Hello Chris,
Thanks for your time and comments.
I would like to give a brief history of the patch.
We tried to apply this optimization by default, and check all the EDID
read based on the live status. But not all developers agreed to have
this by default, with following reasons:
1. live_status was not very reliable for all platforms, so live_status
based solution shouldn't be added.
2. they dint want EDID caching to be by default, as few old platforms
were not even HPD capable.
So we came up with this intermediate solution to have:
1. Timeout based EDID caching, where cached EDID will be cleared after
one minute.
2. An alternative code flow to support HDMI compliance (will be based
on the live_status check, and will be only enabled for commercial
platforms like android) which need HDMI compliance support. So the
kernel command line parameter is to support the need to add this
alternative EDID read method.
Please le me know your opinion about this, considering the background.
Regards
Shashank
On 8/12/2014 6:17 PM, Chris Wilson wrote:
On Tue, Aug 12, 2014 at 06:08:21PM +0530, shashank.sharma@xxxxxxxxx wrote:
From: Shashank Sharma <shashank.sharma@xxxxxxxxx>
During the HDMI complaince tests, most of the HDMI analyzers
issue a soft HPD, to automate the tests. This process keeps
the HDMI cable connected, and DDC chhanel alive.
HDMI detect() logic relies on EDID readability, to decide if
its a HDMI connect or disconnect event. But in this case, the
EDID is always readable, as HDMI cable is physically connected
and DDC channel is UP, so detect() always reports a HDMI connect
even if its intended to be a HDMI disconnect call.
So if HDMI compliance is enabled, we should rely on the live status
register, than EDID availability. This patch adds:
1. One kernel command line parameter for i915 module, which indicates
if we want to support HDMI compliance, for this platform.
I would rather have this as an output property. In fact, I would like
the hotplug detection method exposed (and even selectable, but other
than this compliance testing, I can't think of a scenario where the
kernel shouldn't be able to figure things out for itself).
2. A new function to read EDID, which gets called only in case of
HDMI compliance support is required. This function reads EDID only
if live status register is up. The normal code flow doesn't get effected
if kernel command line parameter is not enabled.
3. After various experiments on VLV2 board, with various HDMI monitors
we have seen that, with few monitors, the live status register gets
set after a slight delay, and then stays reliably. To support such
monitors, there is a busy-loop added, with a max delay upto 50ms,
with a status check after every 10ms. Please see the comment in
intel_hdmi_get_edid.
Wouldn't a tidier solution be to delay the hpd by 50-100ms after the
hotplug interrupt? That may overcome the issue with the live status for
all connectors...
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
