ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

My apologies for doing this outside of the usual time restrictions, but we need to get this in so we can make progress.

Просмотр исходного кода 
Move the ORTE-level debugger code back into orterun and out of the ORTE library to resolve symbol conflicts.

This commit was SVN r25713.
Этот коммит содержится в:
Ralph Castain 2012-01-11 15:53:09 +00:00
родитель 686ee387c8
Коммит bf103de66c
35 изменённых файлов: 587 добавлений и 1557 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

5
ompi/debuggers/debuggers.h
Просмотреть файл

@ -27,7 +27,6 @@
#define OMPI_DEBUGGERS_H
#include "ompi_config.h"
#include "orte/mca/debugger/base/base.h"
BEGIN_C_DECLS
@ -43,8 +42,10 @@ extern void ompi_debugger_notify_abort(char *string);
/**
* Breakpoint function for parallel debuggers.
* This function is also defined in orterun for the starter.
* It should never conflict with this one
*/
OMPI_DECLSPEC extern void MPIR_Breakpoint(void);
OMPI_DECLSPEC void* MPIR_Breakpoint(void);
END_C_DECLS

15
ompi/debuggers/ompi_debuggers.c
Просмотреть файл

@ -74,7 +74,6 @@
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/runtime/orte_globals.h"
#if defined(OMPI_MSGQ_DLL)
@ -126,8 +125,8 @@ OMPI_DECLSPEC opal_datatype_t* opal_datatype_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_datatype_t* ompi_datatype_t_type_force_inclusion = NULL;
OMPI_DECLSPEC volatile int MPIR_debug_gate = 0;
OMPI_DECLSPEC extern volatile int MPIR_being_debugged;
OMPI_DECLSPEC extern volatile int MPIR_debug_state;
OMPI_DECLSPEC volatile int MPIR_being_debugged = 0;
OMPI_DECLSPEC volatile int MPIR_debug_state = 0;
OMPI_DECLSPEC char *MPIR_debug_abort_string = "";
/* Check for a file in few direct ways for portability */
@ -283,3 +282,13 @@ void ompi_debugger_notify_abort(char *reason)
/* Now tell the debugger */
MPIR_Breakpoint();
}
/*
* Breakpoint function for parallel debuggers. This function is also
* defined in orterun for the starter. It should never conflict with
* this
*/
void* MPIR_Breakpoint(void)
{
return NULL;
}

10
ompi/tools/ompi_info/components.c
Просмотреть файл

@ -110,8 +110,6 @@
#if !ORTE_DISABLE_FULL_SUPPORT
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/debugger/debugger.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/iof/iof.h"
#include "orte/mca/iof/base/base.h"
#include "orte/mca/oob/oob.h"
@ -431,14 +429,6 @@ void ompi_info_open_components(void)
map->components = &orte_notifier_base_components_available;
opal_pointer_array_add(&component_map, map);
if (ORTE_SUCCESS != orte_debugger_base_open()) {
goto error;
}
map = OBJ_NEW(ompi_info_component_map_t);
map->type = strdup("debugger");
map->components = &orte_debugger_base_components_available;
opal_pointer_array_add(&component_map, map);
if (ORTE_SUCCESS != mca_oob_base_open()) {
goto error;
}

1
ompi/tools/ompi_info/ompi_info.c
Просмотреть файл

@ -233,7 +233,6 @@ int main(int argc, char *argv[])
#endif
#if !ORTE_DISABLE_FULL_SUPPORT
opal_pointer_array_add(&mca_types, "debugger");
opal_pointer_array_add(&mca_types, "iof");
opal_pointer_array_add(&mca_types, "oob");
opal_pointer_array_add(&mca_types, "odls");

28
orte/mca/debugger/Makefile.am
Просмотреть файл

@ -1,28 +0,0 @@
#
# Copyright (c) 2010 Cisco Systems, Inc.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# main library setup
noinst_LTLIBRARIES = libmca_debugger.la
libmca_debugger_la_SOURCES =
# local files
headers = debugger.h
libmca_debugger_la_SOURCES += $(headers)
# Conditionally install the header files
if WANT_INSTALL_HEADERS
ortedir = $(includedir)/openmpi/$(subdir)
nobase_orte_HEADERS = $(headers)
endif
include base/Makefile.am
distclean-local:
rm -f base/static-components.h

18
orte/mca/debugger/base/Makefile.am
Просмотреть файл

@ -1,18 +0,0 @@
#
# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
headers += \
base/base.h
libmca_debugger_la_SOURCES += \
base/debugger_base_close.c \
base/debugger_base_select.c \
base/debugger_base_open.c \
base/debugger_base_fns.c

82
orte/mca/debugger/base/base.h
Просмотреть файл

@ -1,82 +0,0 @@
/*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/** @file:
*/
#ifndef MCA_DEBUGGER_BASE_H
#define MCA_DEBUGGER_BASE_H
/*
* includes
*/
#include "orte_config.h"
#include "opal/class/opal_list.h"
#include "orte/mca/debugger/debugger.h"
BEGIN_C_DECLS
typedef struct {
int output;
bool dump_proctable;
char *test_daemon;
bool test_attach;
} orte_debugger_base_t;
ORTE_DECLSPEC extern orte_debugger_base_t orte_debugger_base;
/*
* function definitions
*/
ORTE_DECLSPEC int orte_debugger_base_open(void);
ORTE_DECLSPEC int orte_debugger_base_close(void);
ORTE_DECLSPEC int orte_debugger_base_select(void);
ORTE_DECLSPEC void orte_debugger_base_run_debugger(char *basename, opal_cmd_line_t *cmd_line,
int argc, char *argv[], int num_procs);
ORTE_DECLSPEC void orte_debugger_base_init_after_spawn(orte_job_t *jdata);
ORTE_DECLSPEC void orte_debugger_base_dump(void);
ORTE_DECLSPEC extern opal_list_t orte_debugger_base_components_available;
/* +++ begin MPICH/TotalView std debugger interface definitions */
#define MPIR_MAX_PATH_LENGTH 512
#define MPIR_MAX_ARG_LENGTH 1024
struct MPIR_PROCDESC {
char *host_name; /* something that can be passed to inet_addr */
char *executable_name; /* name of binary */
int pid; /* process pid */
};
ORTE_DECLSPEC extern struct MPIR_PROCDESC *MPIR_proctable;
ORTE_DECLSPEC extern int MPIR_proctable_size;
ORTE_DECLSPEC extern volatile int MPIR_being_debugged;
ORTE_DECLSPEC extern volatile int MPIR_debug_state;
ORTE_DECLSPEC extern int MPIR_i_am_starter;
ORTE_DECLSPEC extern int MPIR_partial_attach_ok;
ORTE_DECLSPEC extern char MPIR_executable_path[MPIR_MAX_PATH_LENGTH];
ORTE_DECLSPEC extern char MPIR_server_arguments[MPIR_MAX_ARG_LENGTH];
ORTE_DECLSPEC extern volatile int MPIR_forward_output;
ORTE_DECLSPEC extern volatile int MPIR_forward_comm;
ORTE_DECLSPEC extern char MPIR_attach_fifo[MPIR_MAX_PATH_LENGTH];
ORTE_DECLSPEC extern int MPIR_force_to_main;
typedef void (*orte_debugger_breakpoint_fn_t)(void);
ORTE_DECLSPEC void MPIR_Breakpoint(void);
ORTE_DECLSPEC void orte_debugger_base_pull_mpir_breakpoint(void);
/* --- end MPICH/TotalView std debugger interface definitions */
END_C_DECLS
#endif

33
orte/mca/debugger/base/debugger_base_close.c
Просмотреть файл

@ -1,33 +0,0 @@
/*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "opal/mca/mca.h"
#include "opal/mca/base/base.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/debugger/debugger.h"
int orte_debugger_base_close(void)
{
#if !ORTE_DISABLE_FULL_SUPPORT
if (NULL != orte_debugger.finalize) {
orte_debugger.finalize();
}
/* Close all remaining available components */
mca_base_components_close(orte_debugger_base.output,
&orte_debugger_base_components_available, NULL);
#endif
/* All done */
return ORTE_SUCCESS;
}

201
orte/mca/debugger/base/debugger_base_fns.c
Просмотреть файл

@ -1,201 +0,0 @@
/*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "opal/util/output.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include <stdio.h>
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif /* HAVE_STDLIB_H */
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif /* HAVE_STRINGS_H */
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#include <ctype.h>
#include "opal/util/argv.h"
#include "opal/util/os_path.h"
#include "opal/util/path.h"
#include "opal/util/opal_environ.h"
#include "orte/util/name_fns.h"
#include "orte/util/show_help.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/debugger/base/base.h"
#define DUMP_INT(X) fprintf(stderr, " %s = %d\n", # X, X);
#if !ORTE_DISABLE_FULL_SUPPORT
void orte_debugger_base_dump(void)
{
int i;
DUMP_INT(MPIR_being_debugged);
DUMP_INT(MPIR_debug_state);
DUMP_INT(MPIR_partial_attach_ok);
DUMP_INT(MPIR_i_am_starter);
DUMP_INT(MPIR_forward_output);
DUMP_INT(MPIR_proctable_size);
fprintf(stderr, " MPIR_proctable:\n");
for (i = 0; i < MPIR_proctable_size; i++) {
fprintf(stderr,
" (i, host, exe, pid) = (%d, %s, %s, %d)\n",
i,
MPIR_proctable[i].host_name,
MPIR_proctable[i].executable_name,
MPIR_proctable[i].pid);
}
fprintf(stderr, "MPIR_executable_path: %s\n",
('\0' == MPIR_executable_path[0]) ?
"NULL" : (char*) MPIR_executable_path);
fprintf(stderr, "MPIR_server_arguments: %s\n",
('\0' == MPIR_server_arguments[0]) ?
"NULL" : (char*) MPIR_server_arguments);
}
/*
* Initialization of data structures for running under a debugger
* using the MPICH/TotalView parallel debugger interface. This stage
* of initialization must occur after spawn
*
* NOTE: We -always- perform this step to ensure that any debugger
* that attaches to us post-launch of the application can get a
* completed proctable
*/
void orte_debugger_base_init_after_spawn(orte_job_t *jdata)
{
orte_proc_t *proc;
orte_app_context_t *appctx;
orte_vpid_t i, j;
opal_buffer_t buf;
orte_process_name_t rank0;
int rc;
/* if we couldn't get thru the mapper stage, we might
* enter here with no procs. Avoid the "zero byte malloc"
* message by checking here
*/
if (MPIR_proctable || 0 == jdata->num_procs) {
/* already initialized */
opal_output_verbose(5, orte_debugger_base.output,
"%s: debugger already initialized or zero procs",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return;
}
/* fill in the proc table for the application processes */
opal_output_verbose(5, orte_debugger_base.output,
"%s: Setting up debugger process table for applications",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
MPIR_debug_state = 1;
/* set the total number of processes in the job */
MPIR_proctable_size = jdata->num_procs;
/* allocate MPIR_proctable */
MPIR_proctable = (struct MPIR_PROCDESC *) malloc(sizeof(struct MPIR_PROCDESC) *
MPIR_proctable_size);
if (MPIR_proctable == NULL) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return;
}
if (orte_debugger_base.dump_proctable) {
opal_output(orte_clean_output, "MPIR Proctable for job %s", ORTE_JOBID_PRINT(jdata->jobid));
}
/* initialize MPIR_proctable */
for (j=0; j < jdata->num_procs; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, j))) {
continue;
}
/* store this data in the location whose index
* corresponds to the proc's rank
*/
i = proc->name.vpid;
if (NULL == (appctx = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, proc->app_idx))) {
continue;
}
MPIR_proctable[i].host_name = strdup(proc->node->name);
if ( 0 == strncmp(appctx->app, OPAL_PATH_SEP, 1 )) {
MPIR_proctable[i].executable_name =
opal_os_path( false, appctx->app, NULL );
} else {
MPIR_proctable[i].executable_name =
opal_os_path( false, appctx->cwd, appctx->app, NULL );
}
MPIR_proctable[i].pid = proc->pid;
if (orte_debugger_base.dump_proctable) {
opal_output(orte_clean_output, "%s: Host %s Exe %s Pid %d",
ORTE_VPID_PRINT(i), MPIR_proctable[i].host_name,
MPIR_proctable[i].executable_name, MPIR_proctable[i].pid);
}
}
if (0 < opal_output_get_verbosity(orte_debugger_base.output)) {
orte_debugger_base_dump();
}
/* if we are being launched under a debugger, then we must wait
* for it to be ready to go and do some things to start the job
*/
if (MPIR_being_debugged) {
/* wait for all procs to have reported their contact info - this
* ensures that (a) they are all into mpi_init, and (b) the system
* has the contact info to successfully send a message to rank=0
*/
ORTE_PROGRESSED_WAIT(false, jdata->num_reported, jdata->num_procs);
MPIR_Breakpoint();
/* send a message to rank=0 to release it */
OBJ_CONSTRUCT(&buf, opal_buffer_t); /* don't need anything in this */
rank0.jobid = jdata->jobid;
rank0.vpid = 0;
if (0 > (rc = orte_rml.send_buffer(&rank0, &buf, ORTE_RML_TAG_DEBUGGER_RELEASE, 0))) {
opal_output(0, "Error: could not send debugger release to MPI procs - error %s", ORTE_ERROR_NAME(rc));
}
OBJ_DESTRUCT(&buf);
}
}
#endif
/*
* Dummy function so that the linker can pull in all the symbols from
* this file.
*/
void orte_debugger_base_pull_mpir_breakpoint(void)
{
return;
}
/*
* Breakpoint function for parallel debuggers
*/
void MPIR_Breakpoint(void)
{
return;
}

105
orte/mca/debugger/base/debugger_base_open.c
Просмотреть файл

@ -1,105 +0,0 @@
/*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "opal/mca/mca.h"
#include "opal/util/output.h"
#include "opal/mca/base/base.h"
#include "opal/mca/base/mca_base_param.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include "orte/mca/debugger/base/base.h"
/*
* The following file was created by configure. It contains extern
* statements and the definition of an array of pointers to each
* component's public mca_base_component_t struct.
*/
#include "orte/mca/debugger/base/static-components.h"
/*
* Global variables
*/
orte_debugger_base_t orte_debugger_base;
opal_list_t orte_debugger_base_components_available;
orte_debugger_base_module_t orte_debugger;
/* instance the standard MPIR interfaces */
struct MPIR_PROCDESC *MPIR_proctable = NULL;
int MPIR_proctable_size = 0;
volatile int MPIR_being_debugged = 0;
volatile int MPIR_debug_state = 0;
int MPIR_i_am_starter = 0;
int MPIR_partial_attach_ok = 1;
char MPIR_executable_path[MPIR_MAX_PATH_LENGTH];
char MPIR_server_arguments[MPIR_MAX_ARG_LENGTH];
volatile int MPIR_forward_output = 0;
volatile int MPIR_forward_comm = 0;
char MPIR_attach_fifo[MPIR_MAX_PATH_LENGTH];
int MPIR_force_to_main = 0;
#if ORTE_DISABLE_FULL_SUPPORT
int orte_debugger_base_open(void)
{
return ORTE_SUCCESS;
}
#else
/**
* Function for finding and opening either all MCA components, or the one
* that was specifically requested via a MCA parameter.
*/
int orte_debugger_base_open(void)
{
int value;
/* Debugging / verbose output. Always have stream open, with
verbose set by the mca open system... */
orte_debugger_base.output = opal_output_open(NULL);
mca_base_param_reg_int_name("orte",
"output_debugger_proctable",
"Whether or not to output the debugger proctable after launch (default: false)",
true, false, 0, &value);
orte_debugger_base.dump_proctable = OPAL_INT_TO_BOOL(value);
mca_base_param_reg_string_name("orte", "debugger_test_daemon",
"Name of the executable to be used to simulate a debugger colaunch (relative or absolute path)",
false, false, NULL, &orte_debugger_base.test_daemon);
mca_base_param_reg_int_name("orte",
"debugger_test_attach",
"Test debugger colaunch after debugger attachment",
false, false, 0, &value);
orte_debugger_base.test_attach = OPAL_INT_TO_BOOL(value);
/* Open up all available components */
if (ORTE_SUCCESS !=
mca_base_components_open("debugger", orte_debugger_base.output,
mca_debugger_base_static_components,
&orte_debugger_base_components_available,
true)) {
return ORTE_ERROR;
}
/* All done */
return ORTE_SUCCESS;
}
#endif

54
orte/mca/debugger/base/debugger_base_select.c
Просмотреть файл

@ -1,54 +0,0 @@
/*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include "opal/mca/mca.h"
#include "opal/mca/base/base.h"
#include "opal/util/output.h"
#include "orte/mca/debugger/base/base.h"
int orte_debugger_base_select(void)
{
#if !ORTE_DISABLE_FULL_SUPPORT
orte_debugger_base_module_t *best_module=NULL;
orte_debugger_base_component_t *best_component=NULL;
int ret;
/*
* Select the best component
*/
if( OPAL_SUCCESS != mca_base_select("debugger", orte_debugger_base.output,
&orte_debugger_base_components_available,
(mca_base_module_t **) &best_module,
(mca_base_component_t **) &best_component) ) {
/* This will only happen if no component was selected */
ret = ORTE_ERR_NOT_FOUND;
goto cleanup;
}
/* Save the winner */
/* No global component structure */
orte_debugger = *best_module;
ret = orte_debugger.init();
cleanup:
return ret;
#else
return ORTE_ERR_NOT_IMPLEMENTED;
#endif
}

77
orte/mca/debugger/debugger.h
Просмотреть файл

@ -1,77 +0,0 @@
/*
* Copyright (c) 2010 Cisco Systems, Inc. All Rights Reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef MCA_DEBUGGER_H
#define MCA_DEBUGGER_H
/*
* includes
*/
#include "orte_config.h"
#include "opal/mca/mca.h"
#include "orte/runtime/orte_globals.h"
BEGIN_C_DECLS
/*
* Component functions - all MUST be provided!
*/
/* initialize the selected module */
typedef int (*orte_debugger_base_module_init_fn_t)(void);
/* finalize the selected module */
typedef void (*orte_debugger_base_module_finalize_fn_t)(void);
/* init debuggers before spawn */
typedef void (*orte_debugger_base_module_init_before_spawn_fn_t)(orte_job_t *jdata);
/* init debuggers after spawn */
typedef void (*orte_debugger_base_module_init_after_spawn_fn_t)(orte_job_t *jdata);
/*
* Ver 1.0
*/
struct orte_debugger_base_module_1_0_0_t {
orte_debugger_base_module_init_fn_t init;
orte_debugger_base_module_finalize_fn_t finalize;
orte_debugger_base_module_init_before_spawn_fn_t init_before_spawn;
orte_debugger_base_module_init_after_spawn_fn_t init_after_spawn;
};
typedef struct orte_debugger_base_module_1_0_0_t orte_debugger_base_module_1_0_0_t;
typedef orte_debugger_base_module_1_0_0_t orte_debugger_base_module_t;
ORTE_DECLSPEC extern orte_debugger_base_module_t orte_debugger;
/*
* the standard component data structure
*/
struct orte_debugger_base_component_1_0_0_t {
mca_base_component_t base_version;
mca_base_component_data_t base_data;
};
typedef struct orte_debugger_base_component_1_0_0_t orte_debugger_base_component_1_0_0_t;
typedef orte_debugger_base_component_1_0_0_t orte_debugger_base_component_t;
/*
* Macro for use in components that are of type debugger v1.0.0
*/
#define ORTE_DEBUGGER_BASE_VERSION_1_0_0 \
/* debugger v1.0 is chained to MCA v2.0 */ \
MCA_BASE_VERSION_2_0_0, \
/* debugger v1.0 */ \
"debugger", 1, 0, 0
END_C_DECLS
#endif /* MCA_DEBUGGER_H */

12
orte/mca/debugger/mpir/.windows
Просмотреть файл

@ -1,12 +0,0 @@
#
# Copyright (c) 2008-2010 High Performance Computing Center Stuttgart,
# University of Stuttgart. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# Specific to this module
mca_link_libraries=libopen-rte Ws2_32.lib

36
orte/mca/debugger/mpir/Makefile.am
Просмотреть файл

@ -1,36 +0,0 @@
#
# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
EXTRA_DIST = .windows
sources = \
mpir.h \
mpir.c \
mpir_component.c
# Make the output library in this directory, and name it either
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
# (for static builds).
if MCA_BUILD_orte_debugger_mpir_DSO
component_noinst =
component_install = mca_debugger_mpir.la
else
component_noinst = libmca_debugger_mpir.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_debugger_mpir_la_SOURCES = $(sources)
mca_debugger_mpir_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_debugger_mpir_la_SOURCES =$(sources)
libmca_debugger_mpir_la_LDFLAGS = -module -avoid-version

19
orte/mca/debugger/mpir/configure.m4
Просмотреть файл

@ -1,19 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_debugger_mpir_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_debugger_mpir_CONFIG], [
AC_CONFIG_FILES([orte/mca/debugger/mpir/Makefile])
AS_IF([test "$orte_without_full_support" = 0],
[$1],
[$2])
])

196
orte/mca/debugger/mpir/mpir.c
Просмотреть файл

@ -1,196 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2006 The University of Tennessee and The University
* 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) 2007 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2007 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2007-2010 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2010 Oracle and/or its affiliates. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
* Debugger support for orterun
*
* We interpret the MPICH debugger interface as follows:
*
* a) The launcher
* - spawns the other processes,
* - fills in the table MPIR_proctable, and sets MPIR_proctable_size
* - sets MPIR_debug_state to MPIR_DEBUG_SPAWNED ( = 1)
* - calls MPIR_Breakpoint() which the debugger will have a
* breakpoint on.
*
* b) Applications start and then spin until MPIR_debug_gate is set
* non-zero by the debugger.
*
* This file implements (a).
*
**************************************************************************
*
* Note that we have presently tested both TotalView and DDT parallel
* debuggers. They both nominally subscribe to the Etnus attaching
* interface, but there are differences between the two.
*
* TotalView: user launches "totalview mpirun -a ...<mpirun args>...".
* TV launches mpirun. mpirun launches the application and then calls
* MPIR_Breakpoint(). This is the signal to TV that it's a parallel
* MPI job. TV then reads the proctable in mpirun and attaches itself
* to all the processes (it takes care of launching itself on the
* remote nodes). Upon attaching to all the MPI processes, the
* variable MPIR_being_debugged is set to 1. When it has finished
* attaching itself to all the MPI processes that it wants to,
* MPIR_Breakpoint() returns.
*
* DDT: user launches "ddt bin -np X <mpi app name>". DDT fork/exec's
* mpirun to launch ddt-debugger on the back-end nodes via "mpirun -np
* X ddt-debugger" (not the lack of other arguments -- we can't pass
* anything to mpirun). This app will eventually fork/exec the MPI
* app. DDT does not current set MPIR_being_debugged in the MPI app.
*
**************************************************************************
*
* We support two ways of waiting for attaching debuggers. The
* implementation spans this file and ompi/debuggers/ompi_debuggers.c.
*
* 1. If using orterun: MPI processes will have the
* orte_in_parallel_debugger MCA param set to true (because not all
* debuggers consistently set MPIR_being_debugged in both the launcher
* and in the MPI procs). The HNP will call MPIR_Breakpoint() and
* then RML send a message to VPID 0 (MCW rank 0) when it returns
* (MPIR_Breakpoint() doesn't return until the debugger has attached
* to all relevant processes). Meanwhile, VPID 0 blocks waiting for
* the RML message. All other VPIDs immediately call the grpcomm
* barrier (and therefore block until the debugger attaches). Once
* VPID 0 receives the RML message, we know that the debugger has
* attached to all processes that it cares about, and VPID 0 then
* joins the grpcomm barrier, allowing the job to continue. This
* scheme has the side effect of nicely supporting partial attaches by
* parallel debuggers (i.e., attaching to only some of the MPI
* processes; not necessarily all of them).
*
* 2. If not using orterun: in this case, ORTE_DISABLE_FULL_SUPPORT
* will be true, and we know that there will not be an RML message
* sent to VPID 0. So we have to look for a magic environment
* variable from the launcher to know if the jobs will be attached by
* a debugger (e.g., set by yod, srun, ...etc.), and if so, spin on
* MPIR_debug_gate. These environment variable names must be
* hard-coded in the OMPI layer (see ompi/debuggers/ompi_debuggers.c).
*/
#include "orte_config.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include <stdio.h>
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif /* HAVE_STDLIB_H */
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif /* HAVE_STRINGS_H */
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#include <ctype.h>
#include "opal/util/argv.h"
#include "opal/util/output.h"
#include "opal/util/path.h"
#include "opal/util/os_path.h"
#include "opal/class/opal_pointer_array.h"
#include "opal/mca/base/mca_base_param.h"
#include "opal/util/opal_getcwd.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/rmaps/rmaps_types.h"
#include "orte/runtime/orte_globals.h"
#include "orte/runtime/orte_wait.h"
#include "orte/util/show_help.h"
#include "orte/util/name_fns.h"
#include "orte/mca/debugger/base/base.h"
#include "mpir.h"
#include "mpir.h"
/* Static API's */
static int init(void);
static void finalize(void);
static void init_before_spawn(orte_job_t *jdata);
/* Module definition */
orte_debugger_base_module_t orte_debugger_mpir_module = {
init,
finalize,
init_before_spawn,
orte_debugger_base_init_after_spawn
};
/* local globals */
static int init(void)
{
return ORTE_SUCCESS;
}
/**
* Release resources associated with data structures for running under
* a debugger using the MPICH/TotalView parallel debugger interface.
*/
void finalize(void)
{
if (MPIR_proctable) {
free(MPIR_proctable);
MPIR_proctable = NULL;
}
}
/**
* Initialization of data structures for running under a debugger
* using the MPICH/TotalView parallel debugger interface. Before the
* spawn we need to check if we are being run under a TotalView-like
* debugger; if so then inform applications via an MCA parameter.
*/
void init_before_spawn(orte_job_t *jdata)
{
char *env_name;
orte_app_context_t *app;
int i;
if (!MPIR_being_debugged && !orte_in_parallel_debugger) {
return;
}
opal_output_verbose(1, orte_debugger_base.output, "Info: Spawned by a debugger");
/* tell the procs they are being debugged */
env_name = mca_base_param_environ_variable("orte",
"in_parallel_debugger", NULL);
for (i=0; i < jdata->apps->size; i++) {
if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
continue;
}
opal_setenv(env_name, "1", true, &app->env);
}
free(env_name);
}

34
orte/mca/debugger/mpir/mpir.h
Просмотреть файл

@ -1,34 +0,0 @@
/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2009 The University of Tennessee and The University
* 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) 2007-2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef DEBUGGER_MPIR_H
#define DEBUGGER_MPIR_H
#include "orte_config.h"
#include "orte/mca/debugger/debugger.h"
BEGIN_C_DECLS
ORTE_MODULE_DECLSPEC extern orte_debugger_base_component_t mca_debugger_mpir_component;
extern orte_debugger_base_module_t orte_debugger_mpir_module;
END_C_DECLS
#endif /* ORTE_DEBUGGERS_H */

47
orte/mca/debugger/mpir/mpir_component.c
Просмотреть файл

@ -1,47 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "mpir.h"
static int component_query(mca_base_module_t **module, int *priority);
/*
* Struct of function pointers that need to be initialized
*/
orte_debugger_base_component_t mca_debugger_mpir_component = {
{
ORTE_DEBUGGER_BASE_VERSION_1_0_0,
"mpir", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
ORTE_MINOR_VERSION, /* MCA module minor version */
ORTE_RELEASE_VERSION, /* MCA module release version */
NULL,
NULL,
component_query /* module query */
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
};
static int component_query(mca_base_module_t **module, int *priority)
{
*priority = 100;
*module = (mca_base_module_t *)&orte_debugger_mpir_module;
return ORTE_SUCCESS;
}

34
orte/mca/debugger/mpirx/Makefile.am
Просмотреть файл

@ -1,34 +0,0 @@
#
# Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
sources = \
mpirx.h \
mpirx.c \
mpirx_component.c
# Make the output library in this directory, and name it either
# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
# (for static builds).
if MCA_BUILD_orte_debugger_mpirx_DSO
component_noinst =
component_install = mca_debugger_mpirx.la
else
component_noinst = libmca_debugger_mpirx.la
component_install =
endif
mcacomponentdir = $(pkglibdir)
mcacomponent_LTLIBRARIES = $(component_install)
mca_debugger_mpirx_la_SOURCES = $(sources)
mca_debugger_mpirx_la_LDFLAGS = -module -avoid-version
noinst_LTLIBRARIES = $(component_noinst)
libmca_debugger_mpirx_la_SOURCES =$(sources)
libmca_debugger_mpirx_la_LDFLAGS = -module -avoid-version

19
orte/mca/debugger/mpirx/configure.m4
Просмотреть файл

@ -1,19 +0,0 @@
# -*- shell-script -*-
#
# Copyright (c) 2011 Los Alamos National Security, LLC.
# All rights reserved.
# $COPYRIGHT$
#
# Additional copyrights may follow
#
# $HEADER$
#
# MCA_debugger_mpirx_CONFIG([action-if-found], [action-if-not-found])
# -----------------------------------------------------------
AC_DEFUN([MCA_orte_debugger_mpirx_CONFIG], [
AC_CONFIG_FILES([orte/mca/debugger/mpirx/Makefile])
AS_IF([test "$orte_without_full_support" = 0],
[$1],
[$2])
])

366
orte/mca/debugger/mpirx/mpirx.c
Просмотреть файл

@ -1,366 +0,0 @@
/*
* Copyright (c) 2010-2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2010 Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2010-2011 Los Alamos National Security, LLC.
* All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
#include <stdio.h>
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif /* HAVE_STDLIB_H */
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif /* HAVE_STRINGS_H */
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#include <sys/stat.h>
#include <ctype.h>
#include <fcntl.h>
#include <errno.h>
#include "opal/util/argv.h"
#include "opal/util/output.h"
#include "opal/util/path.h"
#include "opal/util/os_path.h"
#include "opal/class/opal_pointer_array.h"
#include "opal/mca/base/mca_base_param.h"
#include "opal/util/opal_getcwd.h"
#include "opal/mca/event/event.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/plm_private.h"
#include "orte/mca/rmaps/rmaps_types.h"
#include "orte/runtime/orte_globals.h"
#include "orte/runtime/orte_wait.h"
#include "orte/util/show_help.h"
#include "orte/util/name_fns.h"
#include "orte/mca/debugger/base/base.h"
#include "mpirx.h"
#define FILE_MODE (S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)
/* Static API's */
static int init(void);
static void finalize(void);
static void init_before_spawn(orte_job_t *jdata);
/* Module definition */
orte_debugger_base_module_t orte_debugger_mpirx_module = {
init,
finalize,
init_before_spawn,
orte_debugger_base_init_after_spawn
};
/* local globals and functions */
static void attach_debugger(int fd, short event, void *arg);
static void build_debugger_args(orte_app_context_t *debugger);
static void open_fifo(void);
static opal_event_t attach;
static int attach_fd = -1;
static bool fifo_active=false;
static int init(void)
{
return ORTE_SUCCESS;
}
/**
* Release resources associated with data structures for running under
* a debugger using the MPICH/TotalView parallel debugger interface.
*/
void finalize(void)
{
if (fifo_active) {
opal_event_del(&attach);
close(attach_fd);
}
if (MPIR_proctable) {
free(MPIR_proctable);
MPIR_proctable = NULL;
}
}
/**
* Initialization of data structures for running under a debugger
* using an extended MPICH/TotalView parallel debugger interface. Before the
* spawn we need to check if we are being run under a TotalView-like
* debugger; if so then inform applications via an MCA parameter.
*/
void init_before_spawn(orte_job_t *jdata)
{
char *env_name;
orte_app_context_t *app;
int i;
int32_t ljob;
char *attach_fifo;
if (!MPIR_being_debugged && !orte_in_parallel_debugger) {
/* if we were given a test debugger, then we still want to
* colaunch it
*/
if (NULL != orte_debugger_base.test_daemon) {
opal_output_verbose(2, orte_debugger_base.output,
"%s No debugger test daemon specified",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto launchit;
}
/* if we were given an auto-detect rate, then we want to setup
* an event so we periodically do the check
*/
if (0 < orte_debugger_mpirx_check_rate) {
opal_output_verbose(2, orte_debugger_base.output,
"%s Setting debugger attach check rate for %d seconds",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orte_debugger_mpirx_check_rate);
ORTE_TIMER_EVENT(orte_debugger_mpirx_check_rate, 0, attach_debugger);
} else {
/* create the attachment FIFO and put it into MPIR, setup readevent */
/* create a FIFO name in the session dir */
attach_fifo = opal_os_path(false, orte_process_info.job_session_dir, "debugger_attach_fifo", NULL);
if ((mkfifo(attach_fifo, FILE_MODE) < 0) && errno != EEXIST) {
opal_output(0, "CANNOT CREATE FIFO %s: errno %d", attach_fifo, errno);
free(attach_fifo);
return;
}
strncpy(MPIR_attach_fifo, attach_fifo, MPIR_MAX_PATH_LENGTH - 1);
free (attach_fifo);
open_fifo ();
}
return;
}
launchit:
opal_output_verbose(2, orte_debugger_base.output,
"%s: Spawned by a debugger",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* tell the procs they are being debugged */
env_name = mca_base_param_environ_variable("orte",
"in_parallel_debugger", NULL);
for (i=0; i < jdata->apps->size; i++) {
if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
continue;
}
opal_setenv(env_name, "1", true, &app->env);
}
free(env_name);
/* check if we need to co-spawn the debugger daemons */
if ('\0' != MPIR_executable_path[0] || NULL != orte_debugger_base.test_daemon) {
/* can only have one debugger */
if (NULL != orte_debugger_daemon) {
opal_output(0, "-------------------------------------------\n"
"Only one debugger can be used on a job.\n"
"-------------------------------------------\n");
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
return;
}
opal_output_verbose(2, orte_debugger_base.output,
"%s Cospawning debugger daemons %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == orte_debugger_base.test_daemon) ?
MPIR_executable_path : orte_debugger_base.test_daemon);
/* add debugger info to launch message */
orte_debugger_daemon = OBJ_NEW(orte_job_t);
/* create a jobid for these daemons - this is done solely
* to avoid confusing the rest of the system's bookkeeping
*/
orte_plm_base_create_jobid(orte_debugger_daemon);
/* flag the job as being debugger daemons */
orte_debugger_daemon->controls |= ORTE_JOB_CONTROL_DEBUGGER_DAEMON;
/* unless directed, we do not forward output */
if (!MPIR_forward_output) {
orte_debugger_daemon->controls &= ~ORTE_JOB_CONTROL_FORWARD_OUTPUT;
}
/* add it to the global job pool */
ljob = ORTE_LOCAL_JOBID(orte_debugger_daemon->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, orte_debugger_daemon);
/* create an app_context for the debugger daemon */
app = OBJ_NEW(orte_app_context_t);
if (NULL != orte_debugger_base.test_daemon) {
app->app = strdup(orte_debugger_base.test_daemon);
} else {
app->app = strdup((char*)MPIR_executable_path);
}
opal_argv_append_nosize(&app->argv, app->app);
build_debugger_args(app);
opal_pointer_array_add(orte_debugger_daemon->apps, app);
orte_debugger_daemon->num_apps = 1;
}
return;
}
static void open_fifo (void)
{
if (attach_fd > 0) {
close(attach_fd);
}
attach_fd = open(MPIR_attach_fifo, O_RDONLY | O_NONBLOCK, 0);
if (attach_fd < 0) {
opal_output(0, "%s unable to open debugger attach fifo",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return;
}
opal_output_verbose(2, orte_debugger_base.output,
"%s Monitoring debugger attach fifo %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
MPIR_attach_fifo);
opal_event_set(opal_event_base, &attach, attach_fd, OPAL_EV_READ, attach_debugger, NULL);
fifo_active = true;
opal_event_add(&attach, 0);
}
static void attach_debugger(int fd, short event, void *arg)
{
orte_app_context_t *app;
unsigned char fifo_cmd;
int rc;
int32_t ljob;
orte_job_t *jdata;
/* read the file descriptor to clear that event, if necessary */
if (fifo_active) {
opal_event_del(&attach);
fifo_active = false;
rc = read(attach_fd, &fifo_cmd, sizeof(fifo_cmd));
if (!rc) {
/* reopen device to clear hangup */
open_fifo();
return;
}
if (1 != fifo_cmd) {
/* ignore the cmd */
goto RELEASE;
}
}
if (!MPIR_being_debugged && !orte_debugger_base.test_attach) {
/* false alarm */
goto RELEASE;
}
opal_output_verbose(1, orte_debugger_base.output,
"%s Attaching debugger %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == orte_debugger_base.test_daemon) ? MPIR_executable_path : orte_debugger_base.test_daemon);
/* a debugger has attached! All the MPIR_Proctable
* data is already available, so we only need to
* check to see if we should spawn any daemons
*/
if ('\0' != MPIR_executable_path[0] || NULL != orte_debugger_base.test_daemon) {
/* can only have one debugger */
if (NULL != orte_debugger_daemon) {
opal_output(0, "-------------------------------------------\n"
"Only one debugger can be used on a job.\n"
"-------------------------------------------\n");
goto RELEASE;
}
opal_output_verbose(2, orte_debugger_base.output,
"%s Spawning debugger daemons %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == orte_debugger_base.test_daemon) ?
MPIR_executable_path : orte_debugger_base.test_daemon);
/* this will be launched just like a regular job,
* so we do not use the global orte_debugger_daemon
* as this is reserved for co-location upon startup
*/
jdata = OBJ_NEW(orte_job_t);
/* create a jobid for these daemons - this is done solely
* to avoid confusing the rest of the system's bookkeeping
*/
orte_plm_base_create_jobid(jdata);
/* flag the job as being debugger daemons */
jdata->controls |= ORTE_JOB_CONTROL_DEBUGGER_DAEMON;
/* unless directed, we do not forward output */
if (!MPIR_forward_output) {
jdata->controls &= ~ORTE_JOB_CONTROL_FORWARD_OUTPUT;
}
/* add it to the global job pool */
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* create an app_context for the debugger daemon */
app = OBJ_NEW(orte_app_context_t);
if (NULL != orte_debugger_base.test_daemon) {
app->app = strdup(orte_debugger_base.test_daemon);
} else {
app->app = strdup((char*)MPIR_executable_path);
}
jdata->state = ORTE_JOB_STATE_INIT;
opal_argv_append_nosize(&app->argv, app->app);
build_debugger_args(app);
opal_pointer_array_add(jdata->apps, app);
jdata->num_apps = 1;
/* setup the mapping policy to pernode so we get one
* daemon on each node
*/
jdata->map = OBJ_NEW(orte_job_map_t);
jdata->map->mapping = ORTE_MAPPING_PPR;
jdata->map->ppr = strdup("1:n");
/* now go ahead and spawn this job */
if (ORTE_SUCCESS != (rc = orte_plm.spawn(jdata))) {
ORTE_ERROR_LOG(rc);
}
}
RELEASE:
/* reset the read or timer event */
if (0 == orte_debugger_mpirx_check_rate) {
fifo_active = true;
opal_event_add(&attach, 0);
} else if (!MPIR_being_debugged) {
ORTE_TIMER_EVENT(orte_debugger_mpirx_check_rate, 0, attach_debugger);
}
/* notify the debugger that all is ready */
MPIR_Breakpoint();
}
static void build_debugger_args(orte_app_context_t *debugger)
{
int i, j;
char mpir_arg[MPIR_MAX_ARG_LENGTH];
if ('\0' != MPIR_server_arguments[0]) {
j=0;
memset(mpir_arg, 0, MPIR_MAX_ARG_LENGTH);
for (i=0; i < MPIR_MAX_ARG_LENGTH; i++) {
if (MPIR_server_arguments[i] == '\0') {
if (0 < j) {
opal_argv_append_nosize(&debugger->argv, mpir_arg);
memset(mpir_arg, 0, MPIR_MAX_ARG_LENGTH);
j=0;
}
} else {
mpir_arg[j] = MPIR_server_arguments[i];
j++;
}
}
}
}

26
orte/mca/debugger/mpirx/mpirx.h
Просмотреть файл

@ -1,26 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
*/
#ifndef DEBUGGER_MPIRX_H
#define DEBUGGER_MPIRX_H
#include "orte_config.h"
#include "orte/mca/debugger/debugger.h"
BEGIN_C_DECLS
ORTE_MODULE_DECLSPEC extern orte_debugger_base_component_t mca_debugger_mpirx_component;
extern int orte_debugger_mpirx_check_rate;
extern orte_debugger_base_module_t orte_debugger_mpirx_module;
END_C_DECLS
#endif

58
orte/mca/debugger/mpirx/mpirx_component.c
Просмотреть файл

@ -1,58 +0,0 @@
/* -*- C -*-
*
* Copyright (c) 2010 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "orte_config.h"
#include "orte/constants.h"
#include "mpirx.h"
int orte_debugger_mpirx_check_rate=0;
static int component_open(void);
static int component_query(mca_base_module_t **module, int *priority);
/*
* Struct of function pointers that need to be initialized
*/
orte_debugger_base_component_t mca_debugger_mpirx_component = {
{
ORTE_DEBUGGER_BASE_VERSION_1_0_0,
"mpirx", /* MCA module name */
ORTE_MAJOR_VERSION, /* MCA module major version */
ORTE_MINOR_VERSION, /* MCA module minor version */
ORTE_RELEASE_VERSION, /* MCA module release version */
component_open,
NULL,
component_query /* module query */
},
{
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
}
};
static int component_open(void)
{
mca_base_component_t *c = &mca_debugger_mpirx_component.base_version;
mca_base_param_reg_int(c, "check_rate",
"Set rate (in secs) for auto-detect of debugger attachment (0 => do not check)",
false, false, 0, &orte_debugger_mpirx_check_rate);
return ORTE_SUCCESS;
}
static int component_query(mca_base_module_t **module, int *priority)
{
*priority = 10;
*module = (mca_base_module_t *)&orte_debugger_mpirx_module;
return ORTE_SUCCESS;
}

4
orte/mca/errmgr/default_hnp/errmgr_default_hnp.c
Просмотреть файл

@ -41,7 +41,6 @@
#include "orte/mca/rmaps/rmaps_types.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/mca/routed/routed.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/grpcomm/grpcomm.h"
#include "orte/mca/ess/ess.h"
@ -526,9 +525,6 @@ static void default_hnp_abort(orte_jobid_t job, orte_exit_code_t exit_code)
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(job), exit_code));
/* if debuggers are running, clean up */
orte_debugger.finalize();
/* set control params to indicate we are terminating */
orte_job_term_ordered = true;
orte_abnormal_term_ordered = true;

20
orte/mca/ess/hnp/ess_hnp_module.c
Просмотреть файл

@ -58,8 +58,6 @@
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/sensor/base/base.h"
#include "orte/mca/sensor/sensor.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/debugger/debugger.h"
#include "orte/mca/rmaps/base/base.h"
#if OPAL_ENABLE_FT_CR == 1
#include "orte/mca/snapc/base/base.h"
@ -622,18 +620,6 @@ static int rte_init(void)
/* start the local sensors */
orte_sensor.start(ORTE_PROC_MY_NAME->jobid);
/* start the debuggers */
if (ORTE_SUCCESS != (ret = orte_debugger_base_open())) {
ORTE_ERROR_LOG(ret);
error = "orte_debugger_open";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_debugger_base_select())) {
ORTE_ERROR_LOG(ret);
error = "orte_debugger_select";
goto error;
}
/* if a tool has launched us and is requesting event reports,
* then set its contact info into the comm system
*/
@ -701,9 +687,6 @@ static int rte_finalize(void)
signals_set = false;
}
/* stop the debuggers */
orte_debugger_base_close();
/* stop the local sensors */
orte_sensor.stop(ORTE_PROC_MY_NAME->jobid);
@ -1003,9 +986,6 @@ static void abort_exit_callback(int fd, short ign, void *arg)
* to terminate!
*/
if (!orte_never_launched) {
/* if the debuggers were run, clean up */
orte_debugger.finalize();
/*
* Turn off the process recovery functionality, if it was enabled.
* This keeps the errmgr from trying to recover from the shutdown

8
orte/mca/plm/base/plm_base_launch_support.c
Просмотреть файл

@ -41,7 +41,6 @@
#include "opal/mca/hwloc/hwloc.h"
#include "orte/util/show_help.h"
#include "orte/mca/debugger/debugger.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/iof/iof.h"
@ -230,8 +229,6 @@ int orte_plm_base_launch_apps(orte_jobid_t job)
rc = ORTE_ERR_BAD_PARAM;
goto WAKEUP;
}
/* setup for debugging */
orte_debugger.init_before_spawn(jdata);
}
/* setup the buffer */
@ -310,10 +307,7 @@ int orte_plm_base_launch_apps(orte_jobid_t job)
ORTE_ERROR_LOG(rc);
goto WAKEUP;
}
/* complete debugger interface */
orte_debugger.init_after_spawn(jdata);
OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
"%s plm:base:launch completed for job %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),

3
orte/mca/plm/base/plm_base_receive.c
Просмотреть файл

@ -41,7 +41,6 @@
#include "orte/types.h"
#include "orte/util/proc_info.h"
#include "orte/util/error_strings.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/ess/ess.h"
#include "orte/mca/rml/rml.h"
@ -263,7 +262,7 @@ static void process_msg(int fd, short event, void *data)
job = jdata->jobid;
/* output debugger proctable, if requested */
if (orte_debugger_base.dump_proctable) {
if (orte_debugger_dump_proctable && !jdata->map->display_map) {
char *output;
opal_dss.print(&output, NULL, jdata->map, ORTE_JOB_MAP);
if (orte_xml_output) {

1
orte/mca/plm/rsh/plm_rsh_module.c
Просмотреть файл

@ -80,7 +80,6 @@
#include "orte/util/nidmap.h"
#include "orte/util/proc_info.h"
#include "orte/mca/debugger/debugger.h"
#include "orte/mca/rml/rml.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/mca/ess/ess.h"

5
orte/runtime/orte_globals.c
Просмотреть файл

@ -91,6 +91,11 @@ char **orte_fork_agent=NULL;
/* debugger job */
orte_job_t *orte_debugger_daemon=NULL;
bool orte_debugger_dump_proctable;
char *orte_debugger_test_daemon;
bool orte_debugger_test_attach;
bool orte_debugger_enable_fifo_attach;
int orte_debugger_check_rate;
/* exit flags */
int orte_exit_status = 0;

7
orte/runtime/orte_globals.h
Просмотреть файл

@ -78,8 +78,6 @@ ORTE_DECLSPEC extern orte_process_name_t orte_name_invalid; /** instantiated in
/* define the name of my daemon */
#define ORTE_PROC_MY_DAEMON (&orte_process_info.my_daemon)
/* See comment in orte/tools/orterun/debuggers.c about this MCA
param */
ORTE_DECLSPEC extern bool orte_in_parallel_debugger;
/* error manager callback function */
@ -559,6 +557,11 @@ ORTE_DECLSPEC extern char **orte_fork_agent;
/* debugger job */
ORTE_DECLSPEC extern orte_job_t *orte_debugger_daemon;
ORTE_DECLSPEC extern bool orte_debugger_dump_proctable;
ORTE_DECLSPEC extern char *orte_debugger_test_daemon;
ORTE_DECLSPEC extern bool orte_debugger_test_attach;
ORTE_DECLSPEC extern bool orte_debugger_enable_fifo_attach;
ORTE_DECLSPEC extern int orte_debugger_check_rate;
/* exit flags */
ORTE_DECLSPEC extern bool orte_abnormal_term_ordered;

27
orte/runtime/orte_mca_params.c
Просмотреть файл

@ -154,6 +154,33 @@ int orte_register_params(void)
true, false, 0, &value);
orte_in_parallel_debugger = OPAL_INT_TO_BOOL(value);
mca_base_param_reg_int_name("orte",
"output_debugger_proctable",
"Whether or not to output the debugger proctable after launch (default: false)",
false, false, 0, &value);
orte_debugger_dump_proctable = OPAL_INT_TO_BOOL(value);
mca_base_param_reg_string_name("orte", "debugger_test_daemon",
"Name of the executable to be used to simulate a debugger colaunch (relative or absolute path)",
false, false, NULL, &orte_debugger_test_daemon);
mca_base_param_reg_int_name("orte",
"debugger_test_attach",
"Test debugger colaunch after debugger attachment",
false, false, 0, &value);
orte_debugger_test_attach = OPAL_INT_TO_BOOL(value);
mca_base_param_reg_int_name("orte",
"debugger_fifo_attach",
"Create a fifo to support debugger attachment",
false, false, 0, &value);
orte_debugger_enable_fifo_attach = OPAL_INT_TO_BOOL(value);
mca_base_param_reg_int_name("orte",
"debugger_check_rate",
"Set rate (in secs) for auto-detect of debugger attachment (0 => do not check)",
false, false, 0, &orte_debugger_check_rate);
mca_base_param_reg_int_name("orte", "do_not_launch",
"Perform all necessary operations to prepare to launch the application, but do not actually launch it",
false, false, (int)false, &value);

4
orte/runtime/orte_quit.c
Просмотреть файл

@ -49,7 +49,6 @@
#include "orte/mca/plm/plm.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/debugger/debugger.h"
#include "orte/mca/routed/routed.h"
#include "orte/util/session_dir.h"
@ -128,9 +127,6 @@ void orte_jobs_complete(void)
}
}
/* if the debuggers were run, clean up */
orte_debugger.finalize();
if (0 < orte_routed.num_routes()) {
orte_plm.terminate_orteds();
}

10
orte/tools/orte-info/components.c
Просмотреть файл

@ -69,8 +69,6 @@
#if !ORTE_DISABLE_FULL_SUPPORT
#include "orte/mca/notifier/notifier.h"
#include "orte/mca/notifier/base/base.h"
#include "orte/mca/debugger/debugger.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/iof/iof.h"
#include "orte/mca/iof/base/base.h"
#include "orte/mca/oob/oob.h"
@ -373,14 +371,6 @@ void orte_info_open_components(void)
map->components = &orte_notifier_base_components_available;
opal_pointer_array_add(&component_map, map);
if (ORTE_SUCCESS != orte_debugger_base_open()) {
goto error;
}
map = OBJ_NEW(orte_info_component_map_t);
map->type = strdup("debugger");
map->components = &orte_debugger_base_components_available;
opal_pointer_array_add(&component_map, map);
if (ORTE_SUCCESS != mca_oob_base_open()) {
goto error;
}

1
orte/tools/orte-info/orte-info.c
Просмотреть файл

@ -205,7 +205,6 @@ int main(int argc, char *argv[])
opal_pointer_array_add(&mca_types, "event");
#if !ORTE_DISABLE_FULL_SUPPORT
opal_pointer_array_add(&mca_types, "debugger");
opal_pointer_array_add(&mca_types, "iof");
opal_pointer_array_add(&mca_types, "oob");
opal_pointer_array_add(&mca_types, "odls");

578
orte/tools/orterun/orterun.c
Просмотреть файл

@ -12,7 +12,7 @@
* All rights reserved.
* Copyright (c) 2006-2011 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2007-2009 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2007 Los Alamos National Security, LLC. All rights
* Copyright (c) 2007-2011 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
@ -28,6 +28,12 @@
#include <string.h>
#endif
#include <stdio.h>
#ifdef HAVE_STDLIB_H
#include <stdlib.h>
#endif /* HAVE_STDLIB_H */
#ifdef HAVE_STRINGS_H
#include <strings.h>
#endif /* HAVE_STRINGS_H */
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
@ -46,6 +52,10 @@
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif /* HAVE_SYS_TIME_H */
#include <fcntl.h>
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#include "opal/mca/event/event.h"
#include "opal/mca/installdirs/installdirs.h"
@ -75,7 +85,6 @@
#include "orte/util/session_dir.h"
#include "orte/util/hnp_contact.h"
#include "orte/mca/debugger/base/base.h"
#include "orte/mca/odls/odls.h"
#include "orte/mca/plm/plm.h"
#include "orte/mca/plm/base/plm_private.h"
@ -99,6 +108,37 @@
#include "orterun.h"
/* instance the standard MPIR interfaces */
#define MPIR_MAX_PATH_LENGTH 512
#define MPIR_MAX_ARG_LENGTH 1024
struct MPIR_PROCDESC *MPIR_proctable = NULL;
int MPIR_proctable_size = 0;
volatile int MPIR_being_debugged = 0;
volatile int MPIR_debug_state = 0;
int MPIR_i_am_starter = 0;
int MPIR_partial_attach_ok = 1;
char MPIR_executable_path[MPIR_MAX_PATH_LENGTH];
char MPIR_server_arguments[MPIR_MAX_ARG_LENGTH];
volatile int MPIR_forward_output = 0;
volatile int MPIR_forward_comm = 0;
char MPIR_attach_fifo[MPIR_MAX_PATH_LENGTH];
int MPIR_force_to_main = 0;
static void orte_debugger_dump(void);
static void orte_debugger_init_before_spawn(orte_job_t *jdata);
static void orte_debugger_init_after_spawn(orte_job_t *jdata);
static void attach_debugger(int fd, short event, void *arg);
static void build_debugger_args(orte_app_context_t *debugger);
static void open_fifo (void);
ORTE_DECLSPEC void* MPIR_Breakpoint(void);
/*
* Breakpoint function for parallel debuggers
*/
void* MPIR_Breakpoint(void)
{
return NULL;
}
/*
* Globals
*/
@ -549,49 +589,6 @@ int orterun(int argc, char *argv[])
true);
}
/* force the debugger symbols to be included in orterun.
* this is required since the symbols are instantiated in
* the orte library, yet they need to be accessed
* prior to orte_init when a debugger wants to launch
* us
*/
if (NULL == MPIR_proctable) {
rc = ORTE_SUCCESS;
}
if (0 == MPIR_proctable_size) {
rc = ORTE_SUCCESS;
}
if (0 == MPIR_being_debugged) {
rc = ORTE_SUCCESS;
}
if (0 == MPIR_debug_state) {
rc = ORTE_SUCCESS;
}
if (0 == MPIR_i_am_starter) {
rc = ORTE_SUCCESS;
}
if (1 == MPIR_partial_attach_ok) {
rc = ORTE_SUCCESS;
}
if (NULL == MPIR_executable_path) {
rc = ORTE_SUCCESS;
}
if (NULL == MPIR_server_arguments) {
rc = ORTE_SUCCESS;
}
if (0 == MPIR_forward_output) {
rc = ORTE_SUCCESS;
}
if (0 == MPIR_forward_comm) {
rc = ORTE_SUCCESS;
}
MPIR_force_to_main = 0;
memset(MPIR_attach_fifo, 0, MPIR_MAX_PATH_LENGTH);
/* This function call simply ensures that all the symbols --
including MPIR_Breakpoint -- are pulled in via the linker from
orte/mca/debugger/base/debugger_base_fns.c. */
orte_debugger_base_pull_mpir_breakpoint();
/* Check for some "global" command line params */
parse_globals(argc, argv, &cmd_line);
OBJ_DESTRUCT(&cmd_line);
@ -848,9 +845,15 @@ int orterun(int argc, char *argv[])
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* setup for debugging */
orte_debugger_init_before_spawn(jdata);
/* spawn the job and its daemons */
rc = orte_plm.spawn(jdata);
/* complete debugger interface */
orte_debugger_init_after_spawn(jdata);
/* now wait until the termination event fires */
opal_event_dispatch(opal_event_base);
@ -2110,3 +2113,488 @@ static void run_debugger(char *basename, opal_cmd_line_t *cmd_line,
opal_argv_free(new_argv);
exit(1);
}
/**** DEBUGGER CODE ****/
/*
* Debugger support for orterun
*
* We interpret the MPICH debugger interface as follows:
*
* a) The launcher
* - spawns the other processes,
* - fills in the table MPIR_proctable, and sets MPIR_proctable_size
* - sets MPIR_debug_state to MPIR_DEBUG_SPAWNED ( = 1)
* - calls MPIR_Breakpoint() which the debugger will have a
* breakpoint on.
*
* b) Applications start and then spin until MPIR_debug_gate is set
* non-zero by the debugger.
*
* This file implements (a).
*
**************************************************************************
*
* Note that we have presently tested both TotalView and DDT parallel
* debuggers. They both nominally subscribe to the Etnus attaching
* interface, but there are differences between the two.
*
* TotalView: user launches "totalview mpirun -a ...<mpirun args>...".
* TV launches mpirun. mpirun launches the application and then calls
* MPIR_Breakpoint(). This is the signal to TV that it's a parallel
* MPI job. TV then reads the proctable in mpirun and attaches itself
* to all the processes (it takes care of launching itself on the
* remote nodes). Upon attaching to all the MPI processes, the
* variable MPIR_being_debugged is set to 1. When it has finished
* attaching itself to all the MPI processes that it wants to,
* MPIR_Breakpoint() returns.
*
* DDT: user launches "ddt bin -np X <mpi app name>". DDT fork/exec's
* mpirun to launch ddt-debugger on the back-end nodes via "mpirun -np
* X ddt-debugger" (not the lack of other arguments -- we can't pass
* anything to mpirun). This app will eventually fork/exec the MPI
* app. DDT does not current set MPIR_being_debugged in the MPI app.
*
**************************************************************************
*
* We support two ways of waiting for attaching debuggers. The
* implementation spans this file and ompi/debuggers/ompi_debuggers.c.
*
* 1. If using orterun: MPI processes will have the
* orte_in_parallel_debugger MCA param set to true (because not all
* debuggers consistently set MPIR_being_debugged in both the launcher
* and in the MPI procs). The HNP will call MPIR_Breakpoint() and
* then RML send a message to VPID 0 (MCW rank 0) when it returns
* (MPIR_Breakpoint() doesn't return until the debugger has attached
* to all relevant processes). Meanwhile, VPID 0 blocks waiting for
* the RML message. All other VPIDs immediately call the grpcomm
* barrier (and therefore block until the debugger attaches). Once
* VPID 0 receives the RML message, we know that the debugger has
* attached to all processes that it cares about, and VPID 0 then
* joins the grpcomm barrier, allowing the job to continue. This
* scheme has the side effect of nicely supporting partial attaches by
* parallel debuggers (i.e., attaching to only some of the MPI
* processes; not necessarily all of them).
*
* 2. If not using orterun: in this case, ORTE_DISABLE_FULL_SUPPORT
* will be true, and we know that there will not be an RML message
* sent to VPID 0. So we have to look for a magic environment
* variable from the launcher to know if the jobs will be attached by
* a debugger (e.g., set by yod, srun, ...etc.), and if so, spin on
* MPIR_debug_gate. These environment variable names must be
* hard-coded in the OMPI layer (see ompi/debuggers/ompi_debuggers.c).
*/
/* local globals and functions */
static void attach_debugger(int fd, short event, void *arg);
static void build_debugger_args(orte_app_context_t *debugger);
static void open_fifo(void);
static opal_event_t attach;
static int attach_fd = -1;
static bool fifo_active=false;
#define DUMP_INT(X) fprintf(stderr, " %s = %d\n", # X, X);
#define FILE_MODE (S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH)
struct MPIR_PROCDESC {
char *host_name; /* something that can be passed to inet_addr */
char *executable_name; /* name of binary */
int pid; /* process pid */
};
static void orte_debugger_dump(void)
{
int i;
DUMP_INT(MPIR_being_debugged);
DUMP_INT(MPIR_debug_state);
DUMP_INT(MPIR_partial_attach_ok);
DUMP_INT(MPIR_i_am_starter);
DUMP_INT(MPIR_forward_output);
DUMP_INT(MPIR_proctable_size);
fprintf(stderr, " MPIR_proctable:\n");
for (i = 0; i < MPIR_proctable_size; i++) {
fprintf(stderr,
" (i, host, exe, pid) = (%d, %s, %s, %d)\n",
i,
MPIR_proctable[i].host_name,
MPIR_proctable[i].executable_name,
MPIR_proctable[i].pid);
}
fprintf(stderr, "MPIR_executable_path: %s\n",
('\0' == MPIR_executable_path[0]) ?
"NULL" : (char*) MPIR_executable_path);
fprintf(stderr, "MPIR_server_arguments: %s\n",
('\0' == MPIR_server_arguments[0]) ?
"NULL" : (char*) MPIR_server_arguments);
}
/**
* Initialization of data structures for running under a debugger
* using the MPICH/TotalView parallel debugger interface. Before the
* spawn we need to check if we are being run under a TotalView-like
* debugger; if so then inform applications via an MCA parameter.
*/
static void orte_debugger_init_before_spawn(orte_job_t *jdata)
{
char *env_name;
orte_app_context_t *app;
int i;
int32_t ljob;
char *attach_fifo;
if (!MPIR_being_debugged && !orte_in_parallel_debugger) {
/* if we were given a test debugger, then we still want to
* colaunch it
*/
if (NULL != orte_debugger_test_daemon) {
opal_output_verbose(2, orte_debug_output,
"%s No debugger test daemon specified",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
goto launchit;
}
/* if we were given an auto-detect rate, then we want to setup
* an event so we periodically do the check
*/
if (0 < orte_debugger_check_rate) {
opal_output_verbose(2, orte_debug_output,
"%s Setting debugger attach check rate for %d seconds",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
orte_debugger_check_rate);
ORTE_TIMER_EVENT(orte_debugger_check_rate, 0, attach_debugger);
} else if (orte_debugger_enable_fifo_attach) {
/* create the attachment FIFO and put it into MPIR, setup readevent */
/* create a FIFO name in the session dir */
attach_fifo = opal_os_path(false, orte_process_info.job_session_dir, "debugger_attach_fifo", NULL);
if ((mkfifo(attach_fifo, FILE_MODE) < 0) && errno != EEXIST) {
opal_output(0, "CANNOT CREATE FIFO %s: errno %d", attach_fifo, errno);
free(attach_fifo);
return;
}
strncpy(MPIR_attach_fifo, attach_fifo, MPIR_MAX_PATH_LENGTH - 1);
free(attach_fifo);
open_fifo();
}
return;
}
launchit:
opal_output_verbose(1, orte_debug_output, "Info: Spawned by a debugger");
/* tell the procs they are being debugged */
env_name = mca_base_param_environ_variable("orte",
"in_parallel_debugger", NULL);
for (i=0; i < jdata->apps->size; i++) {
if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
continue;
}
opal_setenv(env_name, "1", true, &app->env);
}
free(env_name);
/* check if we need to co-spawn the debugger daemons */
if ('\0' != MPIR_executable_path[0] || NULL != orte_debugger_test_daemon) {
/* can only have one debugger */
if (NULL != orte_debugger_daemon) {
opal_output(0, "-------------------------------------------\n"
"Only one debugger can be used on a job.\n"
"-------------------------------------------\n");
ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
return;
}
opal_output_verbose(2, orte_debug_output,
"%s Cospawning debugger daemons %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == orte_debugger_test_daemon) ?
MPIR_executable_path : orte_debugger_test_daemon);
/* add debugger info to launch message */
orte_debugger_daemon = OBJ_NEW(orte_job_t);
/* create a jobid for these daemons - this is done solely
* to avoid confusing the rest of the system's bookkeeping
*/
orte_plm_base_create_jobid(orte_debugger_daemon);
/* flag the job as being debugger daemons */
orte_debugger_daemon->controls |= ORTE_JOB_CONTROL_DEBUGGER_DAEMON;
/* unless directed, we do not forward output */
if (!MPIR_forward_output) {
orte_debugger_daemon->controls &= ~ORTE_JOB_CONTROL_FORWARD_OUTPUT;
}
/* add it to the global job pool */
ljob = ORTE_LOCAL_JOBID(orte_debugger_daemon->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, orte_debugger_daemon);
/* create an app_context for the debugger daemon */
app = OBJ_NEW(orte_app_context_t);
if (NULL != orte_debugger_test_daemon) {
app->app = strdup(orte_debugger_test_daemon);
} else {
app->app = strdup((char*)MPIR_executable_path);
}
opal_argv_append_nosize(&app->argv, app->app);
build_debugger_args(app);
opal_pointer_array_add(orte_debugger_daemon->apps, app);
orte_debugger_daemon->num_apps = 1;
}
}
/*
* Initialization of data structures for running under a debugger
* using the MPICH/TotalView parallel debugger interface. This stage
* of initialization must occur after spawn
*
* NOTE: We -always- perform this step to ensure that any debugger
* that attaches to us post-launch of the application can get a
* completed proctable
*/
static void orte_debugger_init_after_spawn(orte_job_t *jdata)
{
orte_proc_t *proc;
orte_app_context_t *appctx;
orte_vpid_t i, j;
opal_buffer_t buf;
orte_process_name_t rank0;
int rc;
/* if we couldn't get thru the mapper stage, we might
* enter here with no procs. Avoid the "zero byte malloc"
* message by checking here
*/
if (MPIR_proctable || 0 == jdata->num_procs) {
/* already initialized */
opal_output_verbose(5, orte_debug_output,
"%s: debugger already initialized or zero procs",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return;
}
/* fill in the proc table for the application processes */
opal_output_verbose(5, orte_debug_output,
"%s: Setting up debugger process table for applications",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
MPIR_debug_state = 1;
/* set the total number of processes in the job */
MPIR_proctable_size = jdata->num_procs;
/* allocate MPIR_proctable */
MPIR_proctable = (struct MPIR_PROCDESC *)malloc(sizeof(struct MPIR_PROCDESC) *
MPIR_proctable_size);
if (MPIR_proctable == NULL) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return;
}
if (orte_debugger_dump_proctable) {
opal_output(orte_clean_output, "MPIR Proctable for job %s", ORTE_JOBID_PRINT(jdata->jobid));
}
/* initialize MPIR_proctable */
for (j=0; j < jdata->num_procs; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, j))) {
continue;
}
/* store this data in the location whose index
* corresponds to the proc's rank
*/
i = proc->name.vpid;
if (NULL == (appctx = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, proc->app_idx))) {
continue;
}
MPIR_proctable[i].host_name = strdup(proc->node->name);
if ( 0 == strncmp(appctx->app, OPAL_PATH_SEP, 1 )) {
MPIR_proctable[i].executable_name =
opal_os_path( false, appctx->app, NULL );
} else {
MPIR_proctable[i].executable_name =
opal_os_path( false, appctx->cwd, appctx->app, NULL );
}
MPIR_proctable[i].pid = proc->pid;
if (orte_debugger_dump_proctable) {
opal_output(orte_clean_output, "%s: Host %s Exe %s Pid %d",
ORTE_VPID_PRINT(i), MPIR_proctable[i].host_name,
MPIR_proctable[i].executable_name, MPIR_proctable[i].pid);
}
}
if (0 < opal_output_get_verbosity(orte_debug_output)) {
orte_debugger_dump();
}
/* if we are being launched under a debugger, then we must wait
* for it to be ready to go and do some things to start the job
*/
if (MPIR_being_debugged) {
/* wait for all procs to have reported their contact info - this
* ensures that (a) they are all into mpi_init, and (b) the system
* has the contact info to successfully send a message to rank=0
*/
ORTE_PROGRESSED_WAIT(false, jdata->num_reported, jdata->num_procs);
MPIR_Breakpoint();
/* send a message to rank=0 to release it */
OBJ_CONSTRUCT(&buf, opal_buffer_t); /* don't need anything in this */
rank0.jobid = jdata->jobid;
rank0.vpid = 0;
if (0 > (rc = orte_rml.send_buffer(&rank0, &buf, ORTE_RML_TAG_DEBUGGER_RELEASE, 0))) {
opal_output(0, "Error: could not send debugger release to MPI procs - error %s", ORTE_ERROR_NAME(rc));
}
OBJ_DESTRUCT(&buf);
}
}
static void open_fifo (void)
{
if (attach_fd > 0) {
close(attach_fd);
}
attach_fd = open(MPIR_attach_fifo, O_RDONLY | O_NONBLOCK, 0);
if (attach_fd < 0) {
opal_output(0, "%s unable to open debugger attach fifo",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return;
}
opal_output_verbose(2, orte_debug_output,
"%s Monitoring debugger attach fifo %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
MPIR_attach_fifo);
opal_event_set(opal_event_base, &attach, attach_fd, OPAL_EV_READ, attach_debugger, NULL);
fifo_active = true;
opal_event_add(&attach, 0);
}
static void attach_debugger(int fd, short event, void *arg)
{
orte_app_context_t *app;
unsigned char fifo_cmd;
int rc;
int32_t ljob;
orte_job_t *jdata;
/* read the file descriptor to clear that event, if necessary */
if (fifo_active) {
opal_event_del(&attach);
fifo_active = false;
rc = read(attach_fd, &fifo_cmd, sizeof(fifo_cmd));
if (!rc) {
/* reopen device to clear hangup */
open_fifo();
return;
}
if (1 != fifo_cmd) {
/* ignore the cmd */
goto RELEASE;
}
}
if (!MPIR_being_debugged && !orte_debugger_test_attach) {
/* false alarm */
goto RELEASE;
}
opal_output_verbose(1, orte_debug_output,
"%s Attaching debugger %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == orte_debugger_test_daemon) ? MPIR_executable_path : orte_debugger_test_daemon);
/* a debugger has attached! All the MPIR_Proctable
* data is already available, so we only need to
* check to see if we should spawn any daemons
*/
if ('\0' != MPIR_executable_path[0] || NULL != orte_debugger_test_daemon) {
/* can only have one debugger */
if (NULL != orte_debugger_daemon) {
opal_output(0, "-------------------------------------------\n"
"Only one debugger can be used on a job.\n"
"-------------------------------------------\n");
goto RELEASE;
}
opal_output_verbose(2, orte_debug_output,
"%s Spawning debugger daemons %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
(NULL == orte_debugger_test_daemon) ?
MPIR_executable_path : orte_debugger_test_daemon);
/* this will be launched just like a regular job,
* so we do not use the global orte_debugger_daemon
* as this is reserved for co-location upon startup
*/
jdata = OBJ_NEW(orte_job_t);
/* create a jobid for these daemons - this is done solely
* to avoid confusing the rest of the system's bookkeeping
*/
orte_plm_base_create_jobid(jdata);
/* flag the job as being debugger daemons */
jdata->controls |= ORTE_JOB_CONTROL_DEBUGGER_DAEMON;
/* unless directed, we do not forward output */
if (!MPIR_forward_output) {
jdata->controls &= ~ORTE_JOB_CONTROL_FORWARD_OUTPUT;
}
/* add it to the global job pool */
ljob = ORTE_LOCAL_JOBID(jdata->jobid);
opal_pointer_array_set_item(orte_job_data, ljob, jdata);
/* create an app_context for the debugger daemon */
app = OBJ_NEW(orte_app_context_t);
if (NULL != orte_debugger_test_daemon) {
app->app = strdup(orte_debugger_test_daemon);
} else {
app->app = strdup((char*)MPIR_executable_path);
}
jdata->state = ORTE_JOB_STATE_INIT;
opal_argv_append_nosize(&app->argv, app->app);
build_debugger_args(app);
opal_pointer_array_add(jdata->apps, app);
jdata->num_apps = 1;
/* setup the mapping policy to pernode so we get one
* daemon on each node
*/
jdata->map = OBJ_NEW(orte_job_map_t);
jdata->map->mapping = ORTE_MAPPING_PPR;
jdata->map->ppr = strdup("1:n");
/* now go ahead and spawn this job */
if (ORTE_SUCCESS != (rc = orte_plm.spawn(jdata))) {
ORTE_ERROR_LOG(rc);
}
}
RELEASE:
/* reset the read or timer event */
if (0 == orte_debugger_check_rate) {
fifo_active = true;
opal_event_add(&attach, 0);
} else if (!MPIR_being_debugged) {
ORTE_TIMER_EVENT(orte_debugger_check_rate, 0, attach_debugger);
}
/* notify the debugger that all is ready */
MPIR_Breakpoint();
}
static void build_debugger_args(orte_app_context_t *debugger)
{
int i, j;
char mpir_arg[MPIR_MAX_ARG_LENGTH];
if ('\0' != MPIR_server_arguments[0]) {
j=0;
memset(mpir_arg, 0, MPIR_MAX_ARG_LENGTH);
for (i=0; i < MPIR_MAX_ARG_LENGTH; i++) {
if (MPIR_server_arguments[i] == '\0') {
if (0 < j) {
opal_argv_append_nosize(&debugger->argv, mpir_arg);
memset(mpir_arg, 0, MPIR_MAX_ARG_LENGTH);
j=0;
}
} else {
mpir_arg[j] = MPIR_server_arguments[i];
j++;
}
}
}
}