To think about what free software licences should do about tivoisation, we have to understand what problems we're trying to prevent, and how it works - so that we can ensure that it doesn't work.
How tivoisation works
Tivoisation is a technique that manufacturers use to produce a computer, to sell to you, whose software they can update but you can't.
There are three elements involved in tivoisation:
The manufacturer puts a chip in the computer which checks any software before it is run and which will only allow authorised software to be run.
The chip can recognise authorised software by, for example, comparing a checksum (like a fingerprint) to a list of authorised checksums, or by checking for an encrypted signature.
The manfucturer withholds the information which you would need in order to make software authorised.
By doing this, the manufacturer can still publish new versions of the software in the future. They just have to embed the encrypted signature in their new version, or send a remote command which would add the checksum of the new version to the list of authorised checksums.
However, if you try to use a modified version of the software, or try to run some third-party software, the computer will refuse to function fully, or will simply not run the software at all.
Controlling your own computer
The name "tivoisation" comes from a computer called the Tivo which comes with the above restrictions (at least from Series2 models onwards). The Tivo contains spyware and blocks the copying of information even when you are legally allowed to copy that information.
The operating system installed on each Tivo is GNU+Linux, so if you buy a Tivo, you have access to the human modifiable source code and permission to modify it. But, if you try to use a modified kernel, the computer will not start. (As described in this article, midway in the 3rd paragraph.) So tivoisation prevents you from being able to use software that doesn't contain spyware or wrongly imposed restrictions.
Sustaining the free software movement
The second reason why free software licences should prohibit tivoisation is that tivoisation burns the environment in which free software flourishes.
Normally, when our software spreads, we gain more developers (individuals plus companies) as some of the users will know how to program, and they will make small or large changes. Also, many of the people who make changes will publish their improvements so that everyone, including the non-programmers, can benefit from the general ability of the community to modify the software. By making computers non-programmable, tivoisation makes free software users non-programmers.
So with tivoisation, the ability of the community to choose the direction the software develops in is inhibited, and the link between the spread of our software and the growth of our developer community is cut. If a million people bought Tivos, there would be an extra million GNU+Linux users in the World, and we would gain zero developers.
This is unfortunate to any degree, but it can also become particularly problematic if it becomes widespread.
If we accept this behaviour from hardware manufacturers, we will get more of it because hardware manufacturers have no reason to turn down the opportunity to have more power over their customers. If tivoised computers become the norm and the era of programmable computers fades into history, free software development and users's control of their computers will be in trouble.
What do we have to think about
While GPLv3 is being drafted, we have to think about how many different ways tivoisation can be done and whether or not there are ways that it can be done, or the same problems can be caused, that the current language could be improved to block.
Of the three components of tivoisation mentioned above, item #3 is the problematic one. If manufacturers implement elements #1 and #2, but told each customer the (possibly unique) encrypted signature, or how to add new checksums to the list of authorised checksums, then there would be no problem. The computer would only run authorised software, but you could decide what is authorised.
Indeed, allowing elements #1 and #2 is important because they can be used for security purposes. I could configure my computer to only run signed software, and then I could sign all the software on my computer. Then, if a virus ever modified the software on my computer or added a new program, it wouldn't run. Or as a network administrator, I might also use this for multiple machines within one organisation. So elements #1 and #2 must not be inhibited by any method of blocking element #3.
What discussion draft 2 of GPLv3 says
So, discussion draft 2 of GPLv3 blocks item #3 by saying that when you are required to distribute a program's source code, you must include:
...any encryption or authorization keys necessary to install and/or execute modified versions from source code in the recommended or principal context of use....
This only applies to people distributing hardware plus software where the hardware is configured as in step #1 above. If you are just distributing software, then the number of keys that are necessary to install and/or execute the software is zero. So this language only applies to a small number of hardware manufacturers, probably less than ten.
That sentence I've quoted is from the definitions of "Corresponding Source" in discussion draft 2 of GPLv3. Richard Stallman has said that in discussion draft 3, this will probably be moved out of that definition and into the section on distributing binaries.
Your comments on this issue are sought at the gplv3.fsf.org comments portal.
For more information about GPLv3, see FSFE's GPLv3 project.