Re: kprobes sanity test fails on next-20190708

[Mark Rutland]

On Tue, Jul 09, 2019 at 11:25:48AM -0400, Steven Rostedt wrote:
> On Tue, 9 Jul 2019 21:36:57 +0900
> Masami Hiramatsu <masami.hiramatsu@xxxxxxxxx> wrote:
> 
> > On Tue, 9 Jul 2019 12:19:15 +0200
> > Anders Roxell <anders.roxell@xxxxxxxxxx> wrote:
> > 
> > > On Mon, 8 Jul 2019 at 17:56, James Morse <james.morse@xxxxxxx> wrote:  
> > > >
> > > > Hi,
> > > >
> > > > On 08/07/2019 15:11, Anders Roxell wrote:  
> > > > > argh... resending, with plaintext... Sorry =/
> > > > >
> > > > > I tried to build a next-201908 defconfig + CONFIG_KPROBES=y and
> > > > > CONFIG_KPROBES_SANITY_TEST=y
> > > > >
> > > > > I get the following Call trace, any ideas?
> > > > > I've tried tags back to next-20190525 and they also failes... I haven't
> > > > > found a commit that works yet.
> > > > >
> > > > > [    0.098694] Kprobe smoke test: started
> > > > > [    0.102001] audit: type=2000 audit(0.088:1): state=initialized
> > > > > audit_enabled=0 res=1
> > > > > [    0.104753] Internal error: aarch64 BRK: f2000004 [#1] PREEMPT SMP  
> > > >
> > > > This sounds like the issue Mark reported:
> > > > https://lore.kernel.org/r/20190702165008.GC34718@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > >
> > > > It doesn't look like Steve's patch has percolated into next yet:
> > > > https://lore.kernel.org/lkml/20190703103715.32579c25@xxxxxxxxxxxxxxxxxx/
> 
> I forgot to push it after sending it. :-(
> 
> > > >
> > > > Could you give that a try to see if this is a new issue?  
> > > 
> > > The patch didn't apply cleanly.
> > > However, when I resolved the issue it works.
> > > I'm a bit embarrassed since I now remembered that I reported it a while back.
> > > https://lore.kernel.org/lkml/20190625191545.245259106@xxxxxxxxxxx/
> > > 
> > > Both patches resolved the issue.
> > > I've tested both.  
> > 
> > In that case, the later one (move postcore to subsys) seems good to me.
> > 
> > Delaying the test is just avoiding the issue that the selftest found,
> > since right after init_kprobes() are called, the kprobe is ready for use.
> > This means that the selftest must be run as the first user of the kprobes
> > and it must be run right after initialize kprobes.
> 
> I agree. I pushed to my repo in the for-next branch. Care to test that?
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace.git

I've just given that a spin with KPROBES and KPROBES_SANITY_TEST
selected, and it boots cleanly for me. FWIW:

Tested-by: Mark Rutland <mark.rutland@xxxxxxx>

Thanks,
Mark.