WSOD will need redisign (and spec)
nicolas.dufresne at gmail.com
Sun Apr 19 19:57:34 CEST 2009
Le dimanche 19 avril 2009 à 12:26 -0500, Nelson Castillo a écrit :
> This testing sounds very good but we shouldn't do it with the (future)
> stable kernel :-)
I don't understand this point. Reverting this patch will cause a second
regression (as observed by Michael Trimarchi) and will not fix the
current "regression" that only exist when using U-Boot.
I have been doing test for almost a day now, and I know that the
regression was introduce a long time ago in beginning of 2.6.29, not in
my previous patch.
If you look into the driver, you'll see that a reset was removed from
probe function. That is most likely to cause the regression, but
personally I will never change the code until I have tested it.
My previous patch has been tested in operation for a full week (with Qi
only). I'm a little sad that it is considered testing stuff that should
be put aside.
I don't think we will get anywhere stable some day if we don't consider
all patches between start point (2.6.28) to end point (2.6.29) when
trying to bisect a regression. Reverting is a great feature in git, but
we should use it with care because it's not a garante that previous code
I've currently started testing a patch that re-implement Andy reset
switch with a slightly different approach. Everything seems to be fine
for now on Qi and I'll move to U-Boot before the end of the day. The
only thing I'm not sure of is the time delays required when setting
Glamo GPIO4 on/off (ping called LCM_RESET in schematics) that I'm
combining with software reset of jbt registers in jbt_init_regs() (as
documented in bug #1841 by Joerg comment). That information can only be
provided by an OM employee, which is why I asked previously.
Now, consider my request not to revert the patch as a time saver for me.
I'm using my own time, with very little resources here because I want my
device to work properly. I'm not really interested in doing the work
twice unless my patch are not correctly formated or are direct cause of
regression, which is not the case at the moment.
More information about the openmoko-kernel