> @@ -525,6 +537,9 @@ struct ibv_send_wr {
> uint32_t remote_qpn;
> uint32_t remote_qkey;
> } ud;
> + struct {
> + uint32_t rkey;
> + } invalidate;
> } wr;
> };
Thinking about this a bit further... this doesn't work for iWARP "RDMA
read with invalidate" work requests, since this is inside a union, so
the invalidate rkey and the RDMA read remote_addr fields stomp on each
other.
And since we have to figure out how to marshall this into the kernel,
that is a bit of a problem.
Does anyone see a problem with putting the invalidate rkey inside the
rdma part of the wr union as a new field? That is,
@@ -513,6 +525,7 @@ struct ibv_send_wr {
struct {
uint64_t remote_addr;
uint32_t rkey;
+ uint32_t invalidate_rkey;
} rdma;
and similar on the kernel side? (And there is no invalidate member of
this union added)
- R.
_______________________________________________
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
