On Thu, 10 Dec 2015 12:49:33 +0100 Petr Mladek <pmladek@xxxxxxxx> wrote: > On Thu 2015-12-10 10:48:04, zhouchengming wrote: > > Now the livepatch register one tracer for each changed function. > > So ftrace will iterates the whole tracers list when each changed > > function be called. And when there are many changed functions, this > > solution can cause performance degradation obviously. > > > > So I think we can register only one tracer for all the changed > > functions. And in this tracer handler we can find the right new > > function list easily using the hash value of the old function > > address. > > I haven't studied this in detail but I believe that ftrace > is more effective these days. It could call the handler > directly if only one is registered. > > I think that it is related to the FTRACE_FL_TRAMP flag and > ftrace_update_trampoline(). I am not sure if it works out of box. > Anyway, the trampoline stuff looks like the most effective > way to go. > Correct, it now supports creating a trampoline specifically for a callback. If there's only one callback hooked to a specific function, it calls the trampoline for that function directly, and not the iterator, even if there's other callbacks for other functions. Now I may need to finally user rcu_task, to free the trampolines for it. -- Steve -- To unsubscribe from this list: send the line "unsubscribe live-patching" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
