The default behaviour of podman is to drop all stdout/err it generates into
the journal, which gets rather noisy.

Signed-off-by: Timo Beckers <timo@isovalent.com>

LoadCollectionSpecFromReader marks all .rodata* sections as data sections,
but loadDataSections() only freezes '.rodata'.

Signed-off-by: Timo Beckers <timo@isovalent.com>

https://lore.kernel.org/bpf/CAK3+h2wcDceeGyFVDU3n7kPm=zgp7r1q4WK0=abxBsj9pyFN-g@mail.gmail.com

If an anonymous value or constant (e.g. array declared in a macro) ends up
in a data section, it will not be accompanied by debug info, resulting in
no BTF being emitted for the value. If the section consists of only anonymous
values, there could be no BTF for the section as a whole.

A workaround was added to LLVM to always emit an empty BTF Datasec for .rodata
specifically, but this doesn't work for special sections like .rodata.cst32
and friends.

Signed-off-by: Timo Beckers <timo@isovalent.com>

Symbols for anonymous constants are somewhat irregular on older versions
of LLVM. This commit relaxes the constraints on symbols those compilers
emit map relocations against.

LLVM 7:
     1: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT    16 .Lconstinit.1

LLVM 9:
     1: 0000000000000000    32 OBJECT  LOCAL  DEFAULT    22 .Lconstinit.1

LLVM 14 has these symbols correctly sanitized.

Signed-off-by: Timo Beckers <timo@isovalent.com>

https://lore.kernel.org/bpf/CAK3+h2wcDceeGyFVDU3n7kPm=zgp7r1q4WK0=abxBsj9pyFN-g@mail.gmail.com

Signed-off-by: Timo Beckers <timo@isovalent.com>

Instead of only looking for constants in .rodata, iterate over all maps with
an .rodata prefix.

Moved the datasec bytes and BTF info extraction into a method on MapSpec.
Lifted patchValue back into RewriteConstants.

Signed-off-by: Timo Beckers <timo@isovalent.com>

This exercises support for constants in custom .rodata* sections.

Signed-off-by: Timo Beckers <timo@isovalent.com>