On Mon, Feb 28, 2022 at 8:21 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote: > > On Mon, 28 Feb 2022 17:48:27 -0800 > Daniel Latypov <dlatypov@xxxxxxxxxx> wrote: > > > He also prototyped a more intrusive alternative to using ftrace and > > kernel livepatch since they don't work on all arches, like UML. > > Perhaps instead of working on a intrusive alternative on archs that do > not support live kernel patching, implement live kernel patching on > those archs! ;-) > > It's probably the same amount of work. Well, really, you only need to > implement the klp_arch_set_pc(fregs, new_function); part. Yeah, that's the only bit we'd need to get working. I called this out in "Open questions:" bit on https://kunit-review.googlesource.com/c/linux/+/5109 As for the amount of work, I know how to do KUnit-y things, I have no idea how to do livepatch things :) Also, we're not aiming for something as "magic" as the ftrace one. David's patch is here: https://kunit-review.googlesource.com/c/linux/+/5129 Here's a snippet from the example in that one: static int add_one(int i) { /* This will trigger the stub if active. */ KUNIT_TRIGGER_STATIC_STUB(add_one, i); return i + 1; } i.e. users just add this one macro in with <func> and <args>. It internally expands to roughly if (<check if current test has registered a replacement>) <invoke replacement with <args> So it's all quite simple. But it'd definitely be interesting to try and get klp_arch_set_pc() working on UML if that's a possibility! Speaking from ignorance, I can see this either being somewhat simple or very painful. > > -- Steve
