On Wed, 10 Nov 2010, Paul Mundt wrote:

On Tue, Nov 02, 2010 at 11:47:47AM +0100, Guennadi Liakhovetski wrote:

quoted

--- a/drivers/video/sh_mobile_hdmi.c
+++ b/drivers/video/sh_mobile_hdmi.c

@@ -750,7 +753,38 @@ static int sh_hdmi_read_edid(struct sh_hdmi *hdmi, unsigned long *hdmi_rate,
 	printk(KERN_CONT "\n");
 #endif
 
-	fb_edid_to_monspecs(edid, &hdmi->monspec);
+	if (!hdmi->edid_blocks) {
+		fb_edid_to_monspecs(edid, &hdmi->monspec);
+		hdmi->edid_blocks = edid[126] + 1;
+
+		dev_dbg(hdmi->dev, "%d main modes, %d extension blocks\n",
+			hdmi->monspec.modedb_len, hdmi->edid_blocks - 1);
+	} else {
+		dev_dbg(hdmi->dev, "Extension %u detected, DTD start %u\n",
+			edid[0], edid[2]);
+		fb_edid_add_monspecs(edid, &hdmi->monspec);
+	}
+

Assuming fb_edid_add_monspecs() is changed to hand error values back,
this would probably benefit from being inverted. The fb_edid_to_monspecs()
case looks like a reasonable fallback if fb_did_add_monspecs() fails.

Sorry, I think, there is a confusion here. I don't think 
fb_edid_to_monspecs() can in any case be a fallback from 
fb_edid_add_monspecs(). The former is only good for the _first_ EDID 
segment, whereas the latter shall be used on the following segments. They 
cannot be applied to the same EDID data.

Thanks
Guennadi
---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/