[RFC xserver] os: log a bug whenever WriteToClient is called from the input thread

[Olivier Fourdan]

Hi,

> > The input thread should generate events, not send them. Make it easier to
> > find the instances where it's doing so.
> > 
> > Signed-off-by: Peter Hutterer <peter.hutterer at who-t.net>
> > ---
> >  include/input.h  | 1 +
> >  os/inputthread.c | 7 +++++++
> >  os/io.c          | 6 ++++++
> >  3 files changed, 14 insertions(+)
> > 
> > diff --git a/include/input.h b/include/input.h
> > index bb58b22..6c9e45d 100644
> > --- a/include/input.h
> > +++ b/include/input.h
> > @@ -722,6 +722,7 @@ extern _X_HIDDEN void
> > input_constrain_cursor(DeviceIntPtr
> > pDev, ScreenPtr screen
> >  extern _X_EXPORT void input_lock(void);
> >  extern _X_EXPORT void input_unlock(void);
> >  extern _X_EXPORT void input_force_unlock(void);
> > +extern _X_EXPORT int in_input_thread(void);
> >  
> >  extern void InputThreadPreInit(void);
> >  extern void InputThreadInit(void);
> > diff --git a/os/inputthread.c b/os/inputthread.c
> > index 4400fba..933bc1c 100644
> > --- a/os/inputthread.c
> > +++ b/os/inputthread.c
> > @@ -90,6 +90,12 @@ static pthread_mutex_t input_mutex;
> >  static Bool input_mutex_initialized;
> >  #endif
> >  
> > +int
> > +in_input_thread(void)
> > +{
> > +    return pthread_equal(pthread_self(), inputThreadInfo->thread);
> > +}
> > +
> >  void
> >  input_lock(void)
> >  {
> > @@ -531,6 +537,7 @@ void input_force_unlock(void) {}
> >  void InputThreadPreInit(void) {}
> >  void InputThreadInit(void) {}
> >  void InputThreadFini(void) {}
> > +int in_input_thread(void) { return 0; }
> >  
> >  int InputThreadRegisterDev(int fd,
> >                             NotifyFdProcPtr readInputProc,
> > diff --git a/os/io.c b/os/io.c
> > index be85226..5494b30 100644
> > --- a/os/io.c
> > +++ b/os/io.c
> > @@ -643,6 +643,9 @@ SetCriticalOutputPending(void)
> >   *    this routine as int.
> >   *****************/
> >  
> > +extern inline int
> > +in_input_thread(void);
> > +
> >  int
> >  WriteToClient(ClientPtr who, int count, const void *__buf)
> >  {
> > @@ -651,6 +654,9 @@ WriteToClient(ClientPtr who, int count, const void
> > *__buf)
> >      int padBytes;
> >      const char *buf = __buf;
> >  
> > +    BUG_RETURN_VAL_MSG(in_input_thread(), 0,
> > +                       "******** %s called from input thread *********\n",
> > __func__);
> > +
> >  #ifdef DEBUG_COMMUNICATION
> >      Bool multicount = FALSE;
> >  #endif
> > --
> > 2.9.3
> 
> It works a treat, brilliant - Within not even a second approaching the stylus
> to the tablet, Xorg was logging messages about calling WriteToClient() from
> the input thread, it's just awesome!
> 
> Tested-by: Olivier Fourdan <ofourdan at redhat.com>
> 
> And if I'd dare, fwiw, it looks good to me.
> 
> Reviewed-by: Olivier Fourdan <ofourdan at redhat.com>
  ^^^^^^^^^^^

Err, nope! I take that back, sorry, looks like it's causing trouble to Xwayland... it crashes.

> +int
> +in_input_thread(void)
> +{
> +    return pthread_equal(pthread_self(), inputThreadInfo->thread);
> +}

If this gets called *before* the InputThread is even created, inputThreadInfo is NULL.

I would add a

     if (!inputThreadInfo)
        return 0;

Otherwise Xwayland crashes at startup. With something like that added, it works :)

Cheers,
Olivier