[NITPICKER] Porting X.org 1.1.1 to use NitPicker

[Norman Feske]

Hi Nicolas,

> We are working on L4, currently especially on the graphical aspect. As 
> we love NitPicker (NitPicker is cool ;)), we want to use it to run 
> L4Linux on top on it.

Great to hear that you are giving Nitpicker a try!

> So we have started with a Linux installation (Gentoo Linux), after what 
> we have installed Fiasco, L4Env and L4Linux (2.6.20, next 2.6.21, on top 
> on our recently installed Gentoo, reusing its filesystem). To test our 
> installation, we first use L4Con, but the mouse moves were extremely 
> slow; so next we use DOpE, and it was cool, all worked (network and hard 
> disk via L4Linux, not Ore... Oh is there exist a multiplexer, as 
> Ore, for hard disk ?). But of course, the desktop-in-window aproach does 
> not fully satisfied us, we want a fullscreen aproach so we start 
> investigating on NitPicker.
>  
> The problem is that all tutorials related to NitPicker + X.org are based 
> on a quite-old version of X.org, before the X.org team adopted the 
> modular approach. So we decide to port the last X.org (1.1.1) to use 
> NitPicker, following instructions found at 
> http://os.inf.tu-dresden.de/pipermail/l4-hackers/2007/002774.html 
>  
> First we compiled the nitpicker-client ovlwm server, nitovlwm. Then we 
> install into the L4Linux the lxevent program, started after Linux kernel 
> boot, with creation of /dev/lxevent, and the ovltrack program, started 
> via .xinitrc. We had to change a little the Makefile of ovltrack by 
> adding the -Bdynamic option in LIBS. We compiled and generated 
> lxevent_drv.o (X.org mouse driver in xorg.conf). The last step was 
> generating the X.org screen driver, ovlscreen_drv.o. The changes toward 
> previous versions occurs at this step, as now X.org is modular and 
> consists in several small packages. But after resolving a lot of include 
> path problems, importing the needed X.org package to compile the driver, 
> we got the driver ovlscreen_drv.o compiled and copy it into the L4Linux 
> filesystem, modifying xorg.conf to use this driver.
> Also we use NitDope, so L4fb driver in L4Linux that creates /dev/fb0.
>  
> Then we boot our L4-based system, NitPicker, NitDOpE and L4Linux, all 
> start, all was right. The problems come when starting X.org. Indeed, 
> X.org started, the window manager also (WindowMaker), in fact all 
> works... excepts refresh  ! Indeed when moving the mouse on a L4Linux 
> window, the window content is not refreshed, resulting in an unusable 
> window. But clicking outside the window forces the refresh.
>  
> We know that with DOpE, the refresh is automatic, periodic and so it was 
> working for us. But with NitPicker, we know that the refresh only occurs 
> if a change is made in the client. What we don't know is how this 
> refresh is done, who is responsible for it (ovlscreen, ovltrack, 
> nitovlwm ?), and which are the impacts of using a modular X.org in this 
> process....

The refresh mechanism is controlled by the ovlscreen_drv driver
that uses the shadow-fb module of X.org. The ovlscreen driver
passes a buffer to the X-Server to which the X-Server
performs all drawing operations. This buffer is shared with
Nitpicker. After each completed redraw onto this buffer, the
X Server calls a callback function ('update_screenarea'
in ovlscreen.c) and passes a list of rectangles describing
the screen regions to update. For each such screen region,
ovlscreen makes an IPC call of

   refresh_screen([in] int x, [in] int y, [in] int w, [in] int h);

to NitOvlWM (defined in overlay_wm/idl/overlay.idl),
which essentially forward this information to Nitpicker
via another IPC.

For finding the reason for your problem, you might do some of
the following steps:

* Check the output of OvlScreen in the X.org log file
   /var/log/Xorg.0.log and see if the messages indicate
   any anomaly. If you are uncertain, feel free to send
   me your log via email and I take a look.

* Check if the shadow-fb mechanism is working by
   instrumenting the 'update_screenarea' with a bit of
   log output and observing the result in the Xorg log.

* Check if the IPC from ovlscreen_drv to nitovlwm got
   properly delivered by instrumenting the function
   'overlay_refresh_screen_component' of the file
   nitpicker/examples/nitovlwm/l4/main.c

> Or details about event forwarding ? Is there other changes to be made 
> that we missed ?

As far as I can see, you did everything correctly.

Regards,
Norman