With the following device (very common on ASUS hardware): Atheros AR8121/AR8113/AR8114 PCI-E Ethernet Controller With PCI ID: 1969-1026-b0-00-1043-14f5 http://pci-ids.ucw.cz/read/PC/1969/1026 We get data corruption when transferring data via NFS over TCP, while data transfered via NFS over UDP seems to be fine. We tested this by md5summing large files (100MB+) over NFS. Turning off all offload features via ethool does not help either. However the RX checksum offload can't be turned off individually, but _seems_ to be turned off when the TX checksum offload is turned off. As indicated when transferring data via UDP the problem disappears, and the md5sums are correct again. The above would suggestion something is possibly wrong the RX checksum offload? Is there any way to force the RX checksum offload off?
(switched to email. Please respond via emailed reply-to-all, not via the bugzilla web interface). On Fri, 28 Jan 2011 09:48:13 GMT bugzilla-daemon@bugzilla.kernel.org wrote: > https://bugzilla.kernel.org/show_bug.cgi?id=27712 > > Summary: atl1e data corruption (via NFS over TCP) > Product: Drivers > Version: 2.5 > Kernel Version: 2.6.38-rc2 > Platform: All > OS/Version: Linux > Tree: Mainline > Status: NEW > Severity: high > Priority: P1 > Component: Network > AssignedTo: drivers_network@kernel-bugs.osdl.org > ReportedBy: pmjdebruijn@pcode.nl > Regression: No > > > With the following device (very common on ASUS hardware): > > Atheros AR8121/AR8113/AR8114 PCI-E Ethernet Controller > > With PCI ID: > > 1969-1026-b0-00-1043-14f5 > http://pci-ids.ucw.cz/read/PC/1969/1026 > > We get data corruption when transferring data via NFS over TCP, while data > transfered via NFS over UDP seems to be fine. > > We tested this by md5summing large files (100MB+) over NFS. > > Turning off all offload features via ethool does not help either. However the > RX checksum offload can't be turned off individually, but _seems_ to be > turned > off when the TX checksum offload is turned off. > > As indicated when transferring data via UDP the problem disappears, and the > md5sums are correct again. > > The above would suggestion something is possibly wrong the RX checksum > offload? > > Is there any way to force the RX checksum offload off? >
I have the same hardware (onboard, asus motherboard) and I can confirm this. I have also seen this happen on large transfers via SCP.
Closing as obsolete, if this is still seen on modern kernels please update