1
1
openmpi/ompi/debuggers/ompi_debuggers.c
Rainer Keller 6c5532072a - Split the datatype engine into two parts: an MPI specific part in
OMPI
   and a language agnostic part in OPAL. The convertor is completely
   moved into OPAL.  This offers several benefits as described in RFC
   http://www.open-mpi.org/community/lists/devel/2009/07/6387.php
   namely:
    - Fewer basic types (int* and float* types, boolean and wchar
    - Fixing naming scheme to ompi-nomenclature.
    - Usability outside of the ompi-layer.
 - Due to the fixed nature of simple opal types, their information is
   completely
   known at compile time and therefore constified
 - With fewer datatypes (22), the actual sizes of bit-field types may be
   reduced
   from 64 to 32 bits, allowing reorganizing the opal_datatype
   structure, eliminating holes and keeping data required in convertor
   (upon send/recv) in one cacheline...
   This has implications to the convertor-datastructure and other parts
   of the code.
 - Several performance tests have been run, the netpipe latency does not
   change with
   this patch on Linux/x86-64 on the smoky cluster.
 - Extensive tests have been done to verify correctness (no new
   regressions) using:
   1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and
    ompi-ddt:
    a. running both trunk and ompi-ddt resulted in no differences
       (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run
       correctly).
    b. with --enable-memchecker and running under valgrind (one buglet
       when run with static found in test-suite, commited)
   2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt:
      all passed (except for the dynamic/ tests failed!! as trunk/MTT)
   3. compilation and usage of HDF5 tests on Jaguar using PGI and
      PathScale compilers.
   4. compilation and usage on Scicortex.
 - Please note, that for the heterogeneous case, (-m32 compiled
   binaries/ompi), neither
   ompi-trunk, nor ompi-ddt branch would successfully launch.

This commit was SVN r21641.
2009-07-13 04:56:31 +00:00

285 строки
9.5 KiB
C

/* -*- Mode: C; c-basic-offset:4 ; -*- */
/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2008 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) 2007-2008 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/*
* MPI portion of debugger support: initially based on the
* TotalView/Etnus API for debuggers to attach to MPI jobs.
*
* There is a lengthy explanation of how OMPI handles parallel
* debuggers attaching to MPI jobs in orte/tools/orterun/debuggers.c.
*/
#include "ompi_config.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif /* HAVE_UNISTD_H */
#ifdef HAVE_DIRENT_H
#include <dirent.h>
#endif
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include "opal/mca/base/base.h"
#include "opal/util/argv.h"
#include "opal/mca/installdirs/installdirs.h"
#include "debuggers.h"
/**
* BEWARE: The following headers are required by optimized builds in order
* to get access to the type information. Some compilers remove all type
* information on optimized build, and as a result we are unable to access
* the fields structure (i.e. to get their displacement). This file is
* included in the optimized build just to provide us with this missing
* informations. Therefore, it always have to be compiled with the -g flag,
* otherwise the type information will be missing and the parallel
* debuggers will be unable to initialize the Open MPI debug library.
*/
#include "opal/class/opal_list.h"
#include "ompi/class/ompi_free_list.h"
#include "ompi/request/request.h"
#include "ompi/mca/pml/base/pml_base_request.h"
#include "ompi/mca/pml/base/pml_base_sendreq.h"
#include "ompi/mca/pml/base/pml_base_recvreq.h"
#include "opal/class/opal_pointer_array.h"
#include "ompi/communicator/communicator.h"
#include "ompi/group/group.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/include/mpi.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/rml/rml.h"
#include "orte/runtime/orte_globals.h"
#if defined(OMPI_MSGQ_DLL)
/* This variable is old/deprecated -- the mpimsgq_dll_locations[]
method is preferred because it's more flexible */
OMPI_DECLSPEC char MPIR_dll_name[] = OMPI_MSGQ_DLL;
#endif /* defined(OMPI_MSGQ_DLL) */
OMPI_DECLSPEC char **mpidbg_dll_locations = NULL;
OMPI_DECLSPEC char **mpimsgq_dll_locations = NULL;
OMPI_DECLSPEC int MPIR_debug_typedefs_sizeof[] = {
sizeof(short),
sizeof(int),
sizeof(long),
sizeof(long long),
sizeof(void*),
sizeof(bool),
sizeof(size_t)
};
/*
* Values defined by the standardized interface; do not change these
* values
*/
#define MPIR_DEBUG_SPAWNED 1
#define MPIR_DEBUG_ABORTING 2
/**
* BEWARE: Try to outsmart some compilers. In some cases, when variables
* are defined but not used, some compilers will optimized them out from
* the build. As we need to be able to access the structure sizes from the
* debugged program (in the case where the MPI library and the application
* is compiled with a different architecture flag than the parallel
* debugger, 32 vs. 64 bits), we have to have these variables defined.
*/
OMPI_DECLSPEC opal_list_item_t* opal_list_item_t_type_force_inclusion = NULL;
OMPI_DECLSPEC opal_list_t* opal_list_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_free_list_item_t* ompi_free_list_item_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_free_list_t* ompi_free_list_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_request_t* ompi_request_t_type_force_inclusion = NULL;
OMPI_DECLSPEC mca_pml_base_request_t* mca_pml_base_request_t_type_force_inclusion = NULL;
OMPI_DECLSPEC mca_pml_base_send_request_t* mca_pml_base_send_request_t_type_force_inclusion = NULL;
OMPI_DECLSPEC mca_pml_base_recv_request_t* mca_pml_base_recv_request_t_type_force_inclusion = NULL;
OMPI_DECLSPEC opal_pointer_array_t* opal_pointer_array_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_communicator_t* ompi_communicator_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_group_t* ompi_group_t_type_force_inclusion = NULL;
OMPI_DECLSPEC ompi_status_public_t* ompi_status_public_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 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 */
static void check(char *dir, char *file, char **locations)
{
char *str;
asprintf(&str, "%s/%s.so", dir, file);
#if defined(HAVE_SYS_STAT_H)
{
struct stat buf;
/* Use stat() */
if (0 == stat(str, &buf)) {
opal_argv_append_nosize(&locations, file);
}
}
#else
{
FILE *fp;
/* Just try to open the file */
if (NULL != (fp = fopen(str, "r"))) {
fclose(fp);
opal_argv_append_nosize(&locations, file);
}
}
#endif /* defined(HAVE_SYS_STAT_H) */
free(str);
}
/*
* Wait for a debugger if asked. We support two ways of waiting for
* attaching debuggers -- see big comment in
* orte/tools/orterun/debuggers.c explaning the two scenarios.
*/
void ompi_wait_for_debugger(void)
{
int i, debugger, rc;
char *a, *b, **dirs, **tmp1 = NULL, **tmp2 = NULL;
opal_buffer_t buf;
/* See lengthy comment in orte/tools/orterun/debuggers.c about
orte_in_parallel_debugger */
#if ORTE_DISABLE_FULL_SUPPORT
debugger = 0;
#else
debugger = orte_in_parallel_debugger;
#endif
/* Add in environment variables for other launchers, such as yod,
srun, ...etc. */
if (1 == MPIR_being_debugged) {
debugger = 1;
} else if (NULL != getenv("yod_you_are_being_debugged")) {
debugger = 1;
}
if (1 == MPIR_being_debugged) {
debugger = 1;
}
if (!debugger) {
/* if not, just return */
return;
}
/* if we are being debugged, then we need to find
* the correct plug-ins
*/
a = strdup(opal_install_dirs.pkglibdir);
mca_base_param_reg_string_name("ompi",
"debugger_dll_path",
"List of directories where MPI_INIT should search for debugger plugins",
false, false, a, &b);
free(a);
/* Search the directory for MPI debugger DLLs */
if (NULL != b) {
dirs = opal_argv_split(b, ':');
for (i = 0; dirs[i] != NULL; ++i) {
check(dirs[i], OMPI_MPIHANDLES_DLL_PREFIX, tmp1);
check(dirs[i], OMPI_MSGQ_DLL_PREFIX, tmp2);
}
}
/* Now that we have a full list of directories, assign the argv
arrays to the global variables (since the debugger may read the
global variables at any time, we want to ensure that they have
non-NULL values only when the entire array is ready). */
mpimsgq_dll_locations = tmp1;
mpidbg_dll_locations = tmp2;
if (ORTE_DISABLE_FULL_SUPPORT) {
/* spin until debugger attaches and releases us */
while (MPIR_debug_gate == 0) {
#if defined(__WINDOWS__)
Sleep(100); /* milliseconds */
#elif defined(HAVE_USLEEP)
usleep(100000); /* microseconds */
#else
sleep(1); /* seconds */
#endif
}
} else {
/* only the rank=0 proc waits for either a message from the
* HNP or for the debugger to attach - everyone else will just
* spin in * the grpcomm barrier in ompi_mpi_init until rank=0
* joins them.
*/
if (0 != ORTE_PROC_MY_NAME->vpid) {
return;
}
/* VPID 0 waits for a message from the HNP */
OBJ_CONSTRUCT(&buf, opal_buffer_t);
rc = orte_rml.recv_buffer(ORTE_NAME_WILDCARD, &buf,
ORTE_RML_TAG_DEBUGGER_RELEASE, 0);
OBJ_DESTRUCT(&buf); /* don't care about contents of message */
if (rc < 0) {
/* if it failed for some reason, then we are in trouble -
* for now, just report the problem and give up waiting
*/
opal_output(0, "Debugger_attach[rank=%ld]: could not wait for debugger - error %s!",
(long)ORTE_PROC_MY_NAME->vpid, ORTE_ERROR_NAME(rc));
}
}
}
/*
* Breakpoint function for parallel debuggers. This function is also
* defined in orterun for the starter. It should never conflict with
* this one, but we'll make it static, just to be sure.
*/
static void *MPIR_Breakpoint(void)
{
return NULL;
}
/*
* Tell the debugger that we are about to abort
*/
void ompi_debugger_notify_abort(char *reason)
{
MPIR_debug_state = MPIR_DEBUG_ABORTING;
if (NULL != reason && strlen(reason) > 0) {
MPIR_debug_abort_string = reason;
} else {
MPIR_debug_abort_string = "Unknown";
}
/* Now tell the debugger */
MPIR_Breakpoint();
}