[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#929557: linux: restore __kernel_fpu needed for zfs for AES-NI/AVX support [mainline not in debian yet]



Control: tag -1 wontfix

On Sun, 2019-05-26 at 14:00 +0200, Bastian Blank wrote:
> Hi Chris
> 
> On Sat, May 25, 2019 at 10:03:07PM -0400, Chris Zubrzycki wrote:
> > GKH has been purging __kernel_fpu_{begin,end}() from all kernels
> > including LTS (4.19/5) and it's needed for AES-NI/AVX support in the zfs
> > package.
> 
> The commit also tells you why this was done.  Please bring this up to
> upstream and Greg, not to us.
> 
> >            Arch Linux has already reverted the offending commit in their
> > distro. Patch available at
> > https://raw.githubusercontent.com/Mic92/nixpkgs/7b77c27caa8617c82df5c5af6b9ce6ae010d7f9a/pkgs/os-specific/linux/kernel/export_kernel_fpu_functions.patch
> 
> How is this change not a copyright violation?

It's reverting an upstream commit that doesn't mention anything about
copyright, but I can see that there was an odd inconsistency of GPL-
only status with the preempt-disabling versions.

If the upstream developers consider a feature to be so low-level that
modules shouldn't need it at all, it implies that "the function is
considered an internal implementation issue", which is the main reason
documented for using EXPORT_SYMBOL_GPL.

Debian tries to avoid breaking changes during a stable release or pre-
release freeze, and I would certainly considering restoring exports in
order to do that.  We even have long-standing patches that add exports
to support modular builds of aufs and some Android drivers.  However,
we defer to the upstream developers by only using EXPORT_SYMBOL_GPL
when we do that - which would not help ZoL.

So I agree with Bastian that you will need to persuade upstream to
change this.

Ben.

-- 
Ben Hutchings
73.46% of all statistics are made up.

Attachment: signature.asc
Description: This is a digitally signed message part


Reply to: