On Fri, Dec 06, 2013 at 01:52:29PM +0100, Richard Biener wrote:
> > Note that testing still shows some issues, e.g. because the
> > kThreadDescriptorSize change has not been applied. But it at least builds
> > and for many tests works.
>
> What's wrong with just copying the kernel headers 1:1? ISTR that is what
> kernel folks recommend if you use kernel APIs that don't have a glibc
> correspondent.
Well, for many things that is already the case, sanitizer_common has
apparently it's own __sanitizer_* macros/constants/structures that
are supposed to match kernel ones. The problem is that
the code is comparing them as compile time assertions against the kernel
ones inside of the library code. IMHO that is not a very good idea, much
better would be just say a testcase that would include the sanitizer +
kernel headers, guarded by recent enough LINUX_VERSION_CODE or configure or
similar, so it wouldn't prevent library build on older kernel headers,
the kernel ABI better be stable (only new things added, not size of
structures/magic constants etc. changed from time to time).
But Kostya is apparently not willing to do that, so this patch provides
a workaround in non-compiler-rt maintained files.
> Of course you then need runtime checks on whether the running kernel
> supports the API.
Usually not, say if the structures/magic constants etc. are to wrap syscalls
or ioctls etc., if kernel doesn't support those syscalls/ioctls, either
nothing will use those syscalls/ioctls, or if it does, it will fail from the
kernel, worst case you get some diagnostics from pre-syscall/ioctl wrapper,
otherwise the kernel syscall/ioctl will just fail and post-syscall/ioctl
wrapper will do likely nothing at all.
Jakub
