Re: [PATCH 1/1] perf build: Require at least clang 16.0.6 to build BPF skeletons

[James Clark]

On 10/10/2024 2:20 am, Arnaldo Carvalho de Melo wrote:
> On Wed, Oct 9, 2024, 10:01 PM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
>
> > On Tue, Oct 08, 2024 at 12:27:24AM -0700, Howard Chu wrote:
> > > Hi Alan, Arnaldo and James,
> > >
> > > This problem was solved in [PATCH 0/2] perf trace: Fix support for the
> > > new BPF feature in clang 12 (Link:
> > https://lore.kernel.org/linux-perf-users/20241007051414.2995674-1-howardchu95@xxxxxxxxx/T/#t
> > ),
> > > sorry I forgot to cc you two.
> > >
> > > Alan's thought was correct. Thank you so much! :)
> >
> > It'd be great if any of you can test this change.  Now I only have
> > machines with clang 16.
>
> I'll test this tomorrow.
>
> - Arnaldo
>
> > Thanks,
> > Namhyung

Tested with clang 15:

$ sudo perf trace -e write --max-events=100 -- echo hello

  0.000 ( 0.014 ms): echo/834165 write(fd: 1, buf: hello\10, count: 6) 

                                           =

Tested-by: James Clark <james.clark@xxxxxxxxxx>


Unrelated to this change, I noticed that with older clangs or with 
BUILD_BPF_SKEL=0 that commit b257fac12f38 ("perf trace: Pretty print 
buffer data") changes the buffer address to print nothing, and the '6' 
return value is missing. Not sure if this was intended or not:

 $ sudo perf trace -e write --max-events=100 -- echo hello

Before:
     0.000 ( 0.009 ms): echo/772951 write(fd: 1, buf: 0x58c415257440,
        count: 6)                           = 6

After:
     0.000 ( 0.009 ms): echo/760370 write(fd: 1, buf: , count: 6) 

                                            =

Thanks
James