Project

General

Profile

Bug #5887

hardware_offloading_applyflags sets/unsets most values when already set correctly

Added by Chris Buechler over 5 years ago. Updated almost 5 years ago.

Status:
Confirmed
Priority:
Normal
Assignee:
-
Category:
Interfaces
Target version:
-
Start date:
02/12/2016
Due date:
% Done:

0%

Estimated time:
Plus Target Version:
Release Notes:
Default
Affected Version:
All
Affected Architecture:

Description

hardware_offloading_applyflags sets txcsum, rxcsum, LRO, TSO, etc. whether or not they're already set correctly. That'd be OK, but there are driver issues from time to time that trigger a NIC link cycle upon applying some of those. Then they get re-applied upon link up, making it go link down again. Rinse and repeat indefinitely.

Associated revisions

Revision 7f283fab (diff)
Added by Chris Buechler over 5 years ago

Only set IFCAPS where they're not already set correctly. Ticket #5887

Revision bc4d752b (diff)
Added by Jim Pingle over 5 years ago

This seems to have negatively impacted LAGG/VLAN configs. Revert "Only set IFCAPS where they're not already set correctly. Ticket #5887"

This reverts commit 7f283fab1b18bb575537bb8498b16a6b1b6d0251.

History

#1 Updated by Chris Buechler over 5 years ago

  • Status changed from New to Confirmed

thought this would fix the ix(4) link cycling issue on XG-1540 particularly when a DHCP client (rxcsum/txcsum causes link cycle), though that doesn't seem to do it. haven't had a chance to dig further into it yet.

#2 Updated by Jorge M. Oliveira over 5 years ago

I've seen the changes that you've done, dug a little and found non-initialized variables in another file which uses the same functions:
https://github.com/pfsense/pfsense/blob/master/src/etc/inc/interfaces.inc#L766

    pfSense_interface_capabilities($laggif, -$flags_off);
    pfSense_interface_capabilities($laggif, $flags_on);

Possibly unrelated to this, still worth fixing if that proves to be a bug.

#3 Updated by Jim Pingle over 5 years ago

I had to revert the commit referenced here. In a LAGG+VLAN setup the parent NICs had unexpected flags (POLLING and LRO in my case) which broke connectivity. Reverting the commit brought back connectivity.

I have the status output from each state and a test setup if further testing is needed.

#4 Updated by Chris Buechler over 5 years ago

  • Target version deleted (2.3)

That caused a variety of fallout in a number of cases. encaps either doesn't report correctly, or things don't work without applying the caps again anyway, in various scenarios.

This doesn't help the link cycling issue that I was hoping to resolve by doing so. Leaving this as-is for 2.3.

#5 Updated by Chris Buechler almost 5 years ago

  • Assignee deleted (Chris Buechler)

Also available in: Atom PDF