Messages in this thread | | | Date | Mon, 22 Sep 2008 05:21:11 -0700 (PDT) | Subject | [GIT]: Networking | From | David Miller <> |
| |
Just one regression fix here.
When we are doing multiqueue TX, we accidently try to fetch the protocol inside of IP fragments when computing the hash.
This completely kills performance since what we use for the ports is essentially random garbage, and as a result this causes lots of packet reordering and other nasties.
For drivers that have TX multiqueue support, this creates a serious performance regression against 2.6.26
I know of at least one piece of hardware which has this same bug in it's silicon for doing RX multiqueue hashing :-)
Noticed and fixed by Alexander Duyck.
Please pull, thanks a lot!
The following changes since commit e7272403d2f9be3dbb7cc185fcc390e781b1af6b: Rafael J. Wysocki (1): e100: Use pci_pme_active to clear PME_Status and disable PME#
are available in the git repository at:
master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6.git master
Alexander Duyck (1): netdev: simple_tx_hash shouldn't hash inside fragments
net/core/dev.c | 6 ++++-- 1 files changed, 4 insertions(+), 2 deletions(-)
| |