Re: [lwip-users] Compile errors with lwip 0.7.0 and contrib-20040122

[Peter Graf]

John Taylor wrote:

> > > > Another thing that has puzzled me a bit, is that sys_msleep() is now
> > > > forced to use the overhead of allocating and deallocating a semaphore
> > > > on every call. Wouldn't it be better to allow sys_msleep() to be
> > > > implemented in sys_arch.c?
> > > >   
> > > >
> > > >        
> > > I remember looking at that long ago and I am not sure why but did not
> > > change anything. Maybe the fact that it uses lwip semaphores allows lwip
> > > timers to fire whereas a native sleep would not trigger the timers of
> > > the sleeping thread.You could check and see if that's the case, I'll doit with 
> > > the ecos port sometime soon.
> > >
> > >      
> > IIRC that's not the case. So far sys_msleep() is only used for PPP, 
> > where native sleep had been working OK.
> >    
> No, the sys_msleep() call is required for the lwip timers to fire
> properly.
sys_msleep() is not required for the lwIP timers at all. Just do a find/grep on 
the code and you'll see that it is nowhere called except from PPP. I have lwIP 
running nicely with sys_msleep() compeletely removed.

If there's a technical need for the usage of semaphores here, it must be PPP 
specific, but I doubt even that. AFAICS sys_msleep() seems only to be used 
there during password check, waiting for the link, close, and when handling an 
exceptional return of sio_read.

> Using a native sleep will delay the firing of all 'active'
> timers at the time the sleep call is made.
In a multitasking OS, there's no good reason why a sleeping task should 
delay any timers. And for non-multitasking lwIP, sys_msleep() isn't 
defined. BTW I didn't even say I wish a native call, just sys_arch.c to 
layer it.

>  Now, if there are no
> 'active' timers when the thread is sleeping - no harm, no foul.  Or if
> the amount of time spent sleeping is relatively small compared to the
> timer period, it is probably not noticeable.  However, as good practice,
> I recommend *not* replacing the sys_msleep() with a native sleep call. 
>  
Implementing it in sys_arch.c means not necessarily a direct native 
replacement.

> FYI: This is fresh on my mind, because I too thought the implementation
> of sys_msleep() was "klunky" until I completely understood how lwip
> timers work - then it make perfect sense.
>  
Disagreed :-)

All the best
Peter