Re: [PATCH bpf-next 08/11] samples: bpf: fix shifting unsigned long by 32 positions

[Yonghong Song]

On 4/20/22 10:18 AM, Andrii Nakryiko wrote:
> On Thu, Apr 14, 2022 at 3:46 PM Alexander Lobakin <alobakin@xxxxx> wrote:
> > On 32 bit systems, shifting an unsigned long by 32 positions
> > yields the following warning:
> >
> > samples/bpf/tracex2_kern.c:60:23: warning: shift count >= width of type [-Wshift-count-overflow]
> >          unsigned int hi = v >> 32;
> >                              ^  ~~
>
> long is always 64-bit in BPF, but I suspect this is due to
> samples/bpf/Makefile still using this clang + llc combo, where clang
> is called with native target and llc for -target bpf. Not sure if we
> are ready to ditch that complicated combination. Yonghong, do we still
> need that or can we just use -target bpf in samples/bpf?

Current most bpf programs in samples/bpf do not use vmlinux.h and CO-RE.
They direct use kernel header files. That is why clang C -> IR 
compilation still needs to be native.

We could just use -target bpf for the whole compilation but that needs 
to change the code to use vmlinux.h and CO-RE. There are already a 
couple of sample bpf programs did this.

> > The usual way to avoid this is to shift by 16 two times (see
> > upper_32_bits() macro in the kernel). Use it across the BPF sample
> > code as well.
> >
> > Fixes: d822a1926849 ("samples/bpf: Add counting example for kfree_skb() function calls and the write() syscall")
> > Fixes: 0fb1170ee68a ("bpf: BPF based latency tracing")
> > Fixes: f74599f7c530 ("bpf: Add tests and samples for LWT-BPF")
> > Signed-off-by: Alexander Lobakin <alobakin@xxxxx>
> > ---
> >   samples/bpf/lathist_kern.c      | 2 +-
> >   samples/bpf/lwt_len_hist_kern.c | 2 +-
> >   samples/bpf/tracex2_kern.c      | 2 +-
> >   3 files changed, 3 insertions(+), 3 deletions(-)
>
> [...]