On Mon Feb 9, 2026 at 11:40 PM CET, Timur Tabi wrote:
> GA100 is a compute-only variant of GA102 that boots GSP-RM like a Turing.
>
> Although architecturally like an Ampere, GA100 uses the same GSP-RM
> firmware files as Turing, and therefore must boot it like Turing does.
> In addition, as a compute-only part, GA100 has no display engine,
> no VBIOS, and no graphics (GR) engine.
>
> Since it doesn't have VBIOS, there is no FWSEC firmware, and so there is
> no FRTS region that needs to be reserved or initialized.
>
> Although Nouveau now supports GA100, there is currently no real support
> for this GPU with either Clover or Rusticl/NVK, as both require a 3D
> engine. This is in contrast with GH100, which does contain a GR engine
> and is supported.
>
> Also note that this patch partially reverts commit e8b3627bec35 ("nouveau:
> don't attempt fwsec on sb on newer platforms."), which added fwsec
> pointers to struct ga100_gsp erroneously.
Can you please also explain the consequences of this? It also very much sounds
like the patch needs a Fixes: tag?
Also can you please use scripts/get_maintainer.pl and send the patch to all
maintainers?
> @@ -343,9 +348,12 @@ tu102_gsp_oneinit(struct nvkm_gsp *gsp)
> if (ret)
> return ret;
>
> - ret = nvkm_gsp_fwsec_frts(gsp);
> - if (WARN_ON(ret))
> - return ret;
> + /* Only boot FWSEC-FRTS if it actually exists */
> + if (gsp->fb.wpr2.frts.size) {
> + ret = nvkm_gsp_fwsec_frts(gsp);
> + if (WARN_ON(ret))
Since we are about to change this, I don't think this should be a WARN_ON().
Most of the error paths within nvkm_gsp_fwsec_frts() already print an error
message with nvkm_error() or use WARN_ON() themselves.
Besides that, I don't think WARN_ON() is the correct tool for this, as we are
not violating some invariant when this fails, it's just a normal runtime error.
Thanks,
Danilo
