Nonreproducible kernel modules on stable/12

[Alan Somers]

On Tue, Mar 31, 2020 at 8:08 AM Ed Maste <emaste at freebsd.org> wrote:

> Originally reported off-list, during a stable/12 freebsd-update build
> kernel modules are not reproducible and thus get included in the
> update.
>
> diffoscope reports that the .SUNW_ctf section is responsible:
>
> --- cam.0.ko
> +++ cam.1.ko
> ├── readelf --wide --sections {}
> ...
> │    [30] .shstrtab         STRTAB          0000000000000000 0ea828
> 0001f8 00      0   0  1
> │ -  [31] .SUNW_ctf         PROGBITS        0000000000000000 0f29d8
> 00e2ad 00     32   0  4
> │ +  [31] .SUNW_ctf         PROGBITS        0000000000000000 0f29d8
> 00e2b1 00     32   0  4
> │    [32] .symtab           SYMTAB          0000000000000000 0e12b8
> 009570 18     33 1001  8
> ...
> ├── readelf --wide --decompress --hex-dump=.SUNW_ctf {}
> │ @@ -1,3630 +1,3631 @@
> │
> │  Hex dump of section '.SUNW_ctf':
> │    0x00000000 f1cf0201 00000000 00000000 00000000 ................
> │    0x00000010 08000000 96040000 fc250000 f4310100 .........%...1..
> │ -  0x00000020 6aeb0000 78dabc7d 077c54c5 16f7ccbd j...x..}.|T.....
> │ -  0x00000030 774b42ca 26b48524 b0402001 02440d10 wKB.&..$.@ ..D..
> ...
> │ +  0x00000020 6feb0000 78dabc7d 077c54c5 16f7ccdc o...x..}.|T.....
> │ +  0x00000030 bb252165 13da4242 586a0204 081820d4 .%!e..BBXj.... .
>
> (And build-ID and GNU debuglink differ as a result.)
>
> Running ctfdump and comparing the result shows the issue:
>
> --- a   2020-03-31 10:04:39.766567000 -0400
> +++ b   2020-03-31 10:04:44.663292000 -0400
> @@ -11,11 +11,11 @@
>    cth_funcoff  = 1174
>    cth_typeoff  = 9724
>    cth_stroff   = 78324
> -  cth_strlen   = 60266
> +  cth_strlen   = 60271
>
>  - Label Table
> ----------------------------------------------------------------
>
> -   2407 FreeBSD 12.1-STABLE amd64 1201512
> +   2407 FreeBSD 12.1-FUTUREBUILD amd64 1201512
>
>  - Data Objects
> ---------------------------------------------------------------
>
> Aside, we should add ctfdump to diffoscope.
>

Good sleuthing Ed!  That "FUTUREBUILD" string points to the problem.
Warner's commit "caused" the problem only in the sense that BRANCH_OVERRIDE
didn't actually work beforehand.  So freebsd-update-build was setting
BRANCH_OVERRIDE to no effect.  Now that Warner has fixed BRANCH_OVERRIDE,
we'll have to fix freebsd-update-build, which may be as simple as not
setting BRANCH_OVERRIDE during that phase of the build.  I'll test that
now, but it will take a few hours.
-Alan