The ib_ipath driver depends on /usr/include/infiniband/verbs.h
enum ibv_wr_opcode matching the kernel's ib_verbs.h
enum ib_wr_opcode. The recent change to add IB_WR_LSO breaks this.
Now, you may argue that the kernel should not depend on this equivalence
but you would then need to define IBV_WR_RDMA_WRITE, etc. in some
kernel header file and do a table look up to map from user to
kernel opcode values. Since I don't see any other code which depends
on the value of IB_WR_LSO, I think the following patch is the right
fix.
This should be applied to 2.6.24 and 2.6.25.
Signed-off-by: Ralph Campbell <[EMAIL PROTECTED]>
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 14a51b8..b42fafb 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -616,11 +616,11 @@ enum ib_wr_opcode {
IB_WR_RDMA_WRITE,
IB_WR_RDMA_WRITE_WITH_IMM,
IB_WR_SEND,
- IB_WR_LSO,
IB_WR_SEND_WITH_IMM,
IB_WR_RDMA_READ,
IB_WR_ATOMIC_CMP_AND_SWP,
- IB_WR_ATOMIC_FETCH_AND_ADD
+ IB_WR_ATOMIC_FETCH_AND_ADD,
+ IB_WR_LSO
};
enum ib_send_flags {
_______________________________________________
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
