Skip to content

Releases

kernel 20261107-d2107de6 arm64

Pre-release

Choose a tag to compare

@enavarre-cl enavarre-cl released this 07 Nov 22:56

EmiBSD kernel for arm64, built by just dist-arm64 from commit d2107de6 (main: milestone N5 met, tag n5-vfs, plus N6.0). It replaces 20261106-ac41e133.

What changed from 20261106-ac41e133

  • No abi_openbsd.ko any more. The OpenBSD ABI (crate abi) is linked into the kernel image (decision 47). The boot loader no longer loads a module, and the install media need none. Drop the file from any manifest or image step that expected it. The system-call ABI is unchanged: 331 system calls, 170 of them served by the native kernel.
  • N5 is complete. The buffer cache, vnode, mount, name lookup, files and the VFS system calls are the native kernel's. So are ufs/ffs/mfs, ext2fs, msdosfs, cd9660, udf, tmpfs and fuse.
  • N6.0: the network lock and SMR live in hal. Nothing userland sees changes.
  • The zero DUID reported in 20261106-ac41e133 (newfs_msdos: /dev/r0000000000000000.i during an ISO install) is OpenBSD's own behaviour: a spoofed label shared with cd0 has no DUID. No kernel change was made for it.

One tarball for EmiBSD/OS to assemble the install media with Userland's sets. Fetch it by this tag, check the sum, then unpack:

a6612550f5015fb8fadf1a059562cebd773d5831071b3652474a3b02de8a61f2  kernel-20261107-d2107de6-arm64.tar.zst

Inside kernel-20261107-d2107de6-arm64/, covered by its SHA256 (OpenBSD's format):

File What
bsd.mp the MULTIPROCESSOR kernel with the ABI linked in, release profile (LTO, symbols kept), default features (no qemu)
bsd.RAMDISK the same kernel with an empty 32 MiB ramdisk (65536 sectors); cargo xtask rdsetroot fills it with a miniroot to make bsd.rd
BOOTAA64.EFI the EFI boot loader
BUILDINFO commit, dates, toolchain (rustc 1.99.1), LandingZone pin, profiles

Pre-release: a development build, tested only under QEMU (virt). Before packing, this kernel booted once and passed the user-mode checks of smoke-boot up to init's last one, then hit OpenBSD's "init died" panic as expected (12.0 s). The N5 close ran on n5-vfs: just ci, ci-full and diff-openbsd passed. d2107de6 adds N6.0 on top; it passed one arm64 boot after its merge but has not been through a close's ci. arm64 only until amd64 is homologated (N10).

kernel 20261107-b2a15ba6 arm64

Pre-release

Choose a tag to compare

@enavarre-cl enavarre-cl released this 07 Nov 23:54

EmiBSD kernel for arm64, built by just dist-arm64 from commit b2a15ba6 (main: N5 met, N6.0, decision 48). It replaces 20261107-d2107de6.

What changed from 20261107-d2107de6

  • The kernel also accepts the ELF note name EmiBSD (decision 48), beside OpenBSD, with the same id. Userland's renamed binaries, whose csu ident note and ld.so carry EmiBSD, exec like OpenBSD's. Nothing numeric changes: NT_OPENBSD_*, PT_OPENBSD_*, PF_OPENBSD_MUTABLE, ELFOSABI_OPENBSD and the auxiliary vector are as before. An OpenBSD binary execs exactly as it did.
  • The kernel reads no section name of a user binary. The .openbsd.* sections are mapped to segments by the linker (DevTools' lld), not by the kernel.

One tarball for EmiBSD/OS to assemble the install media with Userland's sets. Fetch it by this tag, check the sum, then unpack:

5479dd94db5d77fb45eed86b252a3868c2d4f52b68d3909ea2ee0f7995047b8a  kernel-20261107-b2a15ba6-arm64.tar.zst

Inside kernel-20261107-b2a15ba6-arm64/, covered by its SHA256 (OpenBSD's format):

File What
bsd.mp the MULTIPROCESSOR kernel with the ABI linked in, release profile (LTO, symbols kept), default features (no qemu)
bsd.RAMDISK the same kernel with an empty 32 MiB ramdisk (65536 sectors); cargo xtask rdsetroot fills it with a miniroot to make bsd.rd
BOOTAA64.EFI the EFI boot loader
BUILDINFO commit, dates, toolchain (rustc 1.99.1), LandingZone pin, profiles

There is no abi_openbsd.ko: the OpenBSD ABI is linked into the kernel (decision 47, since 20261107-d2107de6).

Pre-release: a development build, tested only under QEMU (virt). Before packing, this kernel booted once and passed the user-mode checks of smoke-boot up to init's last one, then hit OpenBSD's "init died" panic as expected (12.7 s). The ELF note change is covered by host tests (exec_elf: 12 passed). No renamed binary has been booted yet, since the ramdisk's are OpenBSD's. The N5 close ran on n5-vfs; N6.0 and decision 48 have not been through a close's ci yet. arm64 only until amd64 is homologated (N10).

kernel 20261106-ea7fc032 arm64

Pre-release

Choose a tag to compare

@enavarre-cl enavarre-cl released this 06 Nov 01:59

EmiBSD kernel for arm64, built by just dist-arm64 from commit ea7fc032 (main, milestone N5 under way), 2026-11-06T01:57:05Z.

One tarball for EmiBSD/OS to assemble the install media with Userland's sets. Fetch it by this tag, check the sum, then unpack:

3a838cd599c175e62c63bfaf0598523b9cdff125a05b517c694f594f2e756eba  kernel-20261106-ea7fc032-arm64.tar.zst

Inside kernel-20261106-ea7fc032-arm64/, covered by its SHA256 (OpenBSD's format):

File What
bsd.mp the MULTIPROCESSOR kernel, release profile (LTO, symbols kept), features default + qemu
bsd.RAMDISK the same kernel with an empty 32 MiB ramdisk (65536 sectors); cargo xtask rdsetroot fills it with a miniroot to make bsd.rd
abi_openbsd.ko the OpenBSD ABI module the boot loader hands the kernel (required)
BOOTAA64.EFI the EFI boot loader
BUILDINFO commit, dates, toolchain (rustc 1.99.1), LandingZone pin, profiles

Pre-release: a development build. The kernel has been tested only under QEMU (virt), and it boots once there (smoke-boot's user-mode checks) before it is packed. arm64 only until amd64 is homologated (N10).

kernel 20261106-ac41e133 arm64

Pre-release

Choose a tag to compare

@enavarre-cl enavarre-cl released this 06 Nov 10:00

EmiBSD kernel for arm64, built by just dist-arm64 from commit ac41e133 (main, milestone N5 under way; N5.8 isofs done). It replaces 20261106-ea7fc032.

What changed from 20261106-ea7fc032

  • The kernels are built without the qemu feature. In ea7fc03, halt ended in QEMU's semihosting exit, which traps ("Unknown kernel exception 0x00") on a machine started without -semihosting-config. Now halt waits for a key, as OpenBSD's does.
  • BUILDINFO no longer holds a clock time, so one commit packs the same tarball that day. Two packs of this commit gave the same sum.
  • The kernel now includes N5.8: cd9660 and udf are the native kernel's.

One tarball for EmiBSD/OS to assemble the install media with Userland's sets. Fetch it by this tag, check the sum, then unpack:

75a1cbd1434743a675f16b6c9e49ce2f54fed32533a62cfe19143696081fd334  kernel-20261106-ac41e133-arm64.tar.zst

Inside kernel-20261106-ac41e133-arm64/, covered by its SHA256 (OpenBSD's format):

File What
bsd.mp the MULTIPROCESSOR kernel, release profile (LTO, symbols kept), default features (no qemu)
bsd.RAMDISK the same kernel with an empty 32 MiB ramdisk (65536 sectors); cargo xtask rdsetroot fills it with a miniroot to make bsd.rd
abi_openbsd.ko the OpenBSD ABI module the boot loader hands the kernel (required)
BOOTAA64.EFI the EFI boot loader
BUILDINFO commit, dates, toolchain (rustc 1.99.1), LandingZone pin, profiles

Pre-release: a development build, tested only under QEMU (virt). Before packing, this very kernel boots once: the user-mode checks of smoke-boot up to init's last one, then OpenBSD's "init died" panic. arm64 only until amd64 is homologated (N10). Known open issue: the installer formats the MSDOS partition through a zero DUID (/dev/r0000000000000000.i). It is being investigated.