On Tue, Nov 13, 2018 at 10:14 AM Cong Wang <xiyou.wangc...@gmail.com> wrote: > > On Tue, Nov 13, 2018 at 10:01 AM Willem de Bruijn > <willemdebruijn.ker...@gmail.com> wrote: > > > > On Mon, Nov 12, 2018 at 2:49 PM Cong Wang <xiyou.wangc...@gmail.com> wrote: > > > > > > Currently netdev_rx_csum_fault() only shows a device name, > > > we need more information about the skb for debugging csum > > > failures. > > > > > > Sample output: > > > > > > ens3: hw csum failure > > > dev features: 0x0000000000014b89 > > > skb len=84 data_len=0 pkt_type=0 gso_size=0 gso_type=0 nr_frags=0 > > > ip_summed=0 csum=0 csum_complete_sw=0 csum_valid=0 csum_level=0 > > > > Recent issues were protocol dependent, including whether vlan headers > > were present. Perhaps also print skb vlan fields and even the first N > > bytes of data to inspect protocol headers? Also skb_iif, esp. if this > > differs from dev->ifindex. > > Pawel's case seems to be vlan related, however, as I mentioned, > my case is neither vlan nor RXFCS related. > > Ideally, we should dump the whole packet in order to verify the > correctness of the checksum. :) It is not easy to do so given > how complex an skb is now. This is why I only select a few skb > fields to dump. I am pretty sure this can't cover all cases, you > can always add more for your need in the future.
Sounds good. This patch is definitely useful as is. Also, instead of adding code to print a (partial) packet header to the kernel log, it may make more sense to export the skb to a drop monitor over netlink or a perf buffer.
