Hello, On Wed, Feb 9, 2022 at 8:27 AM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote: > > On Wed, 09 Feb 2022 15:49:01 +0200 > Jani Nikula <jani.nikula@xxxxxxxxxxxxxxx> wrote: > > > > Because I want to use the lockdep annotation for other purposes. > > > But the workqueue lockdep_map was defined under LOCKDEP > > > only. Please see the description in the cover letter. > > > > > > https://lore.kernel.org/all/20220208184208.79303-1-namhyung@xxxxxxxxxx/ > > > > So lockdep_init_map() might still be there and build just fine for > > CONFIG_LOCKDEP=n, but now we're actually required to wrap all call sites > > in #ifdefs depending on the purpose? I'm not convinced yet. Because work_struct.lockdep_map is there only if CONFIG_LOCKDEP=y. > > I addressed this already. I suggested to add a "raw" variant that turns to > a nop when LOCKDEP is not enabled. That is, for those locations that are > only for working with LOCKDEP, the call will be: > > lockdep_init_map_raw() > > This will differentiate the locations that are for just lockdep and those > that are for both lockdep and tracing. Yep, this should be fine if it's actually defined on CONFIG_LOCKDEP=y. Thanks, Namhyung
