Details

Reviewers

jhb
kib

Commits

rG02f29c1324cf: msi: Support APIC Extended Destination IDs

Summary

Without an IOMMU, the APIC standard only allow 8 bits of Destination ID
for MSI messages, limiting us to 256 CPUs. While IOMMUs can allow for
more than 256 CPUs to be supported, they are not necessarily desireable
in virtualized environments.

Implement the Extended Destination ID standard authored by David
Woodhouse, which uses 7 "Reserved" bits for the high bits of a 15-bit
Extended Destination ID: http://david.woodhou.se/ExtDestId.pdf

This standard is implemented on the host side in Xen, Hyper-V, KVM,
and Bhyve; future commits will add code to detect and enable it
automatically but in the mean time it can be enabled via the loader
tunable machdep.msi_ext_dest_id.

MFC after: 2 weeks

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 70903
Build 67786: arc lint + arc unit

Event Timeline

cperciva created this revision.Feb 22 2026, 4:10 AM

Herald added a subscriber: imp. · View Herald TranscriptFeb 22 2026, 4:10 AM

cperciva requested review of this revision.Feb 22 2026, 4:10 AM

cperciva added a child revision: D55427: KVM: Detect Extended Destination ID support.

Harbormaster completed remote builds in B70865: Diff 172425.Feb 22 2026, 4:11 AM

cperciva added reviewers: jhb, kib.Feb 22 2026, 4:12 AM

Made ext_dest_id static

Harbormaster completed remote builds in B70903: Diff 172486.Feb 23 2026, 6:01 AM

kib added inline comments.Feb 23 2026, 5:09 PM

sys/x86/x86/msi.c
100	Should it be SYSCTL_BOOL?

cperciva added inline comments.Feb 23 2026, 5:27 PM

sys/x86/x86/msi.c
100	I'm actually about to change this to be tri-state, with -1 meaning "autodetect", so that the loader tunable can set this in either direction. So yes it probably should have been, but no it shouldn't be after that change.

kib accepted this revision.Feb 23 2026, 7:36 PM

This revision is now accepted and ready to land.Feb 23 2026, 7:36 PM

kib added inline comments.Feb 23 2026, 7:36 PM

sys/x86/x86/msi.c
666	This line seems to be over-indented by one space.

Switch machdep.msi_ext_dest_id to tristate

-1 (default) means autodetect

Setting the loader tunable to 0 or 1 forces the state off or on.

This revision now requires review to proceed.Feb 23 2026, 7:52 PM

Harbormaster completed remote builds in B70925: Diff 172537.Feb 23 2026, 7:52 PM

kib added inline comments.Feb 23 2026, 11:08 PM

sys/x86/x86/msi.c
105	Too much indentation for this line too.

cperciva added inline comments.Feb 23 2026, 11:15 PM

sys/x86/x86/msi.c
105	That indentation was there in the original `INTEL_ADDR` which I copied. Although perhaps the second line (which I added) should have the same extra half-indent?
666	That was deliberate based on the nesting of parentheses; I thought it was clearer this way.

kib added inline comments.Feb 23 2026, 11:30 PM

sys/x86/x86/msi.c
105	I believe all lines after the first must have the same indent, per style.
666	Style(9) does not say anything about increasing indent for nested expressions. It is plain +4 spaces, same for all continuation lines.