Re: [PATCH] OMAPDSS: HDMI: Cache EDID

On 25 June 2012 13:41, Tomi Valkeinen [off-list ref] wrote:

quoted

I am perfectly OK to resend as a patch series, if you want.

Yes please.

OK, will do.

quoted

quoted

quoted

 bool ti_hdmi_4xxx_detect(struct hdmi_ip_data *ip_data)
 {
-     return gpio_get_value(ip_data->hpd_gpio);
+     if (gpio_get_value(ip_data->hpd_gpio))
+             return true;
+
+     /* Invalidate EDID Cache */
+     ip_data->edid_len = 0;
+
+     return false;

Why is this needed? The HPD interrupt should handle this already. And if
the HPD interrupt doesn't work for some reason, this won't work either,
as the user can plug and unplug his HDMI monitors a thousand times
between two detect calls.

I thought it is almost impossible to unplug->plug cycle the HDMI cable
even twice a second, let alone 1000 times against the 10Hz .detect()
poll  :)    Or you mean some relay controlled HDMI switching
mechanism?

omapdss doesn't call the detect function, so it can't presume it's used
in some certain frequency. Also, last time I tested omapdrm, I think
detect was called once in 5 secs or so.

It's not omapdss.  It's the DRM stack, via the omapdrm, that polls
every 10 secs (not 5).
Sorry I said 10Hz instead of 1/10Hz.

quoted

 Anyways, that is not the point of this patch. This patch only aims to
avoid un-ncessary EDID reads while doing its best to make sure we
invalidate EDID 'as soon as possible'.

I'm not sure I understand your point. If the HPD interrupt works
properly, EDID is invalidated there, and the code in detect is not
needed. And if the HPD interrupt doesn't work (although I don't see why
it wouldn't), the code in detect doesn't work. In either case it's not
correct.

Well, the idea was to tie edid-cache invalidating with de-asserted
HPD, wherever we read HPD. I will drop it.

thnx