On Tue, Oct 13, 2026 at 12:50:42PM +0530, Aswin Murugan wrote:
> 
> On 10/12/2026 11:20 PM, Tom Rini wrote:
> Hi Tom,
> > On Mon, Oct 12, 2026 at 10:37:49PM +0530, Aswin Murugan wrote:
> > > This series adds dynamic device tree selection from FIT images for
> > > Qualcomm Snapdragon platforms, enabling U-Boot to select the
> > > appropriate DTB based on hardware parameters detected at boot.
> > > 
> > > Qualcomm FIT-based DTB format is documented in [1].
> > > 
> > > The FIT image contains only DTBs, while the kernel is provided
> > > separately as part of a UKI image.
> > > 
> > > The implementation consists of:
> > > 
> > > 1. Hardware detection infrastructure: Introduces the qcom,hwinfo
> > >     driver and qcom_hwdetect helper to collect hardware identification
> > >     data from SMEM, IMEM and TCSR into a single qcom_hw_params
> > >     structure for consumers such as FIT multi-DTB selection.
> > > 
> > > 2. FIT multi-DTB selection: Implements DTB selection based on the
> > >     aggregated hardware information. The code parses metadata DTBs,
> > >     matches FIT configurations against detected hardware attributes,
> > >     applies overlays, and loads the selected DTB for EFI boot.
> > > 
> > > [1] https://github.com/qualcomm-linux/qcom-dtb-metadata/blob/main/
> > >      Documentation.md
> > > 
> > > ---
> > > 
> > > Changes in v4:
> > ... which was sent as v5, and this is a re-send of?
> Yes, this is a resend of v4, which was sent as v5
> > > 1. Drop the SMEM cache infrastructure.
> > > 
> > > 2. misc: qcom: add qcom,hwinfo driver to provide a common interface
> > >     for reading IMEM boot cookie and TCSR SOC version information.
> > > 
> > > 3. arm: snapdragon: add qcom_hwdetect helper to aggregate hardware
> > >     information from SMEM, IMEM and TCSR into a single
> > >     struct qcom_hw_params.
> > > 
> > > 4. Refactor FIT multi-DTB selection to consume qcom_hw_params instead
> > >     of accessing SMEM, IMEM and TCSR data directly.
> > > 
> > > 5. Separate hardware discovery from FIT parsing and DTB matching
> > >     logic, reducing complexity and improving code organization.
> > > 
> > > 6. Load FIT images directly from a raw GPT partition and remove the
> > >     FAT filesystem image based loading approach.
> > > 
> > > 7. Drop IH_TYPE_FATFS, mkimage FAT image generation support and
> > >     related documentation updates.
> > > 
> > > 8. arm: dts: enable qcom,hwinfo on qcs9100-ride-r3, qcs615-ride and
> > >     qcs6490-rb3gen2 by adding the required qcom,hwinfo DT nodes and
> > >     board-specific offsets.
> > > 
> > > Link to v3:
> > > https://lore.kernel.org/u-boot/[email protected]/
> > The FAT comments in v3 were addressed. I'm not really sure the discussion 
> > about if
> > this approach makes sense as really was addressed however?
> 
> Yes, the FAT-related comments from v3 were addressed in this
> revision by dropping the FAT filesystem image-based loading
> approach and loading FIT image directly from a raw GPT
> partition.

Right, thanks.

> Regarding the broader design, the reasoning for using a
> DTB-only FIT was discussed in the v3 thread [1]. The intention
> is to support firmware-side DT selection while keeping DTB
> management independent of kernel packaging, matching existing
> Qualcomm deployment models. This is intended to coexist with,
> rather than replace, approaches such as UKI DT bundling.
> 
> [1]: https://lore.kernel.org/u-boot/afSXa7q7hJC_ZW07@sumit-xelite/

Yes, and I'd appreciate some of the people that were discussing things
in that thread chiming in and saying they're happy enough, as it wasn't
clear to me in re-reading the thread yesterday that the discussion ended
with agreement and instead just stopped.

-- 
Tom

Attachment: signature.asc
Description: PGP signature

Reply via email to