1
1
openmpi/ompi/mca/btl/udapl/btl_udapl.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

713 строки
22 KiB
C

/*
* 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 <string.h>
#include "opal/util/output.h"
#include "opal/util/if.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/mca/btl/btl.h"
#include "btl_udapl.h"
#include "btl_udapl_frag.h"
#include "btl_udapl_proc.h"
#include "btl_udapl_endpoint.h"
#include "ompi/datatype/convertor.h"
#include "ompi/datatype/datatype.h"
#include "ompi/mca/mpool/base/base.h"
#include "ompi/mca/mpool/udapl/mpool_udapl.h"
#include "ompi/proc/proc.h"
mca_btl_udapl_module_t mca_btl_udapl_module = {
{
&mca_btl_udapl_component.super,
0, /* max size of first fragment */
0, /* min send fragment size */
0, /* max send fragment size */
0, /* min rdma fragment size */
0, /* max rdma fragment size */
0, /* exclusivity */
0, /* latency */
0, /* bandwidth */
MCA_BTL_FLAGS_SEND,
mca_btl_udapl_add_procs,
mca_btl_udapl_del_procs,
mca_btl_udapl_register,
mca_btl_udapl_finalize,
mca_btl_udapl_alloc,
mca_btl_udapl_free,
mca_btl_udapl_prepare_src,
NULL, /* prepare_dst */
mca_btl_udapl_send,
NULL, /* put */
NULL, /* get */
mca_btl_base_dump
}
};
/**
* Initialize module module resources.
*/
int
mca_btl_udapl_init(DAT_NAME_PTR ia_name, mca_btl_udapl_module_t* btl)
{
mca_mpool_base_resources_t res;
DAT_CONN_QUAL port;
DAT_IA_ATTR attr;
DAT_RETURN rc;
/* open the uDAPL interface */
btl->udapl_evd_async = DAT_HANDLE_NULL;
rc = dat_ia_open(ia_name, mca_btl_udapl_component.udapl_evd_qlen,
&btl->udapl_evd_async, &btl->udapl_ia);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_ia_open");
return OMPI_ERROR;
}
/* create a protection zone */
rc = dat_pz_create(btl->udapl_ia, &btl->udapl_pz);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_pz_create");
goto failure;
}
/* query to get address information */
/* TODO - we only get the address, but there's other useful stuff here */
rc = dat_ia_query(btl->udapl_ia, &btl->udapl_evd_async,
DAT_IA_FIELD_IA_ADDRESS_PTR, &attr, 0, NULL);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_ia_query");
goto failure;
}
memcpy(&btl->udapl_addr.addr, attr.ia_address_ptr, sizeof(DAT_SOCK_ADDR));
/* set up evd's */
rc = dat_evd_create(btl->udapl_ia,
mca_btl_udapl_component.udapl_evd_qlen, DAT_HANDLE_NULL,
DAT_EVD_DTO_FLAG | DAT_EVD_RMR_BIND_FLAG, &btl->udapl_evd_dto);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_evd_create (dto)");
goto failure;
}
rc = dat_evd_create(btl->udapl_ia,
mca_btl_udapl_component.udapl_evd_qlen, DAT_HANDLE_NULL,
DAT_EVD_CR_FLAG | DAT_EVD_CONNECTION_FLAG, &btl->udapl_evd_conn);
if(DAT_SUCCESS != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_evd_create (conn)");
goto failure;
}
/* create our public service point */
/* We have to specify a port, so we go through a range until we
find a port that works */
for(port = mca_btl_udapl_component.udapl_port_low;
port <= mca_btl_udapl_component.udapl_port_high; port++) {
rc = dat_psp_create(btl->udapl_ia, port, btl->udapl_evd_conn,
DAT_PSP_CONSUMER_FLAG, &btl->udapl_psp);
if(DAT_SUCCESS == rc) {
break;
} else if(DAT_CONN_QUAL_IN_USE != rc) {
MCA_BTL_UDAPL_ERROR(rc, "dat_psp_create");
goto failure;
}
}
if(port == mca_btl_udapl_component.udapl_port_high) {
goto failure;
}
/* Save the port with the address information */
/* TODO - since we're doing the hack below, do we need our own port? */
btl->udapl_addr.port = port;
/* TODO - big bad evil hack! */
/* uDAPL doesn't ever seem to keep track of ports with addresses. This
becomes a problem when we use dat_ep_query() to obtain a remote address
on an endpoint. In this case, both the DAT_PORT_QUAL and the sin_port
field in the DAT_SOCK_ADDR are 0, regardless of the actual port. This is
a problem when we have more than one uDAPL process per IA - these
processes will have exactly the same address, as the port is all
we have to differentiate who is who. Thus, our uDAPL EP -> BTL EP
matching algorithm will break down.
So, we insert the port we used for our PSP into the DAT_SOCK_ADDR for
this IA. uDAPL then conveniently propagates this to where we need it.
*/
((struct sockaddr_in*)attr.ia_address_ptr)->sin_port = htons(port);
((struct sockaddr_in*)&btl->udapl_addr.addr)->sin_port = htons(port);
/* initialize the memory pool */
res.udapl_ia = btl->udapl_ia;
res.udapl_pz = btl->udapl_pz;
btl->super.btl_mpool = mca_mpool_base_module_create(
mca_btl_udapl_component.udapl_mpool_name, &btl->super, &res);
/* initialize objects */
OBJ_CONSTRUCT(&btl->udapl_frag_eager, ompi_free_list_t);
OBJ_CONSTRUCT(&btl->udapl_frag_max, ompi_free_list_t);
OBJ_CONSTRUCT(&btl->udapl_frag_user, ompi_free_list_t);
OBJ_CONSTRUCT(&btl->udapl_lock, opal_mutex_t);
/* initialize free lists */
ompi_free_list_init(&btl->udapl_frag_eager,
sizeof(mca_btl_udapl_frag_eager_t) +
mca_btl_udapl_component.udapl_eager_frag_size,
OBJ_CLASS(mca_btl_udapl_frag_eager_t),
mca_btl_udapl_component.udapl_free_list_num,
mca_btl_udapl_component.udapl_free_list_max,
mca_btl_udapl_component.udapl_free_list_inc,
btl->super.btl_mpool);
ompi_free_list_init(&btl->udapl_frag_max,
sizeof(mca_btl_udapl_frag_max_t) +
mca_btl_udapl_component.udapl_max_frag_size,
OBJ_CLASS(mca_btl_udapl_frag_max_t),
mca_btl_udapl_component.udapl_free_list_num,
mca_btl_udapl_component.udapl_free_list_max,
mca_btl_udapl_component.udapl_free_list_inc,
btl->super.btl_mpool);
ompi_free_list_init(&btl->udapl_frag_user,
sizeof(mca_btl_udapl_frag_user_t),
OBJ_CLASS(mca_btl_udapl_frag_user_t),
mca_btl_udapl_component.udapl_free_list_num,
mca_btl_udapl_component.udapl_free_list_max,
mca_btl_udapl_component.udapl_free_list_inc,
NULL);
/* TODO - Set up SRQ when it is supported */
return OMPI_SUCCESS;
failure:
dat_ia_close(btl->udapl_ia, DAT_CLOSE_ABRUPT_FLAG);
return OMPI_ERROR;
}
/*
* Cleanup/release module resources.
*/
int mca_btl_udapl_finalize(struct mca_btl_base_module_t* base_btl)
{
mca_btl_udapl_module_t* udapl_btl = (mca_btl_udapl_module_t*) base_btl;
/* release uDAPL resources */
dat_evd_free(udapl_btl->udapl_evd_dto);
dat_evd_free(udapl_btl->udapl_evd_conn);
dat_pz_free(udapl_btl->udapl_pz);
dat_ia_close(udapl_btl->udapl_ia, DAT_CLOSE_GRACEFUL_FLAG);
/* destroy objects */
OBJ_DESTRUCT(&udapl_btl->udapl_lock);
OBJ_DESTRUCT(&udapl_btl->udapl_frag_eager);
OBJ_DESTRUCT(&udapl_btl->udapl_frag_max);
OBJ_DESTRUCT(&udapl_btl->udapl_frag_user);
free(udapl_btl);
return OMPI_SUCCESS;
}
/*
*
*/
int mca_btl_udapl_add_procs(
struct mca_btl_base_module_t* btl,
size_t nprocs,
struct ompi_proc_t **ompi_procs,
struct mca_btl_base_endpoint_t** peers,
ompi_bitmap_t* reachable)
{
mca_btl_udapl_module_t* udapl_btl = (mca_btl_udapl_module_t*)btl;
int i, rc;
for(i = 0; i < (int) nprocs; i++) {
struct ompi_proc_t* ompi_proc = ompi_procs[i];
mca_btl_udapl_proc_t* udapl_proc;
mca_btl_base_endpoint_t* udapl_endpoint;
if(ompi_proc == ompi_proc_local())
continue;
if(NULL == (udapl_proc = mca_btl_udapl_proc_create(ompi_proc))) {
continue;
}
/*
* Check to make sure that the peer has at least as many interface
* addresses exported as we are trying to use. If not, then
* don't bind this BTL instance to the proc.
*/
OPAL_THREAD_LOCK(&udapl_proc->proc_lock);
/* The btl_proc datastructure is shared by all uDAPL BTL
* instances that are trying to reach this destination.
* Cache the peer instance on the btl_proc.
*/
udapl_endpoint = OBJ_NEW(mca_btl_udapl_endpoint_t);
if(NULL == udapl_endpoint) {
OPAL_THREAD_UNLOCK(&udapl_proc->proc_lock);
return OMPI_ERR_OUT_OF_RESOURCE;
}
udapl_endpoint->endpoint_btl = udapl_btl;
rc = mca_btl_udapl_proc_insert(udapl_proc, udapl_endpoint);
if(rc != OMPI_SUCCESS) {
OBJ_RELEASE(udapl_endpoint);
OPAL_THREAD_UNLOCK(&udapl_proc->proc_lock);
continue;
}
ompi_bitmap_set_bit(reachable, i);
OPAL_THREAD_UNLOCK(&udapl_proc->proc_lock);
peers[i] = udapl_endpoint;
}
return OMPI_SUCCESS;
}
int mca_btl_udapl_del_procs(struct mca_btl_base_module_t* btl,
size_t nprocs,
struct ompi_proc_t **procs,
struct mca_btl_base_endpoint_t ** peers)
{
/* TODO */
return OMPI_SUCCESS;
}
/**
* Register callback function to support send/recv semantics
*/
int mca_btl_udapl_register(
struct mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_module_recv_cb_fn_t cbfunc,
void* cbdata)
{
mca_btl_udapl_module_t* udapl_btl = (mca_btl_udapl_module_t*) btl;
udapl_btl->udapl_reg[tag].cbfunc = cbfunc;
udapl_btl->udapl_reg[tag].cbdata = cbdata;
return OMPI_SUCCESS;
}
/**
* Allocate a segment.
*
* @param btl (IN) BTL module
* @param size (IN) Request segment size.
*/
mca_btl_base_descriptor_t* mca_btl_udapl_alloc(
struct mca_btl_base_module_t* btl,
size_t size)
{
mca_btl_udapl_module_t* udapl_btl = (mca_btl_udapl_module_t*) btl;
mca_btl_udapl_frag_t* frag;
int rc;
if(size <= btl->btl_eager_limit) {
MCA_BTL_UDAPL_FRAG_ALLOC_EAGER(udapl_btl, frag, rc);
frag->segment.seg_len =
size <= btl->btl_eager_limit ?
size : btl->btl_eager_limit;
} else if(size <= btl->btl_max_send_size) {
MCA_BTL_UDAPL_FRAG_ALLOC_MAX(udapl_btl, frag, rc);
frag->segment.seg_len =
size <= btl->btl_max_send_size ?
size : btl->btl_max_send_size;
} else {
return NULL;
}
/* Set up the LMR triplet from the frag segment */
/* Note that this triplet defines a sub-region of a registered LMR */
frag->triplet.virtual_address = (DAT_VADDR)frag->hdr;
frag->triplet.segment_length =
frag->segment.seg_len + sizeof(mca_btl_base_header_t);
assert(frag->triplet.lmr_context ==
((mca_mpool_udapl_registration_t*)frag->registration)->lmr_triplet.lmr_context);
frag->btl = udapl_btl;
frag->base.des_src = &frag->segment;
frag->base.des_src_cnt = 1;
frag->base.des_dst = NULL;
frag->base.des_dst_cnt = 0;
frag->base.des_flags = 0;
return &frag->base;
}
/**
* Return a segment
*/
int mca_btl_udapl_free(
struct mca_btl_base_module_t* btl,
mca_btl_base_descriptor_t* des)
{
mca_btl_udapl_frag_t* frag = (mca_btl_udapl_frag_t*)des;
if(frag->size == 0) {
btl->btl_mpool->mpool_release(btl->btl_mpool, frag->registration);
MCA_BTL_UDAPL_FRAG_RETURN_USER(btl, frag);
} else if(frag->size == mca_btl_udapl_component.udapl_eager_frag_size) {
MCA_BTL_UDAPL_FRAG_RETURN_EAGER(btl, frag);
} else if(frag->size == mca_btl_udapl_component.udapl_max_frag_size) {
MCA_BTL_UDAPL_FRAG_RETURN_MAX(btl, frag);
} else {
OPAL_OUTPUT((0, "[%s:%d] mca_btl_udapl_free: invalid descriptor\n", __FILE__,__LINE__));
return OMPI_ERR_BAD_PARAM;
}
return OMPI_SUCCESS;
}
/**
* Pack data and return a descriptor that can be
* used for send/put.
*
* @param btl (IN) BTL module
* @param peer (IN) BTL peer addressing
*/
mca_btl_base_descriptor_t* mca_btl_udapl_prepare_src(
struct mca_btl_base_module_t* btl,
struct mca_btl_base_endpoint_t* endpoint,
struct mca_mpool_base_registration_t* registration,
struct ompi_convertor_t* convertor,
size_t reserve,
size_t* size
)
{
mca_btl_udapl_frag_t* frag;
struct iovec iov;
uint32_t iov_count = 1;
size_t max_data = *size;
int32_t free_after;
int rc;
#if 0
/*
* If the data has already been pinned and is contigous than we can
* use it in place.
*/
if (NULL != registration && 0 == ompi_convertor_need_buffers(convertor)) {
size_t reg_len;
OPAL_OUTPUT((0, "udapl_prepare_src 1\n"));
MCA_BTL_UDAPL_FRAG_ALLOC_USER(btl, frag, rc);
if(NULL == frag){
return NULL;
}
iov.iov_len = max_data;
iov.iov_base = NULL;
ompi_convertor_pack(convertor, &iov,
&iov_count, &max_data, &free_after);
frag->segment.seg_len = max_data;
frag->segment.seg_addr.pval = iov.iov_base;
frag->triplet.segment_length = max_data;
frag->triplet.virtual_address = (DAT_VADDR)iov.iov_base;
reg_len = (unsigned char*)registration->bound -
(unsigned char*)iov.iov_base + 1;
/* bump reference count as so that the registration
* doesn't go away when the operation completes
*/
btl->btl_mpool->mpool_retain(btl->btl_mpool, registration);
frag->registration = registration;
frag->triplet.lmr_context =
((mca_mpool_udapl_registration_t*)registration)->lmr_triplet.lmr_context;
/*
* if the data is not already pinned - but the leave pinned option is set,
* then go ahead and pin contigous data. however, if a reserve is required
* then we must allocate a fragment w/ buffer space
*/
} else if (max_data > btl->btl_max_send_size &&
ompi_convertor_need_buffers(convertor) == 0 &&
reserve == 0) {
mca_mpool_base_module_t* mpool = btl->btl_mpool;
MCA_BTL_UDAPL_FRAG_ALLOC_USER(btl, frag, rc);
if(NULL == frag){
return NULL;
}
OPAL_OUTPUT((0, "udapl_prepare_src 2\n"));
iov.iov_len = max_data;
iov.iov_base = NULL;
ompi_convertor_pack(convertor, &iov,
&iov_count, &max_data, &free_after);
rc = mpool->mpool_register(
mpool,
iov.iov_base,
max_data,
0,
&registration);
if(rc != OMPI_SUCCESS) {
MCA_BTL_UDAPL_FRAG_RETURN_USER(btl,frag);
return NULL;
}
frag->registration = registration;
frag->triplet.lmr_context =
((mca_mpool_udapl_registration_t*)registration)->lmr_triplet.lmr_context;
/* TODO - should our base addr be frag->hdr? */
frag->segment.seg_len = max_data;
frag->segment.seg_addr.pval = iov.iov_base;
frag->triplet.segment_length = max_data;
frag->triplet.virtual_address = (DAT_VADDR)iov.iov_base;
}
/*
* if we aren't pinning the data and the requested size is less
* than the eager limit pack into a fragment from the eager pool
*/
else
#endif
if(max_data + reserve <= btl->btl_eager_limit) {
MCA_BTL_UDAPL_FRAG_ALLOC_EAGER(btl, frag, rc);
if(NULL == frag) {
return NULL;
}
iov.iov_len = max_data;
iov.iov_base = (unsigned char*) frag->segment.seg_addr.pval + reserve;
rc = ompi_convertor_pack(convertor,
&iov, &iov_count, &max_data, &free_after);
*size = max_data;
if(rc < 0) {
MCA_BTL_UDAPL_FRAG_RETURN_EAGER(btl, frag);
return NULL;
}
frag->segment.seg_len = max_data + reserve;
frag->triplet.segment_length =
max_data + reserve + sizeof(mca_btl_base_header_t);
frag->triplet.virtual_address = (DAT_VADDR)frag->hdr;
}
/*
* otherwise pack as much data as we can into a fragment
* that is the max send size.
*/
else {
MCA_BTL_UDAPL_FRAG_ALLOC_MAX(btl, frag, rc);
if(NULL == frag) {
return NULL;
}
if(max_data + reserve > btl->btl_max_send_size){
max_data = btl->btl_max_send_size - reserve;
}
iov.iov_len = max_data;
iov.iov_base = (unsigned char*) frag->segment.seg_addr.pval + reserve;
rc = ompi_convertor_pack(convertor,
&iov, &iov_count, &max_data, &free_after);
*size = max_data;
if(rc < 0) {
MCA_BTL_UDAPL_FRAG_RETURN_MAX(btl, frag);
return NULL;
}
/* TODO - pull this out of the if statements. */
frag->segment.seg_len = max_data + reserve;
frag->triplet.segment_length =
max_data + reserve + sizeof(mca_btl_base_header_t);
frag->triplet.virtual_address = (DAT_VADDR)frag->hdr;
}
frag->base.des_src = &frag->segment;
frag->base.des_src_cnt = 1;
frag->base.des_dst = NULL;
frag->base.des_dst_cnt = 0;
frag->base.des_flags = 0;
return &frag->base;
}
/**
* Prepare a descriptor for send/rdma using the supplied
* convertor. If the convertor references data that is contigous,
* the descriptor may simply point to the user buffer. Otherwise,
* this routine is responsible for allocating buffer space and
* packing if required.
*
* @param btl (IN) BTL module
* @param endpoint (IN) BTL peer addressing
* @param convertor (IN) Data type convertor
* @param reserve (IN) Additional bytes requested by upper layer to precede user data
* @param size (IN/OUT) Number of bytes to prepare (IN), number of bytes actually prepared (OUT)
*/
#if 0
mca_btl_base_descriptor_t* mca_btl_udapl_prepare_dst(
struct mca_btl_base_module_t* btl,
struct mca_btl_base_endpoint_t* endpoint,
struct mca_mpool_base_registration_t* registration,
struct ompi_convertor_t* convertor,
size_t reserve,
size_t* size)
{
mca_btl_udapl_frag_t* frag;
mca_mpool_base_module_t* mpool = btl->btl_mpool;
long lb;
int rc;
OPAL_OUTPUT((0, "udapl_prepare_dst\n"));
MCA_BTL_UDAPL_FRAG_ALLOC_USER(btl, frag, rc);
if(NULL == frag) {
return NULL;
}
ompi_ddt_type_lb(convertor->pDesc, &lb);
frag->segment.seg_len = *size;
frag->segment.seg_addr.pval = convertor->pBaseBuf + lb + convertor->bConverted;
frag->base.des_src = NULL;
frag->base.des_src_cnt = 0;
frag->base.des_dst = &frag->segment;
frag->base.des_dst_cnt = 1;
frag->base.des_flags = 0;
if(NULL != registration) {
/* bump reference count as so that the registration
* doesn't go away when the operation completes
*/
mpool->mpool_retain(mpool,
(mca_mpool_base_registration_t*) registration);
frag->registration = registration;
} else {
rc = mpool->mpool_register(
mpool,
frag->segment.seg_addr.pval,
frag->segment.seg_len,
0,
&registration);
if(rc != OMPI_SUCCESS) {
MCA_BTL_UDAPL_FRAG_RETURN_USER(btl,frag);
return NULL;
}
frag->registration = registration;
}
return &frag->base;
}
#endif
/**
* Initiate an asynchronous send.
*
* @param btl (IN) BTL module
* @param endpoint (IN) BTL addressing information
* @param descriptor (IN) Description of the data to be transfered
* @param tag (IN) The tag value used to notify the peer.
*/
int mca_btl_udapl_send(
struct mca_btl_base_module_t* btl,
struct mca_btl_base_endpoint_t* endpoint,
struct mca_btl_base_descriptor_t* des,
mca_btl_base_tag_t tag)
{
mca_btl_udapl_frag_t* frag = (mca_btl_udapl_frag_t*)des;
frag->btl = (mca_btl_udapl_module_t*)btl;
frag->endpoint = endpoint;
frag->hdr->tag = tag;
frag->type = MCA_BTL_UDAPL_SEND;
/* TODO - will inlining this give worthwhile performance? */
return mca_btl_udapl_endpoint_send(endpoint, frag);
}
/**
* Initiate an asynchronous put.
*
* @param btl (IN) BTL module
* @param endpoint (IN) BTL addressing information
* @param descriptor (IN) Description of the data to be transferred
*/
int mca_btl_udapl_put(
mca_btl_base_module_t* btl,
mca_btl_base_endpoint_t* endpoint,
mca_btl_base_descriptor_t* des)
{
OPAL_OUTPUT((0, "udapl_put\n"));
return OMPI_ERR_NOT_IMPLEMENTED;
}
/**
* Initiate an asynchronous get.
*
* @param btl (IN) BTL module
* @param endpoint (IN) BTL addressing information
* @param descriptor (IN) Description of the data to be transferred
*
*/
int mca_btl_udapl_get(
mca_btl_base_module_t* btl,
mca_btl_base_endpoint_t* endpoint,
mca_btl_base_descriptor_t* des)
{
OPAL_OUTPUT((0, "udapl_get\n"));
return OMPI_ERR_NOT_IMPLEMENTED;
}