On 5/2/16 23:35, Alexander Potapenko wrote: > On Mon, May 2, 2016 at 5:13 PM, Chen Gang <chengang@xxxxxxxxxxxxxxxx> wrote: >> >> OK. But it does not look quite easy to use kasan_disable_current() for >> INIT_KASAN which is used in INIT_TASK. >> >> If we have to set "kasan_depth == 1", we have to use kasan_depth-- in >> kasan_enable_current(). > Agreed, decrementing the counter in kasan_enable_current() is more natural. > I can fix this together with the comments. OK, thanks. And need I also send patch v2 for include/linux/kasan.h? (or you will fix them together). >> >> If we don't prevent the overflow, it will have negative effect with the >> caller. When we issue an warning, it means the caller's hope fail, but >> can not destroy the caller's original work. In our case: >> >> - Assume "kasan_depth-- for kasan_enable_current()", the first enable >> will let kasan_depth be 0. > Sorry, I'm not sure I follow. > If we start with kasan_depth=0 (which is the default case for every > task except for the init, which also gets kasan_depth=0 short after > the kernel starts), > then the first call to kasan_disable_current() will make kasan_depth > nonzero and will disable KASAN. > The subsequent call to kasan_enable_current() will enable KASAN back. > > There indeed is a problem when someone calls kasan_enable_current() > without previously calling kasan_disable_current(). > In this case we need to check that kasan_depth was zero and print a > warning if it was. > It actually does not matter whether we modify kasan_depth after that > warning or not, because we are already in inconsistent state. > But I think we should modify kasan_depth anyway to ease the debugging. > For me, BUG_ON() will be better for debugging, but it is really not well for using. For WARN_ON(), it already print warnings, so I am not quite sure "always modifying kasan_depth will be ease the debugging". When we are in inconsistent state, for me, what we can do is: - Still try to do correct things within our control: "when the caller make a mistake, if kasan_enable_current() notices about it, it need issue warning, and prevent itself to make mistake (causing disable). - "try to let negative effect smaller to user", e.g. let users "loose hope" (call enable has no effect) instead of destroying users' original work (call enable, but get disable). Thanks. -- Chen Gang (陈刚) Managing Natural Environments is the Duty of Human Beings. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>