DSS: OMAP3530 vs DM3730

DSS: OMAP3530 vs DM3730

[Gary Thomas]

[-- Attachment #1: Type: text/plain, Size: 1049 bytes --]

I'm running the public 3.0 kernel on my boards.  I have some boards
which can have either OMAP3530 or DM3730 (newer boards have the
newer part, but everything else is the same).

On the OMAP3530, I can enable the graphics overlay on the TV (S-Video)
output via the attached script.  On the DM3730 boards, this crashes
the kernel hard :-(

This same operation seems to be fixed on the 3.2 (public) kernel, but
it's difficult for me to upgrade to that kernel for some other reasons.

Does anyone have an idea why this might happen?  where to look to
isolate the bug(s)?

I tried to back-port the 3.2 DSS code into my 3.0 kernel, but my
simple attempt was pretty much a failure...  If I can, I'd like to
figure out how to just fix the 3.0 code, rather than the full-scale
import of the 3.2 code.

Thanks for any ideas/pointers

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

[-- Attachment #2: tv_video --]
[-- Type: text/plain, Size: 982 bytes --]

#! /bin/sh

#
# This script will put the TV (S-Video) into graphics mode,
# mapped onto /dev/fb2
#

ovl0=/sys/devices/platform/omapdss/overlay0
ovl1=/sys/devices/platform/omapdss/overlay1
ovl2=/sys/devices/platform/omapdss/overlay2

mgr0=/sys/devices/platform/omapdss/manager0
mgr1=/sys/devices/platform/omapdss/manager1

lcd=/sys/devices/platform/omapdss/display0
tv=/sys/devices/platform/omapdss/display1

fb0=/sys/class/graphics/fb0
fb1=/sys/class/graphics/fb1
fb2=/sys/class/graphics/fb2

#MODE=${MODE:-ntsc}
MODE=${MODE:-pal}
echo "0" > $tv/enabled
echo ${MODE} >$tv/timings
w=`cat $tv/timings | cut -d "," -f 2 | cut -d "/" -f 1`
h=`cat $tv/timings | cut -d "," -f 3 | cut -d "/" -f 1`
echo "... Setting ${MODE} mode, Width = ${w}, Height = ${h}"

echo "0" > $ovl2/enabled
echo "2" > $fb2/overlays

echo "$w,$h" > $ovl2/output_size
echo "tv" > $ovl2/manager

mem=$((w*h*4))
echo $mem > $fb2/size

echo "1" > $ovl2/enabled
echo "1" > $tv/enabled

fbset -fb /dev/fb2 -nonstd 1

RE: OMAP3530 vs DM3730

[Hiremath, Vaibhav]

On Sat, Jan 21, 2012 at 21:46:38, Gary Thomas wrote:
> I'm running the public 3.0 kernel on my boards.  I have some boards
> which can have either OMAP3530 or DM3730 (newer boards have the
> newer part, but everything else is the same).
> 
> On the OMAP3530, I can enable the graphics overlay on the TV (S-Video)
> output via the attached script.  On the DM3730 boards, this crashes
> the kernel hard :-(
> 
Can you share the kernel crash log here?

Thanks,
Vaibhav

> This same operation seems to be fixed on the 3.2 (public) kernel, but
> it's difficult for me to upgrade to that kernel for some other reasons.
> 
> Does anyone have an idea why this might happen?  where to look to
> isolate the bug(s)?
> 
> I tried to back-port the 3.2 DSS code into my 3.0 kernel, but my
> simple attempt was pretty much a failure...  If I can, I'd like to
> figure out how to just fix the 3.0 code, rather than the full-scale
> import of the 3.2 code.
> 
> Thanks for any ideas/pointers
> 
> -- 
> ------------------------------------------------------------
> Gary Thomas                 |  Consulting for the
> MLB Associates              |    Embedded world
> ------------------------------------------------------------
> 

Re: OMAP3530 vs DM3730

[Gary Thomas]

On 2012-01-22 22:06, Hiremath, Vaibhav wrote:
> On Sat, Jan 21, 2012 at 21:46:38, Gary Thomas wrote:
>> I'm running the public 3.0 kernel on my boards.  I have some boards
>> which can have either OMAP3530 or DM3730 (newer boards have the
>> newer part, but everything else is the same).
>>
>> On the OMAP3530, I can enable the graphics overlay on the TV (S-Video)
>> output via the attached script.  On the DM3730 boards, this crashes
>> the kernel hard :-(
>>
> Can you share the kernel crash log here?

Sadly, no, as it just locks up tight.  At this point, I can only
reset manually.  I've tried sifting through the dmesg log (_log_buf)
in U-Boot, but I don't see any tell-tale trace of why it crashes.

Further investigation shows that it's not really a problem between 3530 and
3730, sorry.  I thought I had a 3530 system where it worked, but trying it
again today, it also fails. It seems to always fail when I enable fb2 as an
overlay using the TV as the manager.

I know this worked as far back as 2.6.28 and it works again in 3.2. I've
isolated where the kernel fails to the function dispc_pclk_rate().  This
function is quite different in 3.2 than 3.0, so I back-stitched it like
this:

unsigned long dispc_pclk_rate(enum omap_channel channel)
{
         unsigned long r;

         if (dispc_mgr_is_lcd(channel)) {
                 int pcd;
                 u32 l;

                 l = dispc_read_reg(DISPC_DIVISORo(channel));
                 pcd = FLD_GET(l, 7, 0);
                 r = dispc_lclk_rate(channel);

                 return r / pcd;
         } else {
                 unsigned long venc_get_pixel_clock(void);
                 return venc_get_pixel_clock();
         }
}

The kernel no longer crashes hard, but now I get these error messages
printed when I enable the overlay:
   omapdss DISPC error: GFX_FIFO_UNDERFLOW, disabling GFX
   omapdss DISPC error: GO bit not down for channel 1
   omapdss DISPC error: VID2_FIFO_UNDERFLOW, disabling VID2
   omapdss DISPC error: SYNC_LOST_DIGIT, disabling TV

Is there any hope of getting this operation to work in 3.0?

>> This same operation seems to be fixed on the 3.2 (public) kernel, but
>> it's difficult for me to upgrade to that kernel for some other reasons.
>>
>> Does anyone have an idea why this might happen?  where to look to
>> isolate the bug(s)?
>>
>> I tried to back-port the 3.2 DSS code into my 3.0 kernel, but my
>> simple attempt was pretty much a failure...  If I can, I'd like to
>> figure out how to just fix the 3.0 code, rather than the full-scale
>> import of the 3.2 code.
>>
>> Thanks for any ideas/pointers

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: OMAP3530 vs DM3730

[Tomi Valkeinen]

[-- Attachment #1: Type: text/plain, Size: 3313 bytes --]

On Mon, 2012-01-23 at 06:46 -0700, Gary Thomas wrote:
> On 2012-01-22 22:06, Hiremath, Vaibhav wrote:
> > On Sat, Jan 21, 2012 at 21:46:38, Gary Thomas wrote:
> >> I'm running the public 3.0 kernel on my boards.  I have some boards
> >> which can have either OMAP3530 or DM3730 (newer boards have the
> >> newer part, but everything else is the same).
> >>
> >> On the OMAP3530, I can enable the graphics overlay on the TV (S-Video)
> >> output via the attached script.  On the DM3730 boards, this crashes
> >> the kernel hard :-(
> >>
> > Can you share the kernel crash log here?
> 
> Sadly, no, as it just locks up tight.  At this point, I can only
> reset manually.  I've tried sifting through the dmesg log (_log_buf)
> in U-Boot, but I don't see any tell-tale trace of why it crashes.
> 
> Further investigation shows that it's not really a problem between 3530 and
> 3730, sorry.  I thought I had a 3530 system where it worked, but trying it
> again today, it also fails. It seems to always fail when I enable fb2 as an
> overlay using the TV as the manager.
> 
> I know this worked as far back as 2.6.28 and it works again in 3.2. I've
> isolated where the kernel fails to the function dispc_pclk_rate().  This
> function is quite different in 3.2 than 3.0, so I back-stitched it like
> this:
> 
> unsigned long dispc_pclk_rate(enum omap_channel channel)
> {
>          unsigned long r;
> 
>          if (dispc_mgr_is_lcd(channel)) {
>                  int pcd;
>                  u32 l;
> 
>                  l = dispc_read_reg(DISPC_DIVISORo(channel));
>                  pcd = FLD_GET(l, 7, 0);
>                  r = dispc_lclk_rate(channel);
> 
>                  return r / pcd;
>          } else {
>                  unsigned long venc_get_pixel_clock(void);
>                  return venc_get_pixel_clock();
>          }
> }

I think this commit fixes the crash:
c3dc6a7afb47735b82a4c0061e814454a649dbfc

Your changes above seems to be ok for omap3.

> The kernel no longer crashes hard, but now I get these error messages
> printed when I enable the overlay:
>    omapdss DISPC error: GFX_FIFO_UNDERFLOW, disabling GFX
>    omapdss DISPC error: GO bit not down for channel 1
>    omapdss DISPC error: VID2_FIFO_UNDERFLOW, disabling VID2
>    omapdss DISPC error: SYNC_LOST_DIGIT, disabling TV

So does everything work if you use the LCD? And the problems only appear
when you enable the tv-output?

You could try a bit more step-by-step approach in your script. First
enable GFX overlay on the LCD, and resize it so that it fits to both LCD
and TV. See that it works ok. Then enable the TV-out. Then move the GFX
overlay to the TV. etc. 

Also, if you have your test setup working on newer kernel, and it
doesn't work on older, you could get dumps of the dss registers and
compare those. (debugfs/omapdss/dispc and venc).

Hmm. Actually, your script is missing something: it doesn't set the
input_size, i.e. the framebuffer size for fb2. You should set it before
enabling the overlay (fbset -xres $w -vxres $w -yres $h -vyres $h).

I haven't tested tv-out for a while, but I could give it a try on
BeagleBoard. Although I don't have an s-video cable, but I guess it's
enough if I don't see the underflows =).

 Tomi


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: OMAP3530 vs DM3730

[Gary Thomas]

On 2012-01-25 04:07, Tomi Valkeinen wrote:
> On Mon, 2012-01-23 at 06:46 -0700, Gary Thomas wrote:
>> On 2012-01-22 22:06, Hiremath, Vaibhav wrote:
>>> On Sat, Jan 21, 2012 at 21:46:38, Gary Thomas wrote:
>>>> I'm running the public 3.0 kernel on my boards.  I have some boards
>>>> which can have either OMAP3530 or DM3730 (newer boards have the
>>>> newer part, but everything else is the same).
>>>>
>>>> On the OMAP3530, I can enable the graphics overlay on the TV (S-Video)
>>>> output via the attached script.  On the DM3730 boards, this crashes
>>>> the kernel hard :-(
>>>>
>>> Can you share the kernel crash log here?
>>
>> Sadly, no, as it just locks up tight.  At this point, I can only
>> reset manually.  I've tried sifting through the dmesg log (_log_buf)
>> in U-Boot, but I don't see any tell-tale trace of why it crashes.
>>
>> Further investigation shows that it's not really a problem between 3530 and
>> 3730, sorry.  I thought I had a 3530 system where it worked, but trying it
>> again today, it also fails. It seems to always fail when I enable fb2 as an
>> overlay using the TV as the manager.
>>
>> I know this worked as far back as 2.6.28 and it works again in 3.2. I've
>> isolated where the kernel fails to the function dispc_pclk_rate().  This
>> function is quite different in 3.2 than 3.0, so I back-stitched it like
>> this:
>>
>> unsigned long dispc_pclk_rate(enum omap_channel channel)
>> {
>>           unsigned long r;
>>
>>           if (dispc_mgr_is_lcd(channel)) {
>>                   int pcd;
>>                   u32 l;
>>
>>                   l = dispc_read_reg(DISPC_DIVISORo(channel));
>>                   pcd = FLD_GET(l, 7, 0);
>>                   r = dispc_lclk_rate(channel);
>>
>>                   return r / pcd;
>>           } else {
>>                   unsigned long venc_get_pixel_clock(void);
>>                   return venc_get_pixel_clock();
>>           }
>> }
>
> I think this commit fixes the crash:
> c3dc6a7afb47735b82a4c0061e814454a649dbfc
>
> Your changes above seems to be ok for omap3.
>
>> The kernel no longer crashes hard, but now I get these error messages
>> printed when I enable the overlay:
>>     omapdss DISPC error: GFX_FIFO_UNDERFLOW, disabling GFX
>>     omapdss DISPC error: GO bit not down for channel 1
>>     omapdss DISPC error: VID2_FIFO_UNDERFLOW, disabling VID2
>>     omapdss DISPC error: SYNC_LOST_DIGIT, disabling TV
>
> So does everything work if you use the LCD? And the problems only appear
> when you enable the tv-output?
>
> You could try a bit more step-by-step approach in your script. First
> enable GFX overlay on the LCD, and resize it so that it fits to both LCD
> and TV. See that it works ok. Then enable the TV-out. Then move the GFX
> overlay to the TV. etc.
>
> Also, if you have your test setup working on newer kernel, and it
> doesn't work on older, you could get dumps of the dss registers and
> compare those. (debugfs/omapdss/dispc and venc).
>
> Hmm. Actually, your script is missing something: it doesn't set the
> input_size, i.e. the framebuffer size for fb2. You should set it before
> enabling the overlay (fbset -xres $w -vxres $w -yres $h -vyres $h).
>
> I haven't tested tv-out for a while, but I could give it a try on
> BeagleBoard. Although I don't have an s-video cable, but I guess it's
> enough if I don't see the underflows =).

Thanks, I played around with this a bit and found an error in the
script I was running.  I have some systems which have only TV, so
their kernels only expose one display(*).  My other systems have both
LCD and TV.  I was trying the [TV only] script on a system with both
which led to the underflow errors.  When I use the right script on
the right machine, it's working fine.

Sorry for the noise

(*) The choice to expose only valid displays was done to minimize the
possible confusion for our [less sophisticated] users.

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------