On Thu, Dec 10, 2020 at 05:16:28PM +0100, Marco d'Itri wrote:
> On Dec 10, Paul Wise <pabs@debian.org> wrote:
> > Both of these situations sound like things that should get solved by
> > rewriting the vendor/O-O-T code and including it in mainline
> > Linux/etc, is there any chance of that happening? Or alternatively,
> The Fibre Channel drivers ARE all upstreamed, it's just that the inbox
> drivers (i.e. distributed with the OS, in vendors speak) are not
> qualified to receive support (and may be buggy, even in RHEL).
> Storage vendors provide a support matrix with supported releases of the
> storage firmware, FC switches firmware, network adapter firmware, OS and
> OS driver. Anything else may not receive support.
>
> > for significant future hardware acquisitions, require mainline support
> > before purchase.
> Obviously, but I am not aware of any such FC/FCoE hardware (not just the
> network adapters, but also the storages).
Acknowledge on that problem.
Do know that it can and must be solved by wallet.
So do talk with your purchase department.
Being out of Kernel means it is out of support or an approved support exception, one of the benefits of paid RHEL is hardware vendors both reach out to us, and we reach out to them to ensure certification and capability. Generally prior to something going to market.
I hear your pain tho, I have a heap of Dell servers with deprecated HBA controllers in RH8 that I need to schlep in elrepo modules for drives to show up. That decision however was made because no-one in mainline is able to solve vendor issues and the vendor has disappeared (related to a PowerPC chip design flaws in a heap of Sandybridge era Raid controllers). We went down the big red warnings route in RH7 and people ignored them and had data lost and then blamed RH. Which is untenable ; a lot of hardware specific support falls into this category; and 100% this is something careful selection solves and is part of your Risk assessment as an Org. It also puts a LTS CentOS at odds of being able to track our source trees, one of the problems this is attempting to ameloirate.
So in all those situations given here, yes, CentOS drivers that exist only in CentOS land really put the entire premise of a Stable Long Lived and robust OS at odds with the reality yes, absolutely.
The ability to compile modules against the RHEL sources is not going away. So you can still run your junk with RHEL (out of support). The streams kernel also likely will solve this issue.