From: Dan Carpenter <dan.carpen...@oracle.com> Sent: Friday, October 27, 2017 2:52 PM
>On Fri, Oct 27, 2017 at 05:32:42PM +0800, Yunsheng Lin wrote: >> > iwarp_info = &p_hwfn->p_rdma_info->iwarp; >> > @@ -2696,6 +2696,7 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, >> > if (rc) >> > goto err; >> > >> > + rc = -ENOMEM; >> > iwarp_info->partial_fpdus = kcalloc((u16)p_hwfn->p_rdma_info->num_qps, >> > sizeof(*iwarp_info->partial_fpdus), >> > GFP_KERNEL); >> >> Does the memory allocated here need to be freed when error happens below? >> > >Hm... I think you're right that it leaks. Also I'm confused by the >qed_iwarp_ll2_alloc_buffers() allocation. The comment in there says >that /* buffers will be deallocated by qed_ll2 */ but qed_ll2 is not >a function name or something which is useful to grep. Thanks Dan, partial_fpdus is released during qed_iwarp_resc_free, called from qed_rdma_resc_free called on qed_rdma_stop and on failure during qed_rdma_start. Regarding ll2 buffers. For each successfully allocated buffer we call qed_iwarp_ll2_post_rx->qed_ll2_post_rx_buffer. These buffers will get released and freed when we call qed_ll2_terminate_connection called from qed_iwarp_ll2_stop ( which is called during qed_iwarp_ll2_start on error ). thanks, Michal
