Skip to contentSkip to conversation

drm/nouveau: Remove DRM_NOUVEAU_GSP_DEFAULT config

PATCH Superseded Mel Henning 3 participants 10 messages Jun 23, 2025 – Jul 14, 2025

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover drm/nouveau: Remove DRM_NOUVEAU_GSP_DEFAULT config

Following earlier discussion at
https://lists.freedesktop.org/archives/nouveau/2025-June/047887.html
this series removes DRM_NOUVEAU_GSP_DEFAULT.

Mel Henning (2):
drm/nouveau: Remove DRM_NOUVEAU_GSP_DEFAULT config drm/nouveau: Remove nvkm_gsp_fwif.enable

 drivers/gpu/drm/nouveau/Kconfig                 | 8 --------
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/ad102.c | 4 ++--
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gb100.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gb202.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h  | 1 -
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c | 6 +-----
 7 files changed, 6 insertions(+), 19 deletions(-)
Signature · 1 line
2.50.0
#1 lore raw

PATCH 1/2 drm/nouveau: Remove DRM_NOUVEAU_GSP_DEFAULT config

This option was originally intoduced because the GSP code path was not well tested and we wanted to leave it up to distros which code path they shipped by default. By now though, the GSP path is probably better tested than the old firmware eg. Fedora ships GSP by default and we generally run CTS on GSP. We've always been GSP-only on Ada and later.

So, this path removes the option and effectively sets the option to always on. We still fall back to the old firmware if GSP is not found. This change only affects Turing and Ampere.

Users can still set nouveau.config=NvGspRm=0 on the kernel command line to force using the old firmware on Turing/Ampere.

---
 drivers/gpu/drm/nouveau/Kconfig                 | 8 --------
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c | 6 +-----
 2 files changed, 1 insertion(+), 13 deletions(-)
drivers/gpu/drm/nouveau/Kconfig +0 -8
index d1587639ebb0..c88776d1e784 100644
@@ -102,14 +102,6 @@ config DRM_NOUVEAU_SVM
 	  Say Y here if you want to enable experimental support for
 	  Shared Virtual Memory (SVM).
 
-config DRM_NOUVEAU_GSP_DEFAULT
-	bool "Use GSP firmware for Turing/Ampere (needs firmware installed)"
-	depends on DRM_NOUVEAU
-	default n
-	help
-	  Say Y here if you want to use the GSP codepaths by default on
-	  Turing and Ampere GPUs.
-
 config DRM_NOUVEAU_CH7006
 	tristate "Chrontel ch7006 TV encoder"
 	depends on DRM_NOUVEAU
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/tu102.c +1 -5
index 58e233bc53b1..81e56da0474a 100644
@@ -383,13 +383,9 @@ int
 tu102_gsp_load_rm(struct nvkm_gsp *gsp, const struct nvkm_gsp_fwif *fwif)
 {
 	struct nvkm_subdev *subdev = &gsp->subdev;
-	bool enable_gsp = fwif->enable;
 	int ret;
 
-#if IS_ENABLED(CONFIG_DRM_NOUVEAU_GSP_DEFAULT)
-	enable_gsp = true;
-#endif
-	if (!nvkm_boolopt(subdev->device->cfgopt, "NvGspRm", enable_gsp))
+	if (!nvkm_boolopt(subdev->device->cfgopt, "NvGspRm", true))
 		return -EINVAL;
 
 	ret = nvkm_gsp_load_fw(gsp, "gsp", fwif->ver, &gsp->fws.rm);
Signature · 1 line
2.50.0
#2 lore raw

PATCH 2/2 drm/nouveau: Remove nvkm_gsp_fwif.enable

In reply to Mel Henning

This struct element is no longer used.

---
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/ad102.c | 4 ++--
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gb100.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gb202.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c | 2 +-
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h  | 1 -
 5 files changed, 5 insertions(+), 6 deletions(-)
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/ad102.c +2 -2
index eb765da0876e..35d1fcef520b 100644
@@ -41,8 +41,8 @@ ad102_gsp = {
 
 static struct nvkm_gsp_fwif
 ad102_gsps[] = {
-	{ 1, tu102_gsp_load, &ad102_gsp, &r570_rm_ga102, "570.144", true },
-	{ 0, tu102_gsp_load, &ad102_gsp, &r535_rm_ga102, "535.113.01", true },
+	{ 1, tu102_gsp_load, &ad102_gsp, &r570_rm_ga102, "570.144" },
+	{ 0, tu102_gsp_load, &ad102_gsp, &r535_rm_ga102, "535.113.01" },
 	{}
 };
 
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gb100.c +1 -1
index 12a3f2c1ed82..1b3b31b95ce4 100644
@@ -20,7 +20,7 @@ gb100_gsp = {
 
 static struct nvkm_gsp_fwif
 gb100_gsps[] = {
-	{ 0, gh100_gsp_load, &gb100_gsp, &r570_rm_gb10x, "570.144", true },
+	{ 0, gh100_gsp_load, &gb100_gsp, &r570_rm_gb10x, "570.144" },
 	{}
 };
 
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gb202.c +1 -1
index c1d718172ddf..51384c63148c 100644
@@ -20,7 +20,7 @@ gb202_gsp = {
 
 static struct nvkm_gsp_fwif
 gb202_gsps[] = {
-	{ 0, gh100_gsp_load, &gb202_gsp, &r570_rm_gb20x, "570.144", true },
+	{ 0, gh100_gsp_load, &gb202_gsp, &r570_rm_gb20x, "570.144" },
 	{}
 };
 
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/gh100.c +1 -1
index ce31e8248807..b0dd5fce7bad 100644
@@ -344,7 +344,7 @@ gh100_gsp_load(struct nvkm_gsp *gsp, int ver, const struct nvkm_gsp_fwif *fwif)
 
 static struct nvkm_gsp_fwif
 gh100_gsps[] = {
-	{ 0, gh100_gsp_load, &gh100_gsp, &r570_rm_gh100, "570.144", true },
+	{ 0, gh100_gsp_load, &gh100_gsp, &r570_rm_gh100, "570.144" },
 	{}
 };
 
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h +0 -1
index 4f14e85fc69e..c3494b7ac572 100644
@@ -14,7 +14,6 @@ struct nvkm_gsp_fwif {
 	const struct nvkm_gsp_func *func;
 	const struct nvkm_rm_impl *rm;
 	const char *ver;
-	bool enable;
 };
 
 int nvkm_gsp_load_fw(struct nvkm_gsp *, const char *name, const char *ver,
Signature · 1 line
2.50.0
#3 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h struct nvkm_gsp_fwif { · line 17

Quoted 11 lines · Mel Henning
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/priv.h +0 -1
index 4f14e85fc69e..c3494b7ac572 100644
@@ -14,7 +14,6 @@ struct nvkm_gsp_fwif {
  	const struct nvkm_gsp_func *func;
  	const struct nvkm_rm_impl *rm;
  	const char *ver;
-	bool enable;
  };

Instead of removing it, I think we should rename it to indicate whether GSP-RM is required. You cannot boot Ada or later without GSP-RM, so on those platforms, nouveau.config=NvGspRm=0 should be ignored, and a pr_warn should be issued that it is being ignored.

author
#4 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

Quoted 3 lines · Timur Tabi

Instead of removing it, I think we should rename it to indicate whether GSP-RM is required. You cannot boot Ada or later without GSP-RM, so on those platforms, nouveau.config=NvGspRm=0 should be ignored, and a pr_warn should be issued that it is being ignored.

We did no such error checking before this series (in fact, most of these options have almost no error checking). Are you saying you want to see this added in this patch series?

#5 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

Quoted 3 lines · M Henning

We did no such error checking before this series (in fact, most of these options have almost no error checking). Are you saying you want to see this added in this patch series?

You have a good point, but I think your change, in effect, necessitates my request. Previously, the default was no GSP-RM unless needed. Now it's yes GSP-RM, and the concept of "need" has been removed. So there's no indication any more that some GPUs need GSP-RM and some do not.

So to address that, I think it makes sense to add a warning if someone tries disable GSP-RM on a GPU that is not supported in that configuration.

Now, whether or not we should ignore NvGspRm=0 on Ada+ is up for debate. If I understand the code correctly, today (and still with your patches), Ada+ would fail to boot. I can't say whether or not that's a good idea. But I think a warning should be printed either way.

author
#6 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

Quoted 10 lines · Timur Tabi

You have a good point, but I think your change, in effect, necessitates my request. Previously, the default was no GSP-RM unless needed. Now it's yes GSP-RM, and the concept of "need" has been removed. So there's no indication any more that some GPUs need GSP-RM and some do not.

So to address that, I think it makes sense to add a warning if someone tries disable GSP-RM on a GPU that is not supported in that configuration.

Now, whether or not we should ignore NvGspRm=0 on Ada+ is up for debate. If I understand the code correctly, today (and still with your patches), Ada+ would fail to boot. I can't say whether or not that's a good idea. But I think a warning should be printed either way.

This patch behaves exactly the same as DRM_NOUVEAU_GSP_DEFAULT=y kernels already behave.

That being said, I'm not against the additional error checking here and can add it to the next version of this series.

Reviewed-by
#7 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

Quoted 8 lines · M Henning
Quoted 7 lines · Timur Tabi

You have a good point, but I think your change, in effect, necessitates my request. Previously, the default was no GSP-RM unless needed. Now it's yes GSP-RM, and the concept of "need" has been removed. So there's no indication any more that some GPUs need GSP-RM and some do not.

So to address that, I think it makes sense to add a warning if someone tries disable GSP-RM on a GPU that is not supported in that configuration.

Quoted 8 lines
Quoted 3 lines

Now, whether or not we should ignore NvGspRm=0 on Ada+ is up for debate. If I understand the code correctly, today (and still with your patches), Ada+ would fail to boot. I can't say whether or not that's a good idea. But I think a warning should be printed either way.

This patch behaves exactly the same as DRM_NOUVEAU_GSP_DEFAULT=y kernels already behave.

That being said, I'm not against the additional error checking here and can add it to the next version of this series.

Yeah, the GPUs that don't support GSP-RM can't hit the code that used fwif.enable anyway, so the series should be fine as it is.

Feel free to add my:

author
#8 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

Quoted 2 lines · Ben Skeggs

Yeah, the GPUs that don't support GSP-RM can't hit the code that used fwif.enable anyway, so the series should be fine as it is.

We're actually talking about the reverse case. That is, on Ada what happens if you set nouveau.config=NvGspRm=0 on the kernel command line?

#9 lore raw

Re: [PATCH 2/2] drm/nouveau: Remove nvkm_gsp_fwif.enable

Quoted 6 lines · M Henning
Quoted 2 lines · Ben Skeggs

Yeah, the GPUs that don't support GSP-RM can't hit the code that used fwif.enable anyway, so the series should be fine as it is.

We're actually talking about the reverse case. That is, on Ada what happens if you set nouveau.config=NvGspRm=0 on the kernel command line?

The last entry in fwif[] for GPUs prior to Ada is one that always succeeds, and sets up some minimal GSP stuff that's needed for booting the pre-GSP-RM firmwares.  Ada onwards lack this, and the driver will cleanly fail to load as if no firmware were present.

The error message could be better for sure, and you can just add an nvkm_error(&gsp->subdev, ...) after the attempt to call nvkm_firmware_load() in subdev/gsp/base.c.

This should work fine without needing to add any kind of "needs GSP-RM" field anywhere.

Ben.