Messages in this thread | | | Date | Tue, 11 Nov 2008 10:09:15 +0100 | From | Ingo Molnar <> | Subject | Re: [PATCH 2/2] ring-buffer: replace most bug ons with warn on and disable buffer |
| |
* Steven Rostedt <rostedt@goodmis.org> wrote:
> +#define RB_WARN_ON_RET_INT(buffer, cond) \ > + do { \ > + if (unlikely(cond)) { \ > + atomic_inc(&buffer->record_disabled); \ > + WARN_ON(1); \ > return -1; \ > } \
btw., the _RET() methods are rather ugly as they include an implicit code flow change (a 'return' statement).
Please change it to a more readable form that preserves the code flow, something like:
if (RB_WARN_ON(buffer, cond)) return -1;
See kernel/lockdep.c about how to do this cleanly: introduce a _single_ global "oh, we are broken" flag which is increased once and never decreased again.
In the case of lockdep that's the debug_locks flag. It's used in various code-flow-preserving forms of debug checks:
... if (DEBUG_LOCKS_WARN_ON(depth >= 20)) return; ... if (!debug_locks_off_graph_unlock()) return NULL; ... if (!debug_locks_off_graph_unlock()) return 0; ... if (!__raw_spin_is_locked(&lockdep_lock)) return DEBUG_LOCKS_WARN_ON(1);
etc.
Ingo
| |