btl/openib: fix cq resize calculation

Before dynamic add_procs the openib_btl_size_queues was called exactly
once for non-dynamic jobs. Now the function is called on each new
connection so the calculation was wrong. Re-wrote the function to
correctly calculate the CQ size and only attempt to adjust the CQ if
the requested size has changed. This fixes a bug when using the openib
btl on psm2 hardware that is caused by the time needed to resize a
CQ. The overhead was causing udcm to timeout and fail.

Signed-off-by: Nathan Hjelm <[email protected]>
(cherry picked from commit open-mpi/ompi@dd519c5)

Author: hjelmn

opal/mca/btl/openib/btl_openib.c

@@ -421,13 +421,20 @@ static int openib_btl_prepare(struct mca_btl_openib_module_t* openib_btl)
 static int openib_btl_size_queues(struct mca_btl_openib_module_t* openib_btl)
 {
     uint32_t send_cqes, recv_cqes;
-    int rc = OPAL_SUCCESS, qp;
+    int rc = OPAL_SUCCESS;
     mca_btl_openib_device_t *device = openib_btl->device;
+    uint32_t requested[BTL_OPENIB_MAX_CQ];
+    bool need_resize = false;
 
     opal_mutex_lock(&openib_btl->ib_lock);
+
+    for (int cq = 0 ; cq < BTL_OPENIB_MAX_CQ ; ++cq) {
+        requested[cq] = 0;
+    }
+
     /* figure out reasonable sizes for completion queues */
-    for(qp = 0; qp < mca_btl_openib_component.num_qps; qp++) {
-        if(BTL_OPENIB_QP_TYPE_SRQ(qp)) {
+    for (int qp = 0 ; qp < mca_btl_openib_component.num_qps ; qp++) {
+        if (BTL_OPENIB_QP_TYPE_SRQ(qp)) {
             send_cqes = mca_btl_openib_component.qp_infos[qp].u.srq_qp.sd_max;
             recv_cqes = mca_btl_openib_component.qp_infos[qp].rd_num;
         } else {
@@ -436,24 +443,30 @@ static int openib_btl_size_queues(struct mca_btl_openib_module_t* openib_btl)
             recv_cqes = send_cqes;
         }
 
-        opal_mutex_lock(&openib_btl->device->device_lock);
-        openib_btl->device->cq_size[qp_cq_prio(qp)] += recv_cqes;
-        openib_btl->device->cq_size[BTL_OPENIB_LP_CQ] += send_cqes;
-        opal_mutex_unlock(&openib_btl->device->device_lock);
+        requested[qp_cq_prio(qp)] += recv_cqes;
+        requested[BTL_OPENIB_LP_CQ] += send_cqes;
     }
 
-    rc = adjust_cq(device, BTL_OPENIB_HP_CQ);
-    if (OPAL_SUCCESS != rc) {
-        goto out;
-    }
+    opal_mutex_lock (&openib_btl->device->device_lock);
+    for (int cq = 0 ; cq < BTL_OPENIB_MAX_CQ ; ++cq) {
+        if (requested[cq] < mca_btl_openib_component.ib_cq_size[cq]) {
+            requested[cq] = mca_btl_openib_component.ib_cq_size[cq];
+        } else if (requested[cq] > openib_btl->device->ib_dev_attr.max_cqe) {
+            requested[cq] = openib_btl->device->ib_dev_attr.max_cqe;
+        }
 
-    rc = adjust_cq(device, BTL_OPENIB_LP_CQ);
-    if (OPAL_SUCCESS != rc) {
-        goto out;
-    }
+        if (openib_btl->device->cq_size[cq] < requested[cq]) {
+            openib_btl->device->cq_size[cq] = requested[cq];
 
-out:
+            rc = adjust_cq (device, cq);
+            if (OPAL_SUCCESS != rc) {
+                break;
+            }
+        }
+    }
+    opal_mutex_unlock (&openib_btl->device->device_lock);
     opal_mutex_unlock(&openib_btl->ib_lock);
+
     return rc;
 }
 
@@ -1081,7 +1094,7 @@ int mca_btl_openib_add_procs(
     }
 
     if (nprocs_new) {
-        OPAL_THREAD_ADD32(&openib_btl->num_peers, nprocs_new);
+        opal_atomic_add_32 (&openib_btl->num_peers, nprocs_new);
 
         /* adjust cq sizes given the new procs */
         rc = openib_btl_size_queues (openib_btl);
@@ -1191,7 +1204,7 @@ struct mca_btl_base_endpoint_t *mca_btl_openib_get_ep (struct mca_btl_base_modul
 
         /* this is a new process to this openib btl
          * account this procs if need */
-        OPAL_THREAD_ADD32(&openib_btl->num_peers, 1);
+        opal_atomic_add_32 (&openib_btl->num_peers, 1);
         rc = openib_btl_size_queues(openib_btl);
         if (OPAL_SUCCESS != rc) {
             BTL_ERROR(("error creating cqs"));

opal/mca/btl/openib/btl_openib.h

@@ -83,6 +83,12 @@ BEGIN_C_DECLS
  * Infiniband (IB) BTL component.
  */
 
+enum {
+    BTL_OPENIB_HP_CQ,
+    BTL_OPENIB_LP_CQ,
+    BTL_OPENIB_MAX_CQ,
+};
+
 typedef enum {
     MCA_BTL_OPENIB_TRANSPORT_IB,
     MCA_BTL_OPENIB_TRANSPORT_IWARP,
@@ -204,7 +210,7 @@ struct mca_btl_openib_component_t {
     uint32_t reg_mru_len;    /**< Length of the registration cache most recently used list */
     uint32_t use_srq;        /**< Use the Shared Receive Queue (SRQ mode) */
 
-    uint32_t ib_cq_size[2];  /**< Max outstanding CQE on the CQ */
+    uint32_t ib_cq_size[BTL_OPENIB_MAX_CQ];  /**< Max outstanding CQE on the CQ */
 
     int      ib_max_inline_data; /**< Max size of inline data */
     unsigned int ib_pkey_val;
@@ -374,8 +380,8 @@ typedef struct mca_btl_openib_device_t {
 #endif
     struct ibv_device_attr ib_dev_attr;
     struct ibv_pd *ib_pd;
-    struct ibv_cq *ib_cq[2];
-    uint32_t cq_size[2];
+    struct ibv_cq *ib_cq[BTL_OPENIB_MAX_CQ];
+    uint32_t cq_size[BTL_OPENIB_MAX_CQ];
     mca_mpool_base_module_t *mpool;
     /* MTU for this device */
     uint32_t mtu;
@@ -848,11 +854,6 @@ extern void mca_btl_openib_frag_progress_pending_put_get(
  */
 void mca_btl_openib_show_init_error(const char *file, int line,
                                     const char *func, const char *dev);
-
-#define BTL_OPENIB_HP_CQ 0
-#define BTL_OPENIB_LP_CQ 1
-
-
 /**
  * Post to Shared Receive Queue with certain priority
  *