On 1/26/20 2:22 PM, Michal Kubecek wrote:
On Sun, Jan 26, 2020 at 02:12:38PM -0800, Shannon Nelson wrote:
On 1/26/20 1:24 PM, Leon Romanovsky wrote:
On Sun, Jan 26, 2020 at 01:17:52PM -0800, Shannon Nelson wrote:
On 1/26/20 1:08 PM, Leon Romanovsky wrote:
The long-standing policy in kernel that we don't really care about
out-of-tree code.
That doesn't mean we need to be aggressively against out-of-tree code. One
of the positive points about Linux and loadable modules has always been the
flexibility that allows and encourages innovation, and helps enable more
work and testing before a driver can become a fully-fledged part of the
kernel. This move actively discourages part of that flexibility and I think
it is breaking part of the usefulness of modules.
You are mixing definitions, nothing stops those people to innovate and
develop their code inside kernel and as standalone modules too.
It just stops them to put useless driver version string inside ethtool.
If they feel that their life can't be without something from 90s, they
have venerable MODULE_VERSION() macro to print anything they want.
Part of the pain of supporting our users is getting them to give us useful
information about their problem. The more commands I need them to run to
get information about the environment, the less likely I will get anything
useful. We've been training our users over the years to use "ethtool -i" to
get a good chunk of that info, with the knowledge that the driver version is
only a hint, based upon the distro involved. I don't want to lose that
hint. If anything, I'd prefer that we added a field for UTS_RELEASE in the
ethtool output, but I know that's too much to ask.
At the same time, I've been trying to explain both our L1/L2 support
guys and our customers that "driver version" information reported by
"ethtool -i" is almost useless and that if they really want to identify
driver version, they should rather use srcversion as reported by modinfo
or sysfs.
Michal
So as I suggested elsewhere, can we compromise by not bashing the driver
string in the caller stack, but require the in-kernel drivers to use a
particular macro that will put the kernel/git version into the string?
This allows out-of-tree drivers the option of overriding the version
with some other string that can be meaningful in any other given old or
new distro kernel. This should be easy to enforce mechanically with
checkpatch, and easy enough to do a sweeping coccinelle change on the
existing drivers.
sln
