On Fri, 2018-03-23 at 18:50 +0100, bigeasy@xxxxxxxxxxxxx wrote: > On 2018-03-23 17:44:46 [+0000], Bart Van Assche wrote: > > In other words, do we really need to remove these checks? I think that these > > checks are useful as documentation to people who read the SCSI target code. > > The target code is already hard to follow so I think any documentation, > > especially documentation in the form of code that is checked at runtime, is > > welcome. > > so if I remove those two and add a kernel doc comment instead, saying > that the caller needs to ensure that "lun->lun_tg_pt_gp_lock" is held > then we would remove the obvious runtime check and add a piece of > documentation. Would that work? Comments are not verified at runtime and hence can become outdated if the code is modified. assert_spin_locked() and lockdep_assert_held() assertions however are verified at runtime with the proper kernel configuration options enabled. Hence my preference for assert_spin_locked()/lockdep_assert_held() over source code comments. Thanks, Bart.
