On Mon, Feb 23, 2026 at 04:07:23PM -0800, Song Liu wrote:
> > > +void iterate_global_symbol_by_demangled_name(const struct elf *elf,
> > > + const char *demangled_name,
> > > + void (*process)(struct symbol
> > > *sym, void *data),
> > > + void *data)
> > > +{
> > > + struct symbol *sym;
> > > +
> > > + elf_hash_for_each_possible(symbol_name, sym, name_hash,
> > > str_hash(demangled_name)) {
> > > + if (!strcmp(sym->demangled_name, demangled_name) &&
> > > !is_local_sym(sym))
> > > + process(sym, data);
> > > + }
> > > +}
> > > +
> >
> > I think a saner interface would be something like:
> >
> > struct symbol *find_global_demangled_symbol(const struct elf *elf, const
> > char *demangled_name)
> > {
> > struct symbol *ret = NULL;
> >
> > elf_hash_for_each_possible(symbol_name, sym, name_hash,
> > str_hash(demangled_name)) {
> > if (!is_local_sym(sym) && !strcmp(sym->demangled_name,
> > demangled_name)) {
> > if (ret)
> > return ERR_PTR(-EEXIST);
> > ret = sym;
> > }
> > }
> >
> > return ret;
> > }
>
> I had something similar to this initially. However, we need to check
> sym->twin, and skip symbols that already have correlations. For
> example, if we have foo.llvm.123 and foo.llvm.456 in the original
> kernel, and foo.llvm.123 and foo.llvm.789 in the patched kernel,
> we will match foo.llvm.456 to foo.llvm.789 without ambiguity.
> Since elf.c doesn't touch sym->twin at all, I think it is cleaner to
> keep this logic in klp-diff.c. If you think it is OK to have elf.c
> handle this, we can do something like:
>
> struct symbol *find_global_demangled_symbol(const struct elf *elf,
> const char *demangled_name)
> {
> struct symbol *ret = NULL;
>
> elf_hash_for_each_possible(symbol_name, sym, name_hash,
> str_hash(demangled_name)) {
> if (!is_local_sym(sym) &&
> !strcmp(sym->demangled_name, demangled_name) &&
> !sym->twin) { /* We need to add this */
> if (ret)
> return ERR_PTR(-EEXIST);
> ret = sym;
> }
> }
>
> return ret;
> }
>
> I personally like v2 patch better. But I wouldn't mind changing to the
> above version in v3.
Ah, I see. Yeah, the v2 implementation seems ok.
--
Josh
