"Gary Guo" <gary@xxxxxxxxxxx> writes: > A new Kconfig option, `RUST_INLINE_HELPERS` is added to allow C helpers > (which was created to allow Rust to call into inline/macro C functions > without having to re-implement the logic in Rust) to be inlined into > Rust crates without performing a global LTO. > > If the option is enabled, the following is performed: > * For helpers, instead of compiling them to object file to be linked > into vmlinux, we compile them to LLVM IR. > * The LLVM IR is patched to add `linkonce_odr` linkage. This linkage > means that the function is inlineable (effect of `_odr`), and the > symbols generated will have weak linkage if emitted into object file > (important since as later described, we might have multiple copies of > the same symbol) and it will may be discarded if it is not invoked or > all invocations are inlined. > * The LLVM IR is compiled to bitcode (This is step is not necessary, but > is a performance optimisation to prevent LLVM from always have to > reparse the same IR). > * When a Rust crate is compiled, instead of generating object file, we > ask LLVM bitcode to be generated. > * llvm-link is invoked to combine the helper bitcode with the crate > bitcode. This step is similar to LTO, but this is much faster since it > only needs to inline the helpers. > * clang is invoked to turn the combined bitcode into object file. > > Some caveats with the option: > * clang and Rust doesn't have the exact target string. Manual inspection > shows that they should be compatible, but since they are not exactly > the same LLVM seems to prefer not inlining them. This is bypassed with > `--ignore-tti-inline-compatible`. > * LLVM doesn't want to inline functions combined with > `-fno-delete-null-pointer-checks` with code compiled without. So we > remove this command when compiling helpers. I think this should be > okay since this is one of the hardening features and we shouldn't have > null pointer dereferences in these helpers. > > The checks can also be bypassed with force inlining (`__always_inline`) > but the behaviour is the same with extra options. > > Co-developed-by: Boqun Feng <boqun.feng@xxxxxxxxx> > Signed-off-by: Boqun Feng <boqun.feng@xxxxxxxxx> > Signed-off-by: Gary Guo <gary@xxxxxxxxxxx> Reviewed-by: Andreas Hindborg <a.hindborg@xxxxxxxxxx> Tested-by: Andreas Hindborg <a.hindborg@xxxxxxxxxx> Tested with downstream rnull tree, works as expected. Best regards, Andreas Hindborg
