ACK/Cmnt: [[SCRIPT=remove_re|Re: [SRU jammy:linux-*-6.2] Revert variable symbol length modversion patch]]

[Andrea Righi]

On Tue, Jun 27, 2023 at 03:01:04PM +0200, Stefan Bader wrote:
> On 27.06.23 11:45, Andrea Righi wrote:
> > NOTE: this SRU doesn't have any patch, it explains a special action to
> > do if you are cranking a Jammy kernel that is derived from Lunar.
> > 
> > [Impact]
> > 
> > In order to support Rust in the kernel starting with Lunar we had to
> > apply the following UBUNTU SAUCE patch:
> > 
> >    27c4fe11712c ("UBUNTU: SAUCE: modpost: support arbitrary symbol length in modversion")
> > 
> > This patch can potentially introduce a user-space regression, because it
> > alters how the module names are stored in the modversion_info area.
> > Realistically it doesn't break anything, since this area is used only by
> > the kmod tools (and they've been verified to work fine also with this
> > change applied).
> > 
> > However, to be 100% that we don't introduce regressions it is safer to
> > simply revert this patch, considering that we don't provide kernel Rust
> > support in Jammy and without Rust this patch doesn't provide any
> > benefit.
> > 
> > [Fix]
> > 
> > In order to fix this we need to explicitly revert the patch in all the
> > Jammy kernels that are derived from any Lunar kernel.
> > 
> > To do so, after `cranky rebase`, run:
> > 
> >   git revert -s 27c4fe11712c
> > 
> > [ The sha1 should correspond to the proper patch to revert, otherwise
> >    make sure that the patch "UBUNTU: SAUCE: modpost: support arbitrary
> >    symbol length in modversion" is not applied, or it's already reverted ]
> > 
> > The comment for the revert should report something like this:
> > ```
> > 
> >   Revert "UBUNTU: SAUCE: modpost: support arbitrary symbol length in modversion"
> > 
> >   This patch is required by Rust and it can potentially break user-space.
> >   It is safer to revert this in all the kernel backported to old releases.
> > ```
> > 
> > [Regression potential]
> > 
> > This revert is not very critical, realistically even with the variable
> > modversions patch applied we won't regress any known tool, however we
> > don't to risk to introduce potential user-space ABI changes. Fixing this
> > requires a slightly different action, respect to our usual workflow, so
> > if some kernels are shipped with the variable modversion patch (for any
> > reason), we may experience regressions with kmod-related tools (e.g.,
> > scripts that are using modinfo, modprobe and similar or custom tools
> > that are parsing .ko sections directly).
> > 
> I think this misses (for reference in the reverts):
> 
> BugLink: https://bugs.launchpad.net/bugs/2025134

Oh yes, indeed! Thanks for adding the buglink.

-Andrea

> 
> Acked-by: Stefan Bader <stefan.bader at canonical.com>
>