mfi: default to using mrsas(4) for newer cards

[emaste]

Summary

The hw.mfi.mrsas_enable tunable was introduced in commit 96f9425f06ecc183958094ebfa079dfb1aa696d0 in 2014, to allow the (at the time) out-of-tree mrsas driver to claim these devices instead. The mrsas driver has been in the tree for quite some time now and is the preferred driver for these devices.

Event:          Kitchener-Waterloo Hackathon 202406

[markj]

Please be sure to document this in RELNOTES and UPDATING.

[emaste]

How about this text:

The hw.mfi.mrsas_enable tunable is now set to 1, so that mrsas(4) will
attach to hardware supported by both it and mfi(4).  Users that wish to
keep using mfi(4) with such hardware should set it to 0 in loader.conf.

[markj]

In D45476#1037286, @emaste wrote:

How about this text:

The hw.mfi.mrsas_enable tunable is now set to 1

"by default"?

, so that mrsas(4) will
attach to hardware supported by both it and mfi(4).  Users that wish to
keep using mfi(4) with such hardware should set it to 0 in loader.conf.

Looks ok to me.

[emaste]

A wrinkle: PR196820 has a report of a system where using mrsas instead of mfi corrupted the volumes.

[markj]

In D45476#1037353, @emaste wrote:

A wrinkle: PR196820 has a report of a system where using mrsas instead of mfi corrupted the volumes.

One possibility is that that was due to a bug in mrsas that has since been fixed. Hard to say.

A look at share/misc/pci_vendors suggests that the controller referenced in that particular PR is a thunderbolt, which mfi has supported for a long time. I was wondering if we should only default to mrsas on the newer invader and fury cards.

[kp]

This causes my Dell T640 boxes to use the mrsas driver without having to explicitly set the loader.conf variable (as I was doing previously), so as far as I'm concerned this is an improvement.