On September 25, 2025 thus sayeth Aristo Chen: > Aristo Chen <jj251510319...@gmail.com> 於 2025年9月24日 週三 下午10:33寫道： > > > > Nishanth Menon <n...@ti.com> 於 2025年9月24日 週三 下午10:05寫道： > > > > > > On 07:51-20250924, Tom Rini wrote: > > > > On Wed, Sep 24, 2025 at 06:31:23AM -0700, E Shattow wrote: > > > > > Hi Aristo, > > > > > > > > > > On 9/24/25 04:43, Nishanth Menon wrote: > > > > > > On 06:37-20250924, Nishanth Menon wrote: > > > > > >> On 10:59-20250914, Aristo Chen wrote: > > > > > >>> This patch series enhances FIT image robustness by adding **memory > > > > > >>> region overlap detection** to `mkimage` and fixing existing > > > > > >>> overlaps > > > > > >>> in DTS files and `binman` tests. > > > > > > > > > > > > [...] > > > > > >>> > > > > > >> > > > > > >> Looks like i see a build regression in linux-next after this > > > > > >> series. > > > > > > > > > > > > I fat fingered that one.. sorry, I meant u-boot next. > > > > > > > > > > > > > > > > > > Fails at commit 4d84fa1261eb, last pass was on commit d81c1118580f > > > > > > > > > > > >> > > > > > >> https://gist.github.com/nmenon/b2fc9e7680cc296062c7dced94105f76 > > > > > > > > > > > > I believe there are outstanding comments on V1 that have'nt been > > > > > > addressed either. Can we revert/drop this series for now while the > > > > > > comments are addressed? > > > > > > > > > > > > > > > > Similar to Nishanth, I am seeing a build regression, itb.map: > > > > > > > > Sorry for the noise all, I've reverted this in next now. > > > > > > > > > > Thanks Tom. > > > -- > > > Regards, > > > Nishanth Menon > > > Key (0xDDB5849D1736249D) / Fingerprint: F8A2 8693 54EB 8232 17A3 1A34 > > > DDB5 849D 1736 249D > > > https://ti.com/opensource > > > > Apology for introducing the regression, I will find time to fix them, > > and thanks Tom for reverting my patch set to unblock others > > > > For TIFS stub entries on AM6x platforms and the starfive visionfive2, > > sounds like the image in the FIT will be selected during the runtime, > > so they are not really overlapping each other. I will need to figure > > out other way instead > > > Thank you all for the feedback on my FIT image overlap detection patch > series, and apology again for the issues that I introduced > > Currently what I have in my mind is to have the following 2 changes in > my patch set > ### 1. Add support for mutually exclusive image groups > Some platforms load only one image from a group at runtime. In the > device tree, we may have something like this > ``` > tifsstub-hs { > // ... existing properties ... > mutually-exclusive-group = "tifsstub"; > }; > tifsstub-fs { > // ... existing properties ... > mutually-exclusive-group = "tifsstub"; > }; > ``` > The overlap checker would then skip validation between images in the > same mutually-exclusive-group, while still catching genuine > overlaps
This was the direction I was leaning as well. A way to opt-out or group the binaries we know should overlap. > > ### 2. Change from build error to warning > Given that there may be edge cases(such as > starfive_visionfive2_defconfig) not covered by current CI pipelines, > changing overlap > detection from a build-stopping error to a warning initially. This would: > - Alert developers to potential memory layout issues > - Avoid breaking existing working configurations > - Allow time to refine the detection logic based on community feedback > - Enable a gradual transition to stricter validation once the approach is > proven > The warning could later become an error once we have confidence in the > detection accuracy across all supported platforms. > I haven't looked into what is needed for the visionfive2 board, but I liked having this as a build error. If we know the build will not work on for the board I think we should fail to build it. Though it's really just for the selfish reason of not having to debug other people's builds that lead to the unsatisfying conclusion of memory corruption. ~Bryan
