Re: [GIT PULL] sched.h split-up

[Linus Torvalds]

On Thu, Mar 2, 2017 at 5:36 PM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
> I tried to test as many build configurations as possible, but some build breakage
> is probably still left - but it should be mostly limited to architectures that
> have no cross-compiler binaries available on kernel.org, and non-default
> configurations.

So I merged this, and it's pushed out, but it showed a real annoyance
almost immediately.

That annoyance is also pushed out - it's the pull of the overlayfs
updates. And the reason that was annoying was that that pull caused
this build error due to a subtle semantic clash:

   fs/overlayfs/util.c: In function âovl_copy_up_startâ:
   ./include/linux/wait.h:634:7: error: implicit declaration of
function âsignal_pendingâ [-Werror=implicit-function-declaration]
      if (signal_pending(current)) {    \
          ^
   ./include/linux/wait.h:679:9: note: in expansion of macro
â__wait_event_interruptible_lockedâ
      ? 0 : __wait_event_interruptible_locked(wq, condition, 0, 0))
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   fs/overlayfs/util.c:275:8: note: in expansion of macro
âwait_event_interruptible_lockedâ
     err = wait_event_interruptible_locked(ofs->copyup_wq, !oe->copying);
           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

and that's just complete bullshit.

The thing is, fs/overlayfs/util.c included <linux/fs.h>, which in turn
includes <linux/wait.h>

So using "wait_event_interruptible_locked()" should damn well have worked.

The fact that you can include <linux/wait.h>, and then cannot use the
wait event functions because you're missing "signal_pending()" is
complete garbage. This needs to be fixed.

The most obvious way to fix it is likely to just shrink that
__wait_event_interruptible_locked() macro.

Yes, yes, some of it absolutely does need to remain as a macro,
because it has that

   do { .. } while (condition);

part of it, but I think the stuff inside that while loop could be made
into a function of its own. Actually, maybe two different functions
(for the "irq" case on/off).

Anyway, I fixed the semantic merge error by just including thar
<linux/sched/signal.h> file, but this is just not acceptable.

Having to have files know that if they use the wait-event functins
(well, _some_ of them), they not only need to include <linux/wait.h>,
they need to completely illogically also include
<linux/sched/signal.h> is just not maintainable.

And who knows what similar semantic cases I _won't_ notice, because
they occur in code I don't build even with my allmodconfig builds?

                     Linus