ipod4g instability fixed with 20070301 daily

[Shovas]

Since April, 06, I've been downloading daily builds of rockbox for my Ipod 4g 20GB. Initially I downloaded once every few days, then moved to once every few weeks. Finally, I started downloading the builds at the start of the month. I've run 21 builds beginning April 24, 06.

Now, there's a common problem with ipod4g's with instability. People who have this model and have used rockbox for a while tend to agree there was hardware instability (triggered by some code, but who knows where it was) when modulating the CPU speed. So the common fix was to jack up the CPU speed. I always ran mine up to 10 from the CPU Frequency debug menu option.

With the 2007-03-01 build, however, I tried again to run without updating the CPU Frequency. I've been running for quite some time now (a few hours) and no crashes, "data corrupt at...", errors, freezes, etc. at all. Previously, these errors would have triggered within 10 minutes or so.

It looks to me like the issue is resolved. No more jacking up the CPU _every_time_ you boot your ipod!

If I see any more problems I'll be sure to post them here.

Can anyone confirm these results?

[Llorean]

There was never any reason to increase the boost by more than 1. It's an on/off thing. Setting it to 10 just means "On, 10 times" which means it'd have to be turned off 10 times before it goes off. It doesn't make it any more boosted than just 1.

[Shovas]

There was never any reason to increase the boost by more than 1. It's an on/off thing. Setting it to 10 just means "On, 10 times" which means it'd have to be turned off 10 times before it goes off. It doesn't make it any more boosted than just 1.

Actually, early on when I started boosting, I would boost it by one, as you say. What seemed to happen is that rockbox would modulate it down to 0 (30mhz) when it felt it didn't need 70mhz. And so you'd get the crashing. The errors disappeared when the system was at 70mhz all the time. So, that's why I've been jacking it up to at least 10. Maybe that behaviour changed at some point.

[Llorean]

If it was going down to 0, there were more unboosts than there were boosts, and it should've been reported.

It only ever needed to be increased by 1. Not set to 1. Increased by 1. As long as the rest of the boost-tracking portion of the code keeps count properly.

[Shovas]

If it was going down to 0, there were more unboosts than there were boosts, and it should've been reported.

It only ever needed to be increased by 1. Not set to 1. Increased by 1. As long as the rest of the boost-tracking portion of the code keeps count properly.

Interesting. I do remember rechecking the cpu frequency after I initially set it to 10 to see if rockbox was changing it. I remember becoming confident it wasn't changing (the last time I would've checked was Fall 06). But I'm also confident I remember it changing the cpu freq when I set it to 1 (this would have been a lot earlier, April/May/June 06).

I wonder if more-unboosts-than-boosts was the bug. I at one point took a look at the boosting code (to see if I could make it a constant) and couldn't see any problematic cases.

[Shovas]

Well, I've been running the ipod serious for a long time now and I can't say I've seen a crash like there was before. So I think something got fixed.

The only problem now is that the ipod's UI is so slow when it's dynamically modulating the cpu freq that I've started jacking up the CPU anyway! lol

Maybe now that the problem is fixed, the developers might be given another reason to force the CPU to 75mhz on ipod4gs! Of course that won't happen because that would have a negative impact on battery life. Oh well. Such is life.

[JdGordon]

If it was going down to 0, there were more unboosts than there were boosts, and it should've been reported.

iirc the schedular has some magic in unboosting which could cause this..


but if you no longer need to boost then great and welcome back to the fun of updating every few days

[bzbire]

I concur... after 24 hours using yesterday's build, everything works great with no crashes... You might want to add this to the "Major Changes" as this is ALL-important to us Ipod 4g users- This adds 90 minutes of playtime now that I no longer need the Mikeage No Frequency Scaling build

[Llorean]

The problem isn't solved for everyone yet, it's just reduced overall.

Secondly, this isn't necessarily a Major Change, as it's just a bug fix if/when it does solve the problem. People regularly updating to help with testing should notice this rather quickly.

[Shovas]

The problem isn't solved for everyone yet, it's just reduced overall.

Yep, the latest current I tred was 2007-03-18 14:16:45, and while it can go for a while, after a bit it now just crashes. Trying another current...