Page MenuHomeFreeBSD

Temporarily force IFF_NEEDSEPOCH until drivers have been resolved.
ClosedPublic

Authored by jeff on Feb 5 2020, 2:48 AM.
Tags
None
Referenced Files
Unknown Object (File)
Thu, Oct 16, 3:04 PM
Unknown Object (File)
Thu, Oct 16, 3:04 PM
Unknown Object (File)
Thu, Oct 16, 3:04 PM
Unknown Object (File)
Thu, Oct 16, 3:38 AM
Unknown Object (File)
Tue, Oct 7, 8:10 PM
Unknown Object (File)
Sep 17 2025, 7:03 PM
Unknown Object (File)
Sep 16 2025, 11:47 AM
Unknown Object (File)
Sep 16 2025, 1:44 AM
Subscribers

Details

Summary

There are some pending fixes to network taskqueue, callout, and some thread based packet handling. Until those are adequately reviewed flip the default back to acquiring epoch in ether_input(). This should be a minor performance penalty that ensures that no more broken drivers are encountered for now. Conveniently someone abstracted flag accesses so that I only had to stop drivers from unintentionally clearing the value in one place.

We have discussed this with a subset of core and we believe this will take some urgency out of the situation so that a more healthy technical discussion can take place.

Diff Detail

Repository
rS FreeBSD src repository - subversion
Lint
Lint Not Applicable
Unit
Tests Not Applicable

Event Timeline

jeff set the repository for this revision to rS FreeBSD src repository - subversion.

This is the least sucky way to fix much of the breakage as a stop gap to the final solution

This revision is now accepted and ready to land.Feb 5 2020, 3:49 AM
hselasky added inline comments.
sys/net/if.c
4158 ↗(On Diff #67808)

The "struct ifnet *" cast on ifp can be removed I think?