We have run into an issue with perftest and the user verbs libraries.
An {qib,hfi1} create_cq fails because of a driver limit in created a recv_cq for perftest. We found this with new UD and UC tests.
We can increase the limit, but.
perftest runs on after having failed to create a recv_cq:
and passes a NULL attr->recv_cq to the user libraries:
if (create_cqs(ctx, user_param)) {
fprintf(stderr, "Failed to create CQs\n");
/* no return FAILUE */
}
recv_cq is then NULL in a create_qp exposing issues with:
int ibv_cmd_create_qp(struct ibv_pd *pd,
struct ibv_qp *qp, struct ibv_qp_init_attr *attr,
struct ibv_create_qp *cmd, size_t cmd_size,
struct ibv_create_qp_resp *resp, size_t resp_size)
{
IBV_INIT_CMD_RESP(cmd, cmd_size, CREATE_QP, resp, resp_size);
cmd->user_handle = (uintptr_t) qp;
cmd->pd_handle = pd->handle;
cmd->send_cq_handle = attr->send_cq->handle; <- potential NULL derefs
cmd->recv_cq_handle = attr->recv_cq->handle;
If either the send_cq or recv_cq are NULL, the job will seqfault.
This could have been checked before the deref here, but other spots in the code lack similar parameter validation.
Comments?
Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
