On Sun, 14 Mar 2010, Paul Brook wrote: > > On Sat, 13 Mar 2010, Paul Brook wrote: > > > > a) elfload.c:859 > > > > > > > > #define TARGET_ELF_PAGESTART(_v) ((_v) & ~(unsigned > > > > long)(TARGET_ELF_EXEC_PAGESIZE-1)) > > > > > > > > This means that for 64bit guest on a 32bit host the _v's value is > > > > silently reduced to 32bit, the cast should be abi_ulong. > > > > > > > > b) mmap.c:428 > > > > > > > > real_start = start & qemu_host_page_mask; > > > > > > > > Same thing basically qemu_host_page_mask is unsigned long and so > > > > the upper bits are sliently cleared. Again qemu_host_page_mask > > > > should probably be abi_ulong. > > > > > > > > The above two make elf binary with one of the segments above 4G load > > > > and run on 32bit guest only to fail well into execution and without > > > > any indication that it shouldn't have been allowed to run in the first > > > > place. > > > > > > I'd be amazed if these are the only two issues, and expect 64-bit guests > > > on 32-bit hosts to be generally unsafe. > > > > These are the two issues i've been bitten by while trying to solve > > x86_64's far jumps, if you are okay with code silently misbehaving > > fine by me. > > I never said this was ok behavior. Only that you've probably only discovered > the tip of the iceberg.
Even on this path (mmap one) it's indeed not, further bellow there are yet more truncations due to improperly sized and's. Furthermore the abi_ulong sized `start' argument is passed to mmap without ever checking if it fits 32bit, question is what to do about it. -- mailto:av1...@comtv.ru
