Re: Re: [PATCH v5 2/2] [BUGFIX] kprobes: Fix "Failed to find blacklist" error on ia64 and ppc64

[Michael Ellerman <mpe@ellerman.id.au>]

On Tue, 2014-07-15 at 11:24 +0900, Masami Hiramatsu wrote:
> (2014/07/15 11:11), Michael Ellerman wrote:
> > On Mon, 2014-07-14 at 10:17 -0700, Tony Luck wrote:
> >> On Tue, Jul 8, 2014 at 5:07 AM, Masami Hiramatsu
> >> <masami.hiramatsu.pt@hitachi.com> wrote:
> >>> Ping?
> >>>
> >>> This patch can be applied without 1/2, and will fix ia64/ppc64 problem.
> >>
> >> Is somebody going to push this upstream? Another week has gone by,
> >> we are at -rc5, and I'm still seeing the
> >>
> >>   Failed to find blacklist a00000010133b150
> >>
> >> messages on ia64.
> > 
> > I don't see those messages on ppc64, I don't know where the original report
> > that it was broken on ppc64 came from. So I'm a bit lukewarm on the patch.
> 
> Right, on ppc64(ABIv1) it may be silently failed. Because each function
> descriptor has another entry on kallsyms, original code can't detect
> that. 

OK, that would have been good to know :)

It's actually much worse than you describe. On ppc64 (ABIv1) we are
successfully blacklisting the function descriptors. But that doesn't prevent
you from probing the text address. So basically NOKPROBE_SYMBOL() does nothing
useful for us.

$ head -2 ../kprobes/blacklist 
0xc000000000d4cff8-0xc000000000d4d010	notify_die
0xc000000000d4cf80-0xc000000000d4cf98	atomic_notifier_call_chain
$ echo "p:atomic_notifier_call_chain .atomic_notifier_call_chain" > kprobe_events
$ echo 1 > events/kprobes/enable 
$ ls
available_events      instances       saved_cmdlines       trace_options
available_tracers     kprobe_events   saved_cmdlines_size  trace_pipe
buffer_size_kb        kprobe_profile  set_event            tracing_cpumask
buffer_total_size_kb  options         snapshot             tracing_max_latency
current_tracer        per_cpu         trace                tracing_on
events                printk_formats  trace_clock          tracing_thresh
free_buffer           README          trace_marker
$ Dumping ftrace buffer:
cpu 0x2: Vector: 400 (Instruction Access) at [c0000001defaf830]
    pc: 0000000000000000
    lr: 0000000000000001
    sp: c0000001defafab0
   msr: 8000000140009032
  current = 0xc0000001def57e40
  paca    = 0xc00000000fe00800	 softe: 0	 irq_happened: 0x01
    pid   = 1, comm = swapper/2
cpu 0x3: Vector: 400 (Instruction Access) at [c0000001ddbcc640]
    pc: 0000000000000000
    lr: 0000000000000000
    sp: c0000001ddbcc8c0
   msr: 8000000040001032
  current = 0xc0000001def5a100
  paca    = 0xc00000000fe00c00	 softe: 0	 irq_happened: 0x01
    pid   = -554326528, comm = 

Dead machine.


With your patch:

$ head -2 kprobes/blacklist 
0xc0000000000bf860-0xc0000000000bf8b0	.notify_die
0xc0000000000bf750-0xc0000000000bf780	.atomic_notifier_call_chain

$ echo "p:notify_die .notify_die" > tracing/kprobe_events
-bash: echo: write error: Invalid argument

So that is much better.

cheers