On Tue, Nov 15, 2022 at 10:39:04PM IST, Alexei Starovoitov wrote:
> On Tue, Nov 8, 2022 at 6:07 AM Toke Høiland-Jørgensen <toke@xxxxxxxxxx> wrote:
> >
> > From: Kumar Kartikeya Dwivedi <memxor@xxxxxxxxx>
> >
> > BPF ABI always uses 64-bit return value, but so far __bpf_prog_run and
> > higher level wrappers always truncated the return value to 32-bit. We
> > want to be able to introduce a new BPF program type that returns a
> > PTR_TO_BTF_ID or NULL from the BPF program to the caller context in the
> > kernel. To be able to use this returned pointer value, the bpf_prog_run
> > invocation needs to be able to return a 64-bit value, so update the
> > definitions to allow this.
>
> ...
>
> > -static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
> > +static __always_inline u64 __bpf_prog_run(const struct bpf_prog *prog,
> > const void *ctx,
> > bpf_dispatcher_fn dfunc)
> > {
> > - u32 ret;
> > + u64 ret;
> >
> > cant_migrate();
> > if (static_branch_unlikely(&bpf_stats_enabled_key)) {
> > @@ -602,7 +602,7 @@ static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog,
> > return ret;
> > }
> >
> > -static __always_inline u32 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
> > +static __always_inline u64 bpf_prog_run(const struct bpf_prog *prog, const void *ctx)
> > {
> > return __bpf_prog_run(prog, ctx, bpf_dispatcher_nop_func);
> > }
>
> I suspect 32-bit archs with JITs are partially broken with this change.
> As long as progs want to return pointers it's ok, but actual
> 64-bit values will be return garbage in upper 32-bit, since 32-bit
> JITs won't populate the upper bits.
> I don't think changing u32->long retval is a good idea either,
> since BPF ISA has to be stable regardless of underlying arch.
> The u32->u64 transition is good long term, but let's add
> full 64-bit tests to lib/test_bpf and fix JITs.
>
I will resubmit with these tests and revisiting the 32-bit JITs (but it will
take some time, since my backlog is full).
