atomic.S patch by mav

I am in for the assembler code patch. I will prepare an updated version.

I am definitely going to do that. But in the meantime we need a buildable stable/13 for i386. The solution I am proposing does (temporarily) place an additional file into cddl/lib/libzfs what is outside of sys/contrib/openzfs.

I assume Brian just forgot to merge that commit or intentionally didn't merge that commit and did not inspect your commit for changes in that direction. The i386 build should be broken for Linux as well.

In D26311#676085, @avg wrote:

@mm , do you have amdgpio driver? Has it attached?

I get currently the following reply on stable/13 with an APU2 bios 4.13.0.4:
gpioleds0: <ACPI GPIO LEDs> on acpi0
gpioleds0: GPIO controller device \134_SB_.PCI0.GPIO is not attached
gpioleds0: GPIO controller device \134_SB_.PCI0.GPIO is not attached
gpioleds0: GPIO controller device \134_SB_.PCI0.GPIO is not attached
gpioleds0: detached

Thank you. We are testing Linux, FreeBSD, macOS (via GitHub Actions) and OmniOS (via Jenkins) on GitHub so that will tell how it behaves on other platforms.

If you don't know I am a core libarchive developer and release manager so getting the fix upstream is up to me.
I don't like modifying the framework this way, we should fix the individual test instead.

What about we develop a fix for this at upstream?

In D28892#647126, @nyan wrote:

In D28892#646718, @ak wrote:

Also, I just noticed that in half of return statements we don't use parentheses, which is violation of style(7).

Yes, but it's a difference of style between FreeBSD and NetBSD.
So I don't change it.

I have merged lastest OpenZFS to main including the new "compatibility" property, you can close this now. Thanks for getting the change upstream!

I have updated the vendor branch to bf156c966, so technically we can merge from it. The idea is that we keep as low diff to upstream as possible. Have you already done merging from a vendor branch?

Yes, looks like Matthew Ahrens has inlined this code into abd_get_offset_scatter()
Thank you @rlibby for submitting the upstream PR.

Rebase and update.

Rebased to eac22dd48079d5cccecb07f4e60b44bb692167f8

To be true guys, what is much more interesting than this diff is the diff between vendor/openzfs and sys/contrib/openzfs - I guess we want to keep it as small as possible (long-term). And currently there is already some difference.
So if you have no objections I might throw this in but I am not really into having this in 13.0-RELEASE, 13-stable after a reasonable amount of time (at least 2 weeks) would be o.k. for me.

The zfs test suite was not very happy with this in my bhyve, I guess we postpone this or someone else could continue with the merge. I have cherry-picked two important commits for 13.0-RELEASE.

@mjg I don't think much has changed under the hood because:

Github version: https://github.com/mmatuska/freebsd-src/commits/openzfs_gc1c31a835
@mjg please review the modification of zfs_freebsd_readlink() in sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c