Re: [BK FBDEV] A few more updates.

On Wed, 2003-03-26 at 13:34, James Simmons wrote:

quoted

3.  BTW, there are too many kmalloc's/kfree's in accel_cursor() and
softcursor().  Personally, I would rather have 2 64-byte buffers for the
mask and the data in the info->cursor structure than allocating/freeing
memory each time the cursor flashes.  However, if you prefer doing it
this way, the patch also includes changes so kmallocs are only done when
necessary.  Still, accel_cursor() has unnecessary work being done, such
as always creating the mask bitmap, when a simple flag to monitor cursor
shape changes could prevent all this.

I agree. The problem is the upper layer of the console system is to brain 
dead. Its either erase the cursor or redraw it again. There is no way to 
just say cursor just moved. There is a CM_MOVE but the upper layer doesn't 
even use it :-( If you look at vgacon and friends you will see they 
recreate the cursor every time the cursor blinks. Yes even vgacon.c does 
this. It is stupid and brain dead but that is the way the upper layers of 
the console work. The correct solution would be to use actually use 
CM_MOVE in the upper layers.

Even so, (and I don't really fault the console cursor as it only needs
to show, hide and move the cursor), accel_cursor() can easily monitor
shape changes.  We can use a bitfield somewhere in fb_cursor(perhaps the
high 8 bits of info->fb_cursor.set?) to "remember" the current cursor
shape.

Tony




-------------------------------------------------------
This SF.net email is sponsored by:
The Definitive IT and Networking Event. Be There!
NetWorld+Interop Las Vegas 2003 -- Register today!
http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en