On Fri, Dec 04, 2020 at 11:00:24AM -0800, Andrii Nakryiko wrote:
> On Fri, Dec 4, 2020 at 1:41 AM Brendan Jackman <jackmanb@xxxxxxxxxx> wrote:
> >
> > On Thu, Dec 03, 2020 at 01:01:27PM -0800, Andrii Nakryiko wrote:
> > > On Thu, Dec 3, 2020 at 8:07 AM Brendan Jackman <jackmanb@xxxxxxxxxx> wrote:
> > > >
> > > > This is somewhat cargo-culted from the libbpf build. It will be used
> > > > in a subsequent patch to query for Clang BPF atomics support.
> > > >
> > > > Change-Id: I9318a1702170eb752acced35acbb33f45126c44c
> > >
> > > Haven't seen this before. What's this Change-Id business?
> >
> > Argh, apologies. Looks like it's time for me to adopt a less error-prone
> > workflow for sending patches.
> >
> > (This is noise from Gerrit, which we sometimes use for internal reviews)
> >
> > > > Signed-off-by: Brendan Jackman <jackmanb@xxxxxxxxxx>
> > > > ---
> > > > tools/testing/selftests/bpf/.gitignore | 1 +
> > > > tools/testing/selftests/bpf/Makefile | 38 ++++++++++++++++++++++++++
> > > > 2 files changed, 39 insertions(+)
> > >
> > > All this just to detect the support for clang atomics?... Let's not
> > > pull in the entire feature-detection framework unnecessarily,
> > > selftests Makefile is complicated enough without that.
> >
> > Then the test build would break for people who haven't updated Clang.
> > Is that acceptable?
> >
> > I'm aware of cases where you need to be on a pretty fresh Clang for
> > tests to _pass_ so maybe it's fine.
>
> I didn't mean to drop any detection of this new feature. I just didn't
> want a new dependency on tools' feature probing framework. See
> IS_LITTLE_ENDIAN and get_sys_includes, we already have various feature
> detection-like stuff in there. So we can do this with a one-liner. I
> just want to keep it simple. Thanks.
Ah right gotcha. Then yeah I think we can do this:
BPF_ATOMICS_SUPPORTED = $(shell \
echo "int x = 0; int foo(void) { return __sync_val_compare_and_swap(&x, 1, 2); }" \
| $(CLANG) -x cpp-output -S -target bpf -mcpu=v3 - -o /dev/null && echo 1 || echo 0)
