2015-03-06 06:50:44 +03:00
|
|
|
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
2005-07-02 18:08:19 +04:00
|
|
|
/*
|
2008-05-06 22:08:45 +04:00
|
|
|
* Copyright (c) 2004-2008 The Trustees of Indiana University and Indiana
|
2005-11-05 22:57:48 +03:00
|
|
|
* University Research and Technology
|
|
|
|
* Corporation. All rights reserved.
|
|
|
|
* Copyright (c) 2004-2005 The University of Tennessee and The University
|
|
|
|
* of Tennessee Research Foundation. All rights
|
|
|
|
* reserved.
|
2015-06-24 06:59:57 +03:00
|
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
2005-07-02 18:08:19 +04:00
|
|
|
* University of Stuttgart. All rights reserved.
|
|
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
|
|
* All rights reserved.
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* Copyright (c) 2008-2012 Cisco Systems, Inc. All rights reserved.
|
2015-03-06 06:50:44 +03:00
|
|
|
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
|
|
|
|
* reserved.
|
2005-07-02 18:08:19 +04:00
|
|
|
* $COPYRIGHT$
|
2015-06-24 06:59:57 +03:00
|
|
|
*
|
2005-07-02 18:08:19 +04:00
|
|
|
* Additional copyrights may follow
|
2015-06-24 06:59:57 +03:00
|
|
|
*
|
2005-07-02 18:08:19 +04:00
|
|
|
* $HEADER$
|
|
|
|
*/
|
2015-06-24 06:59:57 +03:00
|
|
|
/**
|
|
|
|
* @file
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
|
|
|
* Top-level interface for \em all MCA components.
|
2008-07-29 02:40:57 +04:00
|
|
|
*
|
|
|
|
* Historical notes:
|
|
|
|
*
|
|
|
|
* Open MPI originally used a v1.0.0 of the MCA component structs, but
|
|
|
|
* did not have a version number in the struct name. If I recall
|
|
|
|
* correctly, this is because we simply didn't think through (or never
|
|
|
|
* envisioned) changing the MCA base component struct itself. Oops.
|
|
|
|
*
|
|
|
|
* We made some changes in the base struct in Open MPI v1.3, and
|
|
|
|
* decided the following at the same time:
|
|
|
|
*
|
|
|
|
* - Bump the MCA version number to 2.0.0 and add some "reserved"
|
|
|
|
* space at the end of the struct.
|
|
|
|
* - The major MCA version number is essentially tied to the space
|
|
|
|
* that the struct occupies; if we make future changes in the struct
|
|
|
|
* by just using some of the reserved space, it may be possible to
|
|
|
|
* just increment the minor version number (depending on the scope of
|
|
|
|
* the change). If we need to add more space to the struct, we'll
|
|
|
|
* increment the major version number.
|
|
|
|
* - The MCA base component struct now has a version number in it
|
2015-06-24 06:59:57 +03:00
|
|
|
* (starting with Open MPI v1.3, it is 2.0.0).
|
2008-07-29 02:40:57 +04:00
|
|
|
* - As was an unstated assumption in prior versions of Open MPI, the
|
|
|
|
* unversioned versions of struct names (both in the MCA base and in
|
|
|
|
* individual framework bases) are intended for components who want
|
|
|
|
* to be forward source-compatible. That is, the unversioned struct
|
|
|
|
* name always represents the most recent interface version. If you
|
|
|
|
* need to use an older version, you can explicitly use that older
|
|
|
|
* struct version name. Please note, however, the Open MPI
|
|
|
|
* developers may not generally provide older versions of framework
|
|
|
|
* interface structs unless they know if someone outside of the Open
|
2015-06-24 06:59:57 +03:00
|
|
|
* MPI community needs it.
|
2008-07-29 02:40:57 +04:00
|
|
|
*
|
|
|
|
* ***IF YOU NEED BACKWARDS SOURCE OR BINARY COMPATIBILITY, you must
|
|
|
|
* let us know!***
|
|
|
|
*
|
|
|
|
* - We are currently only aware of one external developer making Open
|
|
|
|
* MPI components for the v1.2 series. He already knows that there
|
|
|
|
* are major changes coming in the v1.3 series, and does not expect to
|
|
|
|
* be able to use his v1.2 DSO binaries in v1.3. As such, we are
|
|
|
|
* breaking backwards binary compatibility in v1.3: there is no
|
|
|
|
* possibility of loading an MCA v1.0 binary component in Open MPI
|
|
|
|
* v1.3 or beyond (source compatibility is much easier -- the binary
|
|
|
|
* "refuse to load MCA components <v2.0.0" policy is enforced in
|
|
|
|
* mca_base_component_find.c).
|
|
|
|
*
|
|
|
|
* ***IF YOU NEED BACKWARDS BINARY COMPATIBILITY, please let us
|
|
|
|
* know!***
|
|
|
|
*
|
|
|
|
* - Note that we decided that framework version numbers are *not*
|
|
|
|
* related to the MCA version number. It is permissible to bump the
|
|
|
|
* MCA version number and leave all the framework version numbers
|
|
|
|
* they same. Specifically: a component is uniquely identified by
|
|
|
|
* its (MCA version, framework version, component version) tuple.
|
|
|
|
* So a component that is simply compiled with two different MCA
|
|
|
|
* base versions is still considered "different" because the tuple
|
|
|
|
* first member is different.
|
|
|
|
* - Per the discussion above, we decided to have MCA v2.0 no longer
|
|
|
|
* load <v2.0.0 components, and therefore avoided the "how to upcast
|
|
|
|
* a component in memory" issue. After v2.0.0, it is slightly
|
|
|
|
* easier because the MCA component structs have "reserved" space at
|
|
|
|
* the end that may account for future version data fields.
|
2005-07-02 18:08:19 +04:00
|
|
|
*/
|
|
|
|
|
2007-03-29 05:00:33 +04:00
|
|
|
#ifndef OPAL_MCA_H
|
|
|
|
#define OPAL_MCA_H
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2006-02-12 04:33:29 +03:00
|
|
|
#include "opal_config.h"
|
2005-07-02 18:08:19 +04:00
|
|
|
|
|
|
|
|
2008-05-06 22:08:45 +04:00
|
|
|
/**
|
|
|
|
* Common type for all MCA modules.
|
|
|
|
*
|
|
|
|
* An instance of this type is always the first element in MCA
|
|
|
|
* modules, allowing the module to be associated with a
|
|
|
|
* particular version of a specific framework, and to publish its own
|
|
|
|
* name and version.
|
|
|
|
*/
|
2008-07-29 02:40:57 +04:00
|
|
|
struct mca_base_module_2_0_0_t {
|
2008-05-06 22:08:45 +04:00
|
|
|
int dummy_value;
|
|
|
|
};
|
2008-07-29 02:40:57 +04:00
|
|
|
/** Unversioned convenience typedef; use this name in
|
|
|
|
frameworks/components to stay forward source-compatible */
|
|
|
|
typedef struct mca_base_module_2_0_0_t mca_base_module_t;
|
|
|
|
/** Versioned convenience typedef */
|
|
|
|
typedef struct mca_base_module_2_0_0_t mca_base_module_2_0_0_t;
|
2008-05-06 22:08:45 +04:00
|
|
|
|
|
|
|
|
2005-07-02 18:08:19 +04:00
|
|
|
/**
|
|
|
|
* MCA component open function.
|
|
|
|
*
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* @retval OPAL_SUCCESS This component can be used in the process.
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
2010-01-12 22:29:12 +03:00
|
|
|
* @retval OPAL_ERR_NOT_AVAILABLE Silently ignore this component for
|
|
|
|
* the duration of the process (it may even be unloaded from the
|
|
|
|
* process).
|
|
|
|
*
|
|
|
|
* @retval anything_else The MCA base will print an error message
|
|
|
|
* ignore this component for the duration of the process (it may even
|
|
|
|
* be unloaded from the process).
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
|
|
|
* All MCA components can have an "open" function that is invoked once
|
2015-06-24 06:59:57 +03:00
|
|
|
* per process, when the component is located and loaded.
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
*
|
|
|
|
* This function should avoid registering MCA parameters (use the
|
|
|
|
* component "register" function for that; i.e.,
|
|
|
|
* mca_base_register_component_params_2_0_0_fn_t for that). Legacy
|
|
|
|
* components still register MCA params in their component "open"
|
|
|
|
* function, but their authors should update them to use the component
|
|
|
|
* "register" function.
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
|
|
|
* This function can also be used to allocate any resources necessary
|
|
|
|
* for the component (e.g., heap memory).
|
|
|
|
*
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* This function should return OPAL_SUCCESS if it wishes to remain
|
2005-07-02 18:08:19 +04:00
|
|
|
* loaded in the process. Any other return value will cause the MCA
|
|
|
|
* base to unload the component. Although most components do not use
|
|
|
|
* this mechanism to force themselves to be unloaded (because if they
|
|
|
|
* are immediately unloaded, ompi_info will not display them), the
|
|
|
|
* mechanism is available should the need arise.
|
|
|
|
*
|
|
|
|
* If the component a) has no MCA parameters to register, b) no
|
|
|
|
* resources to allocate, and c) can always be used in a process
|
|
|
|
* (albiet perhaps not selected), it may provide NULL for this
|
|
|
|
* function. In this cause, the MCA will act as if it called the open
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* function and it returned OPAL_SUCCESS.
|
2005-07-02 18:08:19 +04:00
|
|
|
*/
|
2008-07-29 02:40:57 +04:00
|
|
|
typedef int (*mca_base_open_component_1_0_0_fn_t)(void);
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2015-06-24 06:59:57 +03:00
|
|
|
/**
|
2008-07-29 02:40:57 +04:00
|
|
|
* MCA component close function.
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* @retval OPAL_SUCCESS The component successfully shut down.
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
|
|
|
* @retval any_other_value Some error occurred, but is likely to be
|
|
|
|
* ignored.
|
|
|
|
*
|
|
|
|
* This function is invoked on a component after all of its modules
|
|
|
|
* have been finalized (according to the rules of its framework) and
|
|
|
|
* the component will never be used in the process again; the
|
|
|
|
* component may be unloaded from the process memory after the close
|
|
|
|
* function has been invoked.
|
|
|
|
*
|
|
|
|
* This function is typically used to release any resources still in
|
|
|
|
* use by the component.
|
|
|
|
*
|
|
|
|
* If the component has no resources to free, it may provide NULL for
|
|
|
|
* this function. In this case, the MCA will act as if it called the
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* close function and it returned OPAL_SUCCESS.
|
2005-07-02 18:08:19 +04:00
|
|
|
*/
|
2008-07-29 02:40:57 +04:00
|
|
|
typedef int (*mca_base_close_component_1_0_0_fn_t)(void);
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2015-06-24 06:59:57 +03:00
|
|
|
/**
|
2008-07-29 02:40:57 +04:00
|
|
|
* MCA component query function.
|
2008-05-06 22:08:45 +04:00
|
|
|
*
|
|
|
|
* @retval OPAL_SUCCESS The component successfully queried.
|
|
|
|
*
|
|
|
|
* @retval any_other_value Some error occurred, but is likely to be
|
|
|
|
* ignored.
|
|
|
|
*
|
|
|
|
* @param module The module to be used if this component is selected.
|
|
|
|
*
|
|
|
|
* @param priority The priority of this component.
|
|
|
|
*
|
2008-07-29 02:40:57 +04:00
|
|
|
* This function is used by the mca_base_select function to find the
|
|
|
|
* highest priority component to select. Frameworks are free to
|
|
|
|
* implement their own query function, but must also implment their
|
|
|
|
* own select function as a result.
|
|
|
|
*/
|
|
|
|
typedef int (*mca_base_query_component_2_0_0_fn_t)(mca_base_module_2_0_0_t **module, int *priority);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* MCA component parameter registration function.
|
|
|
|
*
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* @retval OPAL_SUCCESS This component successfully registered its
|
2008-07-29 02:40:57 +04:00
|
|
|
* parameters and can be used in this process.
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* @retval OPAL_ERR_BAD_PARAM Indicates that the register function
|
2015-06-24 06:59:57 +03:00
|
|
|
* failed because an MCA parameter got an invalid/incorrect value.
|
2008-07-29 02:40:57 +04:00
|
|
|
*
|
|
|
|
* @retval anything_else The MCA will ignore this component for the
|
|
|
|
* duration of the process.
|
|
|
|
*
|
|
|
|
* If a component has a non-NULL parameter registration function, it
|
|
|
|
* will be invoked to register all MCA parameters associated with the
|
|
|
|
* component. This function is invoked *before* the component "open"
|
|
|
|
* function is invoked.
|
|
|
|
*
|
|
|
|
* The registration function should not allocate any resources that
|
|
|
|
* need to be freed (aside from registering MCA parameters).
|
|
|
|
* Specifically, strings that are passed to the MCA parameter
|
|
|
|
* registration functions are all internally copied; there's no need
|
|
|
|
* for the caller to keep them after registering a parameter. Hence,
|
|
|
|
* it is possible that the registration function will be the *only*
|
|
|
|
* function invoked on a component; component authors should take care
|
|
|
|
* that no resources are leaked in this case.
|
|
|
|
*
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* This function should return OPAL_SUCCESS if it wishes to remain
|
2008-07-29 02:40:57 +04:00
|
|
|
* loaded in the process. Any other return value will cause the MCA
|
|
|
|
* base to unload the component. Although most components do not use
|
|
|
|
* this mechanism to force themselves to be unloaded (because if they
|
|
|
|
* are immediately unloaded, ompi_info will not display them), the
|
|
|
|
* mechanism is available should the need arise.
|
|
|
|
*
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* Note that if the function returns OPAL_ERR_BAD_PARAM, it is
|
|
|
|
* possible (likely?) that the component didn't register all of its
|
|
|
|
* parameters. When this happens, ompi_info (and friends) will stop
|
|
|
|
* execution and print out all existing registered parameters from the
|
|
|
|
* entire framework (since ompi_info doesn't track individual
|
|
|
|
* component register failures). This allows a user to know exactly
|
|
|
|
* what value is incorrect, and from where it was set (e.g., via an
|
|
|
|
* MCA params file).
|
|
|
|
*
|
2008-07-29 02:40:57 +04:00
|
|
|
* If the component a) has no MCA parameters to register, b) no
|
|
|
|
* resources to allocate, and c) can always be used in a process
|
|
|
|
* (albiet perhaps not selected), it may provide NULL for this
|
|
|
|
* function. In this cause, the MCA will act as if it called the
|
Refs trac:3275.
We ran into a case where the OMPI SVN trunk grew a new acceptable MCA
parameter value, but this new value was not accepted on the v1.6
branch (hwloc_base_mem_bind_failure_action -- on the trunk it accepts
the value "silent", but on the older v1.6 branch, it doesn't). If you
set "hwloc_base_mem_bind_failure_action=silent" in the default MCA
params file and then accidentally ran with the v1.6 branch, every OMPI
executable (including ompi_info) just failed because hwloc_base_open()
would say "hey, 'silent' is not a valid value for
hwloc_base_mem_bind_failure_action!". Kaboom.
The only problem is that it didn't give you any indication of where
this value was being set. Quite maddening, from a user perspective.
So we changed the ompi_info handles this case. If any framework open
function return OMPI_ERR_BAD_PARAM (either because its base MCA params
got a bad value or because one of its component register/open
functions return OMPI_ERR_BAD_PARAM), ompi_info will stop, print out
a warning that it received and error, and then dump out the parameters
that it has received so far in the framework that had a problem.
At a minimum, this will show the user the MCA param that had an error
(it's usually the last one), and ''where it was set from'' (so that
they can go fix it).
We updated ompi_info to check for O???_ERR_BAD_PARAM from each from
the framework opens. Also updated the doxygen docs in mca.h for this
O???_BAD_PARAM behavior. And we noticed that mca.h had MCA_SUCCESS
and MCA_ERR_??? codes. Why? I think we used them in exactly one
place in the code base (mca_base_components_open.c). So we deleted
those and just used the normal OPAL_* codes instead.
While we were doing this, we also cleaned up a little memory
management during ompi_info/orte-info/opal-info finalization.
Valgrind still reports a truckload of memory still in use at ompi_info
termination, but they mostly look to be components not freeing
memory/resources properly (and outside the scope of this fix).
This commit was SVN r27306.
The following Trac tickets were found above:
Ticket 3275 --> https://svn.open-mpi.org/trac/ompi/ticket/3275
2012-09-12 00:47:24 +04:00
|
|
|
* registration function and it returned OPAL_SUCCESS.
|
2008-05-06 22:08:45 +04:00
|
|
|
*/
|
2008-07-29 02:40:57 +04:00
|
|
|
typedef int (*mca_base_register_component_params_2_0_0_fn_t)(void);
|
|
|
|
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2015-03-06 06:50:44 +03:00
|
|
|
/**
|
|
|
|
* Maximum length of MCA project string names.
|
|
|
|
*/
|
|
|
|
#define MCA_BASE_MAX_PROJECT_NAME_LEN 15
|
2005-07-02 18:08:19 +04:00
|
|
|
/**
|
|
|
|
* Maximum length of MCA framework string names.
|
|
|
|
*/
|
2008-06-06 04:59:08 +04:00
|
|
|
#define MCA_BASE_MAX_TYPE_NAME_LEN 31
|
2005-07-02 18:08:19 +04:00
|
|
|
/**
|
|
|
|
* Maximum length of MCA component string names.
|
|
|
|
*/
|
2008-06-06 04:59:08 +04:00
|
|
|
#define MCA_BASE_MAX_COMPONENT_NAME_LEN 63
|
2005-07-02 18:08:19 +04:00
|
|
|
|
|
|
|
/**
|
|
|
|
* Common type for all MCA components.
|
|
|
|
*
|
|
|
|
* An instance of this type is always the first element in MCA
|
|
|
|
* components, allowing the component to be associated with a
|
|
|
|
* particular version of a specific framework, and to publish its own
|
|
|
|
* name and version.
|
|
|
|
*/
|
2015-03-06 06:50:44 +03:00
|
|
|
struct mca_base_component_2_1_0_t {
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2015-06-24 06:59:57 +03:00
|
|
|
int mca_major_version;
|
2005-07-02 18:08:19 +04:00
|
|
|
/**< Major number of the MCA. */
|
|
|
|
int mca_minor_version;
|
|
|
|
/**< Minor number of the MCA. */
|
|
|
|
int mca_release_version;
|
|
|
|
/**< Release number of the MCA. */
|
|
|
|
|
2015-03-06 06:50:44 +03:00
|
|
|
char mca_project_name[MCA_BASE_MAX_PROJECT_NAME_LEN + 1];
|
|
|
|
/**< String name of the project that this component belongs to. */
|
|
|
|
int mca_project_major_version;
|
|
|
|
/**< Major version number of the project that this component
|
|
|
|
belongs to. */
|
|
|
|
int mca_project_minor_version;
|
|
|
|
/**< Minor version number of the project that this component
|
|
|
|
belongs to. */
|
|
|
|
int mca_project_release_version;
|
|
|
|
/**< Release version number of the project that this component
|
|
|
|
belongs to. */
|
|
|
|
|
2008-06-06 04:59:08 +04:00
|
|
|
char mca_type_name[MCA_BASE_MAX_TYPE_NAME_LEN + 1];
|
2005-07-02 18:08:19 +04:00
|
|
|
/**< String name of the framework that this component belongs to. */
|
|
|
|
int mca_type_major_version;
|
|
|
|
/**< Major version number of the framework that this component
|
|
|
|
belongs to. */
|
|
|
|
int mca_type_minor_version;
|
|
|
|
/**< Minor version number of the framework that this component
|
|
|
|
belongs to. */
|
|
|
|
int mca_type_release_version;
|
|
|
|
/**< Release version number of the framework that this component
|
|
|
|
belongs to. */
|
|
|
|
|
2008-06-06 04:59:08 +04:00
|
|
|
char mca_component_name[MCA_BASE_MAX_COMPONENT_NAME_LEN + 1];
|
2005-07-02 18:08:19 +04:00
|
|
|
/**< This comopnent's string name. */
|
|
|
|
int mca_component_major_version;
|
|
|
|
/**< This component's major version number. */
|
|
|
|
int mca_component_minor_version;
|
|
|
|
/**< This component's minor version number. */
|
|
|
|
int mca_component_release_version;
|
|
|
|
/**< This component's release version number. */
|
2015-06-24 06:59:57 +03:00
|
|
|
|
2008-07-29 02:40:57 +04:00
|
|
|
mca_base_open_component_1_0_0_fn_t mca_open_component;
|
2005-07-02 18:08:19 +04:00
|
|
|
/**< Method for opening this component. */
|
2008-07-29 02:40:57 +04:00
|
|
|
mca_base_close_component_1_0_0_fn_t mca_close_component;
|
2005-07-02 18:08:19 +04:00
|
|
|
/**< Method for closing this component. */
|
2008-07-29 02:40:57 +04:00
|
|
|
mca_base_query_component_2_0_0_fn_t mca_query_component;
|
2008-05-06 22:08:45 +04:00
|
|
|
/**< Method for querying this component. */
|
2008-07-29 02:40:57 +04:00
|
|
|
mca_base_register_component_params_2_0_0_fn_t mca_register_component_params;
|
|
|
|
/**< Method for registering the component's MCA parameters */
|
|
|
|
|
|
|
|
/** Extra space to allow for expansion in the future without
|
|
|
|
breaking older components. */
|
|
|
|
char reserved[32];
|
2005-07-02 18:08:19 +04:00
|
|
|
};
|
2008-07-29 02:40:57 +04:00
|
|
|
/** Unversioned convenience typedef; use this name in
|
|
|
|
frameworks/components to stay forward source-compatible */
|
2015-03-06 06:50:44 +03:00
|
|
|
typedef struct mca_base_component_2_1_0_t mca_base_component_t;
|
2008-07-29 02:40:57 +04:00
|
|
|
/** Versioned convenience typedef */
|
2015-03-06 06:50:44 +03:00
|
|
|
typedef struct mca_base_component_2_1_0_t mca_base_component_2_1_0_t;
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2007-03-17 02:11:45 +03:00
|
|
|
/*
|
|
|
|
* Metadata Bit field parameters
|
|
|
|
*/
|
|
|
|
#define MCA_BASE_METADATA_PARAM_NONE (uint32_t)0x00 /**< No Metadata flags */
|
|
|
|
#define MCA_BASE_METADATA_PARAM_CHECKPOINT (uint32_t)0x02 /**< Checkpoint enabled Component */
|
|
|
|
#define MCA_BASE_METADATA_PARAM_DEBUG (uint32_t)0x04 /**< Debug enabled/only Component */
|
|
|
|
|
2005-07-02 18:08:19 +04:00
|
|
|
/**
|
2008-07-29 02:40:57 +04:00
|
|
|
* Meta data for MCA v2.0.0 components.
|
2005-07-02 18:08:19 +04:00
|
|
|
*/
|
2008-07-29 02:40:57 +04:00
|
|
|
struct mca_base_component_data_2_0_0_t {
|
2007-03-17 02:11:45 +03:00
|
|
|
uint32_t param_field;
|
2008-07-29 02:40:57 +04:00
|
|
|
/**< Metadata parameter bit field filled in by the parameters
|
|
|
|
defined above */
|
|
|
|
|
|
|
|
/** Extra space to allow for expansion in the future without
|
|
|
|
breaking older components. */
|
|
|
|
char reserved[32];
|
2005-07-02 18:08:19 +04:00
|
|
|
};
|
2008-07-29 02:40:57 +04:00
|
|
|
/** Unversioned convenience typedef; use this name in
|
|
|
|
frameworks/components to stay forward source-compatible */
|
|
|
|
typedef struct mca_base_component_data_2_0_0_t mca_base_component_data_t;
|
|
|
|
/** Versioned convenience typedef */
|
|
|
|
typedef struct mca_base_component_data_2_0_0_t mca_base_component_data_2_0_0_t;
|
2005-07-02 18:08:19 +04:00
|
|
|
|
|
|
|
/**
|
2015-06-24 06:59:57 +03:00
|
|
|
* Macro for framework author convenience.
|
2005-07-02 18:08:19 +04:00
|
|
|
*
|
|
|
|
* This macro is used by frameworks defining their component types,
|
2008-07-29 02:40:57 +04:00
|
|
|
* indicating that they subscribe to the MCA version 2.0.0. See
|
2005-07-02 18:08:19 +04:00
|
|
|
* component header files (e.g., coll.h) for examples of its usage.
|
|
|
|
*/
|
2008-07-29 02:40:57 +04:00
|
|
|
#define MCA_BASE_VERSION_MAJOR 2
|
2015-03-06 06:50:44 +03:00
|
|
|
#define MCA_BASE_VERSION_MINOR 1
|
2008-07-29 02:40:57 +04:00
|
|
|
#define MCA_BASE_VERSION_RELEASE 0
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2015-04-17 22:45:14 +03:00
|
|
|
#define MCA_BASE_MAKE_VERSION(level, MAJOR, MINOR, RELEASE) \
|
|
|
|
.mca_## level ##_major_version = MAJOR, \
|
|
|
|
.mca_## level ##_minor_version = MINOR, \
|
|
|
|
.mca_## level ##_release_version = RELEASE
|
|
|
|
|
|
|
|
|
|
|
|
#define MCA_BASE_VERSION_2_1_0(PROJECT, project_major, project_minor, project_release, TYPE, type_major, type_minor, type_release) \
|
2015-03-06 06:50:44 +03:00
|
|
|
.mca_major_version = MCA_BASE_VERSION_MAJOR, \
|
|
|
|
.mca_minor_version = MCA_BASE_VERSION_MINOR, \
|
|
|
|
.mca_release_version = MCA_BASE_VERSION_RELEASE, \
|
2015-04-17 22:45:14 +03:00
|
|
|
.mca_project_name = PROJECT, \
|
|
|
|
MCA_BASE_MAKE_VERSION(project, project_major, project_minor, project_release), \
|
|
|
|
.mca_type_name = TYPE, \
|
|
|
|
MCA_BASE_MAKE_VERSION(type, type_major, type_minor, type_release)
|
2015-03-06 06:50:44 +03:00
|
|
|
|
|
|
|
#define OPAL_MCA_BASE_VERSION_2_1_0(type, type_major, type_minor, type_release) \
|
|
|
|
MCA_BASE_VERSION_2_1_0("opal", OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, OPAL_RELEASE_VERSION, type, type_major, type_minor, type_release)
|
2005-07-02 18:08:19 +04:00
|
|
|
|
2007-03-29 05:00:33 +04:00
|
|
|
#endif /* OPAL_MCA_H */
|