Messages in this thread | | | From | Dmitry Vyukov <> | Date | Tue, 15 Dec 2015 14:11:12 +0100 | Subject | Re: use-after-free in __perf_install_in_context |
| |
On Fri, Dec 11, 2015 at 9:14 AM, Ingo Molnar <mingo@kernel.org> wrote: > > * Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote: > >> On Thu, Dec 10, 2015 at 10:02:51AM +0100, Peter Zijlstra wrote: >> > On Wed, Dec 09, 2015 at 07:54:35PM -0800, Alexei Starovoitov wrote: >> > > Freeing memory is a requirement regardless. >> > > Even when kernel running with kasan, there must be a way to stop >> > > stack collection and free that memory. >> > > You cannot treat kernel as your test program or 'device under test'. >> > >> > Relax, that is exactly what lockdep does. It cannot dynamically allocate >> > things because allocators use lock etc.. >> > >> > Its fine to build up state for debug bits, esp. if its bounded, like the >> > number of unique callchains. >> >> except the code in question is doing unbounded alloc_pages() > > Yes, but the trick is to still have a bound sized debug pool - which runs out of > entries gracefully. > > Which in practice is plenty enough for most types of testing, and is a lot more > robust than any dynamic scheme.
A hard upper bound on consumed memory would work for us without introducing any slowdown and without increasing code complexity. So it sounds good to me.
| |