Re: [PATCH v2] x86/dumpstack: allow preemption in show_stack_log_lvl() and dump_trace()

[Josh Poimboeuf]

On Tue, Sep 13, 2016 at 08:29:57PM +0200, Ingo Molnar wrote:
> 
> * Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> 
> > show_stack_log_lvl() and dump_trace() are already preemption safe:
> > 
> > - If they're running in irq or exception context, preemption is already
> >   disabled and the percpu stack pointers can be trusted.
> > 
> > - If they're running with preemption enabled, they must be running on
> >   the task stack anyway, so it doesn't matter if they're comparing the
> >   stack pointer against a percpu stack pointer from this CPU or another
> >   one: either way it won't match.
> 
> Yeah, so I'm having second thoughts about this patch. My worry here is: what if we 
> get preempted in this sequence?
> 
> If the kernel is borked real bad then we could get technically correct but really, 
> really weird looking stack traces if for example the task stack is getting 
> corrupted or something like that.

If it's in the oops or BUG path, there can't be preemption anyway
because oops_begin() disables interrupts.

It does look like the WARN path could get preempted.  Not to mention all
the other callers of show_regs(), dump_stack(), show_stack_log_lvl(),
etc.  In those cases, if the stack dump got preempted in the middle, and
then another task dumped its stack, the two dumps could be interspersed
a bit which would indeed be a little confusing.

But that would be quite rare.  And anyway, we already have the same
issue today when two CPUs are dumping the stack at the same time.  So I
don't think it's much of an issue.

> Dunno. How long does the worst-case processing here take on a typical x86 system, 
> does it really matter to scheduling latency?

I haven't heard any complaints about latency.  The goal was just to try
to simplify the code a bit.

-- 
Josh