On Tue, Oct 03, 2017 at 09:41:36AM -0500, Josh Poimboeuf wrote:
> On Tue, Oct 03, 2017 at 09:31:47AM -0500, Josh Poimboeuf wrote:
> > On Tue, Oct 03, 2017 at 10:06:34PM +0800, Fengguang Wu wrote:
> > > Hi Byungchul,
> > >
> > > This patch triggers a NULL-dereference bug at update_stack_state().
> > > Although its parent commit also has a NULL-dereference bug, however
> > > the call stack looks rather different. Both dmesg files are attached.
> > >
> > > It also triggers this warning, which is being discussed in another
> > > thread, so CC Josh. The full dmesg attached, too.
> > >
> > > Please press Enter to activate this console.
> > > [ 138.605622] WARNING: kernel stack regs at be299c9a in
> > > procd:340 has bad 'bp' value 000001be
> > > [ 138.605627] unwind stack type:0 next_sp: (null) mask:0x2
> > > graph_idx:0
> > > [ 138.605631] be299c9a: 299ceb00 (0x299ceb00)
> > > [ 138.605633] be299c9e: 2281f1be (0x2281f1be)
> > > [ 138.605634] be299ca2: 299cebb6 (0x299cebb6)
> >
> > I suspect the bug is in:
> >
> > ce07a9415f26 ("locking/lockdep: Make check_prev_add() able to handle
> > external stack_trace")
> >
> > It converts the stack-allocated stack_trace struct from static to
> > non-static, yet still adds it to a list. Does this fix it?
>
> Actually, I spoke too soon. It's not actually adding the pointer to the
> list, it's copying its contents. So never mind...
I don't know the lockdep code, but one more comment from the peanut
gallery. This code looks suspect to me:
/*
* Stop saving stack_trace if save_trace() was
* called at least once:
*/
if (save && ret == 2)
save = NULL;
>From looking at check_prev_add(), a return value of 2 doesn't
necessarily imply that save_trace() was called. If the
check_redundant() call returns 0, then check_prev_add() can return 2,
and the trace will still be uninitialized, but save will be set to NULL
even though save_trace() hasn't been called. Then a subsequent call to
check_prev_add() could add an uninitialized stack_trace struct to the
dependency list.
I could be wrong, but it's at least something the lockdep folks might
want to look at.
--
Josh
