The QP's need to be flushed and processed via EVD's during
the EP (QP) destroy to avoid an error on poll_cq. IBAL
provider was not moving to ERR state during QP destroy.
Better flush CQ processing was added and pushed down to the provider
level via dapls_ib_qp_free() where it can move QP to ERR, flush CQ,
and then free QP after flushing. Because there is no QP_ERR_FLUSH
state on a QP the spin on poll_cq (until empty) after modify_qp
to ERR could return empty and before all WQE's are flushed. This
could result in a CQE being added to CQ with a invalid QP reference.
So, an additional check was added to flush_evds for the recv_q to
poll_cq until all recv's pending are complete. For transmit_q there
is no quarantee that the posted work is signaled and so the best
that can be done is poll_cq until empty.
Signed-off-by: Arlin Davis <arlin.r.da...@intel.com>
Signed-off-by: Sean Hefty <sean.he...@intel.com>
---
dapl/common/dapl_ep_free.c | 25 ++++++++++++++-----------
dapl/common/dapl_ep_util.c | 6 ++----
dapl/ibal/dapl_ibal_qp.c | 14 +++++++++++---
dapl/openib_common/qp.c | 32 ++++++++++++++++++++------------
4 files changed, 47 insertions(+), 30 deletions(-)
diff --git a/dapl/common/dapl_ep_free.c b/dapl/common/dapl_ep_free.c
index 32d50cc..a8deeb2 100644
--- a/dapl/common/dapl_ep_free.c
+++ b/dapl/common/dapl_ep_free.c
@@ -157,16 +157,6 @@ DAT_RETURN DAT_API dapl_ep_free(IN DAT_EP_HANDLE ep_handle)
pz_ref_count);
param->pz_handle = NULL;
}
- if (param->recv_evd_handle != NULL) {
- dapl_os_atomic_dec(&((DAPL_EVD *) param->recv_evd_handle)->
- evd_ref_count);
- param->recv_evd_handle = NULL;
- }
- if (param->request_evd_handle != NULL) {
- dapl_os_atomic_dec(&((DAPL_EVD *) param->request_evd_handle)->
- evd_ref_count);
- param->request_evd_handle = NULL;
- }
if (param->connect_evd_handle != NULL) {
dapl_os_atomic_dec(&((DAPL_EVD *) param->connect_evd_handle)->
evd_ref_count);
@@ -202,7 +192,20 @@ DAT_RETURN DAT_API dapl_ep_free(IN DAT_EP_HANDLE ep_handle)
}
}
- dapls_ep_flush_cqs(ep_ptr);
+ /*
+ * Release the EVD handles after we destroy the QP, so we can flush all
+ * QP entries.
+ */
+ if (param->recv_evd_handle != NULL) {
+ dapl_os_atomic_dec(&((DAPL_EVD *) param->recv_evd_handle)->
+ evd_ref_count);
+ param->recv_evd_handle = NULL;
+ }
+ if (param->request_evd_handle != NULL) {
+ dapl_os_atomic_dec(&((DAPL_EVD *) param->request_evd_handle)->
+ evd_ref_count);
+ param->request_evd_handle = NULL;
+ }
/* Free the resource */
dapl_ep_dealloc(ep_ptr);
diff --git a/dapl/common/dapl_ep_util.c b/dapl/common/dapl_ep_util.c
index fc911a6..6646528 100644
--- a/dapl/common/dapl_ep_util.c
+++ b/dapl/common/dapl_ep_util.c
@@ -620,10 +620,8 @@ static void dapli_ep_flush_evd(DAPL_EVD *evd_ptr)
void dapls_ep_flush_cqs(DAPL_EP * ep_ptr)
{
- if (ep_ptr->param.request_evd_handle)
- dapli_ep_flush_evd((DAPL_EVD *)
ep_ptr->param.request_evd_handle);
-
- if (ep_ptr->param.recv_evd_handle)
+ dapli_ep_flush_evd((DAPL_EVD *) ep_ptr->param.request_evd_handle);
+ while (dapls_cb_pending(&ep_ptr->recv_buffer))
dapli_ep_flush_evd((DAPL_EVD *) ep_ptr->param.recv_evd_handle);
}
diff --git a/dapl/ibal/dapl_ibal_qp.c b/dapl/ibal/dapl_ibal_qp.c
index e843829..4ea57d7 100644
--- a/dapl/ibal/dapl_ibal_qp.c
+++ b/dapl/ibal/dapl_ibal_qp.c
@@ -317,6 +317,7 @@ dapls_ib_qp_free (
IN DAPL_IA *ia_ptr,
IN DAPL_EP *ep_ptr )
{
+ ib_qp_handle_t qp;
UNREFERENCED_PARAMETER(ia_ptr);
@@ -327,12 +328,19 @@ dapls_ib_qp_free (
dapl_os_lock(&ep_ptr->header.lock);
if (( ep_ptr->qp_handle != IB_INVALID_HANDLE ))
{
- ib_destroy_qp ( ep_ptr->qp_handle, ib_sync_destroy );
+ qp = ep_ptr->qp_handle;
+ ep_ptr->qp_handle = IB_INVALID_HANDLE;
+ dapl_os_unlock(&ep_ptr->header.lock);
+
+ dapls_modify_qp_state_to_error(qp);
+ dapls_ep_flush_cqs(ep_ptr);
+
+ ib_destroy_qp ( qp, ib_sync_destroy );
dapl_dbg_log (DAPL_DBG_TYPE_EP, "--> DsQF: freed QP %p\n",
ep_ptr->qp_handle );
- ep_ptr->qp_handle = IB_INVALID_HANDLE;
+ } else {
+ dapl_os_unlock(&ep_ptr->header.lock);
}
- dapl_os_unlock(&ep_ptr->header.lock);
return DAT_SUCCESS;
}
diff --git a/dapl/openib_common/qp.c b/dapl/openib_common/qp.c
index 5c5c10f..8db6f8e 100644
--- a/dapl/openib_common/qp.c
+++ b/dapl/openib_common/qp.c
@@ -215,30 +215,38 @@ dapls_ib_qp_alloc(IN DAPL_IA * ia_ptr,
*/
DAT_RETURN dapls_ib_qp_free(IN DAPL_IA * ia_ptr, IN DAPL_EP * ep_ptr)
{
+ struct ibv_qp *qp;
+ struct ibv_qp_attr qp_attr;
+
#ifdef _OPENIB_CMA_
dp_ib_cm_handle_t cm_ptr = dapl_get_cm_from_ep(ep_ptr);
+ if (!cm_ptr)
+ return DAT_SUCCESS;
+#endif
dapl_os_lock(&ep_ptr->header.lock);
- if (cm_ptr && cm_ptr->cm_id->qp) {
+ if (ep_ptr->qp_handle != NULL) {
+ qp = ep_ptr->qp_handle;
+ dapl_os_unlock(&ep_ptr->header.lock);
+
+ qp_attr.qp_state = IBV_QPS_ERR;
+ ibv_modify_qp(qp, &qp_attr, IBV_QP_STATE);
+ dapls_ep_flush_cqs(ep_ptr);
+
+ ep_ptr->qp_handle = NULL;
+#ifdef _OPENIB_CMA_
rdma_destroy_qp(cm_ptr->cm_id);
cm_ptr->cm_id->qp = NULL;
- ep_ptr->qp_handle = NULL;
- }
#else
- dapl_os_lock(&ep_ptr->header.lock);
- if (ep_ptr->qp_handle != NULL) {
- /* force error state to flush queue, then destroy */
- dapls_modify_qp_state(ep_ptr->qp_handle, IBV_QPS_ERR, 0,0,0);
-
- if (ibv_destroy_qp(ep_ptr->qp_handle)) {
+ if (ibv_destroy_qp(qp)) {
dapl_log(DAPL_DBG_TYPE_ERR,
" qp_free: ibv_destroy_qp error - %s\n",
strerror(errno));
}
- ep_ptr->qp_handle = NULL;
- }
#endif
- dapl_os_unlock(&ep_ptr->header.lock);
+ } else {
+ dapl_os_unlock(&ep_ptr->header.lock);
+ }
return DAT_SUCCESS;
}
--
1.7.3
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
