Re: [PATCH] x86/stacktrace: update kconfig help text for reliable unwinders

[Josh Poimboeuf]

On Wed, Nov 06, 2019 at 06:16:28PM -0500, Joe Lawrence wrote:
> On 11/6/19 6:05 PM, Josh Poimboeuf wrote:
> > On Wed, Nov 06, 2019 at 05:43:44PM -0500, Joe Lawrence wrote:
> > > commit 6415b38bae26 ("x86/stacktrace: Enable HAVE_RELIABLE_STACKTRACE
> > > for the ORC unwinder") marked the ORC unwinder as a "reliable" unwinder.
> > > Update the help text to reflect that change: the frame pointer unwinder
> > > is no longer the only one that provides HAVE_RELIABLE_STACKTRACE.
> > > 
> > > Signed-off-by: Joe Lawrence <joe.lawrence@xxxxxxxxxx>
> > > ---
> > >   arch/x86/Kconfig.debug | 8 ++++----
> > >   1 file changed, 4 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
> > > index bf9cd83de777..69cdf0558c13 100644
> > > --- a/arch/x86/Kconfig.debug
> > > +++ b/arch/x86/Kconfig.debug
> > > @@ -316,10 +316,6 @@ config UNWINDER_FRAME_POINTER
> > >   	  unwinder, but the kernel text size will grow by ~3% and the kernel's
> > >   	  overall performance will degrade by roughly 5-10%.
> > > -	  This option is recommended if you want to use the livepatch
> > > -	  consistency model, as this is currently the only way to get a
> > > -	  reliable stack trace (CONFIG_HAVE_RELIABLE_STACKTRACE).
> > > -
> > >   config UNWINDER_GUESS
> > >   	bool "Guess unwinder"
> > >   	depends on EXPERT
> > > @@ -333,6 +329,10 @@ config UNWINDER_GUESS
> > >   	  useful in many cases.  Unlike the other unwinders, it has no runtime
> > >   	  overhead.
> > > +	  This option is not recommended if you want to use the livepatch
> > > +	  consistency model, as this unwinder cannot guarantee reliable stack
> > > +	  traces.
> > > +
> > 
> > I'm not sure whether this last hunk is helpful.  At the very least the
> > wording of "not recommended" might be confusing because it's not even
> > possible to combine UNWINDER_GUESS+HAVE_RELIABLE_STACKTRACE.
> > 
> > arch/x86/Kconfig:       select HAVE_RELIABLE_STACKTRACE         if X86_64 && (UNWINDER_FRAME_POINTER || UNWINDER_ORC) && STACK_VALIDATION
> > 
> 
> Ah good point.  The alternative would be to copy the recommended note to
> both UNWINDER_FRAME_POINTER and UNWINDER_ORC, or at least remove the "only"
> phrasing.  I dunno, nobody has noticed it yet, so maybe the first hunk would
> be good enough.

The guess unwinder is listed as an expert option, so the user would have
to enable CONFIG_EXPERT to get to it.  Then if they do enable it and try
to load a livepatch, they'll get a runtime warning along with a patch
which never transitions.

And like you said, nobody has complained, so I'd say just dropping the
2nd hunk would be fine.

-- 
Josh