On 2018/06/19 20:47, Dmitry Vyukov wrote: >> If no, we would want a git tree for testing under syzbot. > > Thinking of this, we could setup a sandbox instance that won't report > anything over email at all, but crashes will be available on the web. > We could point this instance to a custom git tree, where additional > patches can be applied as necessary. The main question is: who and > how will manage this tree? The tree needs to be rebased periodically, > patches applied, old patches taken out. > The simplest approach (if Linus can tolerate) would be to merge any debug printk() patches to linux.git with single kernel config option (e.g. CONFIG_DEBUG_AID_FOR_SYZBOT=y) for turning them on/off. Assuming that we will eventually be able to move to saving and analyzing vmcore files, some bit of debug printk() code would be tolerable for now? Other possible approach (if Andrew can tolerate) would be to use linux-next.git and carry debug printk() patches via mmotm tree.
