xserver: Cleaning up memory allocation functions and macros

Magnus Vigerlöf Magnus.Vigerlof at home.se
Wed May 2 10:53:40 PDT 2007


On tisdag 01 maj 2007, Egbert Eich wrote:
> Magnus Vigerlöf writes:
>  > Imo that's a lazy programmers argument. I can't think of any case where
>  > crashing the Xserver with only an 'Out of memory' in the log will be a
>  > meaningful error message for our end-users.
>
> Lazy programmer, yes.
> These should *never* be called in a running Xserver. I would consider
> them only useful at *startup( time when no clients are yet connected
> and for the unliekly case when there is no memory.
> If for example the driver wants to allocate memory for some internal
> structures but doesn't get it.
> In this case the programmer would just do:
>    if (!(foo = (whatever *)xalloc(sizeof(whatever)))) {
>       xf86DrvMsg(...,"Out of memory\n");
>       return NULL;
>    }
> and will continue for a while and may crash some point later when some
> code decides that there is no use to continue when no more memory is
> available at that time.
>    if we do:
>       foo = (whatever *)xnfalloc(sizeof(whatever));
> In this case we fail immediately.
>
>  > If this have been an argument about an X-client I would probably not be
>  > as bothered by this, as that would only affect that program. Annoying,
>  > yes but that wouldn't kill all my other applications that I have
>  > running, which would be the case we're talking about here.
>
> I think I mentioned that the use of xnfalloc() outside of initialization
> would be a *bad* abuse.
> - When you get an out of memory at initiaization there is probably no use
>   continuing.
> - When you get an out of memory *after* initialization you need to be
>   much more careful with what yo do.

Good, then we have the same view of xnf*. The thing I see is that more and 
more devices are getting hot-pluggable. Input devices are next in queue, and 
I'm just waiting for people wanting to use VGA-adapters connected via USB or 
other hot-pluggable busses.

So the code where calling xnf* is not a problem is steadily decreasing and 
making sure we then uses proper fault propagation instead of crashing will 
hopefully avoid future headache and patches.

xf86AddDriver/xf86AddInputDriver are excellent examples of functions that 
hasn't been a 'problem' but can be when hotplugging is working. They also 
doesn't have a return value so we'll have to change these in order to get rid 
of the xnf*-function calls..

Cheers
  Magnus



More information about the xorg mailing list