On Tue, Jul 07, 2015 at 02:10:50PM -0400, Andy Gospodarek wrote: > On Tue, Jul 07, 2015 at 02:02:18PM -0400, Neil Horman wrote: > > vmxnet3's current napi path is built to count every rx descriptor we > > recieve, > > and use that as a count of the napi budget. That means its possible to > > return > > from a napi poll halfway through recieving a fragmented packet accross > > multiple > > dma descriptors. If that happens, the next napi poll will start with the > > descriptor ring in an improper state (e.g. the first descriptor we look at > > may > > have the end-of-packet bit set), which will cause a BUG halt in the driver. > > > > Fix the issue by only counting whole received packets in the napi poll and > > returning that value, rather than the descriptor count. > > > > Tested by the reporter and myself, successfully > > > > Signed-off-by: Neil Horman <nhor...@tuxdriver.com> > > CC: Shreyas Bhatewara <sbhatew...@vmware.com> > > CC: "David S. Miller" <da...@davemloft.net> > > Looks good. I'm now curious how widespread something like this might be > for drivers that use a similar EOP marker.... > Thats a fair question, though It manifests pretty clearly in any driver that does any sort of strict state checking. I think several drivers just punt if they see an EOP desriptor before an SOP descriptor, toss it and keep going, so you might loose a few extra frames while the driver drains the queue.
> Acked-by: Andy Gospodarek <go...@cumulusnetworks.com> > -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
