Merge pull request #6491 from bosilca/v4.0.x

hppritcha · web-flow · commit ceb93d7c0337 · 2019-03-15T17:08:52.000-06:00
v4.0.x: Cherry-pick fixes for issue #6258 from master (vader fixes)
diff --git a/opal/mca/btl/vader/btl_vader.h b/opal/mca/btl/vader/btl_vader.h
@@ -15,6 +15,8 @@
  * Copyright (c) 2010-2018 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * Copyright (c) 2015      Mellanox Technologies. All rights reserved.
+ * Copyright (c) 2018      Triad National Security, LLC. All rights
+ *                         reserved.
  *
  * $COPYRIGHT$
  *
@@ -53,6 +55,7 @@
 #include "opal/mca/rcache/rcache.h"
 #include "opal/mca/rcache/base/base.h"
 #include "opal/mca/btl/base/btl_base_error.h"
+#include "opal/mca/mpool/base/base.h"
 #include "opal/util/proc.h"
 #include "btl_vader_endpoint.h"
 
@@ -112,16 +115,15 @@ struct mca_btl_vader_component_t {
     opal_mutex_t lock;                      /**< lock to protect concurrent updates to this structure's members */
     char *my_segment;                       /**< this rank's base pointer */
     size_t segment_size;                    /**< size of my_segment */
-    size_t segment_offset;                  /**< start of unused portion of my_segment */
     int32_t num_smp_procs;                  /**< current number of smp procs on this host */
     opal_free_list_t vader_frags_eager;     /**< free list of vader send frags */
     opal_free_list_t vader_frags_max_send;  /**< free list of vader max send frags (large fragments) */
     opal_free_list_t vader_frags_user;      /**< free list of small inline frags */
+    opal_free_list_t vader_fboxes;          /**< free list of available fast-boxes */
 
     unsigned int fbox_threshold;            /**< number of sends required before we setup a send fast box for a peer */
     unsigned int fbox_max;                  /**< maximum number of send fast boxes to allocate */
     unsigned int fbox_size;                 /**< size of each peer fast box allocation */
-    unsigned int fbox_count;                /**< number of send fast boxes allocated  */
 
     int single_copy_mechanism;              /**< single copy mechanism to use */
 
@@ -143,6 +145,7 @@ struct mca_btl_vader_component_t {
 #if OPAL_BTL_VADER_HAVE_KNEM
     unsigned int knem_dma_min;              /**< minimum size to enable DMA for knem transfers (0 disables) */
 #endif
+    mca_mpool_base_module_t *mpool;
 };
 typedef struct mca_btl_vader_component_t mca_btl_vader_component_t;
 OPAL_MODULE_DECLSPEC extern mca_btl_vader_component_t mca_btl_vader_component;
diff --git a/opal/mca/btl/vader/btl_vader_component.c b/opal/mca/btl/vader/btl_vader_component.c
@@ -18,6 +18,9 @@
  * Copyright (c) 2014-2018 Intel, Inc. All rights reserved.
  * Copyright (c) 2014-2018 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2018      Amazon.com, Inc. or its affiliates.  All Rights reserved.
+ * Copyright (c) 2018      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -301,6 +304,7 @@ static int mca_btl_vader_component_open(void)
     OBJ_CONSTRUCT(&mca_btl_vader_component.vader_frags_eager, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_vader_component.vader_frags_user, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_vader_component.vader_frags_max_send, opal_free_list_t);
+    OBJ_CONSTRUCT(&mca_btl_vader_component.vader_fboxes, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_vader_component.lock, opal_mutex_t);
     OBJ_CONSTRUCT(&mca_btl_vader_component.pending_endpoints, opal_list_t);
     OBJ_CONSTRUCT(&mca_btl_vader_component.pending_fragments, opal_list_t);
@@ -321,6 +325,7 @@ static int mca_btl_vader_component_close(void)
     OBJ_DESTRUCT(&mca_btl_vader_component.vader_frags_eager);
     OBJ_DESTRUCT(&mca_btl_vader_component.vader_frags_user);
     OBJ_DESTRUCT(&mca_btl_vader_component.vader_frags_max_send);
+    OBJ_DESTRUCT(&mca_btl_vader_component.vader_fboxes);
     OBJ_DESTRUCT(&mca_btl_vader_component.lock);
     OBJ_DESTRUCT(&mca_btl_vader_component.pending_endpoints);
     OBJ_DESTRUCT(&mca_btl_vader_component.pending_fragments);
@@ -517,7 +522,6 @@ static mca_btl_base_module_t **mca_btl_vader_component_init (int *num_btls,
 
     /* no fast boxes allocated initially */
     component->num_fbox_in_endpoints = 0;
-    component->fbox_count = 0;
 
     mca_btl_vader_check_single_copy ();
 
@@ -559,8 +563,6 @@ static mca_btl_base_module_t **mca_btl_vader_component_init (int *num_btls,
         }
     }
 
-    component->segment_offset = 0;
-
     /* initialize my fifo */
     vader_fifo_init ((struct vader_fifo_t *) component->my_segment);
 
diff --git a/opal/mca/btl/vader/btl_vader_endpoint.h b/opal/mca/btl/vader/btl_vader_endpoint.h
@@ -13,6 +13,8 @@
  * Copyright (c) 2006-2007 Voltaire. All rights reserved.
  * Copyright (c) 2012-2016 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2018      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -58,6 +60,7 @@ typedef struct mca_btl_base_endpoint_t {
         uint32_t *startp;      /**< pointer to location storing start offset */
         unsigned int start, end;
         uint16_t seq;
+        opal_free_list_item_t *fbox; /**< fast-box free list item */
     } fbox_out;
 
     int32_t peer_smp_rank;  /**< my peer's SMP process rank.  Used for accessing
@@ -101,13 +104,16 @@ static inline void mca_btl_vader_endpoint_setup_fbox_recv (struct mca_btl_base_e
     endpoint->fbox_in.buffer = base;
 }
 
-static inline void mca_btl_vader_endpoint_setup_fbox_send (struct mca_btl_base_endpoint_t *endpoint, void *base)
+static inline void mca_btl_vader_endpoint_setup_fbox_send (struct mca_btl_base_endpoint_t *endpoint, opal_free_list_item_t *fbox)
 {
+    void *base = fbox->ptr;
+
     endpoint->fbox_out.start = MCA_BTL_VADER_FBOX_ALIGNMENT;
     endpoint->fbox_out.end = MCA_BTL_VADER_FBOX_ALIGNMENT;
     endpoint->fbox_out.startp = (uint32_t *) base;
     endpoint->fbox_out.startp[0] = MCA_BTL_VADER_FBOX_ALIGNMENT;
     endpoint->fbox_out.seq = 0;
+    endpoint->fbox_out.fbox = fbox;
 
     /* zero out the first header in the fast box */
     memset ((char *) base + MCA_BTL_VADER_FBOX_ALIGNMENT, 0, MCA_BTL_VADER_FBOX_ALIGNMENT);
diff --git a/opal/mca/btl/vader/btl_vader_fbox.h b/opal/mca/btl/vader/btl_vader_fbox.h
@@ -2,6 +2,8 @@
 /*
  * Copyright (c) 2011-2018 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2018      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -259,20 +261,17 @@ static inline void mca_btl_vader_try_fbox_setup (mca_btl_base_endpoint_t *ep, mc
         /* protect access to mca_btl_vader_component.segment_offset */
         OPAL_THREAD_LOCK(&mca_btl_vader_component.lock);
 
-        if (mca_btl_vader_component.segment_size >= mca_btl_vader_component.segment_offset + mca_btl_vader_component.fbox_size &&
-            mca_btl_vader_component.fbox_max > mca_btl_vader_component.fbox_count) {
-            /* verify the remote side will accept another fbox */
-            if (0 <= opal_atomic_add_fetch_32 (&ep->fifo->fbox_available, -1)) {
-                void *fbox_base = mca_btl_vader_component.my_segment + mca_btl_vader_component.segment_offset;
-                mca_btl_vader_component.segment_offset += mca_btl_vader_component.fbox_size;
+        /* verify the remote side will accept another fbox */
+        if (0 <= opal_atomic_add_fetch_32 (&ep->fifo->fbox_available, -1)) {
+            opal_free_list_item_t *fbox = opal_free_list_get (&mca_btl_vader_component.vader_fboxes);
 
+            if (NULL != fbox) {
                 /* zero out the fast box */
-                memset (fbox_base, 0, mca_btl_vader_component.fbox_size);
-                mca_btl_vader_endpoint_setup_fbox_send (ep, fbox_base);
+                memset (fbox->ptr, 0, mca_btl_vader_component.fbox_size);
+                mca_btl_vader_endpoint_setup_fbox_send (ep, fbox);
 
                 hdr->flags |= MCA_BTL_VADER_FLAG_SETUP_FBOX;
                 hdr->fbox_base = virtual2relative((char *) ep->fbox_out.buffer);
-                ++mca_btl_vader_component.fbox_count;
             } else {
                 opal_atomic_add_fetch_32 (&ep->fifo->fbox_available, 1);
             }
diff --git a/opal/mca/btl/vader/btl_vader_frag.c b/opal/mca/btl/vader/btl_vader_frag.c
@@ -13,6 +13,8 @@
  * Copyright (c) 2009      Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2011-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2018      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -41,38 +43,9 @@ static inline void mca_btl_vader_frag_constructor (mca_btl_vader_frag_t *frag)
 int mca_btl_vader_frag_init (opal_free_list_item_t *item, void *ctx)
 {
     mca_btl_vader_frag_t *frag = (mca_btl_vader_frag_t *) item;
-    unsigned int data_size = (unsigned int)(uintptr_t) ctx;
-    unsigned int frag_size = data_size + sizeof (mca_btl_vader_hdr_t);
-
-    /* ensure next fragment is aligned on a cache line */
-    frag_size = (frag_size + 63) & ~63;
-
-    OPAL_THREAD_LOCK(&mca_btl_vader_component.lock);
-
-    if (data_size && mca_btl_vader_component.segment_size < mca_btl_vader_component.segment_offset + frag_size) {
-        OPAL_THREAD_UNLOCK(&mca_btl_vader_component.lock);
-        return OPAL_ERR_OUT_OF_RESOURCE;
-    }
-
-    /* Set the list element here so we don't have to set it on the critical path. This only
-     * works if each free list has its own unique fragment size and ALL free lists are initialized
-     * with opal_free_list_init. */
-    if (mca_btl_vader_component.max_inline_send == data_size) {
-        frag->my_list = &mca_btl_vader_component.vader_frags_user;
-    } else if (mca_btl_vader.super.btl_eager_limit == data_size) {
-        frag->my_list = &mca_btl_vader_component.vader_frags_eager;
-    } else if (mca_btl_vader.super.btl_max_send_size == data_size) {
-        frag->my_list = &mca_btl_vader_component.vader_frags_max_send;
-    }
-
-    if (data_size) {
-        item->ptr = mca_btl_vader_component.my_segment + mca_btl_vader_component.segment_offset;
-        mca_btl_vader_component.segment_offset += frag_size;
-    }
-
-    OPAL_THREAD_UNLOCK(&mca_btl_vader_component.lock);
 
-    mca_btl_vader_frag_constructor ((mca_btl_vader_frag_t *) item);
+    /* Set the list element here so we don't have to set it on the critical path */
+    frag->my_list = (opal_free_list_t *) ctx;
 
     return OPAL_SUCCESS;
 }
diff --git a/opal/mca/btl/vader/btl_vader_module.c b/opal/mca/btl/vader/btl_vader_module.c
@@ -17,7 +17,7 @@
  * Copyright (c) 2014-2015 Intel, Inc. All rights reserved.
  * Copyright (c) 2014-2015 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
- * Copyright (c) 2018      Triad National Security, LLC. All rights
+ * Copyright (c) 2018-2019 Triad National Security, LLC. All rights
  *                         reserved.
  * $COPYRIGHT$
  *
@@ -97,19 +97,32 @@ static int vader_btl_first_time_init(mca_btl_vader_t *vader_btl, int n)
         return OPAL_ERR_OUT_OF_RESOURCE;
     }
 
-    component->segment_offset = MCA_BTL_VADER_FIFO_SIZE;
+    component->mpool = mca_mpool_basic_create ((void *) (component->my_segment + MCA_BTL_VADER_FIFO_SIZE),
+                                               (unsigned long) (mca_btl_vader_component.segment_size - MCA_BTL_VADER_FIFO_SIZE), 64);
+    if (NULL == component->mpool) {
+        free (component->endpoints);
+        return OPAL_ERR_OUT_OF_RESOURCE;
+    }
+
+    rc = opal_free_list_init (&component->vader_fboxes, sizeof (opal_free_list_item_t), 8,
+                              OBJ_CLASS(opal_free_list_item_t), mca_btl_vader_component.fbox_size,
+                              opal_cache_line_size, 0, mca_btl_vader_component.fbox_max, 4,
+                              component->mpool, 0, NULL, NULL, NULL);
+    if (OPAL_SUCCESS != rc) {
+        return rc;
+    }
 
     /* initialize fragment descriptor free lists */
     /* initialize free list for small send and inline fragments */
     rc = opal_free_list_init (&component->vader_frags_user,
                               sizeof(mca_btl_vader_frag_t),
                               opal_cache_line_size, OBJ_CLASS(mca_btl_vader_frag_t),
-                              0, opal_cache_line_size,
-                              component->vader_free_list_num,
+                              mca_btl_vader_component.max_inline_send + sizeof (mca_btl_vader_frag_t),
+                              opal_cache_line_size, component->vader_free_list_num,
                               component->vader_free_list_max,
                               component->vader_free_list_inc,
-                              NULL, 0, NULL, mca_btl_vader_frag_init,
-                              (void *)(intptr_t) mca_btl_vader_component.max_inline_send);
+                              component->mpool, 0, NULL, mca_btl_vader_frag_init,
+                              &component->vader_frags_user);
     if (OPAL_SUCCESS != rc) {
         return rc;
     }
@@ -118,12 +131,12 @@ static int vader_btl_first_time_init(mca_btl_vader_t *vader_btl, int n)
     rc = opal_free_list_init (&component->vader_frags_eager,
                               sizeof (mca_btl_vader_frag_t),
                               opal_cache_line_size, OBJ_CLASS(mca_btl_vader_frag_t),
-                              0, opal_cache_line_size,
-                              component->vader_free_list_num,
+                              mca_btl_vader.super.btl_eager_limit + sizeof (mca_btl_vader_frag_t),
+                              opal_cache_line_size, component->vader_free_list_num,
                               component->vader_free_list_max,
                               component->vader_free_list_inc,
-                              NULL, 0, NULL, mca_btl_vader_frag_init,
-                              (void *)(intptr_t) mca_btl_vader.super.btl_eager_limit);
+                              component->mpool, 0, NULL, mca_btl_vader_frag_init,
+                              &component->vader_frags_eager);
     if (OPAL_SUCCESS != rc) {
         return rc;
     }
@@ -133,12 +146,12 @@ static int vader_btl_first_time_init(mca_btl_vader_t *vader_btl, int n)
         rc = opal_free_list_init (&component->vader_frags_max_send,
                                   sizeof (mca_btl_vader_frag_t),
                                   opal_cache_line_size, OBJ_CLASS(mca_btl_vader_frag_t),
-                                  0, opal_cache_line_size,
-                                  component->vader_free_list_num,
+                                  mca_btl_vader.super.btl_max_send_size + sizeof (mca_btl_vader_frag_t),
+                                  opal_cache_line_size, component->vader_free_list_num,
                                   component->vader_free_list_max,
                                   component->vader_free_list_inc,
-                                  NULL, 0, NULL, mca_btl_vader_frag_init,
-                                  (void *)(intptr_t) mca_btl_vader.super.btl_max_send_size);
+                                  component->mpool, 0, NULL, mca_btl_vader_frag_init,
+                                  &component->vader_frags_max_send);
         if (OPAL_SUCCESS != rc) {
             return rc;
         }
@@ -367,6 +380,11 @@ static int vader_finalize(struct mca_btl_base_module_t *btl)
     }
 #endif
 
+    if (component->mpool) {
+        component->mpool->mpool_finalize (component->mpool);
+        component->mpool = NULL;
+    }
+
     return OPAL_SUCCESS;
 }
 
@@ -536,6 +554,7 @@ static void mca_btl_vader_endpoint_constructor (mca_btl_vader_endpoint_t *ep)
     OBJ_CONSTRUCT(&ep->pending_frags, opal_list_t);
     OBJ_CONSTRUCT(&ep->pending_frags_lock, opal_mutex_t);
     ep->fifo = NULL;
+    ep->fbox_out.fbox = NULL;
 }
 
 #if OPAL_BTL_VADER_HAVE_XPMEM
@@ -564,8 +583,12 @@ static void mca_btl_vader_endpoint_destructor (mca_btl_vader_endpoint_t *ep)
         /* disconnect from the peer's segment */
         opal_shmem_segment_detach (&seg_ds);
     }
+    if (ep->fbox_out.fbox) {
+        opal_free_list_return (&mca_btl_vader_component.vader_fboxes, ep->fbox_out.fbox);
+    }
 
     ep->fbox_in.buffer = ep->fbox_out.buffer = NULL;
+    ep->fbox_out.fbox = NULL;
     ep->segment_base = NULL;
     ep->fifo = NULL;
 }
diff --git a/opal/mca/mpool/base/Makefile.am b/opal/mca/mpool/base/Makefile.am
@@ -28,7 +28,8 @@ libmca_mpool_la_SOURCES += \
         base/mpool_base_lookup.c \
         base/mpool_base_alloc.c \
 	base/mpool_base_tree.c \
-	base/mpool_base_default.c
+	base/mpool_base_default.c \
+	base/mpool_base_basic.c
 
 dist_opaldata_DATA += \
         base/help-mpool-base.txt
diff --git a/opal/mca/mpool/base/base.h b/opal/mca/mpool/base/base.h
@@ -53,6 +53,8 @@ OPAL_DECLSPEC OBJ_CLASS_DECLARATION(mca_mpool_base_selected_module_t);
 OPAL_DECLSPEC mca_mpool_base_component_t* mca_mpool_base_component_lookup(const char* name);
 OPAL_DECLSPEC mca_mpool_base_module_t* mca_mpool_base_module_lookup(const char* name);
 
+OPAL_DECLSPEC mca_mpool_base_module_t *mca_mpool_basic_create (void *base, size_t size, unsigned min_align);
+
 /*
  * Globals
  */
diff --git a/opal/mca/mpool/base/mpool_base_basic.c b/opal/mca/mpool/base/mpool_base_basic.c
