Hi !
Current radeonfb is a bit "anal" about accepting CRT modes, it basically only
accepts modes that have the exact resolution, which tends to break with fbcon
on console switches as it provides "approximate" modes. This patch fixes it
by having the driver chose the closest possible mode instead of looking for
an exact match.
Signed-off-by: Benjamin Herrenschmidt <[email protected]>
--- linux-work.orig/drivers/video/aty/radeon_monitor.c 2005-03-11 16:54:25.000000000 +1100
+++ linux-work/drivers/video/aty/radeon_monitor.c 2005-03-11 16:58:04.000000000 +1100
@@ -903,7 +903,7 @@
*/
/*
- * This is used when looking for modes. We assign a "goodness" value
+ * This is used when looking for modes. We assign a "distance" value
* to a mode in the modedb depending how "close" it is from what we
* are looking for.
* Currently, we don't compare that much, we could do better but
@@ -912,13 +912,11 @@
static int radeon_compare_modes(const struct fb_var_screeninfo *var,
const struct fb_videomode *mode)
{
- int goodness = 0;
+ int distance = 0;
- if (var->yres == mode->yres)
- goodness += 10;
- if (var->xres == mode->xres)
- goodness += 9;
- return goodness;
+ distance = mode->yres - var->yres;
+ distance += (mode->xres - var->xres)/2;
+ return distance;
}
/*
@@ -940,7 +938,7 @@
const struct fb_videomode *db = vesa_modes;
int i, dbsize = 34;
int has_rmx, native_db = 0;
- int goodness = 0;
+ int distance = INT_MAX;
const struct fb_videomode *candidate = NULL;
/* Start with a copy of the requested mode */
@@ -976,19 +974,19 @@
/* Now look for a mode in the database */
while (db) {
for (i = 0; i < dbsize; i++) {
- int g;
+ int d;
if (db[i].yres < src->yres)
continue;
if (db[i].xres < src->xres)
continue;
- g = radeon_compare_modes(src, &db[i]);
+ d = radeon_compare_modes(src, &db[i]);
/* If the new mode is at least as good as the previous one,
* then it's our new candidate
*/
- if (g >= goodness) {
+ if (d < distance) {
candidate = &db[i];
- goodness = g;
+ distance = d;
}
}
db = NULL;
--
Benjamin Herrenschmidt <[email protected]>
Benjamin Herrenschmidt a ?crit :
> Hi !
>
> Current radeonfb is a bit "anal" about accepting CRT modes, it basically only
> accepts modes that have the exact resolution, which tends to break with fbcon
> on console switches as it provides "approximate" modes. This patch fixes it
> by having the driver chose the closest possible mode instead of looking for
> an exact match.
Hi Benjamin,
I tried your patch because on recent -mm kernels I see dirty colored
columns during a few seconds when switching from X to radeon fbcon
(looks like remaining colors of X).
I don't know what visible effect your patch is supposed to have.
I didn't see any difference, but I doesn't seem to break anything.
Regards,
Brice
On Mon, 2005-03-28 at 14:03 +0200, Brice Goglin wrote:
> Benjamin Herrenschmidt a ?crit :
> > Hi !
> >
> > Current radeonfb is a bit "anal" about accepting CRT modes, it basically only
> > accepts modes that have the exact resolution, which tends to break with fbcon
> > on console switches as it provides "approximate" modes. This patch fixes it
> > by having the driver chose the closest possible mode instead of looking for
> > an exact match.
>
> Hi Benjamin,
>
> I tried your patch because on recent -mm kernels I see dirty colored
> columns during a few seconds when switching from X to radeon fbcon
> (looks like remaining colors of X).
> I don't know what visible effect your patch is supposed to have.
> I didn't see any difference, but I doesn't seem to break anything.
The effect is that if your console resolution isn't an exact multiple of
the character width or height, radeonfb would fail to set the mode on
console switches. It doesn't happen with 1024x768 and default font but
it does happen with some weird modes, and some monitors (/me lurks
toward IBM) tend to have quite broken default EDID timings.
Ben.