[dove] possibility of thumb2 instructions invalidly being handled

[Michael Casadevall]

Certain apps such as APT and GCC either show weird behavior or segfault on dove, while working fine with the same binaries on imx51. These "bugs" consist of counters running backwards, unusual segfaults, and general instability. It appears to be related to https://bugs.edge.launchpad.net/ubuntu/+source/linux-mvl-dove/+bug/494831 and can easily be reproduced with running apt-get update and observing the counters, or running gcc -O and watching it segfault or hang.

[Michael Casadevall (mcasadevall)]

Here's what I mean by counters running backwards:
mcasadevall@dawn:~$ sudo apt-get update
[sudo] password for mcasadevall:
Get:1 http://ports.ubuntu.com lucid Release.gpg [-189B]
Ign http://ports.ubuntu.com lucid/main Translation-en_US
Ign http://ports.ubuntu.com lucid/restricted Translation-en_US
Ign http://ports.ubuntu.com lucid/universe Translation-en_US
Ign http://ports.ubuntu.com lucid/multiverse Translation-en_US
Hit http://ports.ubuntu.com lucid-updates Release.gpg
Ign http://ports.ubuntu.com lucid-updates/main Translation-en_US
Ign http://ports.ubuntu.com lucid-updates/restricted Translation-en_US
Get:2 http://archive.ubuntu.com lucid Release.gpg [-189B]
Hit http://archive.ubuntu.com lucid-updates Release.gpg
Ign http://ports.ubuntu.com lucid-updates/universe Translation-en_US
Ign http://ports.ubuntu.com lucid-updates/multiverse Translation-en_US
Hit http://ports.ubuntu.com lucid-security Release.gpg
Ign http://ports.ubuntu.com lucid-security/main Translation-en_US
Get:3 http://archive.ubuntu.com lucid Release [-57,242B]
Ign http://ports.ubuntu.com lucid-security/restricted Translation-en_US
Ign http://ports.ubuntu.com lucid-security/universe Translation-en_US
Ign http://ports.ubuntu.com lucid-security/multiverse Translation-en_US
Get:4 http://ports.ubuntu.com lucid Release [-57,242B]
Hit http://ports.ubuntu.com lucid-updates Release
Hit http://ports.ubuntu.com lucid-security Release
Hit http://archive.ubuntu.com lucid-updates Release
Get:5 http://ports.ubuntu.com lucid/main Packages [-1,371,481B]
Get:6 http://archive.ubuntu.com lucid/main Sources [-658,385B]
Get:7 http://archive.ubuntu.com lucid/restricted Sources [-3,534B]
Get:8 http://ports.ubuntu.com lucid/restricted Packages [-552B]
Get:9 http://ports.ubuntu.com lucid/universe Packages [-5,198,403B]
Get:10 http://archive.ubuntu.com lucid/universe Sources [-3,110,220B]
Get:11 http://archive.ubuntu.com lucid/multiverse Sources [-122,686B]
Hit http://archive.ubuntu.com lucid-updates/main Sources
Hit http://archive.ubuntu.com lucid-updates/restricted Sources
Hit http://archive.ubuntu.com lucid-updates/universe Sources
Hit http://archive.ubuntu.com lucid-updates/multiverse Sources
Get:12 http://ports.ubuntu.com lucid/multiverse Packages [-167,351B]
Hit http://ports.ubuntu.com lucid-updates/main Packages
Hit http://ports.ubuntu.com lucid-updates/restricted Packages
Hit http://ports.ubuntu.com lucid-updates/universe Packages
Hit http://ports.ubuntu.com lucid-updates/multiverse Packages
Hit http://ports.ubuntu.com lucid-security/main Packages
Hit http://ports.ubuntu.com lucid-security/restricted Packages
Hit http://ports.ubuntu.com lucid-security/main Sources
Hit http://po...

[Michael Casadevall (mcasadevall)]

Subscribed Marvell's engineers to the bug so we have a common point to communicate on.

[Andy Whitcroft (apw)]

It is rumoured that the latest 2.6.32-200 kernel update may fix this. Once that is in the archive perhaps you could test and confirm/deny.

[Andy Whitcroft (apw)]

Ok this new kernel is now in the archive, could you re-test and report back?

[Andy Whitcroft (apw)]

I believe that this was fixed in 2.6.32-200.7. Testing at sprint with the x0 stepping seemed to be working without clocks running backwards. Could you confirm so we can close this off.