On Sat, 2002-08-24 at 20:33, Paul Mackerras wrote: 

Hi Paul, 

> James Simmons writes:
> 
> > Paul please test the code.
> 
> (The new cfbimgblt.c, that is.)
> 
Thanks for testing the code :)   

> It mostly seems to be fine, except there are some problems with the
> cursor.  I have only tested it with the standard 8x16 font so far
> though.  I had to add a #include <video/fbcon.h> near the top to get
> the definitions of fb_readl and fb_writel.
> 
Can you test it with a bit depth not a multiple of 32/64?  Or just force
the code to always call slow_imageblit?  I'm concerned about
slow_imageblit not correct with big endian machines. 

> It seems to be not erasing the cursor image when it should.  So, if I
> am logged in on the console and I type a few characters and then press
> backspace a few times, it leaves those character positions entirely
> white.  Also, when I press return it leaves the cursor image on that
> line as well as drawing the cursor after the shell prompt on the next
> line.
> 
It looks like a fillrect problem. 

> I just tried with the old cfbimgblt.c and it also does the same
> thing.  So it's not the new cfbimgblt.c that is doing this, it's
> something else in your fbcon changes (or just possibly mine :).  This
> is with atyfb with my patches.
> Paul.
I'm also attaching cfbfillrect.c which hopefully addresses some of the
problems which Geert mentioned before (access/pitch alignment, support
for all bit depths, etc). 

Tony 

PS. Sorry about the attachment, my mailer mangles inline text.