Re: [PATCH 2/6] perf record: Enable off-cpu analysis with BPF

[Namhyung Kim]

On Tue, May 31, 2022 at 11:01:14PM -0700, Namhyung Kim wrote:
> On Tue, May 31, 2022 at 5:00 PM Ian Rogers <irogers@xxxxxxxxxx> wrote:
> >
> > On Wed, May 18, 2022 at 3:47 PM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
> [SNIP]
> > > +
> > > +/*
> > > + * Old kernel used to call it task_struct->state and now it's '__state'.
> > > + * Use BPF CO-RE "ignored suffix rule" to deal with it like below:
> > > + *
> > > + * https://nakryiko.com/posts/bpf-core-reference-guide/#handling-incompatible-field-and-type-changes
> > > + */
> > > +static inline int get_task_state(struct task_struct *t)
> > > +{
> > > +       if (bpf_core_field_exists(t->__state))
> > > +               return BPF_CORE_READ(t, __state);
> > > +
> >
> > When building against a pre-5.14 kernel I'm running into a build issue here:
> >
> > tools/perf/util/bpf_skel/off_cpu.bpf.c:96:31: error: no member named '__
> > state' in 'struct task_struct'; did you mean 'state'?
> >        if (bpf_core_field_exists(t->__state))
> >                                     ^~~~~~~
> >                                     state
> >
> > This isn't covered by Andrii's BPF CO-RE reference guide. I have an
> > #iffy workaround below,but this will be brittle if the 5.14+ kernel
> > code is backported. Suggestions welcomed :-)
> 
> Thanks for the fix.  I think we should not guess the field name
> in the current task struct and check both versions separately.
> I'm afraid the version check won't work with some backported
> kernels.  But do we care?


What about this instead?

----8<----