Re: [PATCH v3 1/5] gpu: nova-core: correct FRTS vidmem offset calculation

On Sat Jul 25, 2026 at 1:52 AM JST, Timur Tabi wrote:
> On Fri, 2026-07-24 at 15:56 +0900, Eliot Courtney wrote:
>> +++ b/drivers/gpu/nova-core/fsp/hal.rs
>> @@ -19,6 +19,10 @@ pub(super) trait FspHal {
>> =C2=A0
>> =C2=A0=C2=A0=C2=A0=C2=A0 /// Returns the FSP Chain of Trust protocol ver=
sion this chipset advertises.
>> =C2=A0=C2=A0=C2=A0=C2=A0 fn cot_version(&self) -> u16;
>> +
>> +=C2=A0=C2=A0=C2=A0 // TODO: consider moving this into the TLV firmware =
metadata when ready
>> +=C2=A0=C2=A0=C2=A0 /// Returns the size reserved at the end of the fram=
ebuffer, in bytes.
>> +=C2=A0=C2=A0=C2=A0 fn fb_end_reserved_size(&self) -> u32;
>
> Do you know where in OpenRM this value is encoded?
>
> There are quite a few places in the code, especially for GA100 stuff, tha=
t are arbitrary and GPU-
> specific.  Without a means to scrape that data from the OpenRM source, it=
 will be impossible for the
> script to obtain the values.

Yerp, for before blackwell it's here [1]. For blackwell it's here [2].
According to [3] it seems to be able to vary based on openrm version and
it's required to boot GSP, so that's why I think TLV is a good spot for
it.

[1]: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/452cec62d827034=
798072827d3866d1881662b77/src/nvidia/src/kernel/gpu/mem_mgr/arch/maxwell/me=
m_mgr_gm107.c#L1829
[2]: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/452cec62d827034=
798072827d3866d1881662b77/src/nvidia/src/kernel/gpu/mem_mgr/arch/blackwell/=
mem_mgr_gb100.c#L64
[3]: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/452cec62d827034=
798072827d3866d1881662b77/src/nvidia/src/kernel/gpu/fsp/arch/hopper/kern_fs=
p_gh100.c#L1399