On Tue, Nov 07, 2017 at 08:32:19AM -0800, Linus Torvalds wrote: > On Tue, Nov 7, 2017 at 1:32 AM, Fengguang Wu <[email protected]> wrote: > > > > [ 7.795097] Kprobe smoke test: started > > [ 7.807563] > > ================================================================== > > [ 7.808007] BUG: KASAN: stack-out-of-bounds in deref_stack_reg+0xb5/0x11a > > Ok, this is apparently with a new enough gcc to make it valid: gcc > version 6.2.0. > > Adding Josh to the list of people, since it's in the unwinder. Josh, > I'll forward the original report with full dmesg and config in > private.
It looks like the unwinder got confused because of a kprobe breakpoint at the beginning of the kprobe_target() function. At least it only read a bad address *on* the stack. It didn't go off beyond the stack -- it has safeguards for that. I don't think there's a good solution for this, unless we had some kind of unwinder integration with generated/patched code. I think I'll just have to disable KASAN warnings in the ORC code. -- Josh
