, Dave Airlie wrote:
>>> On Thu, 5 Feb 2026 at 07:36, Timur Tabi <ttabi@nvidia.com> wrote:
>> ...
>>>> So are you saying that some RPC commands need to have a sequence number set, and some do not?
>>>
>>> I'm copying the behaviour of opengpu here,
>>> src/kernel/gpu/gsp/kernel_gsp.c
>>> if (pSequence)
>>>          vgpu_rpc_message_header_v->sequence = *pSequence = pRpc->sequence++;
>>>      else
>>>          vgpu_rpc_message_header_v->sequence = 0;
>>>
>>> src/kernel/vgpu/rpc.c:_issueRpcAsync
>>> doesn't pass pSequence
>>> _issueRpcAndWait does pass it.
>>>
>>> The SetSystemInfo and SetRegistry are the two async calls in nouveau.
>>>
>>> So I'm not saying some RPC commands need to have a sequence number and
>>> some don't, that would be up to someone who can access GSP source
>>> code, I'm saying that opengpu does this and I'm copying it :-)
>>>
>>
>> lol I love it. If only someone here had access to the GSP code and
>> the GSP team! :)
>>
>> OK, then, I just had a very enlightening call with one of our GSP
>> RPC experts, in order to learn what the sequence number story really
>> is. The notes below are sort of Nova-centric, but it should apply
>> equally to Nouveau.
>>
>> Let me Cc Eliot, because he was also fixing up other aspects
>> of GSP RPC calls on Nova, in case this overlaps.
>>
>> ==========================================================
>> Background
>> ==========
>>
>> Today there are some loose ends and inconsistencies even in the
>> Open RM + GSP scenario, for how sequence numbers are used. And these
>> are being cleaned up and fixed. In fact, I was even able to request,
>> and receive some nice clean behavior, which will be implemented in
>> GSP soon (we'll get it when we upgrade, likely sometime this year).
>>
>> Today, there are 2+ sequence number spaces, one for send-receive pairs
>> (command + response) RPC calls, and another for GSP-initiated ("async")
>> messages to the CPU.
>>
>> The "2+" is because there is an inconsistency (it will be fixed in
>> GSP), leading to the first two very early RPC calls being in yet another
>> unaccounted for number space. These:
>>
>> NovaCore 0000:01:00.0: GSP RPC: send: seq# 0, function=GSP_SET_SYSTEM_INFO, length=0x3f0
>> NovaCore 0000:01:00.0: GSP RPC: send: seq# 1, function=SET_REGISTRY, length=0xc5
>>
>> ...are not included in the counting, by GSP.
>>
>> The GSP finally starts counting up when it gets the first "non-async"
>> (command/response) message, here:
>>
>> NovaCore 0000:01:00.0: GSP RPC: send: seq# 2, function=GET_GSP_STATIC_INFO, length=0x6c8
>> NovaCore 0000:01:00.0: GSP RPC: receive: seq# 0, function=Ok(GetGspStaticInfo), length=0x6c8
>>
>>
>> But even here, it's not what I think we want, because we want the CPU to
>> get back the same seq num that it sent, for command/response pairs. But
>> that's not what actually happens (at least not directly).
>>
>> So for now, seq numbers on Nova and Nouveau are generally "do what
>> you want, it will work ok". But actually, we will soon be able to
>> use them for
>>
>>      a) debugging aids,
>>      b) detecting missing messages, and
>>      c) recovering from "CPU sent a message, timed out waiting for
>>         a response" cases.
>>
>> ==========================================================
>> Next steps for Nova (and maybe Nouveau, if not already done)
>> ============================================================
>>
>> a) Change debug output to print the seq number numeric space,
>> which is either "async message from GSP" or "command response
>> from GSP".
>>
>> b) Put a comment in the code to indicate that GSP_SET_SYSTEM_INFO
>> and SET_REGISTRY do not yet participate in the incrementing seq
>> number system, but will in future GSP versions.
>>
>> I'll send patches for Nova soon, to do the above.
> 
> So just FYI nova gets this wrong as well.
> 
> There are two sequence numbers:
> 
> the one in GSP_MSG_QUEUE_ELEMENT and the one in rpc_message_header_v
> 
> The rules for the first seem to be just increment, the rules for the
> 2nd seems to be along the lines above, two msgs get 0, then start from
> 0.
> 
> Currently we never init the rpc message header sequence to anything but 0.
> 

Yes. That's accurate. To be fixed.

thanks,
-- 
John Hubbard