Page MenuHomeFreeBSD

mxge: Setup mbuf flowid before calling tcp_lro_rx().
ClosedPublic

Authored by sepherosa_gmail.com on May 11 2016, 8:22 AM.
Tags
None
Referenced Files
Unknown Object (File)
Wed, Aug 12, 1:10 PM
Unknown Object (File)
Wed, Aug 12, 1:06 PM
Unknown Object (File)
Wed, Aug 12, 12:49 PM
Unknown Object (File)
Sun, Aug 9, 7:21 AM
Unknown Object (File)
Sun, Aug 9, 3:41 AM
Unknown Object (File)
Sun, Aug 9, 2:51 AM
Unknown Object (File)
Sat, Aug 8, 8:43 PM
Unknown Object (File)
Jun 17 2026, 11:40 PM
Subscribers
None

Diff Detail

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

Event Timeline

sepherosa_gmail.com retitled this revision from to mxge: Setup mbuf flowid before calling tcp_lro_rx()..
sepherosa_gmail.com updated this object.
sepherosa_gmail.com edited the test plan for this revision. (Show Details)
gallatin edited edge metadata.

Note that the hardware is only semi-RSS capable, which is why it is using an opaque flowid derived from the queue id.

By default, a "cheap" hash is done in the NIC, rather than a toeplitz hash, and the hash result is not passed to the driver. The NIC can perform a toeplitz hash, but that cuts packet rate by about 40%.

This revision is now accepted and ready to land.May 11 2016, 2:26 PM

Note that the hardware is only semi-RSS capable, which is why it is using an opaque flowid derived from the queue id.

By default, a "cheap" hash is done in the NIC, rather than a toeplitz hash, and the hash result is not passed to the driver. The NIC can perform a toeplitz hash, but that cuts packet rate by about 40%.

Yeah, I have tested the RSS from NIC as you had told me several years ago, even the bulk data rate is reduced by ~12%, IIRC :P

This revision was automatically updated to reflect the committed changes.