My guess about the WSOD

Christian Rüb christian.rueb at gmx.net
Fri Jul 23 11:06:44 CEST 2010


On Friday, 23. July 2010 10:21:49 Helge Hafting wrote:
> On 21. juli 2010 10:25, Petr Vanek wrote:
> >> With the 2+4+2 qi, WSOD seems to happen more often. And curiously
> >> enough, removing power (as in pulling the battery out), doesn't help.
> >> It is then guaranteed to reboot into WSOD. Strange that the phone is
> >> capable of keeping this state over poweroff - do qi and uboot write
> >> state into flash?
> >>
> >> The solution is to boot into NOR flash, and then select "Poweroff"
> >>from the menu. After that, I get a normal reboot without the WSOD.
> >> Graphics indeed seems a bit faster with this qi. :-) I hope the WSOD
> >> problems can be fixed in the kernel and/or the xserver. It happens
> >> without this qi too - perhaps not as often.
> >
> > i can confirm exactly the same findings. as JaMa stated, perhaps this
> > will help to find the cause of WS altogether...
> 
> A WSOD was always a possibility. But in older images, it was rare.
> I saw it occationally with duke nukem - Duke changes resolution and 
> orientation of the screen.
> 
> The 2.6.32 kernel brought nice speedups. But WSODs happens now and then.
> 
> The qi that changes glamo timing from 4+4+4 to 2+4+2 brings even more 
> speed, and much more WSOD trouble too. At least when combined with 
> 2.6.32, I haven't tried it with 2.6.29 or older kernels.
> 
> It seems to me that the WSOD happen only when the display is turned on, 
> or the resolution and/or orientation is changed.
> 
> So I guess the programming of resolution and/or orientation is 
> timing-sensitive. Existing code worked reasonably well with old kernels 
> and 4+4+4  timing. But speedups break it.
> 
> Maybe someone with a compiler could try adding delays between hardware 
> operations in the modesetting code?  (Preferably both the kernel 
> modesetting, and any screen-on operation the bootloader might do. And if 
> there is any modesetting left in Xorg these days.) Extra delays (or 
> longer delays) should compensate for unknown timing sensitivities in the 
> hardware. Maybe evn 1+4+2 timing can be used eventually.
> 
> The video mode change is a rare operation - it won't matter much
> if it gets slower. Not if graphics painting operations gets faster. :-)
> 
> Helge Hafting

I had tested qi with 2+4+2 and got heaps of WSOD on new SHR-U (2.6.32):
about only every 3rd boot worked (screen was white from beginning otherwise)
about every 2nd display lock caused WSOD

With same qi I did not get any WSOD on my other installations on the same Freerunner (Android on SD and SHR-U old (2.6.29) in NAND).

I switched back to "regular" qi and do not see "real" WSODs anymore, meaning that sometimes when dimming the screen turns white but comes back immediately. With 2+4+2 it sometimes came back after suspend (too bad if you have an app running requesting CPU and thus not being able to suspend).

Christian



More information about the community mailing list