opal: rework mpool and rcache frameworks

This commit rewrites both the mpool and rcache frameworks. Summary of
changes:

 - Before this change a significant portion of the rcache
   functionality lived in mpool components. This meant that it was
   impossible to add a new memory pool to use with rdma networks
   (ugni, openib, etc) without duplicating the functionality of an
   existing mpool component. All the registration functionality has
   been removed from the mpool and placed in the rcache framework.

 - All registration cache mpools components (udreg, grdma, gpusm,
   rgpusm) have been changed to rcache components. rcaches are
   allocated and released in the same way mpool components were.

 - It is now valid to pass NULL as the resources argument when
   creating an rcache. At this time the gpusm and rgpusm components
   support this. All other rcache components require non-NULL
   resources.

 - A new mpool component has been added: hugepage. This component
   supports huge page allocations on linux.

 - Memory pools are now allocated using "hints". Each mpool component
   is queried with the hints and returns a priority. The current hints
   supported are NULL (uses posix_memalign/malloc), page_size=x (huge
   page mpool), and mpool=x.

 - The sm mpool has been moved to common/sm. This reflects that the sm
   mpool is specialized and not meant for any general
   allocations. This mpool may be moved back into the mpool framework
   if there is any objection.

 - The opal_free_list_init arguments have been updated. The unused0
   argument is not used to pass in the registration cache module. The
   mpool registration flags are now rcache registration flags.

 - All components have been updated to make use of the new framework
   interfaces.

As this commit makes significant changes to both the mpool and rcache
frameworks both versions have been bumped to 3.0.0.

Signed-off-by: Nathan Hjelm <[email protected]>

Author: hjelmn

ompi/mca/crcp/base/base.h

@@ -1,3 +1,4 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
  * Copyright (c) 2004-2010 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
@@ -9,6 +10,8 @@
  *                         University of Stuttgart.  All rights reserved.
  * Copyright (c) 2004-2005 The Regents of the University of California.
  *                         All rights reserved.
+ * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -145,7 +148,7 @@ BEGIN_C_DECLS
     ompi_crcp_base_btl_state_t*
     ompi_crcp_base_none_btl_prepare_src( struct mca_btl_base_module_t* btl,
                                          struct mca_btl_base_endpoint_t* endpoint,
-                                         mca_mpool_base_registration_t* registration,
+                                         mca_rcache_base_registration_t* registration,
                                          struct opal_convertor_t* convertor,
                                          size_t reserve,
                                          size_t* size,
@@ -154,7 +157,7 @@ BEGIN_C_DECLS
     ompi_crcp_base_btl_state_t*
     ompi_crcp_base_none_btl_prepare_dst( struct mca_btl_base_module_t* btl,
                                          struct mca_btl_base_endpoint_t* endpoint,
-                                         mca_mpool_base_registration_t* registration,
+                                         mca_rcache_base_registration_t* registration,
                                          struct opal_convertor_t* convertor,
                                          size_t reserve,
                                          size_t* size,

ompi/mca/crcp/base/crcp_base_fns.c

@@ -336,7 +336,7 @@ ompi_crcp_base_none_btl_free( struct mca_btl_base_module_t* btl,
 ompi_crcp_base_btl_state_t*
 ompi_crcp_base_none_btl_prepare_src( struct mca_btl_base_module_t* btl,
                                      struct mca_btl_base_endpoint_t* endpoint,
-                                     mca_mpool_base_registration_t* registration,
+                                     mca_rcache_base_registration_t* registration,
                                      struct opal_convertor_t* convertor,
                                      size_t reserve,
                                      size_t* size,
@@ -349,7 +349,7 @@ ompi_crcp_base_none_btl_prepare_src( struct mca_btl_base_module_t* btl,
 ompi_crcp_base_btl_state_t*
 ompi_crcp_base_none_btl_prepare_dst( struct mca_btl_base_module_t* btl,
                                      struct mca_btl_base_endpoint_t* endpoint,
-                                     mca_mpool_base_registration_t* registration,
+                                     mca_rcache_base_registration_t* registration,
                                      struct opal_convertor_t* convertor,
                                      size_t reserve,
                                      size_t* size,

ompi/mca/crcp/crcp.h

@@ -235,7 +235,7 @@ typedef ompi_crcp_base_btl_state_t* (*mca_crcp_base_btl_module_free_fn_t)
 typedef ompi_crcp_base_btl_state_t* (*mca_crcp_base_btl_module_prepare_fn_t)
      ( struct mca_btl_base_module_t* btl,
        struct mca_btl_base_endpoint_t* endpoint,
-       mca_mpool_base_registration_t* registration,
+       mca_rcache_base_registration_t* registration,
        struct opal_convertor_t* convertor,
        size_t reserve,
        size_t* size,

ompi/mca/pml/base/pml_base_bsend.c

@@ -1,3 +1,4 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
@@ -12,6 +13,8 @@
  * Copyright (c) 2007      Sun Microsystems, Inc.  All rights reserved.
  * Copyright (c) 2015      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -55,10 +58,7 @@ extern char *ompi_pml_base_bsend_allocator_name;
 /*
  * Routine to return pages to sub-allocator as needed
  */
-static void* mca_pml_bsend_alloc_segment(
-    struct mca_mpool_base_module_t* module,
-    size_t* size_inout,
-    mca_mpool_base_registration_t** registration)
+static void* mca_pml_bsend_alloc_segment(void *ctx, size_t *size_inout)
 {
     void *addr;
     size_t size = *size_inout;
@@ -70,7 +70,6 @@ static void* mca_pml_bsend_alloc_segment(
     addr = mca_pml_bsend_addr;
     mca_pml_bsend_addr += size;
     *size_inout = size;
-    if (NULL != registration) *registration = NULL;
     return addr;
 }
 
@@ -232,7 +231,7 @@ int mca_pml_base_bsend_request_start(ompi_request_t* request)
 
         /* allocate a buffer to hold packed message */
         sendreq->req_addr = mca_pml_bsend_allocator->alc_alloc(
-            mca_pml_bsend_allocator, sendreq->req_bytes_packed, 0, NULL);
+            mca_pml_bsend_allocator, sendreq->req_bytes_packed, 0);
         if(NULL == sendreq->req_addr) {
             /* release resources when request is freed */
             sendreq->req_base.req_pml_complete = true;
@@ -287,7 +286,7 @@ int mca_pml_base_bsend_request_alloc(ompi_request_t* request)
 
     /* allocate a buffer to hold packed message */
     sendreq->req_addr = mca_pml_bsend_allocator->alc_alloc(
-        mca_pml_bsend_allocator, sendreq->req_bytes_packed, 0, NULL);
+        mca_pml_bsend_allocator, sendreq->req_bytes_packed, 0);
     if(NULL == sendreq->req_addr) {
         /* release resources when request is freed */
         sendreq->req_base.req_pml_complete = true;
@@ -321,7 +320,7 @@ void*  mca_pml_base_bsend_request_alloc_buf( size_t length )
 
     /* allocate a buffer to hold packed message */
     buf = mca_pml_bsend_allocator->alc_alloc(
-        mca_pml_bsend_allocator, length, 0, NULL);
+        mca_pml_bsend_allocator, length, 0);
     if(NULL == buf) {
         /* release resources when request is freed */
         OPAL_THREAD_UNLOCK(&mca_pml_bsend_mutex);

ompi/mca/pml/ob1/pml_ob1_component.c

@@ -79,12 +79,9 @@ mca_pml_base_component_2_0_0_t mca_pml_ob1_component = {
     .pmlm_finalize = mca_pml_ob1_component_fini,
 };
 
-void *mca_pml_ob1_seg_alloc( struct mca_mpool_base_module_t* mpool,
-                             size_t* size,
-                             mca_mpool_base_registration_t** registration);
+void *mca_pml_ob1_seg_alloc (void *ctx, size_t* size);
 
-void mca_pml_ob1_seg_free( struct mca_mpool_base_module_t* mpool,
-                           void* segment );
+void mca_pml_ob1_seg_free (void *ctx, void *segment);
 
 static inline int mca_pml_ob1_param_register_int(
     const char* param_name,
@@ -354,13 +351,12 @@ int mca_pml_ob1_component_fini(void)
     return OMPI_SUCCESS;
 }
 
-void *mca_pml_ob1_seg_alloc( struct mca_mpool_base_module_t* mpool,
-                             size_t* size,
-                             mca_mpool_base_registration_t** registration) {
+void *mca_pml_ob1_seg_alloc (void *ctx, size_t *size)
+{
     return malloc(*size);
 }
 
-void mca_pml_ob1_seg_free( struct mca_mpool_base_module_t* mpool,
-                           void* segment ) {
+void mca_pml_ob1_seg_free (void *ctx, void *segment)
+{
     free(segment);
 }

ompi/mca/pml/ob1/pml_ob1_recvfrag.h

@@ -81,7 +81,7 @@ do {                                                                    \
         buffers[0].addr = (char*)                                       \
             mca_pml_ob1.allocator->alc_alloc( mca_pml_ob1.allocator,    \
                                               buffers[0].len,           \
-                                              0, NULL);                 \
+                                              0);                       \
         _ptr = (unsigned char*)(buffers[0].addr);                       \
         macro_segments[0].seg_addr.pval = buffers[0].addr;              \
     }                                                                   \

ompi/mca/vprotocol/base/vprotocol_base_request.c

@@ -42,8 +42,8 @@ int mca_vprotocol_base_request_parasite(void)
                                    pml_fl_save.fl_max_to_alloc,
                                    pml_fl_save.fl_num_per_alloc,
                                    pml_fl_save.fl_mpool,
-                                   pml_fl_save.fl_mpool_reg_flags,
-                                   0,
+                                   pml_fl_save.fl_rcache_reg_flags,
+                                   pml_fl_save.fl_rcache,
                                    pml_fl_save.item_init,
                                    pml_fl_save.ctx);
         if(OMPI_SUCCESS != ret) return ret;
@@ -71,8 +71,8 @@ int mca_vprotocol_base_request_parasite(void)
                                    pml_fl_save.fl_max_to_alloc,
                                    pml_fl_save.fl_num_per_alloc,
                                    pml_fl_save.fl_mpool,
-                                   pml_fl_save.fl_mpool_reg_flags,
-                                   0,
+                                   pml_fl_save.fl_rcache_reg_flags,
+                                   pml_fl_save.fl_rcache,
                                    pml_fl_save.item_init,
                                    pml_fl_save.ctx);
         if(OMPI_SUCCESS != ret) return ret;

ompi/mpi/c/alloc_mem.c

@@ -1,3 +1,4 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
  * Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
@@ -12,6 +13,8 @@
  * Copyright (c) 2007      Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2015      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -43,6 +46,8 @@ static const char FUNC_NAME[] = "MPI_Alloc_mem";
 
 int MPI_Alloc_mem(MPI_Aint size, MPI_Info info, void *baseptr)
 {
+    char info_value[MPI_MAX_INFO_VAL + 1];
+    char *mpool_hints = NULL;
 
     if (MPI_PARAM_CHECK) {
         OMPI_ERR_INIT_FINALIZE(FUNC_NAME);
@@ -69,7 +74,16 @@ int MPI_Alloc_mem(MPI_Aint size, MPI_Info info, void *baseptr)
 
     OPAL_CR_ENTER_LIBRARY();
 
-    *((void **) baseptr) = mca_mpool_base_alloc((size_t) size, (struct opal_info_t*)info);
+    if (MPI_INFO_NULL != info) {
+        int flag;
+        (void) ompi_info_get (info, "mpool_hints", MPI_MAX_INFO_VAL, info_value, &flag);
+        if (flag) {
+            mpool_hints = info_value;
+        }
+    }
+
+    *((void **) baseptr) = mca_mpool_base_alloc ((size_t) size, (struct opal_info_t*)info,
+                                                 mpool_hints);
     OPAL_CR_EXIT_LIBRARY();
     if (NULL == *((void **) baseptr)) {
         return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_NO_MEM,

ompi/runtime/ompi_mpi_init.c

@@ -620,13 +620,6 @@ int ompi_mpi_init(int argc, char **argv, int requested, int *provided)
 
     /* Select which MPI components to use */
 
-    if (OMPI_SUCCESS !=
-        (ret = mca_mpool_base_init(OPAL_ENABLE_PROGRESS_THREADS,
-                                   ompi_mpi_thread_multiple))) {
-        error = "mca_mpool_base_init() failed";
-        goto error;
-    }
-
     if (OMPI_SUCCESS !=
         (ret = mca_pml_base_select(OPAL_ENABLE_PROGRESS_THREADS,
                                    ompi_mpi_thread_multiple))) {

opal/class/opal_free_list.c

@@ -28,6 +28,8 @@
 #include "opal/align.h"
 #include "opal/util/output.h"
 #include "opal/mca/mpool/mpool.h"
+#include "opal/mca/mpool/base/base.h"
+#include "opal/mca/rcache/rcache.h"
 
 typedef struct opal_free_list_item_t opal_free_list_memory_t;
 
@@ -49,17 +51,22 @@ static void opal_free_list_construct(opal_free_list_t* fl)
     fl->fl_payload_buffer_alignment = 0;
     fl->fl_frag_class = OBJ_CLASS(opal_free_list_item_t);
     fl->fl_mpool = NULL;
+    fl->fl_rcache = NULL;
     /* default flags */
-    fl->fl_mpool_reg_flags = MCA_MPOOL_FLAGS_CACHE_BYPASS |
-        MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM;
+    fl->fl_rcache_reg_flags = MCA_RCACHE_FLAGS_CACHE_BYPASS |
+        MCA_RCACHE_FLAGS_CUDA_REGISTER_MEM;
     fl->ctx = NULL;
     OBJ_CONSTRUCT(&(fl->fl_allocations), opal_list_t);
 }
 
 static void opal_free_list_allocation_release (opal_free_list_t *fl, opal_free_list_memory_t *fl_mem)
 {
+    if (NULL != fl->fl_rcache) {
+        fl->fl_rcache->rcache_deregister (fl->fl_rcache, fl_mem->registration);
+    }
+
     if (NULL != fl->fl_mpool) {
-        fl->fl_mpool->mpool_free (fl->fl_mpool, fl_mem->ptr, fl_mem->registration);
+        fl->fl_mpool->mpool_free (fl->fl_mpool, fl_mem->ptr);
     } else if (fl_mem->ptr) {
         free (fl_mem->ptr);
     }
@@ -108,8 +115,9 @@ int opal_free_list_init (opal_free_list_t *flist, size_t frag_size, size_t frag_
                          opal_class_t *frag_class, size_t payload_buffer_size,
                          size_t payload_buffer_alignment, int num_elements_to_alloc,
                          int max_elements_to_alloc, int num_elements_per_alloc,
-                         mca_mpool_base_module_t* mpool, int mpool_reg_flags,
-                         void *unused0, opal_free_list_item_init_fn_t item_init, void *ctx)
+                         mca_mpool_base_module_t *mpool, int rcache_reg_flags,
+                         mca_rcache_base_module_t *rcache, opal_free_list_item_init_fn_t item_init,
+                         void *ctx)
 {
     /* alignment must be more than zero and power of two */
     if (frag_alignment <= 1 || (frag_alignment & (frag_alignment - 1))) {
@@ -137,11 +145,12 @@ int opal_free_list_init (opal_free_list_t *flist, size_t frag_size, size_t frag_
     flist->fl_max_to_alloc = max_elements_to_alloc;
     flist->fl_num_allocated = 0;
     flist->fl_num_per_alloc = num_elements_per_alloc;
-    flist->fl_mpool = mpool;
+    flist->fl_mpool = mpool ? mpool : mca_mpool_base_default_module;
+    flist->fl_rcache = rcache;
     flist->fl_frag_alignment = frag_alignment;
     flist->fl_payload_buffer_alignment = payload_buffer_alignment;
     flist->item_init = item_init;
-    flist->fl_mpool_reg_flags |= mpool_reg_flags;
+    flist->fl_rcache_reg_flags |= rcache_reg_flags;
     flist->ctx = ctx;
 
     if (num_elements_to_alloc) {
@@ -153,10 +162,10 @@ int opal_free_list_init (opal_free_list_t *flist, size_t frag_size, size_t frag_
 
 int opal_free_list_grow_st (opal_free_list_t* flist, size_t num_elements)
 {
-    unsigned char *ptr, *mpool_alloc_ptr = NULL, *payload_ptr = NULL;
+    unsigned char *ptr, *payload_ptr = NULL;
     opal_free_list_memory_t *alloc_ptr;
     size_t alloc_size, head_size, elem_size = 0;
-    mca_mpool_base_registration_t *reg = NULL;
+    mca_rcache_base_registration_t *reg = NULL;
     int rc = OPAL_SUCCESS;
 
     if (flist->fl_max_to_alloc && (flist->fl_num_allocated + num_elements) >
@@ -187,30 +196,29 @@ int opal_free_list_grow_st (opal_free_list_t* flist, size_t num_elements)
         assert (elem_size > 0);
 
         /* allocate the rest from the mpool (or use memalign/malloc) */
-        if(flist->fl_mpool != NULL) {
-            payload_ptr = mpool_alloc_ptr =
-                (unsigned char *) flist->fl_mpool->mpool_alloc(flist->fl_mpool,
-                                                               num_elements * elem_size,
-                                                               flist->fl_payload_buffer_alignment,
-                                                               flist->fl_mpool_reg_flags, &reg);
-        } else {
-#ifdef HAVE_POSIX_MEMALIGN
-            posix_memalign ((void **) &mpool_alloc_ptr, flist->fl_payload_buffer_alignment,
-                            num_elements * elem_size);
-            payload_ptr = mpool_alloc_ptr;
-#else
-            mpool_alloc_ptr = (unsigned char *) malloc (num_elements * elem_size +
-                                      flist->fl_payload_buffer_alignment);
-            payload_ptr = (unsigned char *) OPAL_ALIGN((uintptr_t)mpool_alloc_ptr,
-                                            flist->fl_payload_buffer_alignment,
-                                            uintptr_t);
-#endif
-        }
-
-        if(NULL == mpool_alloc_ptr) {
+        payload_ptr =
+            (unsigned char *) flist->fl_mpool->mpool_alloc(flist->fl_mpool,
+                                                           num_elements * elem_size,
+                                                           flist->fl_payload_buffer_alignment, 0);
+        if (NULL == payload_ptr) {
             free(alloc_ptr);
             return OPAL_ERR_TEMP_OUT_OF_RESOURCE;
         }
+
+        if (flist->fl_rcache) {
+            rc = flist->fl_rcache->rcache_register (flist->fl_rcache, payload_ptr, num_elements * elem_size,
+                                                    flist->fl_rcache_reg_flags, MCA_RCACHE_ACCESS_ANY, &reg);
+            if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
+                free (alloc_ptr);
+                if (flist->fl_mpool) {
+                    flist->fl_mpool->mpool_free (flist->fl_mpool, payload_ptr);
+                } else {
+                    free (payload_ptr);
+                }
+
+                return rc;
+            }
+        }
     }
 
     /* make the alloc_ptr a list item, save the chunk in the allocations list,
@@ -219,7 +227,7 @@ int opal_free_list_grow_st (opal_free_list_t* flist, size_t num_elements)
     opal_list_append(&(flist->fl_allocations), (opal_list_item_t*)alloc_ptr);
 
     alloc_ptr->registration = reg;
-    alloc_ptr->ptr = mpool_alloc_ptr;
+    alloc_ptr->ptr = payload_ptr;
 
     ptr = (unsigned char*)alloc_ptr + sizeof(opal_free_list_memory_t);
     ptr = OPAL_ALIGN_PTR(ptr, flist->fl_frag_alignment, unsigned char*);