Eli Cohen wrote:
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -95,7 +95,14 @@ enum ib_device_cap_flags {
+ * devices which publish this capability must support insertion of UDP
+ * and TCP checksum on outgoing packets and can verify the validity of
+ * checksum for incoming packets. Setting this flag implies the driver
+ * may set NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM.
+ IB_DEVICE_IP_CSUM = (1<<18),
enum ib_atomic_cap {
@@ -431,6 +438,7 @@ struct ib_wc {
+ int csum_ok;
Hi Eli,
With the comment in patch #4 at ipoib_ib_handle_rx_wc at hand, the
IB_DEVICE_IP_CSUM and the csum_ok bit are not well defined, since there
are some cases when the HW can not do tx csum and there are some cases
where the hw driver reported csum_ok but some more validation need to be
done in order to make sure its really ok, I suggest you guys discuss
this internally and come up with some solution.
Or.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
