video/modedb: fb_find_nearest_mode: take vmode into account

Previously fb_find_nearest_mode() only searched the modelist for a video mode that best matches the
desired resolution and refresh rate.
With this patch it also takes the vmode into account if there is more then one mode with the same
resolution and refresh rate.
Typical use case is HDMI TVs that support both 1080p60 and 1080i60

Signed-off-by: Floris Bos <redacted>
---
 drivers/video/modedb.c |    4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/video/modedb.c b/drivers/video/modedb.c
index a9a907c..e852371 100644
--- a/drivers/video/modedb.c
+++ b/drivers/video/modedb.c

@@ -913,6 +913,8 @@ const struct fb_videomode *fb_find_best_mode(const struct fb_var_screeninfo *var
  * Finds best matching videomode, smaller or greater in dimension.
  * If more than 1 videomode is found, will return the videomode with
  * the closest refresh rate.
+ * If multiple modes with the same resolution and refresh rate are found
+ * pick the one with the matching vmode (e.g. non-interlaced)
  */
 const struct fb_videomode *fb_find_nearest_mode(const struct fb_videomode *mode,
 					        struct list_head *head)

@@ -939,6 +941,8 @@ const struct fb_videomode *fb_find_nearest_mode(const struct fb_videomode *mode,
 			if (diff_refresh > d) {
 				diff_refresh = d;
 				best = cmode;
+			} else if (diff_refresh = d && cmode->vmode = mode->vmode) {
+				best = cmode;
 			}
 		}
 	}

-- 
1.7.10.4

Hi,

On 02/10/2013 09:23 PM, Floris Bos wrote:

Hmm, my version of this patch was more conservative, only comparing the INTERLACED bit
of vmode. I assume you've tested this, and it works as advertised? If so ack.

Regards,

Hans

On 02/11/2013 10:28 AM, Hans de Goede wrote:

On 02/10/2013 09:23 PM, Floris Bos wrote:

quoted

Previously fb_find_nearest_mode() only searched the modelist for a 
video mode that best matches the
desired resolution and refresh rate.
With this patch it also takes the vmode into account if there is more 
then one mode with the same
resolution and refresh rate.
Typical use case is HDMI TVs that support both 1080p60 and 1080i60

Signed-off-by: Floris Bos <redacted>
---
  drivers/video/modedb.c |    4 ++++
  1 file changed, 4 insertions(+)

diff --git a/drivers/video/modedb.c b/drivers/video/modedb.c
index a9a907c..e852371 100644
--- a/drivers/video/modedb.c
+++ b/drivers/video/modedb.c

@@ -913,6 +913,8 @@ const struct fb_videomode 

*fb_find_best_mode(const struct fb_var_screeninfo *var
   * Finds best matching videomode, smaller or greater in dimension.
   * If more than 1 videomode is found, will return the videomode with
   * the closest refresh rate.
+ * If multiple modes with the same resolution and refresh rate are 
found
+ * pick the one with the matching vmode (e.g. non-interlaced)
   */
  const struct fb_videomode *fb_find_nearest_mode(const struct 
fb_videomode *mode,
                              struct list_head *head)

@@ -939,6 +941,8 @@ const struct fb_videomode 

*fb_find_nearest_mode(const struct fb_videomode *mode,
              if (diff_refresh > d) {
                  diff_refresh = d;
                  best = cmode;
+            } else if (diff_refresh = d && cmode->vmode = 
mode->vmode) {
+                best = cmode;
              }
          }

7 more quoted lines

It works as advertised on my HDMI TV.
fbcon_new_modelist() which uses fb_find_nearest_mode() no longer tries 
to switch from 1080p to 1080i after hot-plugging the same HDMI TV.

But thinking of it, I wonder if it would be better to test on "vmode & 
FB_VMODE_MASK" instead though.
So that it does tests on FB_VMODE_INTERLACED, FB_VMODE_DOUBLE and 
FB_VMODE_ODD_FLD_FIRST
But not on FB_VMODE_YWRAP, FB_VMODE_SMOOTH_XPAN, FB_VMODE_CONUPDATE.


Yours sincerely,

Floris Bos

On 11 February 2013 14:52, Floris Bos [off-list ref] wrote:

On 02/11/2013 10:28 AM, Hans de Goede wrote:

quoted

On 02/10/2013 09:23 PM, Floris Bos wrote:

quoted

Previously fb_find_nearest_mode() only searched the modelist for a video
mode that best matches the
desired resolution and refresh rate.
With this patch it also takes the vmode into account if there is more
then one mode with the same
resolution and refresh rate.
Typical use case is HDMI TVs that support both 1080p60 and 1080i60

Signed-off-by: Floris Bos <redacted>
---
  drivers/video/modedb.c |    4 ++++
  1 file changed, 4 insertions(+)

diff --git a/drivers/video/modedb.c b/drivers/video/modedb.c
index a9a907c..e852371 100644
--- a/drivers/video/modedb.c
+++ b/drivers/video/modedb.c

@@ -913,6 +913,8 @@ const struct fb_videomode *fb_find_best_mode(const

struct fb_var_screeninfo *var
   * Finds best matching videomode, smaller or greater in dimension.
   * If more than 1 videomode is found, will return the videomode with
   * the closest refresh rate.
+ * If multiple modes with the same resolution and refresh rate are found
+ * pick the one with the matching vmode (e.g. non-interlaced)
   */
  const struct fb_videomode *fb_find_nearest_mode(const struct
fb_videomode *mode,
                              struct list_head *head)

@@ -939,6 +941,8 @@ const struct fb_videomode *fb_find_nearest_mode(const

struct fb_videomode *mode,
              if (diff_refresh > d) {
                  diff_refresh = d;
                  best = cmode;
+            } else if (diff_refresh = d && cmode->vmode = mode->vmode)
{
+                best = cmode;

21 more quoted lines

I don't have hardware with interlaced mode support so can't really test this.

I would expect that it would be better to just switch to flagless
modes when available and accept doublescan and interlaced when not.

eg. when you unplug an interlaced TV and plug in a progressive capable
TV the mode would be upgraded. Also when the old TV had 1080i50 and
the new has 1080i50 and 1080p60 I would pick the latter mode as user
unless in a very special situation.

Thanks

Michal

On 02/11/2013 03:06 PM, Michal Suchanek wrote:

On 11 February 2013 14:52, Floris Bos [off-list ref] wrote:

quoted

On 02/11/2013 10:28 AM, Hans de Goede wrote:

quoted

On 02/10/2013 09:23 PM, Floris Bos wrote:

quoted

Previously fb_find_nearest_mode() only searched the modelist for a video
mode that best matches the
desired resolution and refresh rate.
With this patch it also takes the vmode into account if there is more
then one mode with the same
resolution and refresh rate.
Typical use case is HDMI TVs that support both 1080p60 and 1080i60

Hmm, my version of this patch was more conservative, only comparing the
INTERLACED bit
of vmode. I assume you've tested this, and it works as advertised? If so
ack.

It works as advertised on my HDMI TV.
fbcon_new_modelist() which uses fb_find_nearest_mode() no longer tries to
switch from 1080p to 1080i after hot-plugging the same HDMI TV.

But thinking of it, I wonder if it would be better to test on "vmode &
FB_VMODE_MASK" instead though.
So that it does tests on FB_VMODE_INTERLACED, FB_VMODE_DOUBLE and
FB_VMODE_ODD_FLD_FIRST
But not on FB_VMODE_YWRAP, FB_VMODE_SMOOTH_XPAN, FB_VMODE_CONUPDATE.

I don't have hardware with interlaced mode support so can't really test this.

I would expect that it would be better to just switch to flagless
modes when available and accept doublescan and interlaced when not.

eg. when you unplug an interlaced TV and plug in a progressive capable
TV the mode would be upgraded. Also when the old TV had 1080i50 and
the new has 1080i50 and 1080p60 I would pick the latter mode as user
unless in a very special situation.

Problem is that if you go looking for a BETTER mode, the function 
fb_find_NEAREST_mode would no longer behave like the name says.

It also poses problems when the user specifically asked for i50 because 
p60 gives problems, e.g. when the display provides the wrong EDID 
information and advertises modes it does not actually support.
Be aware that the function is not only called when a different display 
is plugged in, but also when the link to the exact same display is 
broken and restored (e.g. due to DPMS power saving).


Yours sincerely,

Floris Bos