Re: [GIT PULL] perf tools changes for v6.4

[Jiri Olsa]

On Wed, May 03, 2023 at 10:51:15PM -0700, Ian Rogers wrote:
> On Wed, May 3, 2023 at 8:12 PM Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Wed, May 3, 2023 at 8:00 PM Linus Torvalds
> > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > I did consider it, but the end result doesn't even build, so I unpulled again..
> > >
> > > I get some libbpf error, and I'm just not interested in trying to
> > > debug it. This has clearly not been tested well enough to be merged.
> >
> > Side note: its' not even about testing.
> >
> > The error message makes it clear that this is garbage and should never
> > be merged even if it were to compile.
> >
> > There is not a way in hell that it is correct that a 'perf' tool build
> > should ever even look at the vmlinux binary to build.
> >
> > The fact that it does shows that something is seriously wrong in
> > perf-tool land, and I will not be touching any pulls until that
> > fundamental mistake is entirely gone.
> >
> > The vmlinux image that is present in my tree (ie
> > /home/torvalds/v2.6/linux/vmlinux) is a test build with an insane
> > config. And the fact that the perf tool even looks at it is seriously
> > broken.
> >
> > Whatever you are doing - stop it right now.
> >
> >                Linus
> 
> I think the error you gave makes it pretty clear what is going on and
> Arnaldo's e-mail explains the motivation. Perhaps we can check a
> vmlinux.h into the perf tree so that we don't default to generating
> it. This would avoid the binary dependency but we may need different
> flavors for different architectures because of structs like pt_regs.

I think we could check that vmlinux with .BTF is present before
allowing to build skeletons

jirka