[PATCH 0/2] OMAPDSS: DISPC: Improvements to DIGIT sync signal selection

[PATCH 0/2] OMAPDSS: DISPC: Improvements to DIGIT sync signal selection

[Ricardo Neri]

Hello,

DSS code wrongly assumes that VENC is always available as source for the external
sync signal for the display controller DIGIT channel. One cannot blindly
rely only on the value of DSS_CONTROL register as certain processors may not
have VENC (e.g., OMAP5).

These patches add logic to correctly set/get the sync signal based on the
available displays in the DIGIT channel.

For the HDMI driver, handle the error in the selection of the source.

BR,

Ricardo


Ricardo Neri (2):
  OMAPDSS: DISPC: Improve logic of selection of external sync signal
  OMAPDSS:HDMI: Improve error handling at power on

 drivers/video/omap2/dss/dss.c  |   19 +++++++++++++++++--
 drivers/video/omap2/dss/dss.h  |    2 +-
 drivers/video/omap2/dss/hdmi.c |   10 ++++++++--
 3 files changed, 26 insertions(+), 5 deletions(-)

-- 
1.7.5.4

[PATCH 1/2] OMAPDSS: DISPC: Improve logic of selection of external sync signal

[Ricardo Neri]

The DIGIT channel of the display controller can take the external sync
signal from VENC or HDMI. The code was wrongly assuming that VENC was always
available, which is not necessarily true (e.g., OMAP5). Now, first verify if
the requested external source sync signal is available before getting/setting
the source.

Signed-off-by: Ricardo Neri <ricardo.neri@ti.com>
---
 drivers/video/omap2/dss/dss.c |   19 +++++++++++++++++--
 drivers/video/omap2/dss/dss.h |    2 +-
 2 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/drivers/video/omap2/dss/dss.c b/drivers/video/omap2/dss/dss.c
index 04b4586..e399b4f 100644
--- a/drivers/video/omap2/dss/dss.c
+++ b/drivers/video/omap2/dss/dss.c
@@ -648,9 +648,21 @@ void dss_set_dac_pwrdn_bgz(bool enable)
 	REG_FLD_MOD(DSS_CONTROL, enable, 5, 5);	/* DAC Power-Down Control */
 }
 
-void dss_select_hdmi_venc_clk_source(enum dss_hdmi_venc_clk_source_select hdmi)
+int dss_select_hdmi_venc_clk_source(enum dss_hdmi_venc_clk_source_select src)
 {
-	REG_FLD_MOD(DSS_CONTROL, hdmi, 15, 15);	/* VENC_HDMI_SWITCH */
+	enum omap_display_type displays;
+
+	displays = dss_feat_get_supported_displays(OMAP_DSS_CHANNEL_DIGIT);
+
+	/* discard invalid cases */
+	if (src == DSS_VENC_TV_CLK && !(displays & OMAP_DISPLAY_TYPE_VENC))
+		return -EINVAL;
+
+	if (src == DSS_HDMI_M_PCLK && !(displays & OMAP_DISPLAY_TYPE_HDMI))
+		return -EINVAL;
+
+	REG_FLD_MOD(DSS_CONTROL, src, 15, 15);	/* VENC_HDMI_SWITCH */
+	return 0;
 }
 
 enum dss_hdmi_venc_clk_source_select dss_get_hdmi_venc_clk_source(void)
@@ -661,6 +673,9 @@ enum dss_hdmi_venc_clk_source_select dss_get_hdmi_venc_clk_source(void)
 	if ((displays & OMAP_DISPLAY_TYPE_HDMI) == 0)
 		return DSS_VENC_TV_CLK;
 
+	if ((displays & OMAP_DISPLAY_TYPE_VENC) == 0)
+		return DSS_HDMI_M_PCLK;
+
 	return REG_GET(DSS_CONTROL, 15, 15);
 }
 
diff --git a/drivers/video/omap2/dss/dss.h b/drivers/video/omap2/dss/dss.h
index f67afe7..7ae763c 100644
--- a/drivers/video/omap2/dss/dss.h
+++ b/drivers/video/omap2/dss/dss.h
@@ -270,7 +270,7 @@ bool dss_ovl_use_replication(struct dss_lcd_mgr_config config,
 int dss_init_platform_driver(void) __init;
 void dss_uninit_platform_driver(void);
 
-void dss_select_hdmi_venc_clk_source(enum dss_hdmi_venc_clk_source_select);
+int dss_select_hdmi_venc_clk_source(enum dss_hdmi_venc_clk_source_select);
 enum dss_hdmi_venc_clk_source_select dss_get_hdmi_venc_clk_source(void);
 const char *dss_get_generic_clk_source_name(enum omap_dss_clk_source clk_src);
 void dss_dump_clocks(struct seq_file *s);
-- 
1.7.5.4

[PATCH 2/2] OMAPDSS:HDMI: Improve error handling at power on

[Ricardo Neri]

Handle the errors that may occur when selecting the source of the
sync signal for the DIGIT channel of the display controller and
when enabling the PHY.

Signed-off-by: Ricardo Neri <ricardo.neri@ti.com>
---
 drivers/video/omap2/dss/hdmi.c |   10 ++++++++--
 1 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/video/omap2/dss/hdmi.c b/drivers/video/omap2/dss/hdmi.c
index 060216f..3caac2a 100644
--- a/drivers/video/omap2/dss/hdmi.c
+++ b/drivers/video/omap2/dss/hdmi.c
@@ -501,13 +501,17 @@ static int hdmi_power_on(struct omap_dss_device *dssdev)
 	r = hdmi.ip_data.ops->phy_enable(&hdmi.ip_data);
 	if (r) {
 		DSSDBG("Failed to start PHY\n");
-		goto err;
+		goto err_phy_enable;
 	}
 
 	hdmi.ip_data.ops->video_configure(&hdmi.ip_data);
 
 	/* Make selection of HDMI in DSS */
-	dss_select_hdmi_venc_clk_source(DSS_HDMI_M_PCLK);
+	r = dss_select_hdmi_venc_clk_source(DSS_HDMI_M_PCLK);
+	if (r) {
+		DSSDBG("Failed to start PHY\n");
+		goto err_src_hdmi_venc;
+	}
 
 	/* Select the dispc clock source as PRCM clock, to ensure that it is not
 	 * DSI PLL source as the clock selected by DSI PLL might not be
@@ -536,7 +540,9 @@ static int hdmi_power_on(struct omap_dss_device *dssdev)
 err_mgr_enable:
 	hdmi.ip_data.ops->video_disable(&hdmi.ip_data);
 err_vid_enable:
+err_src_hdmi_venc:
 	hdmi.ip_data.ops->phy_disable(&hdmi.ip_data);
+err_phy_enable:
 	hdmi.ip_data.ops->pll_disable(&hdmi.ip_data);
 err:
 	hdmi_runtime_put();
-- 
1.7.5.4

Re: [PATCH 0/2] OMAPDSS: DISPC: Improvements to DIGIT sync signal selection

[Tomi Valkeinen]

[-- Attachment #1: Type: text/plain, Size: 1510 bytes --]

On Fri, 2012-07-27 at 19:21 -0500, Ricardo Neri wrote:
> Hello,
> 
> DSS code wrongly assumes that VENC is always available as source for the external
> sync signal for the display controller DIGIT channel. One cannot blindly
> rely only on the value of DSS_CONTROL register as certain processors may not
> have VENC (e.g., OMAP5).
> 
> These patches add logic to correctly set/get the sync signal based on the
> available displays in the DIGIT channel.
> 
> For the HDMI driver, handle the error in the selection of the source.

Is there an actual problem seen that this resolves? If so, It'd be nice
to have a brief description of the problem.

Does the bit 15 in DSS_CONTROL still exist on OMAP5? What does it return
on OMAP5?

I think it's clear that dss_get_hdmi_venc_clk_source() needs to return a
correct value so that DSS operates correctly. But I'm not sure we need
the extra code in dss_select_hdmi_venc_clk_source().

dss_select_hdmi_venc_clk_source() should be called by venc or hdmi, and
there should be any possibility of using it in the wrong way. For extra
sanity checking there could be a BUG_ON() check in the
dss_select_hdmi_venc_clk_source() to point out if there's a bad bug in
venc or hdmi.

I'd like to keep the lowest level dss-internal functions as simple as
possible, and written in the manner that they presume they are called
correctly. Otherwise we'll end up with lots of new code, checking for
errors that can never happen in practice.

 Tomi


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

Re: [PATCH 0/2] OMAPDSS: DISPC: Improvements to DIGIT sync signal selection

[Archit Taneja]

On Monday 30 July 2012 04:41 PM, Tomi Valkeinen wrote:
> On Fri, 2012-07-27 at 19:21 -0500, Ricardo Neri wrote:
>> Hello,
>>
>> DSS code wrongly assumes that VENC is always available as source for the external
>> sync signal for the display controller DIGIT channel. One cannot blindly
>> rely only on the value of DSS_CONTROL register as certain processors may not
>> have VENC (e.g., OMAP5).
>>
>> These patches add logic to correctly set/get the sync signal based on the
>> available displays in the DIGIT channel.
>>
>> For the HDMI driver, handle the error in the selection of the source.
>
> Is there an actual problem seen that this resolves? If so, It'd be nice
> to have a brief description of the problem.

The issue is seen on OMAP5 with a 3.5 kernel. In 3.5 kernel, we had a 
hack where we used to divide the TV manager's height by 2 if the 
connected interface was VENC. On OMAP5, the bit 15 returns VENC (a zero) 
irrespective of what we write.

This won't happen in 3.6 as we have included interlace as a part of the 
omap_video_timings struct, and we don't try to check the bit to see 
whether we are VENC or HDMI.

I guess we should still put a check to not set the bit on OMAP5 as it 
may cause unknown HW behavior, i.e, make the select function like:

void dss_select_hdmi_venc_clk_source()
{
	...

	displays =
		dss_feat_get_supported_displays(OMAP_DSS_CHANNEL_DIGIT);

	if ((displays & OMAP_DISPLAY_TYPE_VENC) == 0)
		return;

	REG_FLD_MOD(DSS_CONTROL, src, 15, 15);	/* VENC_HDMI_SWITCH */
}

Archit

>
> Does the bit 15 in DSS_CONTROL still exist on OMAP5? What does it return
> on OMAP5?
>
> I think it's clear that dss_get_hdmi_venc_clk_source() needs to return a
> correct value so that DSS operates correctly. But I'm not sure we need
> the extra code in dss_select_hdmi_venc_clk_source().
>
> dss_select_hdmi_venc_clk_source() should be called by venc or hdmi, and
> there should be any possibility of using it in the wrong way. For extra
> sanity checking there could be a BUG_ON() check in the
> dss_select_hdmi_venc_clk_source() to point out if there's a bad bug in
> venc or hdmi.
>
> I'd like to keep the lowest level dss-internal functions as simple as
> possible, and written in the manner that they presume they are called
> correctly. Otherwise we'll end up with lots of new code, checking for
> errors that can never happen in practice.
>
>   Tomi
>

Re: [PATCH 0/2] OMAPDSS: DISPC: Improvements to DIGIT sync signal selection

[Ricardo Neri]

Hi Tomi, Archit,

Thanks for your feedback.

On 07/30/2012 06:48 AM, Archit Taneja wrote:
> On Monday 30 July 2012 04:41 PM, Tomi Valkeinen wrote:
>> On Fri, 2012-07-27 at 19:21 -0500, Ricardo Neri wrote:
>>> Hello,
>>>
>>> DSS code wrongly assumes that VENC is always available as source for
>>> the external
>>> sync signal for the display controller DIGIT channel. One cannot blindly
>>> rely only on the value of DSS_CONTROL register as certain processors
>>> may not
>>> have VENC (e.g., OMAP5).
>>>
>>> These patches add logic to correctly set/get the sync signal based on
>>> the
>>> available displays in the DIGIT channel.
>>>
>>> For the HDMI driver, handle the error in the selection of the source.
>>
>> Is there an actual problem seen that this resolves? If so, It'd be nice
>> to have a brief description of the problem.
>
> The issue is seen on OMAP5 with a 3.5 kernel. In 3.5 kernel, we had a
> hack where we used to divide the TV manager's height by 2 if the
> connected interface was VENC. On OMAP5, the bit 15 returns VENC (a zero)
> irrespective of what we write.
>
> This won't happen in 3.6 as we have included interlace as a part of the
> omap_video_timings struct, and we don't try to check the bit to see
> whether we are VENC or HDMI.
>
> I guess we should still put a check to not set the bit on OMAP5 as it
> may cause unknown HW behavior, i.e, make the select function like:
>
> void dss_select_hdmi_venc_clk_source()
> {
>      ...
>
>      displays =
>          dss_feat_get_supported_displays(OMAP_DSS_CHANNEL_DIGIT);
>
>      if ((displays & OMAP_DISPLAY_TYPE_VENC) == 0)
>          return;
>
>      REG_FLD_MOD(DSS_CONTROL, src, 15, 15);    /* VENC_HDMI_SWITCH */
> }

As Tomi mentions, perhaps this could be complemented with a BUG_ON could 
be used if someone tries to use VENC as source when this is not available.
>
> Archit
>
>>
>> Does the bit 15 in DSS_CONTROL still exist on OMAP5? What does it return
>> on OMAP5?
>>
>> I think it's clear that dss_get_hdmi_venc_clk_source() needs to return a
>> correct value so that DSS operates correctly. But I'm not sure we need
>> the extra code in dss_select_hdmi_venc_clk_source().

Yes, the main purpose of these patch was to avoid blindly reading 
DSS_CONTROL[15]. The additions to selection were just for completeness.
>>
>> dss_select_hdmi_venc_clk_source() should be called by venc or hdmi, and
>> there should be any possibility of using it in the wrong way. For extra
>> sanity checking there could be a BUG_ON() check in the
>> dss_select_hdmi_venc_clk_source() to point out if there's a bad bug in
>> venc or hdmi.
>>
>> I'd like to keep the lowest level dss-internal functions as simple as
>> possible, and written in the manner that they presume they are called
>> correctly. Otherwise we'll end up with lots of new code, checking for
>> errors that can never happen in practice.
Well, this was an error that happened in practice :).

I will resubmit based on these comments.

Thanks,

Ricardo
>>
>>   Tomi
>>
>