Hi; this email is intended mainly to raise awareness of an
issue which some people here might care about but be currently
unaware of, and to see if anybody already has a plan to deal with it.
The VM System Specification for ARM Processors:
http://www.linaro.org/docs/vm-system-specification-arm-processors/
defines the contract between a VM implementaton (like KVM/QEMU or
Xen) and an image to run on it (like a distro's ISO installer image).
This specification mandates that we boot by treating the image as
a UEFI ISO image with a FAT32 filesystem. So the VM (and QEMU in
particular) needs to have a custom UEFI firmware image if it wants to
support booting these images, and that UEFI needs to understand FAT32.
Unfortunately the Tianocore UEFI implementation, though mostly BSD
licensed, has a critical piece which is not. The FAT filesystem driver
license is "BSD plus may only be used as part of EFI"; full license
text on this web page:
http://tianocore.sourceforge.net/wiki/Edk2-fat-driver
This is redistributable, but obviously not Free by either the OSI or
Debian Free Software Guidelines definitions. Accordingly it would have
to go in Debian "non-free" and other distro equivalents. (This is
exactly where the x86 Tianocore-derived UEFI implementation, OVMF,
is currently distributed.)
So the default place we're likely to end up if nobody takes any
action is that to use typically distributed ISO images with QEMU
the end-user will need to install the UEFI blob on their host system
from non-free or other distro equivalent (or download it themselves
from the upstream Tianocore website, perhaps).
The question then is: is this outcome unacceptable or undesirable
to anybody? Is anybody planning some other approach?
Some notes:
(1) The VM spec doesn't require that the VM supports *only* the UEFI
ISO image format, so it's not the case that QEMU is useless
without the UEFI firmware. For instance, you could have a Free UEFI
variant which omitted the FAT driver, and pick one of the other
filesystem types that Tianocore supports for your distro's CD images.
(Obviously those images would then probably not be able to boot on h/w,
and conversely "standard" images wouldn't boot on the Free-UEFI.)
(2) When describing or proposing a solution to this it's important
to be clear about the distinction between copyright licensing issues
and patent issues. The problem with the tianocore FAT driver is its
copyright license. Anybody proposing a reimplementation with a different
license needs to consider whether they run into patent problems.
(I'm not going to state an opinion either way about whether such
a reimplementation would potentially intersect any FAT related
patents or whether the MS license for purposes of implementing EFI
would avoid problems or not.)
(3) I imagine that a lot of this ground has already been covered
when considering x86 and OVMF; the only difference for ARM is that
there's no standard Free-licensed alternative firmware, so the problem
is a little more acute. If anybody has a pointer to previous
discussion or conclusions that might save us some argument.
(4) If the "default outcome" is actually at least not unacceptable
to anybody then we can of course simply do nothing...
thanks
-- PMM
Hi All,
This is just a heads-up that we've run into quite a horrible crash +
data corruption issue with MariaDB when running sysbench with a large
number of threads on a couple of AArch64 platforms.
I believe the problem can be attributed to the code used to release
the custom mutexes used by the storage engine.
I have logged a ticket (and attached an emergency fix) here:
https://mariadb.atlassian.net/browse/MDEV-6615
Cheers,
--
Steve Capper
