ZFS On Linux Has Figured Out A Way To Restore SIMD Support On Linux 5.0+

[oiaohm]

Working with another accepted OSI license in a reasonable way isn't abuse.

The kernel developers should grant ZFS an exception like AFS or the Nvidia driver.

Nvidia does not get a special pass. Nvidia to maintain stuff in kernel has to make the open source driver for their hardware use it.

AFS that a horrible story.

https://www.infradead.org/~dhowells/kafs/

The Afs in the Linux kernel IBM paid Redhat to rewrite from base documentation as pure GPLv2.

AFS shows a party biting the bullet and fixing it properly why we have kafs and openafs these days kafs is GPL and openafs is still the incomparable IBM public license.

So what exception are you talking about neither Nvidia or AFS gets a licensing exception.

[oiaohm]

Not to mention that if one reads the licenses, the issues are the derivative works part and the Ubuntu legal review pretty much clears that up and the only other issue is that the CDDL allows using closed-source build systems; when CDDL projects use open-source build systems they're compliant with all the GPL requirements.

No this is not what the Ubuntu legal review showed.

https://ubuntu.com/blog/zfs-licensing-and-linux

Moreover, the CDDL allows relicensing a project once compiled so it could, in theory, be set up in a way that it's CDDL in source-form and it gets converted to the GPL once compiled. It would be really interesting if a major distribution went that route to circumvent the GPL exports.

GPL requires the binary source code to be under GPL license. So this stunt does not work with GPL at all. Yes CDDL allows re-licensing of binary this does not alter the GPL vs CDDL conflict at the source code.

[oiaohm]

Unfortunately that's close to impossible. ZFS (ZoL) has many, many authors and it's not really feasible to relicense it. Sury, it was a shit move by Sun to make CDDL incompatible to GPL on purpose, but people that want to use it can't change that. And unfortunately, there is no alternative to ZFS available either

If that the case the solution is simple.

https://www.infradead.org/~dhowells/kafs/

Do what was done with AFS to get it into the Linux kernel. Rewrite ZFS driver from scratch. Don't come back and say it cannot be done I just gave link a group did just that.

https://lwn.net/Articles/755919/

BPFilter here provides a different solution to the cannot save FPU state in kernel space. Yes you can in fact embed a usermode program inside your kernel mode .ko file. ZoL developers as far as I have seen have not tried this. Usermode program is allowed to use FPU as much as they like. This does not disrupt PREEMPT_RT.

[starshipeleven]

Well it isn't. Whoever Dainese is got their way. Ubuntu's review (and others) is a clear common sense analysis not hype fud or rumors.

The issue here is that in a court that would just be their own interpretation, while the "hype fud or rumours" are also another interpretation, and until there is a legal battle about this issue with a declared winner and a loser, this will remain in a limbo.

[starshipeleven]

The kernel developers should grant ZFS an exception like AFS or the Nvidia driver.

The NVIDIA driver situation is at best "tolerated" afaik, there is no "exception" anywhere.

[skeevy420]

The NVIDIA driver situation is at best "tolerated" afaik, there is no "exception" anywhere.

That. The kernel just dropped code that makes the Nvidia driver not work on s390s because the only one using it was Nvidia. The reality is that out-of-tree is at best tolerated and at worst not even considered.

EDIT: Actually, at worst is actively coded against as-is the case for ZFS.

[starshipeleven]

EDIT: Actually, at worst is actively coded against as-is the case for ZFS.

None is "actively coding against" ZFS.

No developer ever changed or removed stuff "to break ZFS". Breakage is the normal life of an out-of-tree module.

[skeevy420]

None is "actively coding against" ZFS.

No developer ever changed or removed stuff "to break ZFS". Breakage is the normal life of an out-of-tree module.

GPL exports can be seen in that regard; especially after the appeal with examples of exceptions in the past (notably AFS) and them being slammed in the mailing list or them never getting a response when wanting to discuss how both could compromise and what to do so everyone could work together.

It just seems like they've been more hostile since Torvalds was forced into anger management and they had to become "nice" and "friendly" and "professional". Can't call retards retarded so they'll drop this to troll Nvidia or tweak that to slow down ZFS and are all behind closed doors while saying "maintenance" to us.

[oiaohm]

GPL exports can be seen in that regard; especially after the appeal with examples of exceptions in the past (notably AFS) and them being slammed in the mailing list or them never getting a response when wanting to discuss how both could compromise and what to do so everyone could work together.

AFS out of tree got exceptions because was working was doing GPLv2 rewrite for mainline merge. https://www.infradead.org/~dhowells/kafs/ . ZFS is not doing this so it does not qualify under AFS prior example.

Only way to get exceptions:
1) Have code targeted to go mainline that will superseded your code that is the AFS example.
2) Have something in kernel require it like the. This is the Nvidia route.

ZFS on Linux simply does not qualify. Out of tree drivers with no possible of merging get zero consideration.

It just seems like they've been more hostile since Torvalds was forced into anger management and they had to become "nice" and "friendly" and "professional".

No it aligns with PREEMPT_RT finally going though the mainlining process. This means methods of locking stuff in different areas has to change. FPU is one of those.

https://arstechnica.com/information-technology/2006/12/8428/

Hate to binary kernel modules is not new in the Linux kernel world. Heck hate for out of tree drivers is quite high as well. Lot of these rules were in place with the debates after 2006 on banning binary drivers out right. There was another debate after it covering rights of out of tree kernel modules. The ruling was you have bugger all if you are not planning to merge or are unable to merge. So skeevy420 the rules have been in place for a decade.

The horrible part is the way ZFS on Linux has worked around it this time they are still going to run straight into PREEMPT_RT merge work again so all they have done is bought themselves time until it breaks again.

[starshipeleven]

GPL exports can be seen in that regard;

You can see them however you like, but if they rework the code to not need that stuff exported outside of GPL they usually lock it down. This is against any and all out-of-tree modules that don't have a compatible license. It's not picking against ZFS in particular.

them being slammed in the mailing list

It was more like someone posting their own personal opinion. I didn't see much dogpiling or anything.

or them never getting a response when wanting to discuss how both could compromise and what to do so everyone could work together.

There is nothing to discuss with LKML, they are developers, not lawyers.

Relicensing a codebase of that size and history is a massive pain in the ass, and it would still stir up a big shitstorm, while still carrying a significant risk of not solving the license issues (as someone could then start throwing lawsuits over using this codebase that was "allegedly" relicensed wrongly).

It just seems like they've been more hostile since Torvalds was forced into anger management and they had to become "nice" and "friendly" and "professional".

You are reading too much tabloids. The only one in the LKML that had to become "nice" and "friendly" and "professional" was Torvalds himself, and it was not 100% a bad thing, given his temperament.