[Bug target/114523] bpf: ssa-phiopt optimization generates unverifiable code.

["jemarch at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523

Jose E. Marchesi <jemarch at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|RESOLVED                    |NEW
         Resolution|INVALID                     |---
                 CC|                            |jemarch at gcc dot gnu.org
   Last reconfirmed|                            |2024-03-28

--- Comment #9 from Jose E. Marchesi <jemarch at gcc dot gnu.org> ---
Please do not close this bug.  We are well aware this is a limitation in the
kernel verifier, but we need to keep track of this.

In general, like in this case, many of the problems related to unverifiable
generated code can be reduced to "this is a kernel verifier bug, make it
smarter", but that doesn't mean we can't put workarounds in the back-end so we
can actually compile actually-existing BPF programs (such as that systemd
utility) in actually-existing current kernels.  That includes disabling certain
optimizations.  I don't like it any more than you do, trust me on that.

Note that the development of the BPF backends, in both clang and GCC, are by
necessity very much in lock-step with the kernel BPF support.  So this is not
something we will forget to "undo" as soon as the verifier can handle these
instructions.  Very likely we will be doing the kernel patch as well.