drivers/gpu/nova-core/falcon.rs | 315 ++++++++++++++++--- drivers/gpu/nova-core/falcon/hal.rs | 6 +- drivers/gpu/nova-core/firmware.rs | 107 ++++--- drivers/gpu/nova-core/firmware/booter.rs | 65 ++-- drivers/gpu/nova-core/firmware/fwsec.rs | 129 +++----- drivers/gpu/nova-core/firmware/fwsec/bootloader.rs | 348 +++++++++++++++++++++ drivers/gpu/nova-core/gpu.rs | 9 +- drivers/gpu/nova-core/gsp/boot.rs | 17 +- drivers/gpu/nova-core/regs.rs | 30 ++ 9 files changed, 820 insertions(+), 206 deletions(-)
This patchset adds the remaining support required for booting the GSP on
Turing.
We did a deep dive with Eliot looking for the reasons why some fields
involved in the bootloader are ignored or used apparently
inconsistently, and this results in a more documented flow and a few
fixes. Apart from that, this series seems to be stabilizing and
successfully probes my TU106:
NovaCore 0000:08:00.0: NVIDIA (Chipset: TU106, Architecture: Turing, Revision: a.1)
NovaCore 0000:08:00.0: GPU name: NVIDIA GeForce RTX 2070
This series is based on `drm-rust-next`. A tree with all the patches is
available at [1].
[1] https://github.com/Gnurou/linux/tree/b4/turing
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Changes in v11:
- Fix build error/warnings and rustfmt formatting.
- Address incorrect IMEM section start offsets in FalconUCodeDescV2
and better document fields usage and unused fields.
- Use `get`/`get_mut` instead of direct array indexing when accessing
firmware content.
- Link to v10: https://patch.msgid.link/20260301-turing_prep-v10-0-dde5ee437c60@nvidia.com
Changes in v10:
- Store the firmwares into a regular KVec and move them into a DMA
object only when actually loading using DMA.
- Use `try_update` when updating the `NV_PFALCON_FBIF_TRANSCFG` register
array as its index is not build-time proven to be valid.
- Fix alignment issue when processing imem section of the FWSEC
bootloader (thanks Eliot!).
- Link to v9: https://patch.msgid.link/20260212-turing_prep-v9-0-238520ad8799@nvidia.com
Changes in v9:
- Add a few preparatory patches to simplify the actual feature patches.
- Use a wrapping type for the bootloader.
- Simplify the falcon loading code and move the complexity to the
firmware types.
- Add the generic bootloader files to `ModInfoBuilder`.
- Link to v8: https://lore.kernel.org/all/20260122222848.2555890-1-ttabi@nvidia.com/
---
Alexandre Courbot (10):
gpu: nova-core: create falcon firmware DMA objects lazily
gpu: nova-core: falcon: add constant for memory block alignment
gpu: nova-core: falcon: rename load parameters to reflect DMA dependency
gpu: nova-core: falcon: remove FalconFirmware's dependency on FalconDmaLoadable
gpu: nova-core: move brom_params and boot_addr to FalconFirmware
gpu: nova-core: falcon: remove unwarranted safety check in dma_load
gpu: nova-core: firmware: add comments to justify v3 header values
gpu: nova-core: firmware: fix and explain v2 header offsets computations
gpu: nova-core: make Chipset::arch() const
gpu: nova-core: add gen_bootloader firmware to ModInfoBuilder
Timur Tabi (2):
gpu: nova-core: add PIO support for loading firmware images
gpu: nova-core: use the Generic Bootloader to boot FWSEC on Turing
drivers/gpu/nova-core/falcon.rs | 315 ++++++++++++++++---
drivers/gpu/nova-core/falcon/hal.rs | 6 +-
drivers/gpu/nova-core/firmware.rs | 107 ++++---
drivers/gpu/nova-core/firmware/booter.rs | 65 ++--
drivers/gpu/nova-core/firmware/fwsec.rs | 129 +++-----
drivers/gpu/nova-core/firmware/fwsec/bootloader.rs | 348 +++++++++++++++++++++
drivers/gpu/nova-core/gpu.rs | 9 +-
drivers/gpu/nova-core/gsp/boot.rs | 17 +-
drivers/gpu/nova-core/regs.rs | 30 ++
9 files changed, 820 insertions(+), 206 deletions(-)
---
base-commit: 15da5bc9f3adab7242867db0251fe451ac3ddb72
change-id: 20260204-turing_prep-6f6f54fe1850
Best regards,
--
Alexandre Courbot <acourbot@nvidia.com>
On Fri Mar 6, 2026 at 1:52 PM JST, Alexandre Courbot wrote: > gpu: nova-core: create falcon firmware DMA objects lazily [acourbot@nvidia.com: add TODO item to switch back to a coherent allocation when it becomes convenient to do so.] > gpu: nova-core: falcon: add constant for memory block alignment > gpu: nova-core: falcon: rename load parameters to reflect DMA dependency [acourbot@nvidia.com: fixup order of import items.] > gpu: nova-core: falcon: remove FalconFirmware's dependency on FalconDmaLoadable > gpu: nova-core: move brom_params and boot_addr to FalconFirmware > gpu: nova-core: falcon: remove unwarranted safety check in dma_load > gpu: nova-core: make Chipset::arch() const > gpu: nova-core: add gen_bootloader firmware to ModInfoBuilder > gpu: nova-core: add PIO support for loading firmware images > gpu: nova-core: use the Generic Bootloader to boot FWSEC on Turing All the above pushed to drm-rust-next, thanks! > gpu: nova-core: firmware: add comments to justify v3 header values > gpu: nova-core: firmware: fix and explain v2 header offsets computations These two not pushed yet as they were introduced late and are still pending proper review.
On 3/8/26 6:52 PM, Alexandre Courbot wrote: > On Fri Mar 6, 2026 at 1:52 PM JST, Alexandre Courbot wrote: >> gpu: nova-core: create falcon firmware DMA objects lazily > [acourbot@nvidia.com: add TODO item to switch back to a coherent > allocation when it becomes convenient to do so.] >> gpu: nova-core: falcon: add constant for memory block alignment >> gpu: nova-core: falcon: rename load parameters to reflect DMA dependency > [acourbot@nvidia.com: fixup order of import items.] >> gpu: nova-core: falcon: remove FalconFirmware's dependency on FalconDmaLoadable >> gpu: nova-core: move brom_params and boot_addr to FalconFirmware >> gpu: nova-core: falcon: remove unwarranted safety check in dma_load >> gpu: nova-core: make Chipset::arch() const >> gpu: nova-core: add gen_bootloader firmware to ModInfoBuilder >> gpu: nova-core: add PIO support for loading firmware images >> gpu: nova-core: use the Generic Bootloader to boot FWSEC on Turing > > All the above pushed to drm-rust-next, thanks! > Amazing! I'll start testing on Turing locally, in addition to Blackwell and Ampere, now. Exciting! Congratulations to Timur Tabi, and all of the expert reviewers and refactor-ers to! thanks, -- John Hubbard
On Mon Mar 9, 2026 at 11:06 AM JST, John Hubbard wrote: > On 3/8/26 6:52 PM, Alexandre Courbot wrote: >> On Fri Mar 6, 2026 at 1:52 PM JST, Alexandre Courbot wrote: >>> gpu: nova-core: create falcon firmware DMA objects lazily >> [acourbot@nvidia.com: add TODO item to switch back to a coherent >> allocation when it becomes convenient to do so.] >>> gpu: nova-core: falcon: add constant for memory block alignment >>> gpu: nova-core: falcon: rename load parameters to reflect DMA dependency >> [acourbot@nvidia.com: fixup order of import items.] >>> gpu: nova-core: falcon: remove FalconFirmware's dependency on FalconDmaLoadable >>> gpu: nova-core: move brom_params and boot_addr to FalconFirmware >>> gpu: nova-core: falcon: remove unwarranted safety check in dma_load >>> gpu: nova-core: make Chipset::arch() const >>> gpu: nova-core: add gen_bootloader firmware to ModInfoBuilder >>> gpu: nova-core: add PIO support for loading firmware images >>> gpu: nova-core: use the Generic Bootloader to boot FWSEC on Turing >> >> All the above pushed to drm-rust-next, thanks! >> > > Amazing! I'll start testing on Turing locally, in addition to Blackwell > and Ampere, now. Exciting! > > Congratulations to Timur Tabi, and all of the expert reviewers and > refactor-ers to! Note that you still need to cherry-pick one of the two non-merged patches for probe to complete properly: https://lore.kernel.org/rust-for-linux/20260306-turing_prep-v11-9-8f0042c5d026@nvidia.com/ I should maybe have pushed the whole series, but would like to get at least one Reviewed-by before I do so, for good conscience.
On Fri Mar 6, 2026 at 5:52 AM CET, Alexandre Courbot wrote: > This patchset adds the remaining support required for booting the GSP on > Turing. > > We did a deep dive with Eliot looking for the reasons why some fields > involved in the bootloader are ignored or used apparently > inconsistently, and this results in a more documented flow and a few > fixes. Apart from that, this series seems to be stabilizing and > successfully probes my TU106: > > NovaCore 0000:08:00.0: NVIDIA (Chipset: TU106, Architecture: Turing, Revision: a.1) > NovaCore 0000:08:00.0: GPU name: NVIDIA GeForce RTX 2070 > > This series is based on `drm-rust-next`. A tree with all the patches is > available at [1]. > > [1] https://github.com/Gnurou/linux/tree/b4/turing > > Signed-off-by: Alexandre Courbot <acourbot@nvidia.com> > > Changes in v11: > - Fix build error/warnings and rustfmt formatting. > - Address incorrect IMEM section start offsets in FalconUCodeDescV2 > and better document fields usage and unused fields. > - Use `get`/`get_mut` instead of direct array indexing when accessing > firmware content. > - Link to v10: https://patch.msgid.link/20260301-turing_prep-v10-0-dde5ee437c60@nvidia.com > > Changes in v10: > - Store the firmwares into a regular KVec and move them into a DMA > object only when actually loading using DMA. > - Use `try_update` when updating the `NV_PFALCON_FBIF_TRANSCFG` register > array as its index is not build-time proven to be valid. > - Fix alignment issue when processing imem section of the FWSEC > bootloader (thanks Eliot!). > - Link to v9: https://patch.msgid.link/20260212-turing_prep-v9-0-238520ad8799@nvidia.com > > Changes in v9: > - Add a few preparatory patches to simplify the actual feature patches. > - Use a wrapping type for the bootloader. > - Simplify the falcon loading code and move the complexity to the > firmware types. > - Add the generic bootloader files to `ModInfoBuilder`. > - Link to v8: https://lore.kernel.org/all/20260122222848.2555890-1-ttabi@nvidia.com/ > > --- > Alexandre Courbot (10): > gpu: nova-core: create falcon firmware DMA objects lazily > gpu: nova-core: falcon: add constant for memory block alignment > gpu: nova-core: falcon: rename load parameters to reflect DMA dependency > gpu: nova-core: falcon: remove FalconFirmware's dependency on FalconDmaLoadable > gpu: nova-core: move brom_params and boot_addr to FalconFirmware > gpu: nova-core: falcon: remove unwarranted safety check in dma_load > gpu: nova-core: firmware: add comments to justify v3 header values > gpu: nova-core: firmware: fix and explain v2 header offsets computations > gpu: nova-core: make Chipset::arch() const > gpu: nova-core: add gen_bootloader firmware to ModInfoBuilder > > Timur Tabi (2): > gpu: nova-core: add PIO support for loading firmware images > gpu: nova-core: use the Generic Bootloader to boot FWSEC on Turing > > drivers/gpu/nova-core/falcon.rs | 315 ++++++++++++++++--- > drivers/gpu/nova-core/falcon/hal.rs | 6 +- > drivers/gpu/nova-core/firmware.rs | 107 ++++--- > drivers/gpu/nova-core/firmware/booter.rs | 65 ++-- > drivers/gpu/nova-core/firmware/fwsec.rs | 129 +++----- > drivers/gpu/nova-core/firmware/fwsec/bootloader.rs | 348 +++++++++++++++++++++ > drivers/gpu/nova-core/gpu.rs | 9 +- > drivers/gpu/nova-core/gsp/boot.rs | 17 +- > drivers/gpu/nova-core/regs.rs | 30 ++ > 9 files changed, 820 insertions(+), 206 deletions(-) > --- > base-commit: 15da5bc9f3adab7242867db0251fe451ac3ddb72 > change-id: 20260204-turing_prep-6f6f54fe1850 > > Best regards, Hi, I just want to remind that there is still issues for some Turing cards with the firmware used by Nova (570.144) and this patchset still suffer from the issue. I am not able to probe on my GeForce GTX 1650 Mobile : [ 2.246095] NovaCore 0000:01:00.0: NVIDIA (Chipset: TU117, Architecture: Turing, Revision: a.1) [ 2.722681] NovaCore 0000:01:00.0: Booter-load failed with error 0x31 However nouveau does not probe either with this firmware so that's not really this patchset fault. Are there any plans to check this to enable support on all Turing cards ? I already reported this error in the V4 patch [1] for context. Feel free to ask me if you need additional tests or results. [1]: https://lore.kernel.org/rust-for-linux/DFA1CUMND2ME.1D3PAJW641QHM@ik.me/T/#u Regards, Ewan
On 3/9/26 12:48 PM, Ewan Chorynski wrote: > On Fri Mar 6, 2026 at 5:52 AM CET, Alexandre Courbot wrote: ... > I am not able to probe on my GeForce GTX 1650 Mobile : > > [ 2.246095] NovaCore 0000:01:00.0: NVIDIA (Chipset: TU117, Architecture: Turing, Revision: a.1) > [ 2.722681] NovaCore 0000:01:00.0: Booter-load failed with error 0x31 I have that exact card available, so I'll give this a quick test and see what's missing or wrong, now that Alex has pushed the entire Turing support set up to drm-rust-next. > > However nouveau does not probe either with this firmware so that's not > really this patchset fault. > > Are there any plans to check this to enable support on all Turing cards ? Yes, the plan is that Nova will support all Turing and later GPUs. thanks, -- John Hubbard
On Mon, 2026-03-09 at 13:04 -0700, John Hubbard wrote: > > I have that exact card available, so I'll give this a quick test and see > what's missing or wrong, now that Alex has pushed the entire Turing support > set up to drm-rust-next. The TU117 is technically a mobile chip, and its VBIOS is different. My initial version of the Turing patches would "ignore" the problematic VBIOS sections, so perhaps this changed. > > > > > However nouveau does not probe either with this firmware so that's not > > really this patchset fault. Now *that* is interesting. Nouveau does generally work on TU117s.
On 3/9/26 1:18 PM, Timur Tabi wrote:
> On Mon, 2026-03-09 at 13:04 -0700, John Hubbard wrote:
>>
>> I have that exact card available, so I'll give this a quick test and see
>> what's missing or wrong, now that Alex has pushed the entire Turing support
>> set up to drm-rust-next.
>
> The TU117 is technically a mobile chip, and its VBIOS is different. My initial version of the
> Turing patches would "ignore" the problematic VBIOS sections, so perhaps this changed.
>
No repro on the latest drm-rust-next branch:
NovaCore 0000:e1:00.0: Probe Nova Core GPU driver.
NovaCore 0000:e1:00.0: NVIDIA (Chipset: TU117, Architecture: Turing, Revision: a.1)
NovaCore 0000:e1:00.0: Found BIOS image: size: 0xe600, type: Ok(PciAt), last: false
NovaCore 0000:e1:00.0: Found BIOS image: size: 0x11000, type: Ok(Efi), last: false
NovaCore 0000:e1:00.0: Found BIOS image: size: 0xc200, type: Ok(FwSec), last: false
NovaCore 0000:e1:00.0: Found BIOS image: size: 0x22400, type: Ok(FwSec), last: false
NovaCore 0000:e1:00.0: Invalid signature for NpdeStruct: [1, 1, 66, 86]
NovaCore 0000:e1:00.0: Invalid signature for NpdeStruct: [1, 1, 66, 86]
NovaCore 0000:e1:00.0: Found BIOS image: size: 0x1a00, type: Ok(Nbsi), last: true
NovaCore 0000:e1:00.0: PmuLookupTableEntry desc: V2(
FalconUCodeDescV2 {
hdr: 3932673,
stored_size: 39968,
uncompressed_size: 39968,
virtual_entry: 0,
interface_offset: 224,
imem_phys_base: 0,
imem_load_size: 38912,
imem_virt_base: 0,
imem_sec_base: 1024,
imem_sec_size: 37888,
dmem_offset: 38912,
dmem_phys_base: 0,
dmem_load_size: 1056,
alt_imem_load_size: 38912,
alt_dmem_load_size: 26168,
},
)
NovaCore 0000:e1:00.0: FbLayout {
fb: 0x0..0x100000000,
vga_workspace: 0xfff00000..0x100000000,
frts: 0xffe00000..0xfff00000,
boot: 0xffdff000..0xffe00000,
elf: 0xfe2c0000..0xffdf4ea0,
wpr2_heap: 0xf7900000..0xfe200000,
wpr2: 0xf7800000..0xfff00000,
heap: 0xf7700000..0xf7800000,
vf_partition_count: 0x0,
}
NovaCore 0000:e1:00.0: WPR2: 0xffe00000-0xffee0000
NovaCore 0000:e1:00.0: GPU instance built
NovaCore 0000:e1:00.0: GSP RPC: send: seq# 0, function=GspSetSystemInfo, length=0x3f0
NovaCore 0000:e1:00.0: GSP RPC: send: seq# 1, function=SetRegistry, length=0xc5
NovaCore 0000:e1:00.0: GSP MBOX0: 0xffffe000, MBOX1: 0x0
NovaCore 0000:e1:00.0: Using SEC2 to load and run the booter_load firmware...
NovaCore 0000:e1:00.0: SEC2 MBOX0: 0x0, MBOX10x0
NovaCore 0000:e1:00.0: RISC-V active? true
NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspRunCpuSequencer), length=0x820
NovaCore 0000:e1:00.0: Running CPU Sequencer commands
NovaCore 0000:e1:00.0: CPU Sequencer commands completed successfully
NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspPostNoCat), length=0x50c
NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspPostNoCat), length=0x50c
NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspInitDone), length=0x50
NovaCore 0000:e1:00.0: GSP RPC: send: seq# 2, function=GetGspStaticInfo, length=0x6c8
NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GetGspStaticInfo), length=0x6c8
NovaCore 0000:e1:00.0: GPU name: NVIDIA GeForce GTX 1650
>>
>>>
>>> However nouveau does not probe either with this firmware so that's not
>>> really this patchset fault.
>
> Now *that* is interesting. Nouveau does generally work on TU117s.
>
thanks,
--
John Hubbard
On Mon Mar 9, 2026 at 9:29 PM CET, John Hubbard wrote:
> On 3/9/26 1:18 PM, Timur Tabi wrote:
>> On Mon, 2026-03-09 at 13:04 -0700, John Hubbard wrote:
>>>
>>> I have that exact card available, so I'll give this a quick test and see
>>> what's missing or wrong, now that Alex has pushed the entire Turing support
>>> set up to drm-rust-next.
>>
>> The TU117 is technically a mobile chip, and its VBIOS is different. My initial version of the
>> Turing patches would "ignore" the problematic VBIOS sections, so perhaps this changed.
>>
>
> No repro on the latest drm-rust-next branch:
I guess I may have an issue with my linux-firmware. I have no stable
right now so I can't download the latest one but I'll try
soon. On which commit on linux-firmware are you ?
>
> NovaCore 0000:e1:00.0: Probe Nova Core GPU driver.
> NovaCore 0000:e1:00.0: NVIDIA (Chipset: TU117, Architecture: Turing, Revision: a.1)
> NovaCore 0000:e1:00.0: Found BIOS image: size: 0xe600, type: Ok(PciAt), last: false
> NovaCore 0000:e1:00.0: Found BIOS image: size: 0x11000, type: Ok(Efi), last: false
> NovaCore 0000:e1:00.0: Found BIOS image: size: 0xc200, type: Ok(FwSec), last: false
> NovaCore 0000:e1:00.0: Found BIOS image: size: 0x22400, type: Ok(FwSec), last: false
> NovaCore 0000:e1:00.0: Invalid signature for NpdeStruct: [1, 1, 66, 86]
> NovaCore 0000:e1:00.0: Invalid signature for NpdeStruct: [1, 1, 66, 86]
> NovaCore 0000:e1:00.0: Found BIOS image: size: 0x1a00, type: Ok(Nbsi), last: true
> NovaCore 0000:e1:00.0: PmuLookupTableEntry desc: V2(
> FalconUCodeDescV2 {
> hdr: 3932673,
> stored_size: 39968,
> uncompressed_size: 39968,
> virtual_entry: 0,
> interface_offset: 224,
> imem_phys_base: 0,
> imem_load_size: 38912,
> imem_virt_base: 0,
> imem_sec_base: 1024,
> imem_sec_size: 37888,
> dmem_offset: 38912,
> dmem_phys_base: 0,
> dmem_load_size: 1056,
> alt_imem_load_size: 38912,
> alt_dmem_load_size: 26168,
> },
> )
> NovaCore 0000:e1:00.0: FbLayout {
> fb: 0x0..0x100000000,
> vga_workspace: 0xfff00000..0x100000000,
> frts: 0xffe00000..0xfff00000,
> boot: 0xffdff000..0xffe00000,
> elf: 0xfe2c0000..0xffdf4ea0,
> wpr2_heap: 0xf7900000..0xfe200000,
> wpr2: 0xf7800000..0xfff00000,
> heap: 0xf7700000..0xf7800000,
> vf_partition_count: 0x0,
> }
> NovaCore 0000:e1:00.0: WPR2: 0xffe00000-0xffee0000
> NovaCore 0000:e1:00.0: GPU instance built
> NovaCore 0000:e1:00.0: GSP RPC: send: seq# 0, function=GspSetSystemInfo, length=0x3f0
> NovaCore 0000:e1:00.0: GSP RPC: send: seq# 1, function=SetRegistry, length=0xc5
> NovaCore 0000:e1:00.0: GSP MBOX0: 0xffffe000, MBOX1: 0x0
> NovaCore 0000:e1:00.0: Using SEC2 to load and run the booter_load firmware...
> NovaCore 0000:e1:00.0: SEC2 MBOX0: 0x0, MBOX10x0
> NovaCore 0000:e1:00.0: RISC-V active? true
> NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspRunCpuSequencer), length=0x820
> NovaCore 0000:e1:00.0: Running CPU Sequencer commands
> NovaCore 0000:e1:00.0: CPU Sequencer commands completed successfully
> NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspPostNoCat), length=0x50c
> NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspPostNoCat), length=0x50c
> NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GspInitDone), length=0x50
> NovaCore 0000:e1:00.0: GSP RPC: send: seq# 2, function=GetGspStaticInfo, length=0x6c8
> NovaCore 0000:e1:00.0: GSP RPC: receive: seq# 0, function=Ok(GetGspStaticInfo), length=0x6c8
> NovaCore 0000:e1:00.0: GPU name: NVIDIA GeForce GTX 1650
>
>
>>>
>>>>
>>>> However nouveau does not probe either with this firmware so that's not
>>>> really this patchset fault.
>>
>> Now *that* is interesting. Nouveau does generally work on TU117s.
>>
>
> thanks,
On Mon, 2026-03-09 at 22:00 +0100, Ewan Chorynski wrote: > On Mon Mar 9, 2026 at 9:29 PM CET, John Hubbard wrote: > > On 3/9/26 1:18 PM, Timur Tabi wrote: > > > On Mon, 2026-03-09 at 13:04 -0700, John Hubbard wrote: > > > > > > > > I have that exact card available, so I'll give this a quick test and see > > > > what's missing or wrong, now that Alex has pushed the entire Turing support > > > > set up to drm-rust-next. > > > > > > The TU117 is technically a mobile chip, and its VBIOS is different. My initial version of the > > > Turing patches would "ignore" the problematic VBIOS sections, so perhaps this changed. > > > > > > > No repro on the latest drm-rust-next branch: > > I guess I may have an issue with my linux-firmware. I have no stable > right now so I can't download the latest one but I'll try > soon. On which commit on linux-firmware are you ? There's only one version of linux-firmware that works with Nova, and you didn't have it, it wouldn't boot at all. Although, now that I think about it, I'm assuming that on Turing, if gen_bootloader is absent, NovaCore will not even try to boot. That file was added recently and is missing in most distros today. /lib/firmware/nvidia/tu102/gsp/gen_bootloader-570.144.bin (or its zstd compressed version) If you have this file, then you have everything you need to boot NovaCore on Turing.
On Mon Mar 9, 2026 at 10:05 PM CET, Timur Tabi wrote: > On Mon, 2026-03-09 at 22:00 +0100, Ewan Chorynski wrote: >> On Mon Mar 9, 2026 at 9:29 PM CET, John Hubbard wrote: >> > On 3/9/26 1:18 PM, Timur Tabi wrote: >> > > On Mon, 2026-03-09 at 13:04 -0700, John Hubbard wrote: >> > > > >> > > > I have that exact card available, so I'll give this a quick test and see >> > > > what's missing or wrong, now that Alex has pushed the entire Turing support >> > > > set up to drm-rust-next. >> > > >> > > The TU117 is technically a mobile chip, and its VBIOS is different. My initial version of the >> > > Turing patches would "ignore" the problematic VBIOS sections, so perhaps this changed. >> > > >> > >> > No repro on the latest drm-rust-next branch: >> >> I guess I may have an issue with my linux-firmware. I have no stable >> right now so I can't download the latest one but I'll try >> soon. On which commit on linux-firmware are you ? > > There's only one version of linux-firmware that works with Nova, and you didn't have it, it wouldn't > boot at all. > > Although, now that I think about it, I'm assuming that on Turing, if gen_bootloader is absent, > NovaCore will not even try to boot. That file was added recently and is missing in most distros > today. > > /lib/firmware/nvidia/tu102/gsp/gen_bootloader-570.144.bin (or its zstd compressed version) > > If you have this file, then you have everything you need to boot NovaCore on Turing. I had this file installed but I think I broke something when I updated. I tried to redo my installation from the tarball I had and now it is probing, so the issue was indeed on my side with my firmware. Thanks for trying the repro and sorry for the false alarm. Have a good day Ewan
On Mon, 2026-03-09 at 22:16 +0100, Ewan Chorynski wrote: > I had this file installed but I think I broke something when I updated. > I tried to redo my installation from the tarball I had and now it is > probing, so the issue was indeed on my side with my firmware. > > Thanks for trying the repro and sorry for the false alarm. It would be good to know exactly how your broken /lib/firmware caused booter-load to fail.
On Mon, 2026-03-09 at 13:29 -0700, John Hubbard wrote: > On 3/9/26 1:18 PM, Timur Tabi wrote: > > On Mon, 2026-03-09 at 13:04 -0700, John Hubbard wrote: > > > > > > I have that exact card available, so I'll give this a quick test and see > > > what's missing or wrong, now that Alex has pushed the entire Turing support > > > set up to drm-rust-next. > > > > The TU117 is technically a mobile chip, and its VBIOS is different. My initial version of the > > Turing patches would "ignore" the problematic VBIOS sections, so perhaps this changed. > > > > No repro on the latest drm-rust-next branch: > > NovaCore 0000:e1:00.0: Probe Nova Core GPU driver. > NovaCore 0000:e1:00.0: NVIDIA (Chipset: TU117, Architecture: Turing, Revision: a.1) > NovaCore 0000:e1:00.0: Found BIOS image: size: 0xe600, type: Ok(PciAt), last: false > NovaCore 0000:e1:00.0: Found BIOS image: size: 0x11000, type: Ok(Efi), last: false > NovaCore 0000:e1:00.0: Found BIOS image: size: 0xc200, type: Ok(FwSec), last: false > NovaCore 0000:e1:00.0: Found BIOS image: size: 0x22400, type: Ok(FwSec), last: false > NovaCore 0000:e1:00.0: Invalid signature for NpdeStruct: [1, 1, 66, 86] > NovaCore 0000:e1:00.0: Invalid signature for NpdeStruct: [1, 1, 66, 86] So this is the problematic section that gets ignored. It's on my TODO list to fix this, but last time I looked at it, the documentation I had on the VBIOS layout did not align with the VBIOS on my TU117. > [ 2.246095] NovaCore 0000:01:00.0: NVIDIA (Chipset: TU117, Architecture: Turing, Revision: a.1) > [ 2.722681] NovaCore 0000:01:00.0: Booter-load failed with error 0x31 > > However nouveau does not probe either with this firmware so that's not > really this patchset fault. So Booter-load error 0x31 means that Booter technically did start, but it aborted very early. Unfortunately, this is very difficult to debug in the field. Normally what I would do is build custom versions of booter-load to see where it fails. I cannot do this without the card in my hand. The first thing I would do is verify that GspFwWprMeta does not have nonsensical values.
© 2016 - 2026 Red Hat, Inc.