Bug #5649
closed
bce0: Discard frame w/o leading ethernet header (len 0 pkt len 0)
0%
Description
I just migrated my physical pfSesne install to a virtual machine. I am running pfSense 2.5 in kvm and am passing two physical interfaces through. These are each a BCM5709, in a Dell PowerEdge R710.
I have found that in order to work around the issue, I have to disconnect the ethernet cables and allow pfSense to fully boot to the login prompt before reconnecting them, or else all interfaces spam discard frame errors. This is problematic in the case of a power event, as it means I will need to be physically present for the network to come back up.
Updated by Matt Parnell over 8 years ago
I should mention the messages always have a 0 length.
Updated by Chris Buechler over 8 years ago
- Category changed from Interfaces to Operating System
- Status changed from New to Needs Patch
First probably want to try the latest 2.3 snapshot to see if it's still an issue there. If it is a driver issue, it may not be an issue in the newer base OS. https://snapshots.pfsense.org
If it is still an issue there, you'll need to replicate on stock FreeBSD and report to freebsd-net list and/or FreeBSD's bugzilla. We don't modify or develop the bce driver and only fix driver issues relevant to hardware we sell.
might also be a good idea to try a different type of NIC, and the same passthrough to a different OS, to see if it's something inherent in your KVM setup before pursuing upstream.
Updated by Matt Parnell over 8 years ago
I can confirm that the 2.3 Alpha snapshot works as expected. Thank you!
Updated by Matt Parnell over 8 years ago
Bah, upon a subsequent reboot the problem remains.
Updated by Chris Buechler over 8 years ago
Then it's definitely worth putting stock FreeBSD 10.2 in the same situation, and post to freebsd-net list. It'd be good to try a different NIC in the same scenario first as well, so you have a better idea of whether it's specific to bce (I'm guessing it is, but might not be).
Updated by Matt Parnell about 8 years ago
Chris Buechler wrote:
Then it's definitely worth putting stock FreeBSD 10.2 in the same situation, and post to freebsd-net list. It'd be good to try a different NIC in the same scenario first as well, so you have a better idea of whether it's specific to bce (I'm guessing it is, but might not be).
I have just reported the bug on the FreeBSD bugzilla, here's hoping someone can make heads or tails of it.
Updated by Matt Parnell about 8 years ago
As of tonight's nightly build this issue seems to have gone away (pfSense snapshot).
Updated by Matt Parnell about 8 years ago
Well, it appears that this still happens...
I'm not sure if it's the hardware itself and the fact it's passed through via PCI, or just something with the crappy firmware that bce tends to use?
Updated by Matt Parnell over 7 years ago
I believe this issue can now be closed.
After using pci-stub on the Linux host for the two NIC's in question, which prevents the host from loading firmware for the cards, coupled with the article below, I seem to have this issue no longer.
https://doc.pfsense.org/index.php/Tuning_and_Troubleshooting_Network_Cards#Broadcom_bce.284.29_Cards
That saves me buying a dual Intel NIC.