On Mon, Jul 06, 2015 at 13:08:28 -0400, John Ferlan wrote:
v1 here:
http://www.redhat.com/archives/libvir-list/2015-June/msg00814.html
Changes since v1:
- Add doc patch 1 to indicate that this feature may only be supported by
certain kernels
- Adjust former patch 1 to add call to qemuIsSharedHostdev from
qemuSetUnprivSGIO
- Insert patches 7 & 8 which essentially refactor qemuSetUnprivSGIO a bit.
There should be no functional difference
- Patch 9 is now a much slimmer former patch 6
The end result is that 'generically speaking' if any kernel supports
setting the unprivileged SGIO feature, then these patches provide
the capability to do so.
Although as pointed out in the review of v1 only one specific downstream
kernel supports the feature, that doesn't mean other distros couldn't add
support in the same manner. So rather than just remove all traces from
libvirt completely, it seems it would be reasonable to keep the checks
in place and if a kernel then decides to add support this code exists
to assist.
Well, I'm not going to insist that we revert the existing code since
it's possible that the feature might actually make it into the upstream
linux kernel eventually.
Until it's upstream I don't think though we should add support (even if
we document that it will not work) for stuff that is not upstream since
the design of the upstream interface might then differ, which will make
us carry two implementations.
Since there's already existing code that touches the kernel interface
for unpriv_sgio, actually exposing the support will then require us to
carry that part instead of changing it to the actual upstream impl.
If a "hypothetical" downstream distro would add kernel patches to add the
feature, they might as well as carry the downstream libvirt patches too.
NACK series until upstream kernel support is present. (Some of the
cleanup patches may be worth taking until the kernel issue gets settled
though.)
Peter