Re: [PATCH v2 2/2] kernel.h: Disable -Wreturn-stack-address for _THIS_IP_

[Nick Desaulniers]

On Tue, Jul 31, 2018 at 11:58 AM Nick Desaulniers
<ndesaulniers@xxxxxxxxxx> wrote:
>
> On Tue, Jul 31, 2018 at 10:02 AM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> >
> > On Tue, Jul 31, 2018 at 9:48 AM, Nick Desaulniers
> > > On Tue, Jul 31, 2018 at 3:27 AM kbuild test robot <lkp@xxxxxxxxx> wrote:
> > >>    drivers/net//wireless/intel/iwlwifi/iwl-trans.c: In function 'iwl_trans_send_cmd':
> > >> >> drivers/net//wireless/intel/iwlwifi/iwl-trans.c:137:2: warning: this 'if' clause does not guard... [-Wmisleading-indentation]
> > >>      if (!(cmd->flags & CMD_ASYNC))
> > >>      ^~
> > >>    drivers/net//wireless/intel/iwlwifi/iwl-trans.c:138:1: note: ...this statement, but the latter is misleadingly indented as if it were guarded by the 'if'
> > >>       lock_map_acquire_read(&trans->sync_cmd_lockdep_map);
> > >>     ^ ~
> > >>
> > >> vim +/if +137 drivers/net//wireless/intel/iwlwifi/iwl-trans.c
> > >>
> > >> 92fe8343 Emmanuel Grumbach 2015-12-01 @137      if (!(cmd->flags & CMD_ASYNC))
> > >> 92fe8343 Emmanuel Grumbach 2015-12-01  138              lock_map_acquire_read(&trans->sync_cmd_lockdep_map);
> >
> > #define lock_map_acquire_read(l)
> > lock_acquire_shared_recursive(l, 0, 0, NULL, _THIS_IP_)
> >
> > #define lock_acquire_shared_recursive(l, s, t, n, i)
> > lock_acquire(l, s, t, 2, 1, n, i)
> >
> > The config doesn't have CONFIG_LOCKDEP, so it's not:
> >
> > extern void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
> >                          int trylock, int read, int check,
> >                          struct lockdep_map *nest_lock, unsigned long ip);
> >
> > but rather:
> >
> > # define lock_acquire(l, s, t, r, c, n, i)      do { } while (0)
>
> This is tricky, if I preprocess that translation unit with the exact
> flags used during compilation, I get:
>
> ```
>  if (!(cmd->flags & CMD_ASYNC))
>
> #pragma GCC diagnostic push
>
> #pragma GCC diagnostic pop
>  do { } while (0);
> ```
>
> Which is not enough to trigger -Wmisleading-indentation alone.  It is
> curious that if we add braces to that if statement (as Nathan notes in
> a sibling post) or removing the pop (not shippable) seems to fix the
> warning.

Something fishy is going on here: https://godbolt.org/g/b5dsqH

It seems that gcc's warning is technically correct, but it seems to be
a miscompile as puts() in my reduced test case is called
unconditionally.  I've filed:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86765

In the meanwhile, I've reworked the patch to change _THIS_IP_ to a
only contain a function call, to a new static inline function which
does what the statement expression used to.  This now triggers
-Wreturn-local-addr warnings in gcc, which is a warning added in
gcc-4.8, so I need to add another __diag_ignore, and case for gcc 4.8
to include/linux/compiler-gcc.h.

At this point, I think I might as well consolidate current_text_addr()
and _THIS_IP_.  Stay tuned for v3.

--
Thanks,
~Nick Desaulniers