ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
aca71521db
openmpi/ompi/mca/btl/udapl/btl_udapl_component.c
Andrew Friedley c68c6ac122 A number of fixes and the usual cleanup.. 
- Added some basic flow control to limit number of posted sends.
- Merged endpoint send/recv lock into single endpoint lock.
- Set the LMR triplet length in the send path, not at allocation time.
  This has to be done because upper layers might send less than the
  amount allocated.
- Alter the tie-breaker if statement protecting the second call
  to dat_ep_connect().  The logic was reversed compared to the tie-
  breaker for the first dat_ep_connect(), making it possible for
  3 or more processes to form a deadlock loop.
- Some asserts were added for debugging purposes.. leaving them
  in place for now.

This commit was SVN r10317.
2006-06-12 22:42:01 +00:00

619 строки
22 KiB
C
Исходник Ответственный История

/*
* Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana
* 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.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2006 Sandia National Laboratories. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "ompi/constants.h"
#include "opal/event/event.h"
#include "opal/util/if.h"
#include "opal/util/argv.h"
#include "opal/util/output.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/mca/btl/btl.h"
#include "opal/mca/base/mca_base_param.h"
#include "orte/mca/errmgr/errmgr.h"
#include "ompi/mca/mpool/base/base.h"
#include "ompi/mca/mpool/udapl/mpool_udapl.h"
#include "btl_udapl.h"
#include "btl_udapl_frag.h"
#include "btl_udapl_endpoint.h"
#include "btl_udapl_proc.h"
#include "ompi/mca/btl/base/base.h"
#include "ompi/mca/btl/base/btl_base_error.h"
#include "ompi/datatype/convertor.h"
#include "btl_udapl_endpoint.h"
#include "orte/util/proc_info.h"
#include "ompi/mca/pml/base/pml_base_module_exchange.h"
mca_btl_udapl_component_t mca_btl_udapl_component = {
{
/* First, the mca_base_component_t struct containing meta information
about the component itself */
{
/* Indicate that we are a pml v1.0.0 component (which also implies a
specific MCA version) */
MCA_BTL_BASE_VERSION_1_0_0,
"udapl", /* MCA component name */
OMPI_MAJOR_VERSION, /* MCA component major version */
OMPI_MINOR_VERSION, /* MCA component minor version */
OMPI_RELEASE_VERSION, /* MCA component release version */
mca_btl_udapl_component_open, /* component open */
mca_btl_udapl_component_close /* component close */
},
/* Next the MCA v1.0.0 component meta data */
{
/* Whether the component is checkpointable or not */
false
},
mca_btl_udapl_component_init,
mca_btl_udapl_component_progress,
}
};
/**
* Report a uDAPL error - for debugging
*/
#if OMPI_ENABLE_DEBUG
void
mca_btl_udapl_error(DAT_RETURN ret, char* str)
{
char* major;
char* minor;
if(DAT_SUCCESS != dat_strerror(ret,
(const char**)&major, (const char**)&minor))
{
printf("dat_strerror failed! ret is %d\n", ret);
exit(-1);
}
OPAL_OUTPUT((0, "ERROR: %s %s %s\n", str, major, minor));
}
#endif
/*
* Utility routines for parameter registration
*/
static inline char* mca_btl_udapl_param_register_string(
const char* param_name,
const char* default_value)
{
char *param_value;
int id = mca_base_param_register_string("btl","udapl",param_name,NULL,default_value);
mca_base_param_lookup_string(id, &param_value);
return param_value;
}
static inline int mca_btl_udapl_param_register_int(
const char* param_name,
int default_value)
{
int id = mca_base_param_register_int("btl","udapl",param_name,NULL,default_value);
int param_value = default_value;
mca_base_param_lookup_int(id,&param_value);
return param_value;
}
/*
* Called by MCA framework to open the component, registers
* component parameters.
*/
int mca_btl_udapl_component_open(void)
{
int param, value;
/* initialize state */
mca_btl_udapl_component.udapl_num_btls=0;
mca_btl_udapl_component.udapl_btls=NULL;
/* initialize objects */
OBJ_CONSTRUCT(&mca_btl_udapl_component.udapl_procs, opal_list_t);
OBJ_CONSTRUCT(&mca_btl_udapl_component.udapl_lock, opal_mutex_t);
/* register uDAPL component parameters */
mca_btl_udapl_component.udapl_free_list_num =
mca_btl_udapl_param_register_int("free_list_num", 8);
mca_btl_udapl_component.udapl_free_list_max =
mca_btl_udapl_param_register_int("free_list_max", -1);
mca_btl_udapl_component.udapl_free_list_inc =
mca_btl_udapl_param_register_int("free_list_inc", 8);
mca_btl_udapl_component.udapl_mpool_name =
mca_btl_udapl_param_register_string("mpool", "udapl");
mca_btl_udapl_component.udapl_max_btls =
mca_btl_udapl_param_register_int("max_modules", 8);
mca_btl_udapl_component.udapl_evd_qlen =
mca_btl_udapl_param_register_int("evd_qlen", 32);
mca_btl_udapl_component.udapl_num_recvs =
mca_btl_udapl_param_register_int("num_recvs", 8);
mca_btl_udapl_component.udapl_num_sends =
mca_btl_udapl_param_register_int("num_sends", 8);
mca_btl_udapl_component.udapl_port_low =
mca_btl_udapl_param_register_int("port_low", 45000);
mca_btl_udapl_component.udapl_port_high =
mca_btl_udapl_param_register_int("port_high", 49000);
mca_btl_udapl_component.udapl_timeout =
mca_btl_udapl_param_register_int("timeout", 10000000);
/* register uDAPL module parameters */
mca_btl_udapl_module.super.btl_exclusivity =
mca_btl_udapl_param_register_int ("exclusivity",
MCA_BTL_EXCLUSIVITY_DEFAULT - 10);
mca_btl_udapl_module.super.btl_eager_limit =
mca_btl_udapl_param_register_int ("eager_limit", 32*1024);
mca_btl_udapl_module.super.btl_min_send_size =
mca_btl_udapl_param_register_int ("min_send_size", 16*1024);
mca_btl_udapl_module.super.btl_max_send_size =
mca_btl_udapl_param_register_int ("max_send_size", 64*1024);
mca_btl_udapl_module.super.btl_min_rdma_size =
mca_btl_udapl_param_register_int("min_rdma_size", 512*1024);
mca_btl_udapl_module.super.btl_max_rdma_size =
mca_btl_udapl_param_register_int("max_rdma_size", 128*1024);
mca_btl_udapl_module.super.btl_bandwidth =
mca_btl_udapl_param_register_int("bandwidth", 225);
/* compute udapl_eager_frag_size and udapl_max_frag_size */
mca_btl_udapl_component.udapl_eager_frag_size =
mca_btl_udapl_module.super.btl_eager_limit;
mca_btl_udapl_module.super.btl_eager_limit -=
sizeof(mca_btl_base_header_t);
mca_btl_udapl_component.udapl_max_frag_size =
mca_btl_udapl_module.super.btl_max_send_size;
mca_btl_udapl_module.super.btl_max_send_size -=
sizeof(mca_btl_base_header_t);
/* leave pinned option */
value = 0;
param = mca_base_param_find("mpi", NULL, "leave_pinned");
mca_base_param_lookup_int(param, &value);
mca_btl_udapl_component.leave_pinned = value;
return OMPI_SUCCESS;
}
/*
* component cleanup - sanity checking of queue lengths
*/
int mca_btl_udapl_component_close(void)
{
/* TODO - what needs to be done here? */
return OMPI_SUCCESS;
}
/*
* Register uDAPL component addressing information. The MCA framework
* will make this available to all peers.
*/
static int
mca_btl_udapl_modex_send(void)
{
int rc;
size_t i;
size_t size;
mca_btl_udapl_addr_t *addrs = NULL;
size = sizeof(mca_btl_udapl_addr_t) *
mca_btl_udapl_component.udapl_num_btls;
if (0 != size) {
addrs = (mca_btl_udapl_addr_t*)malloc(size);
if (NULL == addrs) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
for (i = 0; i < mca_btl_udapl_component.udapl_num_btls; i++) {
mca_btl_udapl_module_t* btl = mca_btl_udapl_component.udapl_btls[i];
addrs[i] = btl->udapl_addr;
}
}
rc = mca_pml_base_modex_send(
&mca_btl_udapl_component.super.btl_version, addrs, size);
if (NULL != addrs) {
free (addrs);
}
return rc;
}
/*
* Initialize the uDAPL component,
* check how many interfaces are available and create a btl module for each.
*/
mca_btl_base_module_t **
mca_btl_udapl_component_init (int *num_btl_modules,
bool enable_progress_threads,
bool enable_mpi_threads)
{
DAT_PROVIDER_INFO* datinfo;
DAT_PROVIDER_INFO** datinfoptr;
mca_btl_base_module_t **btls;
mca_btl_udapl_module_t *btl;
DAT_COUNT num_ias;
int32_t i;
/* enumerate uDAPL interfaces */
/* Have to do weird pointer stuff to make uDAPL happy -
just an array of DAT_PROVIDER_INFO isn't good enough. */
datinfo = malloc(sizeof(DAT_PROVIDER_INFO) *
mca_btl_udapl_component.udapl_max_btls);
datinfoptr = malloc(sizeof(DAT_PROVIDER_INFO*) *
mca_btl_udapl_component.udapl_max_btls);
if(NULL == datinfo || NULL == datinfoptr) {
return NULL;
}
for(i = 0; i < (int32_t)mca_btl_udapl_component.udapl_max_btls; i++) {
datinfoptr[i] = &datinfo[i];
}
if(DAT_SUCCESS != dat_registry_list_providers(
mca_btl_udapl_component.udapl_max_btls,
(DAT_COUNT*)&num_ias, datinfoptr)) {
free(datinfo);
free(datinfoptr);
return NULL;
}
free(datinfoptr);
/* allocate space for the each possible BTL */
mca_btl_udapl_component.udapl_btls = (mca_btl_udapl_module_t **)
malloc(num_ias * sizeof(mca_btl_udapl_module_t *));
if(NULL == mca_btl_udapl_component.udapl_btls) {
free(datinfo);
return NULL;
}
/* create a BTL module for each interface */
for(mca_btl_udapl_component.udapl_num_btls = i = 0; i < num_ias; i++) {
btl = malloc(sizeof(mca_btl_udapl_module_t));
if(NULL == btl) {
free(datinfo);
free(mca_btl_udapl_component.udapl_btls);
return NULL;
}
/* copy default values into the new BTL */
memcpy(btl, &mca_btl_udapl_module, sizeof(mca_btl_udapl_module_t));
/* initialize this BTL */
/* TODO - make use of the thread-safety info in datinfo also */
if(OMPI_SUCCESS != mca_btl_udapl_init(datinfo[i].ia_name, btl)) {
free(btl);
continue;
}
/* successful btl creation */
mca_btl_udapl_component.udapl_btls[mca_btl_udapl_component.udapl_num_btls] = btl;
if(++mca_btl_udapl_component.udapl_num_btls >=
mca_btl_udapl_component.udapl_max_btls) {
break;
}
}
/* finished with datinfo */
free(datinfo);
/* Make sure we have some interfaces */
if(0 == mca_btl_udapl_component.udapl_num_btls) {
mca_btl_base_error_no_nics("uDAPL", "NIC");
free(mca_btl_udapl_component.udapl_btls);
return NULL;
}
/* publish uDAPL parameters with the MCA framework */
if (OMPI_SUCCESS != mca_btl_udapl_modex_send()) {
free(mca_btl_udapl_component.udapl_btls);
return NULL;
}
/* Post OOB receive */
mca_btl_udapl_endpoint_post_oob_recv();
/* return array of BTLs */
btls = (mca_btl_base_module_t**) malloc(sizeof(mca_btl_base_module_t *) *
mca_btl_udapl_component.udapl_num_btls);
if (NULL == btls) {
free(mca_btl_udapl_component.udapl_btls);
return NULL;
}
memcpy(btls, mca_btl_udapl_component.udapl_btls,
mca_btl_udapl_component.udapl_num_btls *
sizeof(mca_btl_udapl_module_t *));
*num_btl_modules = mca_btl_udapl_component.udapl_num_btls;
return btls;
}
static int mca_btl_udapl_accept_connect(mca_btl_udapl_module_t* btl,
DAT_CR_HANDLE cr_handle)
{
DAT_EP_HANDLE endpoint;
int rc;
rc = dat_ep_create(btl->udapl_ia, btl->udapl_pz,
btl->udapl_evd_dto, btl->udapl_evd_dto,
btl->udapl_evd_conn, NULL, &endpoint);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_ep_create");
return OMPI_ERROR;
}
rc = dat_cr_accept(cr_handle, endpoint, 0, NULL);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_cr_accept");
return OMPI_ERROR;
}
return OMPI_SUCCESS;
}
/*
* uDAPL component progress.
*/
int mca_btl_udapl_component_progress()
{
mca_btl_udapl_module_t* btl;
static int32_t inprogress = 0;
DAT_EVENT event;
int count = 0;
size_t i;
/* prevent deadlock - only one thread should be 'progressing' at a time */
if(OPAL_THREAD_ADD32(&inprogress, 1) > 1) {
OPAL_THREAD_ADD32(&inprogress, -1);
return OMPI_SUCCESS;
}
/* check for work to do on each uDAPL btl */
OPAL_THREAD_LOCK(&mca_btl_udapl_component.udapl_lock);
for(i = 0; i < mca_btl_udapl_component.udapl_num_btls; i++) {
btl = mca_btl_udapl_component.udapl_btls[i];
/* Check DTO EVD */
while(DAT_SUCCESS ==
dat_evd_dequeue(btl->udapl_evd_dto, &event)) {
DAT_DTO_COMPLETION_EVENT_DATA* dto;
mca_btl_udapl_frag_t* frag;
switch(event.event_number) {
case DAT_DTO_COMPLETION_EVENT:
dto = &event.event_data.dto_completion_event_data;
frag = dto->user_cookie.as_ptr;
/* Was the DTO successful? */
if(DAT_DTO_SUCCESS != dto->status) {
OPAL_OUTPUT((0,
"btl_udapl ***** DTO error %d *****\n",
dto->status));
break;
}
switch(frag->type) {
case MCA_BTL_UDAPL_SEND:
{
mca_btl_udapl_endpoint_t* endpoint = frag->endpoint;
/*OPAL_OUTPUT((0, "btl_udapl UDAPL_SEND %d",
dto->transfered_length));*/
assert(frag->base.des_src == &frag->segment);
assert(frag->base.des_src_cnt == 1);
assert(frag->base.des_dst == NULL);
assert(frag->base.des_dst_cnt == 0);
assert(frag->type == MCA_BTL_UDAPL_SEND);
frag->base.des_cbfunc(&btl->super, frag->endpoint,
&frag->base, OMPI_SUCCESS);
if(frag->size ==
mca_btl_udapl_component.udapl_eager_frag_size) {
if(!opal_list_is_empty(
&endpoint->endpoint_eager_frags)) {
DAT_DTO_COOKIE cookie;
frag = (mca_btl_udapl_frag_t*)
opal_list_remove_first(
&endpoint->endpoint_eager_frags);
assert(frag->triplet.segment_length ==
frag->segment.seg_len +
sizeof(mca_btl_base_header_t));
cookie.as_ptr = frag;
dat_ep_post_send(endpoint->endpoint_eager,
1, &frag->triplet, cookie,
DAT_COMPLETION_DEFAULT_FLAG);
} else {
OPAL_THREAD_ADD32(
&endpoint->endpoint_eager_sends, 1);
}
} else {
assert(frag->size ==
mca_btl_udapl_component.udapl_max_frag_size);
if(!opal_list_is_empty(
&endpoint->endpoint_max_frags)) {
DAT_DTO_COOKIE cookie;
frag = (mca_btl_udapl_frag_t*)
opal_list_remove_first(
&endpoint->endpoint_max_frags);
assert(frag->triplet.segment_length ==
frag->segment.seg_len +
sizeof(mca_btl_base_header_t));
cookie.as_ptr = frag;
dat_ep_post_send(endpoint->endpoint_max,
1, &frag->triplet, cookie,
DAT_COMPLETION_DEFAULT_FLAG);
} else {
OPAL_THREAD_ADD32(
&endpoint->endpoint_max_sends, 1);
}
}
break;
}
case MCA_BTL_UDAPL_RECV:
{
mca_btl_base_recv_reg_t* reg =
&btl->udapl_reg[frag->hdr->tag];
assert(frag->base.des_dst == &frag->segment);
assert(frag->base.des_dst_cnt == 1);
assert(frag->base.des_src == NULL);
assert(frag->base.des_src_cnt == 0);
assert(frag->type == MCA_BTL_UDAPL_RECV);
assert(frag->triplet.virtual_address ==
(DAT_VADDR)frag->hdr);
assert(frag->triplet.segment_length == frag->size);
assert(frag->btl == btl);
/*OPAL_OUTPUT((0, "btl_udapl UDAPL_RECV %d",
dto->transfered_length));*/
/* OPAL_OUTPUT((0, "recv from %s %d %p\n",
inet_ntoa(addr->sin_addr), ntohs(addr->sin_port),
frag->endpoint));*/
frag->segment.seg_addr.pval = frag->hdr + 1;
frag->segment.seg_len = dto->transfered_length -
sizeof(mca_btl_base_header_t);
OPAL_THREAD_UNLOCK(&mca_btl_udapl_component.udapl_lock);
reg->cbfunc(&btl->super,
frag->hdr->tag, &frag->base, reg->cbdata);
OPAL_THREAD_LOCK(&mca_btl_udapl_component.udapl_lock);
/* Repost the frag */
frag->segment.seg_addr.pval = frag->hdr;
frag->segment.seg_len =
frag->size - sizeof(mca_btl_base_header_t);
frag->base.des_flags = 0;
if(frag->size ==
mca_btl_udapl_component.udapl_eager_frag_size) {
dat_ep_post_recv(frag->endpoint->endpoint_eager,
1, &frag->triplet, dto->user_cookie,
DAT_COMPLETION_DEFAULT_FLAG);
} else {
assert(frag->size ==
mca_btl_udapl_component.udapl_max_frag_size);
dat_ep_post_recv(frag->endpoint->endpoint_max,
1, &frag->triplet, dto->user_cookie,
DAT_COMPLETION_DEFAULT_FLAG);
}
break;
}
default:
OPAL_OUTPUT((0, "WARNING unknown frag type: %d\n",
frag->type));
}
count++;
break;
default:
OPAL_OUTPUT((0, "WARNING unknown dto event: %d\n",
event.event_number));
}
}
/* Check connection EVD */
while(DAT_SUCCESS ==
dat_evd_dequeue(btl->udapl_evd_conn, &event)) {
switch(event.event_number) {
case DAT_CONNECTION_REQUEST_EVENT:
/* Accept a new connection */
mca_btl_udapl_accept_connect(btl,
event.event_data.cr_arrival_event_data.cr_handle);
count++;
break;
case DAT_CONNECTION_EVENT_ESTABLISHED:
/* Both the client and server side of a connection generate
this event */
mca_btl_udapl_endpoint_finish_connect(btl,
event.event_data.connect_event_data.ep_handle);
count++;
break;
case DAT_CONNECTION_EVENT_PEER_REJECTED:
case DAT_CONNECTION_EVENT_NON_PEER_REJECTED:
case DAT_CONNECTION_EVENT_ACCEPT_COMPLETION_ERROR:
case DAT_CONNECTION_EVENT_DISCONNECTED:
case DAT_CONNECTION_EVENT_BROKEN:
case DAT_CONNECTION_EVENT_TIMED_OUT:
/* handle this case specially? if we have finite timeout,
we might want to try connecting again here. */
case DAT_CONNECTION_EVENT_UNREACHABLE:
/* Need to set the BTL endpoint to MCA_BTL_UDAPL_FAILED
See dat_ep_connect documentation pdf pg 198 */
break;
default:
OPAL_OUTPUT((0, "WARNING unknown conn event: %d\n",
event.event_number));
}
}
/* Check async EVD */
while(DAT_SUCCESS ==
dat_evd_dequeue(btl->udapl_evd_async, &event)) {
switch(event.event_number) {
case DAT_ASYNC_ERROR_EVD_OVERFLOW:
case DAT_ASYNC_ERROR_IA_CATASTROPHIC:
case DAT_ASYNC_ERROR_EP_BROKEN:
case DAT_ASYNC_ERROR_TIMED_OUT:
case DAT_ASYNC_ERROR_PROVIDER_INTERNAL_ERROR:
break;
default:
OPAL_OUTPUT((0, "WARNING unknown async event: %d\n",
event.event_number));
}
}
}
/* unlock and return */
OPAL_THREAD_UNLOCK(&mca_btl_udapl_component.udapl_lock);
OPAL_THREAD_ADD32(&inprogress, -1);
return count;
}
Показать git blame Копировать постоянную ссылку