On Fri, Feb 08, 2019 at 05:12:21PM +1100, Michael Ellerman wrote: > Segher Boessenkool <[email protected]> writes: > > On Thu, Feb 07, 2019 at 10:53:13PM +1100, Michael Ellerman wrote: > >> Chandan reported that fstests' generic/026 test hit a crash: > > > >> The instruction dump decodes as: > >> subfic r6,r5,8 > >> rlwinm r6,r6,3,0,28 > >> ldbrx r9,0,r3 > >> ldbrx r10,0,r4 <- > >> > >> Which shows us doing an 8 byte load from c00000062ac3fff9, which > >> crosses the page boundary at c00000062ac40000 and faults. > >> > >> It's not OK for memcmp to read past the end of the source or > >> destination buffers. > > > > It's not okay to access memory pages unsolicited. Reading past the end > > is fine per se. > > Yeah I guess that's true. > > Things like KASAN/valgrind probably disagree, but KASAN at least > overrides memcmp AIUI. > > I guess I feel better about it not reading past the end of the buffers, > but maybe I'm being paranoid.
Sure, and that may be the best thing to do in the kernel. OTOH, newer GCC will inline many mem* for powerpc, and it will access past the end of strings and buffers (but not past 4kB boundaries). > The other complication is we support multiple page sizes, so detecting a > page boundary is more complicated than it could be. Yeah. > So I guess I'm inclined to stick with this approach, but I can update > the change log. Thanks! I mentioned it because this was the bug that was hit here: reading past the end had no ill effect (as far as we know), but accessing the wrong page did :-) Segher
