2014-02-03 21:01:46 +04:00
|
|
|
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
2012-08-16 23:11:35 +04:00
|
|
|
/*
|
|
|
|
*
|
|
|
|
* Copyright (c) 2009-2012 Oak Ridge National Laboratory. All rights reserved.
|
|
|
|
* Copyright (c) 2009-2012 Mellanox Technologies. All rights reserved.
|
2014-02-03 21:01:46 +04:00
|
|
|
* Copyright (c) 2012-2014 Los Alamos National Security, LLC. All rights
|
|
|
|
* reserved.
|
2014-01-09 07:57:55 +04:00
|
|
|
* Copyright (c) 2014 Intel, Inc. All rights reserved.
|
2014-03-28 03:25:31 +04:00
|
|
|
* Copyright (c) 2014 Research Organization for Information Science
|
|
|
|
* and Technology (RIST). All rights reserved.
|
2012-08-16 23:11:35 +04:00
|
|
|
* $COPYRIGHT$
|
|
|
|
*
|
|
|
|
* Additional copyrights may follow
|
|
|
|
*
|
|
|
|
* $HEADER$
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "ompi_config.h"
|
|
|
|
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <fcntl.h>
|
|
|
|
#include <errno.h>
|
2014-01-09 07:57:55 +04:00
|
|
|
#ifdef HAVE_STRINGS_H
|
|
|
|
#include <strings.h>
|
|
|
|
#endif
|
2012-08-16 23:11:35 +04:00
|
|
|
|
|
|
|
#include "ompi/proc/proc.h"
|
2013-02-06 01:52:55 +04:00
|
|
|
#include "ompi/patterns/comm/coll_ops.h"
|
2014-05-19 19:21:58 +04:00
|
|
|
#include "opal/align.h"
|
2012-08-16 23:11:35 +04:00
|
|
|
|
|
|
|
#include "opal/dss/dss.h"
|
|
|
|
#include "opal/util/error.h"
|
|
|
|
#include "opal/util/output.h"
|
|
|
|
#include "opal/class/opal_list.h"
|
|
|
|
#include "opal/class/opal_hash_table.h"
|
|
|
|
|
|
|
|
#include "bcol_basesmuma.h"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
#define SM_BACKING_FILE_NAME_MAX_LEN 256
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
static bcol_basesmuma_smcm_mmap_t * bcol_basesmuma_smcm_reg_mmap(void *in_ptr, int fd, size_t length,
|
|
|
|
size_t addr_offset, size_t alignment,
|
|
|
|
char *file_name);
|
|
|
|
|
2012-08-16 23:11:35 +04:00
|
|
|
struct file_info_t {
|
|
|
|
uint32_t vpid;
|
|
|
|
uint32_t jobid;
|
|
|
|
uint64_t file_size;
|
|
|
|
uint64_t size_ctl_structure;
|
|
|
|
uint64_t data_seg_alignment;
|
|
|
|
char file_name[SM_BACKING_FILE_NAME_MAX_LEN];
|
|
|
|
};
|
|
|
|
|
|
|
|
/* need to allocate space for the peer */
|
2014-05-13 21:00:31 +04:00
|
|
|
static void bcol_basesmuma_smcm_proc_item_t_construct (bcol_basesmuma_smcm_proc_item_t * item)
|
|
|
|
{
|
|
|
|
memset ((char *) item + sizeof (item->item), 0, sizeof (*item) - sizeof (item->item));
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/* need to free the space for the peer */
|
2014-05-13 21:00:31 +04:00
|
|
|
static void bcol_basesmuma_smcm_proc_item_t_destruct (bcol_basesmuma_smcm_proc_item_t * item)
|
|
|
|
{
|
|
|
|
if (item->sm_mmap) {
|
2014-05-27 22:40:41 +04:00
|
|
|
OBJ_RELEASE(item->sm_mmap);
|
2014-05-13 21:00:31 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
if (item->sm_file.file_name) {
|
|
|
|
free (item->sm_file.file_name);
|
|
|
|
item->sm_file.file_name = NULL;
|
|
|
|
}
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
OBJ_CLASS_INSTANCE(bcol_basesmuma_smcm_proc_item_t,
|
2014-02-03 21:01:46 +04:00
|
|
|
opal_list_item_t,
|
|
|
|
bcol_basesmuma_smcm_proc_item_t_construct,
|
|
|
|
bcol_basesmuma_smcm_proc_item_t_destruct);
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
static void bcol_basesmuma_smcm_mmap_construct (bcol_basesmuma_smcm_mmap_t *smcm_mmap)
|
2012-08-16 23:11:35 +04:00
|
|
|
{
|
2014-05-27 22:40:41 +04:00
|
|
|
memset ((char *) smcm_mmap + sizeof (smcm_mmap->super), 0, sizeof (*smcm_mmap) - sizeof (smcm_mmap->super));
|
|
|
|
}
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
static void bcol_basesmuma_smcm_mmap_destruct (bcol_basesmuma_smcm_mmap_t *smcm_mmap)
|
|
|
|
{
|
|
|
|
if (smcm_mmap->map_seg) {
|
2014-08-11 11:36:46 +04:00
|
|
|
munmap ((void *)smcm_mmap->map_seg, smcm_mmap->map_size);
|
2014-05-27 22:40:41 +04:00
|
|
|
smcm_mmap->map_seg = NULL;
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
if (smcm_mmap->map_path) {
|
|
|
|
free (smcm_mmap->map_path);
|
|
|
|
smcm_mmap->map_path = NULL;
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
OBJ_CLASS_INSTANCE(bcol_basesmuma_smcm_mmap_t, opal_list_item_t,
|
|
|
|
bcol_basesmuma_smcm_mmap_construct,
|
|
|
|
bcol_basesmuma_smcm_mmap_destruct);
|
2012-08-16 23:11:35 +04:00
|
|
|
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* smcm_allgather_connection:
|
|
|
|
This function is called when a shared memory subgroup wants to establish shared memory "connections" among
|
|
|
|
a group of processes.
|
|
|
|
|
2012-08-16 23:11:35 +04:00
|
|
|
This function DOES NOT create any shared memory backing files, it only mmaps already existing files. Shared
|
2014-02-03 21:01:46 +04:00
|
|
|
memory files are created by the shared memory registration function
|
2012-08-16 23:11:35 +04:00
|
|
|
-----------------------------------------------------------------------------------------------------------
|
|
|
|
Input params:
|
|
|
|
|
|
|
|
- sbgp module The subgrouping module contains the list of ranks to wire up.
|
|
|
|
|
|
|
|
- peer_list An opal list containing a list of bcol_basesmuma_smcm_proc_item_t types. This
|
2014-02-03 21:01:46 +04:00
|
|
|
contains a list of peers whose shared memory files I have already mapped.
|
2012-08-16 23:11:35 +04:00
|
|
|
Upon completion of the allgather exchange with all members of the group and depending on the
|
|
|
|
value of "map_all", my peers' shared memory files are mapped into my local virtual memory
|
2014-02-03 21:01:46 +04:00
|
|
|
space, with all pertinent information being stored in an bcol_basesmuma_smcm_proc_item_t which is
|
|
|
|
subsequently appended onto the "peer_list".
|
|
|
|
|
|
|
|
- comm The ompi_communicator_t communicator.
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
- input A data struct that caches the information about my shared memory file.
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
- map_all Bool that determines whether or not to go ahead and map the files from all of the peers
|
2012-08-16 23:11:35 +04:00
|
|
|
defined in the sbgp-ing module. If map_all == true, then go ahead and mmap all of the files
|
|
|
|
obtained in the exchange and append the information to the "peer_list". If map_all == false
|
2014-02-03 21:01:46 +04:00
|
|
|
then make a check and only mmap those peers' files whose vpid/jobid/filename combination do
|
|
|
|
not already exist in the "peer_list". Once mapping is completed, append this peer's information
|
2012-08-16 23:11:35 +04:00
|
|
|
to the "peer_list".
|
|
|
|
-----------------------------------------------------------------------------------------------------------
|
2014-02-03 21:01:46 +04:00
|
|
|
*
|
|
|
|
*/
|
2012-08-16 23:11:35 +04:00
|
|
|
|
|
|
|
|
|
|
|
int bcol_basesmuma_smcm_allgather_connection(
|
2014-02-03 21:01:46 +04:00
|
|
|
mca_bcol_basesmuma_module_t *sm_bcol_module,
|
|
|
|
mca_sbgp_base_module_t *module,
|
|
|
|
opal_list_t *peer_list,
|
|
|
|
bcol_basesmuma_smcm_proc_item_t ***back_files,
|
|
|
|
ompi_communicator_t *comm,
|
|
|
|
bcol_basesmuma_smcm_file_t input,
|
|
|
|
char *base_fname,
|
|
|
|
bool map_all)
|
2012-08-16 23:11:35 +04:00
|
|
|
{
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* define local variables */
|
|
|
|
|
2014-02-08 01:39:24 +04:00
|
|
|
int rc, i, fd;
|
2014-02-03 21:01:46 +04:00
|
|
|
ptrdiff_t mem_offset;
|
|
|
|
ompi_proc_t *proc_temp, *my_id;
|
|
|
|
bcol_basesmuma_smcm_proc_item_t *temp;
|
|
|
|
bcol_basesmuma_smcm_proc_item_t *item_ptr;
|
|
|
|
bcol_basesmuma_smcm_proc_item_t **backing_files;
|
|
|
|
struct file_info_t local_file;
|
|
|
|
struct file_info_t *all_files=NULL;
|
2014-05-27 22:40:41 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* sanity check */
|
|
|
|
if (strlen(input.file_name) > SM_BACKING_FILE_NAME_MAX_LEN-1) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "backing file name too long: %s len :: %d",
|
2014-02-03 21:01:46 +04:00
|
|
|
input.file_name, (int) strlen(input.file_name));
|
|
|
|
return OMPI_ERR_BAD_PARAM;
|
2014-01-22 19:39:19 +04:00
|
|
|
}
|
2014-02-03 21:01:46 +04:00
|
|
|
|
|
|
|
backing_files = (bcol_basesmuma_smcm_proc_item_t **)
|
|
|
|
calloc(module->group_size, sizeof(bcol_basesmuma_smcm_proc_item_t *));
|
|
|
|
if (!backing_files) {
|
|
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
|
|
}
|
|
|
|
|
2014-03-28 03:25:31 +04:00
|
|
|
/* FIXME *back_files might have been already allocated
|
|
|
|
* so free it in order to avoid a memory leak */
|
|
|
|
if (NULL != *back_files) {
|
|
|
|
free (*back_files);
|
|
|
|
}
|
2014-02-03 21:01:46 +04:00
|
|
|
*back_files = backing_files;
|
|
|
|
|
|
|
|
my_id = ompi_proc_local();
|
|
|
|
|
|
|
|
/* Phase One:
|
|
|
|
gather a list of processes that will participate in the allgather - I'm
|
|
|
|
preparing this list from the sbgp-ing module that was passed into the function */
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* fill in local file information */
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
|
|
|
local_file.vpid = ((orte_process_name_t*)&my_id->super.proc_name)->vpid;
|
|
|
|
local_file.jobid = ((orte_process_name_t*)&my_id->super.proc_name)->jobid;
|
2014-02-03 21:01:46 +04:00
|
|
|
local_file.file_size=input.size;
|
|
|
|
local_file.size_ctl_structure=input.size_ctl_structure;
|
|
|
|
local_file.data_seg_alignment=input.data_seg_alignment;
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
strcpy (local_file.file_name, input.file_name);
|
|
|
|
|
|
|
|
/* will exchange this data type as a string of characters -
|
|
|
|
* this routine is first called before MPI_init() completes
|
|
|
|
* and before error handling is setup, so can't use the
|
|
|
|
* MPI data types to send this data */
|
|
|
|
all_files = (struct file_info_t *) calloc(module->group_size,
|
|
|
|
sizeof (struct file_info_t));
|
|
|
|
if (!all_files) {
|
|
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* exchange data */
|
|
|
|
rc = comm_allgather_pml(&local_file,all_files,sizeof(struct file_info_t), MPI_CHAR,
|
|
|
|
sm_bcol_module->super.sbgp_partner_module->my_index,
|
|
|
|
sm_bcol_module->super.sbgp_partner_module->group_size,
|
|
|
|
sm_bcol_module->super.sbgp_partner_module->group_list,
|
|
|
|
sm_bcol_module->super.sbgp_partner_module->group_comm);
|
|
|
|
if( OMPI_SUCCESS != rc ) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "failed in comm_allgather_pml. Error code: %d", rc);
|
2014-02-03 21:01:46 +04:00
|
|
|
goto Error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Phase four:
|
|
|
|
loop through the receive buffer, unpack the data recieved from remote peers */
|
|
|
|
|
|
|
|
for (i = 0; i < module->group_size; i++) {
|
|
|
|
struct file_info_t *rem_file = all_files + i;
|
|
|
|
|
|
|
|
/* check if this is my index or if the file is already mapped (set above). ther
|
|
|
|
* is no reason to look through the peer list again because no two members of
|
|
|
|
* the group will have the same vpid/jobid pair. ignore this previously found
|
|
|
|
* mapping if map_all was requested (NTH: not sure why exactly since we re-map
|
|
|
|
* and already mapped file) */
|
2014-02-08 01:39:24 +04:00
|
|
|
if (sm_bcol_module->super.sbgp_partner_module->my_index == i) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
proc_temp = ompi_comm_peer_lookup(comm,module->group_list[i]);
|
|
|
|
|
|
|
|
OPAL_LIST_FOREACH(item_ptr, peer_list, bcol_basesmuma_smcm_proc_item_t) {
|
|
|
|
/* if the vpid/jobid/filename combination already exists in the list,
|
|
|
|
then do not map this peer's file --- because you already have */
|
2014-07-30 09:38:53 +04:00
|
|
|
if (0 == ompi_rte_compare_name_fields(OMPI_RTE_CMP_ALL,
|
2014-08-01 18:44:11 +04:00
|
|
|
OMPI_CAST_RTE_NAME(&proc_temp->super.proc_name),
|
2014-07-30 09:38:53 +04:00
|
|
|
&item_ptr->peer) &&
|
2014-02-08 01:39:24 +04:00
|
|
|
0 == strcmp (item_ptr->sm_file.file_name, rem_file->file_name)) {
|
2014-05-13 21:00:31 +04:00
|
|
|
++item_ptr->refcnt;
|
2014-02-08 01:39:24 +04:00
|
|
|
/* record file data */
|
|
|
|
backing_files[i] = item_ptr;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!map_all && backing_files[i]) {
|
2014-02-03 21:01:46 +04:00
|
|
|
continue;
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
2014-02-03 21:01:46 +04:00
|
|
|
|
|
|
|
temp = OBJ_NEW(bcol_basesmuma_smcm_proc_item_t);
|
|
|
|
if (!temp) {
|
|
|
|
rc = OMPI_ERR_OUT_OF_RESOURCE;
|
2012-08-16 23:11:35 +04:00
|
|
|
goto Error;
|
|
|
|
}
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
temp->peer.vpid = rem_file->vpid;
|
|
|
|
temp->peer.jobid = rem_file->jobid;
|
|
|
|
|
|
|
|
temp->sm_file.file_name = strdup (rem_file->file_name);
|
|
|
|
if (!temp->sm_file.file_name) {
|
|
|
|
rc = OMPI_ERR_OUT_OF_RESOURCE;
|
|
|
|
OBJ_RELEASE(temp);
|
2012-08-16 23:11:35 +04:00
|
|
|
goto Error;
|
|
|
|
}
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
temp->sm_file.size = (size_t) rem_file->file_size;
|
|
|
|
temp->sm_file.mpool_size = (size_t) rem_file->file_size;
|
|
|
|
temp->sm_file.size_ctl_structure = (size_t) rem_file->size_ctl_structure;
|
|
|
|
temp->sm_file.data_seg_alignment = (size_t) rem_file->data_seg_alignment;
|
2014-05-13 21:00:31 +04:00
|
|
|
temp->refcnt = 1;
|
2014-02-03 21:01:46 +04:00
|
|
|
|
|
|
|
/* Phase Five:
|
|
|
|
If map_all == true, then we map every peer's file
|
|
|
|
else we check to see if I have already mapped this
|
|
|
|
vpid/jobid/filename combination and if I have, then
|
|
|
|
I do not mmap this peer's file.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
fd = open(temp->sm_file.file_name, O_RDWR, 0600);
|
|
|
|
if (0 > fd) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "SMCM Allgather failed to open sm backing file %s. errno = %d",
|
|
|
|
temp->sm_file.file_name, errno);
|
2014-02-03 21:01:46 +04:00
|
|
|
rc = OMPI_ERROR;
|
|
|
|
goto Error;
|
|
|
|
}
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* map the file */
|
2014-05-27 22:40:41 +04:00
|
|
|
temp->sm_mmap = bcol_basesmuma_smcm_reg_mmap (NULL, fd, temp->sm_file.size,
|
|
|
|
temp->sm_file.size_ctl_structure,
|
|
|
|
temp->sm_file.data_seg_alignment,
|
|
|
|
temp->sm_file.file_name);
|
|
|
|
close (fd);
|
2014-02-03 21:01:46 +04:00
|
|
|
if (NULL == temp->sm_mmap) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "mmapping failed to map remote peer's file");
|
2014-02-03 21:01:46 +04:00
|
|
|
OBJ_RELEASE(temp);
|
|
|
|
rc = OMPI_ERROR;
|
|
|
|
goto Error;
|
|
|
|
}
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* compute memory offset */
|
|
|
|
mem_offset = (ptrdiff_t) temp->sm_mmap->data_addr -
|
|
|
|
(ptrdiff_t) temp->sm_mmap->map_seg;
|
|
|
|
temp->sm_mmap->map_seg->seg_offset = mem_offset;
|
|
|
|
temp->sm_mmap->map_seg->seg_size = temp->sm_file.size - mem_offset;
|
|
|
|
/* more stuff to follow */
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* append this peer's info, including shared memory map addr, onto the
|
|
|
|
peer_list */
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* record file data */
|
|
|
|
backing_files[i] = (bcol_basesmuma_smcm_proc_item_t *) temp;
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
opal_list_append(peer_list, (opal_list_item_t*) temp);
|
|
|
|
}
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
rc = OMPI_SUCCESS;
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
Error:
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* error clean-up and return */
|
|
|
|
if (NULL != all_files) {
|
|
|
|
free(all_files);
|
|
|
|
}
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
return rc;
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
|
|
|
|
2014-05-13 21:00:31 +04:00
|
|
|
int bcol_basesmuma_smcm_release_connections (mca_bcol_basesmuma_module_t *sm_bcol_module,
|
|
|
|
mca_sbgp_base_module_t *sbgp_module, opal_list_t *peer_list,
|
|
|
|
bcol_basesmuma_smcm_proc_item_t ***back_files)
|
|
|
|
{
|
|
|
|
bcol_basesmuma_smcm_proc_item_t **smcm_procs = *back_files;
|
|
|
|
|
|
|
|
for (int i = 0 ; i < sbgp_module->group_size ; ++i) {
|
|
|
|
if (smcm_procs[i] && 0 == --smcm_procs[i]->refcnt) {
|
|
|
|
opal_list_remove_item (peer_list, (opal_list_item_t *) smcm_procs[i]);
|
|
|
|
OBJ_RELEASE(smcm_procs[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
free (smcm_procs);
|
|
|
|
*back_files = NULL;
|
|
|
|
|
|
|
|
return OMPI_SUCCESS;
|
|
|
|
}
|
|
|
|
|
2012-08-16 23:11:35 +04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* mmap the specified file as a shared file. No information exchange with other
|
|
|
|
* processes takes place within this routine.
|
|
|
|
* This function assumes that the memory has already been allocated, and only the
|
|
|
|
* mmap needs to be done.
|
|
|
|
*/
|
|
|
|
bcol_basesmuma_smcm_mmap_t *bcol_basesmuma_smcm_mem_reg(void *in_ptr,
|
2014-02-03 21:01:46 +04:00
|
|
|
size_t length,
|
|
|
|
size_t alignment,
|
|
|
|
char* file_name)
|
2012-08-16 23:11:35 +04:00
|
|
|
{
|
2014-02-03 21:01:46 +04:00
|
|
|
/* local variables */
|
|
|
|
int fd = -1;
|
|
|
|
bcol_basesmuma_smcm_mmap_t *map = NULL;
|
2014-05-19 19:21:58 +04:00
|
|
|
int rc;
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* if pointer is not allocated - return error. We have no clue how the user will allocate or
|
|
|
|
* free this memory.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* open the shared memory backing file */
|
|
|
|
|
|
|
|
fd = open(file_name, O_CREAT|O_RDWR,0600);
|
|
|
|
if (fd < 0) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "basesmuma shared memory allocation open failed with errno: %d",
|
2014-02-03 21:01:46 +04:00
|
|
|
errno);
|
|
|
|
} else if (0 != ftruncate(fd,length)) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "basesmuma shared memory allocation ftruncate failed with errno: %d",
|
2014-02-03 21:01:46 +04:00
|
|
|
errno);
|
|
|
|
} else {
|
2014-05-19 19:21:58 +04:00
|
|
|
/* ensure there is enough space for the backing store */
|
|
|
|
rc = ftruncate (fd, length);
|
|
|
|
if (0 > rc) {
|
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "failed to truncate the file to be mapped. errno: %d", errno);
|
|
|
|
return NULL;
|
|
|
|
}
|
2014-02-03 21:01:46 +04:00
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
map = bcol_basesmuma_smcm_reg_mmap(in_ptr, fd, length, 0, alignment, file_name);
|
2014-02-03 21:01:46 +04:00
|
|
|
if (NULL == map) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
}
|
2014-05-01 02:28:08 +04:00
|
|
|
/* no longer need this file descriptor. close it */
|
|
|
|
close (fd);
|
2014-02-03 21:01:46 +04:00
|
|
|
|
|
|
|
/* takes us to the top of the control structure */
|
|
|
|
|
|
|
|
return map;
|
|
|
|
|
2012-08-16 23:11:35 +04:00
|
|
|
}
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
static bcol_basesmuma_smcm_mmap_t * bcol_basesmuma_smcm_reg_mmap(void *in_ptr, int fd, size_t length,
|
|
|
|
size_t addr_offset, size_t alignment,
|
|
|
|
char *file_name)
|
2012-08-16 23:11:35 +04:00
|
|
|
{
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* local variables */
|
|
|
|
bcol_basesmuma_smcm_mmap_t *map;
|
|
|
|
bcol_basesmuma_smcm_file_header_t *seg;
|
|
|
|
unsigned char* myaddr = NULL;
|
2014-05-26 11:18:31 +04:00
|
|
|
int flags = MAP_SHARED;
|
2012-08-16 23:11:35 +04:00
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
/* set up the map object */
|
|
|
|
map = OBJ_NEW(bcol_basesmuma_smcm_mmap_t);
|
|
|
|
if (OPAL_UNLIKELY(NULL == map)) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2014-02-03 21:01:46 +04:00
|
|
|
/* map the file and initialize the segment state */
|
2014-05-26 11:18:31 +04:00
|
|
|
if (NULL != in_ptr) {
|
|
|
|
flags |= MAP_FIXED;
|
|
|
|
}
|
2014-02-03 21:01:46 +04:00
|
|
|
seg = (bcol_basesmuma_smcm_file_header_t *)
|
2014-05-26 11:18:31 +04:00
|
|
|
mmap(in_ptr, length, PROT_READ|PROT_WRITE, flags, fd, 0);
|
2014-02-03 21:01:46 +04:00
|
|
|
if((void*)-1 == seg) {
|
2014-05-27 22:40:41 +04:00
|
|
|
OBJ_RELEASE(map);
|
2014-02-03 21:01:46 +04:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
map->map_path = strdup (file_name);
|
2014-02-03 21:01:46 +04:00
|
|
|
|
|
|
|
/* the first entry in the file is the control structure. the first entry
|
|
|
|
in the control structure is an mca_common_sm_file_header_t element */
|
|
|
|
map->map_seg = seg;
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
myaddr = (unsigned char *) seg + addr_offset;
|
2014-02-03 21:01:46 +04:00
|
|
|
/* if we have a data segment (i.e. if 0 != data_seg_alignement) */
|
|
|
|
|
2014-05-27 22:40:41 +04:00
|
|
|
if (alignment) {
|
2014-02-03 21:01:46 +04:00
|
|
|
myaddr = OPAL_ALIGN_PTR(myaddr, alignment, unsigned char*);
|
|
|
|
|
|
|
|
/* is addr past the end of the file? */
|
|
|
|
if ((unsigned char *) seg+length < myaddr) {
|
2014-05-13 21:00:31 +04:00
|
|
|
opal_output (ompi_bcol_base_framework.framework_output, "mca_bcol_basesmuma_sm_alloc_mmap: memory region too small len %lu add %p",
|
2014-02-03 21:01:46 +04:00
|
|
|
(unsigned long) length, myaddr);
|
2014-05-27 22:40:41 +04:00
|
|
|
OBJ_RELEASE(map);
|
2014-08-11 11:36:46 +04:00
|
|
|
munmap ((void *)seg, length);
|
2014-02-03 21:01:46 +04:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
map->data_addr = (unsigned char*) myaddr;
|
|
|
|
map->map_addr = (unsigned char*) seg;
|
|
|
|
map->map_size = length;
|
|
|
|
|
|
|
|
return map;
|
|
|
|
}
|