2013-11-18 04:58:37 +00:00
|
|
|
/* -*- 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
|
|
|
|
* Corporation. All rights reserved.
|
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 00:47:28 +00:00
|
|
|
* Copyright (c) 2004-2013 The University of Tennessee and The University
|
2013-11-18 04:58:37 +00:00
|
|
|
* of Tennessee Research Foundation. All rights
|
|
|
|
* reserved.
|
|
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
|
|
|
* University of Stuttgart. All rights reserved.
|
|
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
* All rights reserved.
|
|
|
|
* Copyright (c) 2006 Voltaire. All rights reserved.
|
2015-02-19 13:41:41 -07:00
|
|
|
* Copyright (c) 2011-2015 Los Alamos National Security, LLC. All rights
|
2013-11-18 04:58:37 +00:00
|
|
|
* reserved.
|
|
|
|
*
|
|
|
|
* $COPYRIGHT$
|
|
|
|
*
|
|
|
|
* Additional copyrights may follow
|
|
|
|
*
|
|
|
|
* $HEADER$
|
|
|
|
*/
|
|
|
|
/**
|
|
|
|
* @file
|
|
|
|
*/
|
|
|
|
#ifndef MCA_MPOOL_UDREG_H
|
|
|
|
#define MCA_MPOOL_UDREG_H
|
|
|
|
|
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 00:47:28 +00:00
|
|
|
#include "opal_config.h"
|
2013-11-18 04:58:37 +00:00
|
|
|
#include "opal/class/opal_list.h"
|
2015-02-19 13:41:41 -07:00
|
|
|
#include "opal/class/opal_free_list.h"
|
2013-11-18 04:58:37 +00:00
|
|
|
#include "opal/mca/event/event.h"
|
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 00:47:28 +00:00
|
|
|
#include "opal/mca/mpool/mpool.h"
|
2014-07-31 15:29:20 +00:00
|
|
|
#include "opal/util/proc.h"
|
2013-11-18 04:58:37 +00:00
|
|
|
#if HAVE_SYS_MMAN_H
|
|
|
|
#include <sys/mman.h>
|
|
|
|
#endif
|
|
|
|
|
|
|
|
BEGIN_C_DECLS
|
|
|
|
|
|
|
|
struct mca_mpool_udreg_component_t {
|
|
|
|
mca_mpool_base_component_t super;
|
|
|
|
bool print_stats;
|
|
|
|
int leave_pinned;
|
|
|
|
opal_list_t huge_pages;
|
|
|
|
bool use_huge_pages;
|
|
|
|
};
|
|
|
|
typedef struct mca_mpool_udreg_component_t mca_mpool_udreg_component_t;
|
|
|
|
|
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 00:47:28 +00:00
|
|
|
OPAL_DECLSPEC extern mca_mpool_udreg_component_t mca_mpool_udreg_component;
|
2013-11-18 04:58:37 +00:00
|
|
|
|
|
|
|
struct mca_mpool_udreg_module_t;
|
|
|
|
|
|
|
|
struct mca_mpool_base_resources_t {
|
|
|
|
/* the start of this mpool should match grdma */
|
|
|
|
char *pool_name;
|
|
|
|
void *reg_data;
|
|
|
|
size_t sizeof_reg;
|
|
|
|
int (*register_mem)(void *reg_data, void *base, size_t size,
|
|
|
|
mca_mpool_base_registration_t *reg);
|
|
|
|
int (*deregister_mem)(void *reg_data, mca_mpool_base_registration_t *reg);
|
|
|
|
|
|
|
|
/* udreg specific resources */
|
|
|
|
bool use_kernel_cache;
|
|
|
|
bool use_evict_w_unreg;
|
|
|
|
int max_entries;
|
|
|
|
size_t page_size;
|
|
|
|
};
|
|
|
|
typedef struct mca_mpool_base_resources_t mca_mpool_base_resources_t;
|
|
|
|
|
|
|
|
struct mca_mpool_udreg_hugepage_t {
|
|
|
|
opal_list_item_t super;
|
|
|
|
unsigned long page_size;
|
|
|
|
char *path;
|
|
|
|
opal_list_t allocations;
|
|
|
|
int cnt;
|
|
|
|
};
|
|
|
|
typedef struct mca_mpool_udreg_hugepage_t mca_mpool_udreg_hugepage_t;
|
|
|
|
|
|
|
|
OBJ_CLASS_DECLARATION(mca_mpool_udreg_hugepage_t);
|
|
|
|
|
|
|
|
struct mca_mpool_udreg_hugepage_alloc_t {
|
|
|
|
opal_list_item_t super;
|
|
|
|
int fd;
|
|
|
|
char *path;
|
|
|
|
void *ptr;
|
|
|
|
size_t size;
|
|
|
|
mca_mpool_udreg_hugepage_t *huge_table;
|
|
|
|
};
|
|
|
|
typedef struct mca_mpool_udreg_hugepage_alloc_t mca_mpool_udreg_hugepage_alloc_t;
|
|
|
|
|
|
|
|
OBJ_CLASS_DECLARATION(mca_mpool_udreg_hugepage_pool_item_t);
|
|
|
|
|
|
|
|
struct mca_mpool_udreg_module_t {
|
|
|
|
mca_mpool_base_module_t super;
|
|
|
|
struct mca_mpool_base_resources_t resources;
|
2015-02-19 13:41:41 -07:00
|
|
|
opal_free_list_t reg_list;
|
2013-11-18 04:58:37 +00:00
|
|
|
mca_mpool_udreg_hugepage_t *huge_page;
|
2014-10-08 10:10:19 -06:00
|
|
|
opal_mutex_t lock;
|
2013-11-18 04:58:37 +00:00
|
|
|
void *udreg_handle;
|
2015-09-29 15:28:00 -06:00
|
|
|
/** used to communicate the access flags to the underlying registration
|
|
|
|
* function */
|
|
|
|
int requested_access_flags;
|
2013-11-18 04:58:37 +00:00
|
|
|
};
|
|
|
|
typedef struct mca_mpool_udreg_module_t mca_mpool_udreg_module_t;
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Initializes the mpool module.
|
|
|
|
*/
|
|
|
|
int mca_mpool_udreg_module_init(mca_mpool_udreg_module_t *mpool);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Returns base address of shared memory mapping.
|
|
|
|
*/
|
|
|
|
void *mca_mpool_udreg_base(mca_mpool_base_module_t *mpool);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Allocate block of registered memory.
|
|
|
|
*/
|
|
|
|
void* mca_mpool_udreg_alloc(mca_mpool_base_module_t *mpool, size_t size,
|
|
|
|
size_t align, uint32_t flags,
|
|
|
|
mca_mpool_base_registration_t** registration);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* realloc block of registered memory
|
|
|
|
*/
|
|
|
|
void* mca_mpool_udreg_realloc( mca_mpool_base_module_t *mpool, void* addr,
|
|
|
|
size_t size, mca_mpool_base_registration_t** registration);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* register block of memory
|
|
|
|
*/
|
|
|
|
int mca_mpool_udreg_register(mca_mpool_base_module_t* mpool, void *addr,
|
2015-09-29 15:28:00 -06:00
|
|
|
size_t size, uint32_t flags, int32_t access_flags, mca_mpool_base_registration_t **reg);
|
2013-11-18 04:58:37 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* deregister memory
|
|
|
|
*/
|
|
|
|
int mca_mpool_udreg_deregister(mca_mpool_base_module_t *mpool,
|
|
|
|
mca_mpool_base_registration_t *reg);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* free memory allocated by alloc function
|
|
|
|
*/
|
|
|
|
void mca_mpool_udreg_free(mca_mpool_base_module_t *mpool, void * addr,
|
|
|
|
mca_mpool_base_registration_t *reg);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* find registration for a given block of memory
|
|
|
|
*/
|
|
|
|
int mca_mpool_udreg_find(struct mca_mpool_base_module_t* mpool, void* addr,
|
|
|
|
size_t size, mca_mpool_base_registration_t **reg);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* finalize mpool
|
|
|
|
*/
|
|
|
|
void mca_mpool_udreg_finalize(struct mca_mpool_base_module_t *mpool);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Fault Tolerance Event Notification Function
|
|
|
|
* @param state Checkpoint Stae
|
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 00:47:28 +00:00
|
|
|
* @return OPAL_SUCCESS or failure status
|
2013-11-18 04:58:37 +00:00
|
|
|
*/
|
|
|
|
int mca_mpool_udreg_ft_event(int state);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* evict one unused registration from the mpool's lru.
|
|
|
|
* @return true on success, false on failure
|
|
|
|
*/
|
|
|
|
bool mca_mpool_udreg_evict (struct mca_mpool_base_module_t *mpool);
|
|
|
|
|
|
|
|
END_C_DECLS
|
|
|
|
#endif
|