irdma(4): upgrade to 1.2.17-k

[bartosz.sobczak_intel.com]

Summary

Update Intel irdma driver to version 1.2.17-k

Notable changes:
pf-reset handling improvements, including fixes in communication with if_ice(4)
avoid racing when handling various events
adding sw stats sysctls
hand over pe_criterr handling from ice(4) to irdma(4)
debug prints adjustments
fix crash after changes in irdma_add_mqh_ifa_cb

Test Plan

the code has been validated using testing tools in kernel and user space available for freebsd, including krping, rping, pingpong, ucmatose perftest and others.

[erj]

I wish this was uploaded with context, but I'm not an expert on RDMA so I'm not sure it would let me review this code better.

That said, I see some checks for FreeBSD 14.0; is this patch something that is intended to be directly MFC-able to stable/13?

contrib/ofed/libirdma/irdma_uk.c

279

Why is this change here? The existing ! should be enough to make sure the value is 0 or 1.

[pkubaj]

Please don't add hselasky to reviewers. He has died recently: https://forums.freebsd.org/threads/in-memoriam-hans-petter-william-sirevag-selasky.89697/

[jhibbits]

I'm not familiar enough to comment at all. My only role was mechanically converting to IfAPI.

[bartosz.sobczak_intel.com]

In D41425#944908, @erj wrote:

I wish this was uploaded with context, but I'm not an expert on RDMA so I'm not sure it would let me review this code better.

That said, I see some checks for FreeBSD 14.0; is this patch something that is intended to be directly MFC-able to stable/13?

It definitely should help moving it to stable/13
I'm still not quite clear how to deal with stable/13 and 14-cur differences (as they are huge due to massive changes in ofed in 14-cur). I hope to keep clearing those out as soon as i can.

contrib/ofed/libirdma/irdma_uk.c

279

this is removing an issue reported by static code analysis. The claim was that this line results in a 'dubious x & !y' warning.

[bartosz.sobczak_intel.com]

In D41425#945229, @jhibbits wrote:

I'm not familiar enough to comment at all. My only role was mechanically converting to IfAPI.

Nevertheless i'm touching things we've been working on last time, so please take a look if you are ok with those!

[erj]

contrib/ofed/libirdma/irdma_uk.c

279

Looking at the definition for FIELD_PREP, I can see why, but I don't know, I'd try casting !(qp->swe_polarity) to a u64 first instead of introducing the ternary operator.

[erj]

I'll add an MFC period of 1 month as a reminder to check in about merging this update back to stable/13.