Hi Laurent,

Am 09.11.2017 um 04:45 schrieb Laurent Pinchart [off-list ref]:

Hi Nikolaus,

Thank you for the patch.

On Wednesday, 8 November 2017 23:09:32 EET H. Nikolaus Schaller wrote:

quoted

commit d178e034d565 ("drm: omapdrm: Move FEAT_DPI_USES_VDDS_DSI feature to
dpi code")

introduced a new match table which turned out to be wrong, at least
for the 600 MHz OpenPandora using the OMAP3530.

The effect was strange: only the Blue channel of the RGB panel was
driven while Red and Green stayed black. So a coloured picture turned
into blue/black.

The GTA04 with DM3730 didn't show the effect.

It turned out that VDDS_DSI was not properly initialized on OMAP3530,
because the .family string is just "OMAP3" for these processors and
not "OMAP3xxx".

Therefore we match the .machine attribute.

Signed-off-by: H. Nikolaus Schaller <redacted>

I've already submitted a similar patch (but without the problem pointed out 
below) in the mail thread where we discussed the issue. It is customary to use 
the first patch posted (unless it is utterly broken of course).

Ah sorry. My workflow isn't well prepared for that and I already had committed
something to my private branch...

Could you thus 
please include it in this series in replacement of this patch ?

Well, you can as well reject my patch (it is just a proposal) and take yours
as a replacement. Especially as you better understand all the potential values
for .family and .machine than me.

Should be less work for both of us.

quoted

---
drivers/gpu/drm/omapdrm/dss/dpi.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/omapdrm/dss/dpi.c

b/drivers/gpu/drm/omapdrm/dss/dpi.c index 4ed5fde11313..aae3626910bb 100644

--- a/drivers/gpu/drm/omapdrm/dss/dpi.c
+++ b/drivers/gpu/drm/omapdrm/dss/dpi.c

@@ -566,8 +566,7 @@ static int dpi_verify_pll(struct dss_pll *pll)

}

static const struct soc_device_attribute dpi_soc_devices[] = {
-	{ .family = "OMAP3[456]*" },
-	{ .family = "[AD]M37*" },
+	{ .machine = "OMAP3[456]*" },

You also need 

	{ .machine = "[AD]M37*" },

otherwise there will be no match for the OMAP3-like AM37xx and DM37xx SoCs.

Ah, ok. I wasn't aware that there are some AM37 and DM37 chips with and
some without OMAP3 prefix.

Another option would be to match on { .family = "OMAP3*" } but there could be 
spurious matches, even though I haven't identified any.

quoted

	{ /* sentinel */ }
};

BR and thanks,
Nikolaus Schaller