Linus likely believes in the benefits of collaborative development, and that the GPLv2 creates a legal framework to allow for that development.
However he also likely cares more about the software aspects (getting contributions to make better kernel versions and have them be widely used) than the more philosophical aspects of software freedom.
The philosophy on binary drivers for example is one where the Linux kernel does not guarantee a stable binary interface - it isn't that you can't legally distribute a binary driver (with caveats), but he has stated he would like those developers to occasionally "wake up in a cold sweat" over their linux binary drivers breaking on minor releases.
Additionally - because of the velocity of the linux kernel, maintaining a substantial vendor fork is not simple. This creates an atmosphere where maintaining functionality outside the mainline kernel could be a technical/financial burden that justifies any IPR release needed for a L-K contribution.
The number of contributions also means that such forks are likely not a significant fractional contribution compared to the contributions which people are willing to make in the open. The success of the project no longer depends on enforcement of the publication clauses.
Finally, clauses in the GPLv3 and AGPLv3 do influence and even force technology choices. The community likely does not want to lose contributions by pushing people to other kernels, and the commercial interests contributing to Linux would have to account for lost customers.
AGPLv3 is often chosen by commercial entities to force distinct commercial licensing of a published-source code base, rather than an effort to protect a community of contributors. As none of the commercial interests in Linux have exclusive copyright assignment to do a non-AGPLv3 license, there is significantly less commercial benefit for them to support such a move.
[0] https://www.zdnet.com/article/linus-torvaldss-love-hate-rela...
[1] https://www.cio.com/article/3112582/linus-torvalds-says-gpl-...
[2] https://www.youtube.com/watch?v=PaKIZ7gJlRU
[3] https://www.zdnet.com/article/torvalds-battle-with-gplv3/
https://en.wikipedia.org/wiki/Debian_Free_Software_Guideline...
That's mistaken.
If you write Linux kernel code and release it, the GPLv2 obligates you to release your source. That is to say, you do not have the option of releasing only the binary. [0] You have the option of never releasing your work, and you can for instance run a cloud service powered by your modified Linux kernel without having to release your code. [0] (This is how the GPL licences differ from the AGPL licences. [1])
The Linux kernel has benefited enormously from this obligation to share. Torvalds certainly thinks so: [2]
> I really think the license has been one of the defining factors in the success of Linux because it enforced that you have to give back
It's true that you aren't obligated to contact the Linux kernel maintainers and suggest they merge your changes, but I don't see that this would be desirable. As you say, such a requirement would be problematic.
[0] https://www.gnu.org/licenses/old-licenses/gpl-2.0-faq.en.htm...
[1] https://www.gnu.org/licenses/gpl-faq.html#UnreleasedModsAGPL
[2] https://www.zdnet.com/article/linus-torvaldss-love-hate-rela...
You're emphasising the distinction between making GPL'ed source-code available to your customers specifically, and making it available publicly online. In practice there is little difference. Few companies bother hoping no customer will post the source publicly (which the customers are permitted to do under the GPL).
In practical terms the decision is between not releasing, or releasing to the public including the source-code. In a world where the Internet didn't make distribution of software trivially easy, things might be different.
Of course, there are some notable exceptions, the most prominent being nVidia's probably-non-compliant proprietary graphics driver. A fair bit has been written about that.
> Therefore the sharing back that does happen is down to culture not to the license.
No, Torvalds has it right.
Proprietary freeware kernel modules are forbidden outright, as you are not permitted to publicly release the binary without also releasing the source. As I described, 'contained' distribution of source-code isn't a practical option.
The decision of what to merge into the mainline kernel is of course up to the kernel developers. There's no other way it could be.
Companies use and modify his code, building their businesses on them, and then are not distributing their changes back. Arguably the whole cloud revolution has been built on this. Better make a SaaS and freeload on other people's work than make client side software and be forced to share your work (and technical advantage)