static int mca_btl_ugni_endpoint_get_modex (mca_btl_base_endpoint_t *ep)
{
mca_btl_ugni_modex_t *modex;
size_t msg_size;
int rc;
assert (NULL != ep && NULL != ep->peer_proc);
/* Receive the modex */
OPAL_MODEX_RECV(rc, &mca_btl_ugni_component.super.btl_version,
&ep->peer_proc->proc_name, (void **)&modex, &msg_size);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
BTL_ERROR(("error receiving modex"));
return rc;
}
ep->ep_rem_addr = modex->addr;
ep->ep_rem_id = modex->id;
BTL_VERBOSE(("received modex for ep %p. addr: %d, id: %d", (void*)ep, ep->ep_rem_addr, ep->ep_rem_id));
free (modex);
return OPAL_SUCCESS;
}
static int
create_peer_and_endpoint(int interface,
opal_proc_t *proc,
ptl_process_t *phys_peer,
mca_btl_base_endpoint_t **endpoint)
{
int ret;
size_t size;
ptl_process_t *id;
OPAL_MODEX_RECV(ret, &mca_btl_portals4_component.super.btl_version,
&proc->proc_name, (void**) &id, &size);
if (OPAL_ERR_NOT_FOUND == ret) {
OPAL_OUTPUT_VERBOSE((30, opal_btl_base_framework.framework_output,
"btl/portals4: Portals 4 BTL not available on peer: %s", opal_strerror(ret)));
return ret;
} else if (OPAL_SUCCESS != ret) {
opal_output_verbose(0, opal_btl_base_framework.framework_output,
"btl/portals4: opal_modex_recv failed: %s", opal_strerror(ret));
return ret;
}
if (size < sizeof(ptl_process_t)) { /* no available connection */
return OPAL_ERROR;
}
if ((size % sizeof(ptl_process_t)) != 0) {
opal_output_verbose(0, opal_btl_base_framework.framework_output,
"btl/portals4: invalid format in modex");
return OPAL_ERROR;
}
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
"btl/portals4: %d NI(s) declared in the modex", (int) (size/sizeof(ptl_process_t))));
/*
* check if create_endpoint() already created the endpoint.
* if not, create it here.
*/
if (NULL == *endpoint) {
*endpoint = malloc(sizeof(mca_btl_base_endpoint_t));
if (NULL == *endpoint) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
}
/*
* regardless of who created the endpoint, set the rank here
* because we are using logical mapping.
*/
(*endpoint)->ptl_proc.rank = proc->proc_name.vpid;
phys_peer->phys.pid = id[interface].phys.pid;
phys_peer->phys.nid = id[interface].phys.nid;
opal_output_verbose(50, opal_btl_base_framework.framework_output,
"logical: global rank=%d pid=%d nid=%d\n",
proc->proc_name.vpid, phys_peer->phys.pid, phys_peer->phys.nid);
return OPAL_SUCCESS;
}
void usnic_compat_modex_recv(int *rc,
mca_base_component_t *component,
opal_proc_t *proc,
opal_btl_usnic_modex_t **modexes,
size_t *size)
{
OPAL_MODEX_RECV(*rc, component, &proc->proc_name,
(uint8_t**) modexes, size);
}
static int recv_ep_address(ompi_proc_t *proc, void **address_p, size_t *addrlen_p)
{
int rc;
OPAL_MODEX_RECV(rc, &mca_pml_yalla_component.pmlm_version, &proc->super,
address_p, addrlen_p);
if (rc < 0) {
PML_YALLA_ERROR("Failed to receive EP address");
}
return rc;
}
static int init_vader_endpoint (struct mca_btl_base_endpoint_t *ep, struct opal_proc_t *proc, int remote_rank) {
mca_btl_vader_component_t *component = &mca_btl_vader_component;
union vader_modex_t *modex;
size_t msg_size;
int rc;
OBJ_CONSTRUCT(ep, mca_btl_vader_endpoint_t);
ep->peer_smp_rank = remote_rank;
if (remote_rank != MCA_BTL_VADER_LOCAL_RANK) {
OPAL_MODEX_RECV(rc, &component->super.btl_version,
&proc->proc_name, (void **) &modex, &msg_size);
if (OPAL_SUCCESS != rc) {
return rc;
}
/* attatch to the remote segment */
#if OPAL_BTL_VADER_HAVE_XPMEM
if (MCA_BTL_VADER_XPMEM == mca_btl_vader_component.single_copy_mechanism) {
/* always use xpmem if it is available */
ep->segment_data.xpmem.apid = xpmem_get (modex->xpmem.seg_id, XPMEM_RDWR, XPMEM_PERMIT_MODE, (void *) 0666);
ep->segment_data.xpmem.rcache = mca_rcache_base_module_create("vma");
(void) vader_get_registation (ep, modex->xpmem.segment_base, mca_btl_vader_component.segment_size,
MCA_MPOOL_FLAGS_PERSIST, (void **) &ep->segment_base);
} else {
#endif
/* store a copy of the segment information for detach */
ep->segment_data.other.seg_ds = malloc (msg_size);
if (NULL == ep->segment_data.other.seg_ds) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
memcpy (ep->segment_data.other.seg_ds, &modex->seg_ds, msg_size);
ep->segment_base = opal_shmem_segment_attach (ep->segment_data.other.seg_ds);
if (NULL == ep->segment_base) {
return OPAL_ERROR;
}
#if OPAL_BTL_VADER_HAVE_XPMEM
}
#endif
OBJ_CONSTRUCT(&ep->lock, opal_mutex_t);
free (modex);
} else {
/* set up the segment base so we can calculate a virtual to real for local pointers */
ep->segment_base = component->my_segment;
}
ep->fifo = (struct vader_fifo_t *) ep->segment_base;
return OPAL_SUCCESS;
}
static int mca_pml_ucx_recv_worker_address(ompi_proc_t *proc,
ucp_address_t **address_p,
size_t *addrlen_p)
{
int ret;
*address_p = NULL;
OPAL_MODEX_RECV(ret, &mca_pml_ucx_component.pmlm_version, &proc->super.proc_name,
(void**)address_p, addrlen_p);
if (ret < 0) {
PML_UCX_ERROR("Failed to receive EP address");
}
return ret;
}
static int init_vader_endpoint (struct mca_btl_base_endpoint_t *ep, struct opal_proc_t *proc, int remote_rank) {
const int fbox_in_offset = MCA_BTL_VADER_LOCAL_RANK - (MCA_BTL_VADER_LOCAL_RANK > remote_rank);
const int fbox_out_offset = remote_rank - (MCA_BTL_VADER_LOCAL_RANK < remote_rank);
mca_btl_vader_component_t *component = &mca_btl_vader_component;
struct vader_modex_t *modex;
size_t msg_size;
int rc;
ep->peer_smp_rank = remote_rank;
if (remote_rank != MCA_BTL_VADER_LOCAL_RANK) {
OPAL_MODEX_RECV(rc, &component->super.btl_version,
proc, (uint8_t**)&modex, &msg_size);
if (OPAL_SUCCESS != rc) {
return rc;
}
/* attatch to the remote segment */
#if OPAL_BTL_VADER_HAVE_XPMEM
/* always use xpmem if it is available */
ep->apid = xpmem_get (modex->seg_id, XPMEM_RDWR, XPMEM_PERMIT_MODE, (void *) 0666);
ep->rcache = mca_rcache_base_module_create("vma");
(void) vader_get_registation (ep, modex->segment_base, mca_btl_vader_component.segment_size,
MCA_MPOOL_FLAGS_PERSIST, (void **) &ep->segment_base);
#else
msg_size -= offsetof (struct vader_modex_t, seg_ds);
memcpy (&ep->seg_ds, &modex->seg_ds, msg_size);
ep->segment_base = opal_shmem_segment_attach (&ep->seg_ds);
if (NULL == ep->segment_base) {
return rc;
}
#endif
free (modex);
ep->next_fbox_out = 0;
ep->next_fbox_in = 0;
ep->next_sequence = 0;
ep->expected_sequence = 0;
ep->fbox_in = (struct mca_btl_vader_fbox_t * restrict) (ep->segment_base + MCA_BTL_VADER_FIFO_SIZE +
fbox_in_offset * MCA_BTL_VADER_FBOX_PEER_SIZE);
ep->fbox_out = (struct mca_btl_vader_fbox_t * restrict) (component->my_segment + MCA_BTL_VADER_FIFO_SIZE +
fbox_out_offset * MCA_BTL_VADER_FBOX_PEER_SIZE);
} else {
static int
create_endpoint(int interface,
opal_proc_t *proc,
mca_btl_base_endpoint_t **endpoint)
{
int ret;
size_t size;
ptl_process_t *id;
OPAL_MODEX_RECV(ret, &mca_btl_portals4_component.super.btl_version,
&proc->proc_name, (void**) &id, &size);
if (OPAL_ERR_NOT_FOUND == ret) {
OPAL_OUTPUT_VERBOSE((30, opal_btl_base_framework.framework_output,
"btl/portals4: Portals 4 BTL not available on peer: %s", opal_strerror(ret)));
return ret;
} else if (OPAL_SUCCESS != ret) {
opal_output_verbose(0, opal_btl_base_framework.framework_output,
"btl/portals4: opal_modex_recv failed: %s", opal_strerror(ret));
return ret;
}
if (size < sizeof(ptl_process_t)) { /* no available connection */
return OPAL_ERROR;
}
if ((size % sizeof(ptl_process_t)) != 0) {
opal_output_verbose(0, opal_btl_base_framework.framework_output,
"btl/portals4: invalid format in modex");
return OPAL_ERROR;
}
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
"btl/portals4: %d NI(s) declared in the modex", (int) (size/sizeof(ptl_process_t))));
*endpoint = malloc(sizeof(mca_btl_base_endpoint_t));
if (NULL == *endpoint) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
(*endpoint)->ptl_proc = id[interface];
return OPAL_SUCCESS;
}
int opal_common_ugni_endpoint_for_proc (opal_common_ugni_device_t *dev, opal_proc_t *peer_proc,
opal_common_ugni_endpoint_t **ep)
{
opal_common_ugni_endpoint_t *endpoint;
opal_common_ugni_modex_t *modex;
size_t msg_size;
int rc;
assert (NULL != dev && NULL != ep && peer_proc);
endpoint = OBJ_NEW(opal_common_ugni_endpoint_t);
if (OPAL_UNLIKELY(NULL == endpoint)) {
assert (0);
return OPAL_ERR_OUT_OF_RESOURCE;
}
/* Receive the modex */
OPAL_MODEX_RECV(rc, &opal_common_ugni_component,
peer_proc, &modex, &msg_size);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
OPAL_OUTPUT((-1, "btl/ugni error receiving modex"));
return rc;
}
endpoint->ep_rem_addr = modex->addr;
endpoint->ep_rem_id = modex->id;
endpoint->ep_rem_irq_memhndl = modex->irq_memhndl;
endpoint->dev = dev;
*ep = endpoint;
free (modex);
return OPAL_SUCCESS;
}
int
ompi_mtl_psm2_add_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs,
struct ompi_proc_t** procs)
{
int i,j;
int rc;
psm2_epid_t *epids_in = NULL;
int *mask_in = NULL;
psm2_epid_t *epid;
psm2_epaddr_t *epaddrs_out = NULL;
psm2_error_t *errs_out = NULL, err;
size_t size;
int proc_errors[PSM2_ERROR_LAST] = { 0 };
int timeout_in_secs;
assert(mtl == &ompi_mtl_psm2.super);
rc = OMPI_ERR_OUT_OF_RESOURCE;
errs_out = (psm2_error_t *) malloc(nprocs * sizeof(psm2_error_t));
if (errs_out == NULL) {
goto bail;
}
epids_in = (psm2_epid_t *) malloc(nprocs * sizeof(psm2_epid_t));
if (epids_in == NULL) {
goto bail;
}
mask_in = (int *) malloc(nprocs * sizeof(int));
if (mask_in == NULL) {
goto bail;
}
epaddrs_out = (psm2_epaddr_t *) malloc(nprocs * sizeof(psm2_epaddr_t));
if (epaddrs_out == NULL) {
goto bail;
}
rc = OMPI_SUCCESS;
/* Get the epids for all the processes from modex */
for (i = 0; i < (int) nprocs; i++) {
if (NULL != procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL]) {
/* Already connected: don't connect again */
mask_in[i] = 0;
continue;
}
OPAL_MODEX_RECV(rc, &mca_mtl_psm2_component.super.mtl_version,
&procs[i]->super.proc_name, (void**)&epid, &size);
if (rc != OMPI_SUCCESS || size != sizeof(psm2_epid_t)) {
return OMPI_ERROR;
}
epids_in[i] = *epid;
mask_in[i] = 1;
}
timeout_in_secs = max(ompi_mtl_psm2.connect_timeout, 0.5 * nprocs);
psm2_error_register_handler(ompi_mtl_psm2.ep, PSM2_ERRHANDLER_NOP);
err = psm2_ep_connect(ompi_mtl_psm2.ep,
nprocs,
epids_in,
mask_in,
errs_out,
epaddrs_out,
timeout_in_secs * 1e9);
if (err) {
char *errstr = (char *) ompi_mtl_psm2_connect_error_msg(err);
if (errstr == NULL) {
opal_output(0, "PSM2 returned unhandled/unknown connect error: %s\n",
psm2_error_get_string(err));
}
for (i = 0; i < (int) nprocs; i++) {
if (0 == mask_in[i]) {
continue;
}
psm2_error_t thiserr = errs_out[i];
errstr = (char *) ompi_mtl_psm2_connect_error_msg(thiserr);
if (proc_errors[thiserr] == 0) {
proc_errors[thiserr] = 1;
opal_output(0, "PSM2 EP connect error (%s):",
errstr ? errstr : "unknown connect error");
for (j = 0; j < (int) nprocs; j++) {
if (errs_out[j] == thiserr) {
opal_output(0, " %s", (NULL == procs[j]->super.proc_hostname) ?
"unknown" : procs[j]->super.proc_hostname);
}
}
opal_output(0, "\n");
}
}
rc = OMPI_ERROR;
}
else {
/* Default error handling is enabled, errors will not be returned to
* user. PSM2 prints the error and the offending endpoint's hostname
* and exits with -1 */
psm2_error_register_handler(ompi_mtl_psm2.ep, PSM2_ERRHANDLER_DEFAULT);
/* Fill in endpoint data */
for (i = 0; i < (int) nprocs; i++) {
if (0 == mask_in[i]) {
continue;
}
mca_mtl_psm2_endpoint_t *endpoint =
(mca_mtl_psm2_endpoint_t *) OBJ_NEW(mca_mtl_psm2_endpoint_t);
endpoint->peer_epid = epids_in[i];
endpoint->peer_addr = epaddrs_out[i];
procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL] = endpoint;
}
rc = OMPI_SUCCESS;
}
bail:
if (epids_in != NULL) {
free(epids_in);
}
if (mask_in != NULL) {
free(mask_in);
}
if (errs_out != NULL) {
free(errs_out);
}
if (epaddrs_out != NULL) {
free(epaddrs_out);
}
return rc;
}
int
ompi_mtl_portals4_add_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs,
struct ompi_proc_t** procs)
{
int ret, me;
size_t i;
bool new_found = false;
/* Get the list of ptl_process_id_t from the runtime and copy into structure */
for (i = 0 ; i < nprocs ; ++i) {
ptl_process_t *modex_id;
size_t size;
if( procs[i] == ompi_proc_local_proc ) {
me = i;
}
if (procs[i]->super.proc_arch != ompi_proc_local()->super.proc_arch) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"Portals 4 MTL does not support heterogeneous operations.");
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"Proc %s architecture %x, mine %x.",
OMPI_NAME_PRINT(&procs[i]->super.proc_name),
procs[i]->super.proc_arch, ompi_proc_local()->super.proc_arch);
return OMPI_ERR_NOT_SUPPORTED;
}
OPAL_MODEX_RECV(ret, &mca_mtl_portals4_component.mtl_version,
&procs[i]->super, (char**)&modex_id, &size);
if (OMPI_SUCCESS != ret) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: ompi_modex_recv failed: %d\n",
__FILE__, __LINE__, ret);
return ret;
} else if (sizeof(ptl_process_t) != size) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: ompi_modex_recv failed: %d\n",
__FILE__, __LINE__, ret);
return OMPI_ERR_BAD_PARAM;
}
if (NULL == procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_PORTALS4]) {
ptl_process_t *peer_id;
peer_id = malloc(sizeof(ptl_process_t));
if (NULL == peer_id) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: malloc failed: %d\n",
__FILE__, __LINE__, ret);
return OMPI_ERR_OUT_OF_RESOURCE;
}
*peer_id = *modex_id;
procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_PORTALS4] = peer_id;
new_found = true;
} else {
ptl_process_t *proc = (ptl_process_t*) procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_PORTALS4];
if (proc->phys.nid != modex_id->phys.nid ||
proc->phys.pid != modex_id->phys.pid) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: existing peer and modex peer don't match\n",
__FILE__, __LINE__);
return OMPI_ERROR;
}
}
}
#if OMPI_MTL_PORTALS4_FLOW_CONTROL
if (new_found) {
ret = ompi_mtl_portals4_flowctl_add_procs(me, nprocs, procs);
if (OMPI_SUCCESS != ret) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: flowctl_add_procs failed: %d\n",
__FILE__, __LINE__, ret);
return ret;
}
}
#endif
return OMPI_SUCCESS;
}
mca_btl_openib_proc_t* mca_btl_openib_proc_get_locked(opal_proc_t* proc)
{
mca_btl_openib_proc_t *ib_proc = NULL, *ib_proc_ret = NULL;
size_t msg_size;
uint32_t size;
int rc, i, j;
void *message;
char *offset;
int modex_message_size;
mca_btl_openib_modex_message_t dummy;
bool is_new = false;
/* Check if we have already created a IB proc
* structure for this ompi process */
ib_proc = ibproc_lookup_and_lock(proc);
if (NULL != ib_proc) {
/* Gotcha! */
return ib_proc;
}
/* All initialization has to be an atomic operation. we do the following assumption:
* - we let all concurent threads to try to do the initialization;
* - when one has finished it locks ib_lock and checks if corresponding
* process is still missing;
* - if so - new proc is added, otherwise - initialized proc struct is released.
*/
/* First time, gotta create a new IB proc
* out of the opal_proc ... */
ib_proc = OBJ_NEW(mca_btl_openib_proc_t);
if (NULL == ib_proc) {
return NULL;
}
/* Initialize number of peer */
ib_proc->proc_endpoint_count = 0;
ib_proc->proc_opal = proc;
/* query for the peer address info */
OPAL_MODEX_RECV(rc, &mca_btl_openib_component.super.btl_version,
&proc->proc_name, &message, &msg_size);
if (OPAL_SUCCESS != rc) {
BTL_VERBOSE(("[%s:%d] opal_modex_recv failed for peer %s",
__FILE__, __LINE__,
OPAL_NAME_PRINT(proc->proc_name)));
goto no_err_exit;
}
if (0 == msg_size) {
goto no_err_exit;
}
/* Message was packed in btl_openib_component.c; the format is
listed in a comment in that file */
modex_message_size = ((char *) &(dummy.end)) - ((char*) &dummy);
/* Unpack the number of modules in the message */
offset = (char *) message;
unpack8(&offset, &(ib_proc->proc_port_count));
BTL_VERBOSE(("unpack: %d btls", ib_proc->proc_port_count));
if (ib_proc->proc_port_count > 0) {
ib_proc->proc_ports = (mca_btl_openib_proc_modex_t *)
malloc(sizeof(mca_btl_openib_proc_modex_t) *
ib_proc->proc_port_count);
} else {
ib_proc->proc_ports = NULL;
}
/* Loop over unpacking all the ports */
for (i = 0; i < ib_proc->proc_port_count; i++) {
/* Unpack the modex comment message struct */
size = modex_message_size;
memcpy(&(ib_proc->proc_ports[i].pm_port_info), offset, size);
#if !defined(WORDS_BIGENDIAN) && OPAL_ENABLE_HETEROGENEOUS_SUPPORT
MCA_BTL_OPENIB_MODEX_MSG_NTOH(ib_proc->proc_ports[i].pm_port_info);
#endif
offset += size;
BTL_VERBOSE(("unpacked btl %d: modex message, offset now %d",
i, (int)(offset-((char*)message))));
/* Unpack the number of CPCs that follow */
unpack8(&offset, &(ib_proc->proc_ports[i].pm_cpc_data_count));
BTL_VERBOSE(("unpacked btl %d: number of cpcs to follow %d (offset now %d)",
i, ib_proc->proc_ports[i].pm_cpc_data_count,
(int)(offset-((char*)message))));
ib_proc->proc_ports[i].pm_cpc_data = (opal_btl_openib_connect_base_module_data_t *)
calloc(ib_proc->proc_ports[i].pm_cpc_data_count,
sizeof(opal_btl_openib_connect_base_module_data_t));
if (NULL == ib_proc->proc_ports[i].pm_cpc_data) {
goto err_exit;
}
/* Unpack the CPCs */
for (j = 0; j < ib_proc->proc_ports[i].pm_cpc_data_count; ++j) {
uint8_t u8;
opal_btl_openib_connect_base_module_data_t *cpcd;
cpcd = ib_proc->proc_ports[i].pm_cpc_data + j;
unpack8(&offset, &u8);
BTL_VERBOSE(("unpacked btl %d: cpc %d: index %d (offset now %d)",
i, j, u8, (int)(offset-(char*)message)));
cpcd->cbm_component =
opal_btl_openib_connect_base_get_cpc_byindex(u8);
BTL_VERBOSE(("unpacked btl %d: cpc %d: component %s",
i, j, cpcd->cbm_component->cbc_name));
unpack8(&offset, &cpcd->cbm_priority);
unpack8(&offset, &cpcd->cbm_modex_message_len);
BTL_VERBOSE(("unpacked btl %d: cpc %d: priority %d, msg len %d (offset now %d)",
i, j, cpcd->cbm_priority,
cpcd->cbm_modex_message_len,
(int)(offset-(char*)message)));
if (cpcd->cbm_modex_message_len > 0) {
cpcd->cbm_modex_message = malloc(cpcd->cbm_modex_message_len);
if (NULL == cpcd->cbm_modex_message) {
BTL_ERROR(("Failed to malloc"));
goto err_exit;
}
memcpy(cpcd->cbm_modex_message, offset,
cpcd->cbm_modex_message_len);
offset += cpcd->cbm_modex_message_len;
BTL_VERBOSE(("unpacked btl %d: cpc %d: blob unpacked %d %x (offset now %d)",
i, j,
((uint32_t*)cpcd->cbm_modex_message)[0],
((uint32_t*)cpcd->cbm_modex_message)[1],
(int)(offset-((char*)message))));
}
}
}
if (0 == ib_proc->proc_port_count) {
ib_proc->proc_endpoints = NULL;
} else {
ib_proc->proc_endpoints = (volatile mca_btl_base_endpoint_t**)
malloc(ib_proc->proc_port_count *
sizeof(mca_btl_base_endpoint_t*));
}
if (NULL == ib_proc->proc_endpoints) {
goto err_exit;
}
BTL_VERBOSE(("unpacking done!"));
/* Finally add this process to the initialized procs list */
opal_mutex_lock(&mca_btl_openib_component.ib_lock);
ib_proc_ret = ibproc_lookup_no_lock(proc);
if (NULL == ib_proc_ret) {
/* if process can't be found in this list - insert it locked
* it is safe to lock ib_proc here because this thread is
* the only one who knows about it so far */
opal_mutex_lock(&ib_proc->proc_lock);
opal_list_append(&mca_btl_openib_component.ib_procs, &ib_proc->super);
ib_proc_ret = ib_proc;
is_new = true;
} else {
/* otherwise - release module_proc */
OBJ_RELEASE(ib_proc);
}
opal_mutex_unlock(&mca_btl_openib_component.ib_lock);
/* if we haven't insert the process - lock it here so we
* won't lock mca_btl_openib_component.ib_lock */
if( !is_new ){
opal_mutex_lock(&ib_proc_ret->proc_lock);
}
return ib_proc_ret;
err_exit:
BTL_ERROR(("%d: error exit from mca_btl_openib_proc_create", OPAL_PROC_MY_NAME.vpid));
no_err_exit:
OBJ_RELEASE(ib_proc);
return NULL;
}
/*
* Create an opal_btl_usnic_proc_t and initialize it with modex info
* and an empty array of endpoints.
*
* Returns OPAL_ERR_UNREACH if we can't reach the peer (i.e., we can't
* find their modex data).
*/
static int create_proc(opal_proc_t *opal_proc,
opal_btl_usnic_proc_t **usnic_proc)
{
opal_btl_usnic_proc_t *proc = NULL;
size_t size;
int rc;
*usnic_proc = NULL;
/* Create the proc if it doesn't already exist */
proc = OBJ_NEW(opal_btl_usnic_proc_t);
if (NULL == proc) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
/* Initialize number of peers */
proc->proc_endpoint_count = 0;
proc->proc_opal = opal_proc;
/* query for the peer address info */
OPAL_MODEX_RECV(rc, &mca_btl_usnic_component.super.btl_version,
opal_proc, (uint8_t**)&proc->proc_modex, &size);
/* If this proc simply doesn't have this key, then they're not
running the usnic BTL -- just ignore them. Otherwise, show an
error message. */
if (OPAL_ERR_DATA_VALUE_NOT_FOUND == rc) {
OBJ_RELEASE(proc);
return OPAL_ERR_UNREACH;
} else if (OPAL_SUCCESS != rc) {
opal_show_help("help-mpi-btl-usnic.txt",
"internal error during init",
true,
opal_process_info.nodename,
"<none>", "<none>",
"opal_modex_recv() failed", __FILE__, __LINE__,
opal_strerror(rc));
OBJ_RELEASE(proc);
return OPAL_ERROR;
}
if ((size % sizeof(opal_btl_usnic_addr_t)) != 0) {
char msg[1024];
snprintf(msg, sizeof(msg),
"sizeof(modex for peer %s data) == %d, expected multiple of %d",
OPAL_NAME_PRINT(opal_proc->proc_name),
(int) size, (int) sizeof(opal_btl_usnic_addr_t));
opal_show_help("help-mpi-btl-usnic.txt", "internal error during init",
true,
opal_process_info.nodename,
"<none>", 0,
"invalid modex data", __FILE__, __LINE__,
msg);
OBJ_RELEASE(proc);
return OPAL_ERR_VALUE_OUT_OF_BOUNDS;
}
proc->proc_modex_count = size / sizeof(opal_btl_usnic_addr_t);
if (0 == proc->proc_modex_count) {
proc->proc_endpoints = NULL;
OBJ_RELEASE(proc);
return OPAL_ERR_UNREACH;
}
/* Sanity check: ensure that the remote proc agrees with this proc
on whether we're doing UDP or not. Note that all endpoints on
the remote proc will have the same "use_udp" value, so we only
need to check one of them. */
if (proc->proc_modex[0].use_udp !=
mca_btl_usnic_component.use_udp) {
opal_show_help("help-mpi-btl-usnic.txt",
"transport mismatch",
true,
opal_process_info.nodename,
proc->proc_opal->proc_hostname);
OBJ_RELEASE(proc);
return OPAL_ERR_BAD_PARAM;
}
proc->proc_modex_claimed = (bool*)
calloc(proc->proc_modex_count, sizeof(bool));
if (NULL == proc->proc_modex_claimed) {
OPAL_ERROR_LOG(OPAL_ERR_OUT_OF_RESOURCE);
OBJ_RELEASE(proc);
return OPAL_ERR_OUT_OF_RESOURCE;
}
proc->proc_endpoints = (mca_btl_base_endpoint_t**)
calloc(proc->proc_modex_count, sizeof(mca_btl_base_endpoint_t*));
if (NULL == proc->proc_endpoints) {
OPAL_ERROR_LOG(OPAL_ERR_OUT_OF_RESOURCE);
OBJ_RELEASE(proc);
return OPAL_ERR_OUT_OF_RESOURCE;
}
*usnic_proc = proc;
return OPAL_SUCCESS;
}
mca_btl_openib_proc_t* mca_btl_openib_proc_create(opal_proc_t* proc)
{
mca_btl_openib_proc_t* module_proc = NULL;
size_t msg_size;
uint32_t size;
int rc, i, j;
void *message;
char *offset;
int modex_message_size;
mca_btl_openib_modex_message_t dummy;
/* Check if we have already created a IB proc
* structure for this ompi process */
module_proc = mca_btl_openib_proc_lookup_proc(proc);
if (NULL != module_proc) {
/* Gotcha! */
return module_proc;
}
/* Oops! First time, gotta create a new IB proc
* out of the opal_proc ... */
module_proc = OBJ_NEW(mca_btl_openib_proc_t);
/* Initialize number of peer */
module_proc->proc_endpoint_count = 0;
module_proc->proc_opal = proc;
/* query for the peer address info */
OPAL_MODEX_RECV(rc, &mca_btl_openib_component.super.btl_version,
proc, &message, &msg_size);
if (OPAL_SUCCESS != rc) {
BTL_ERROR(("[%s:%d] opal_modex_recv failed for peer %s",
__FILE__, __LINE__,
OPAL_NAME_PRINT(proc->proc_name)));
OBJ_RELEASE(module_proc);
return NULL;
}
if (0 == msg_size) {
return NULL;
}
/* Message was packed in btl_openib_component.c; the format is
listed in a comment in that file */
modex_message_size = ((char *) &(dummy.end)) - ((char*) &dummy);
/* Unpack the number of modules in the message */
offset = (char *) message;
unpack8(&offset, &(module_proc->proc_port_count));
BTL_VERBOSE(("unpack: %d btls", module_proc->proc_port_count));
if (module_proc->proc_port_count > 0) {
module_proc->proc_ports = (mca_btl_openib_proc_modex_t *)
malloc(sizeof(mca_btl_openib_proc_modex_t) *
module_proc->proc_port_count);
} else {
module_proc->proc_ports = NULL;
}
/* Loop over unpacking all the ports */
for (i = 0; i < module_proc->proc_port_count; i++) {
/* Unpack the modex comment message struct */
size = modex_message_size;
memcpy(&(module_proc->proc_ports[i].pm_port_info), offset, size);
#if !defined(WORDS_BIGENDIAN) && OPAL_ENABLE_HETEROGENEOUS_SUPPORT
MCA_BTL_OPENIB_MODEX_MSG_NTOH(module_proc->proc_ports[i].pm_port_info);
#endif
offset += size;
BTL_VERBOSE(("unpacked btl %d: modex message, offset now %d",
i, (int)(offset-((char*)message))));
/* Unpack the number of CPCs that follow */
unpack8(&offset, &(module_proc->proc_ports[i].pm_cpc_data_count));
BTL_VERBOSE(("unpacked btl %d: number of cpcs to follow %d (offset now %d)",
i, module_proc->proc_ports[i].pm_cpc_data_count,
(int)(offset-((char*)message))));
module_proc->proc_ports[i].pm_cpc_data = (opal_btl_openib_connect_base_module_data_t *)
calloc(module_proc->proc_ports[i].pm_cpc_data_count,
sizeof(opal_btl_openib_connect_base_module_data_t));
if (NULL == module_proc->proc_ports[i].pm_cpc_data) {
return NULL;
}
/* Unpack the CPCs */
for (j = 0; j < module_proc->proc_ports[i].pm_cpc_data_count; ++j) {
uint8_t u8;
opal_btl_openib_connect_base_module_data_t *cpcd;
cpcd = module_proc->proc_ports[i].pm_cpc_data + j;
unpack8(&offset, &u8);
BTL_VERBOSE(("unpacked btl %d: cpc %d: index %d (offset now %d)",
i, j, u8, (int)(offset-(char*)message)));
cpcd->cbm_component =
opal_btl_openib_connect_base_get_cpc_byindex(u8);
BTL_VERBOSE(("unpacked btl %d: cpc %d: component %s",
i, j, cpcd->cbm_component->cbc_name));
unpack8(&offset, &cpcd->cbm_priority);
unpack8(&offset, &cpcd->cbm_modex_message_len);
BTL_VERBOSE(("unpacked btl %d: cpc %d: priority %d, msg len %d (offset now %d)",
i, j, cpcd->cbm_priority,
cpcd->cbm_modex_message_len,
(int)(offset-(char*)message)));
if (cpcd->cbm_modex_message_len > 0) {
cpcd->cbm_modex_message = malloc(cpcd->cbm_modex_message_len);
if (NULL == cpcd->cbm_modex_message) {
BTL_ERROR(("Failed to malloc"));
return NULL;
}
memcpy(cpcd->cbm_modex_message, offset,
cpcd->cbm_modex_message_len);
offset += cpcd->cbm_modex_message_len;
BTL_VERBOSE(("unpacked btl %d: cpc %d: blob unpacked %d %x (offset now %d)",
i, j,
((uint32_t*)cpcd->cbm_modex_message)[0],
((uint32_t*)cpcd->cbm_modex_message)[1],
(int)(offset-((char*)message))));
}
}
}
if (0 == module_proc->proc_port_count) {
module_proc->proc_endpoints = NULL;
} else {
module_proc->proc_endpoints = (mca_btl_base_endpoint_t**)
malloc(module_proc->proc_port_count *
sizeof(mca_btl_base_endpoint_t*));
}
if (NULL == module_proc->proc_endpoints) {
OBJ_RELEASE(module_proc);
return NULL;
}
BTL_VERBOSE(("unpacking done!"));
return module_proc;
}
mca_btl_tcp_proc_t* mca_btl_tcp_proc_create(opal_proc_t* proc)
{
mca_btl_tcp_proc_t* btl_proc;
size_t size;
int rc;
OPAL_THREAD_LOCK(&mca_btl_tcp_component.tcp_lock);
rc = opal_proc_table_get_value(&mca_btl_tcp_component.tcp_procs,
proc->proc_name, (void**)&btl_proc);
if(OPAL_SUCCESS == rc) {
OPAL_THREAD_UNLOCK(&mca_btl_tcp_component.tcp_lock);
return btl_proc;
}
do {
btl_proc = OBJ_NEW(mca_btl_tcp_proc_t);
if(NULL == btl_proc) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
break;
}
btl_proc->proc_opal = proc;
OBJ_RETAIN(btl_proc->proc_opal);
/* lookup tcp parameters exported by this proc */
OPAL_MODEX_RECV(rc, &mca_btl_tcp_component.super.btl_version,
&proc->proc_name, (uint8_t**)&btl_proc->proc_addrs, &size);
if(rc != OPAL_SUCCESS) {
if(OPAL_ERR_NOT_FOUND != rc)
BTL_ERROR(("opal_modex_recv: failed with return value=%d", rc));
break;
}
if(0 != (size % sizeof(mca_btl_tcp_addr_t))) {
BTL_ERROR(("opal_modex_recv: invalid size %lu: btl-size: %lu\n",
(unsigned long) size, (unsigned long)sizeof(mca_btl_tcp_addr_t)));
rc = OPAL_ERROR;
break;
}
btl_proc->proc_addr_count = size / sizeof(mca_btl_tcp_addr_t);
/* allocate space for endpoint array - one for each exported address */
btl_proc->proc_endpoints = (mca_btl_base_endpoint_t**)
malloc((1 + btl_proc->proc_addr_count) *
sizeof(mca_btl_base_endpoint_t*));
if(NULL == btl_proc->proc_endpoints) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
break;
}
if(NULL == mca_btl_tcp_component.tcp_local && (proc == opal_proc_local_get())) {
mca_btl_tcp_component.tcp_local = btl_proc;
}
/* convert the OPAL addr_family field to OS constants,
* so we can check for AF_INET (or AF_INET6) and don't have
* to deal with byte ordering anymore.
*/
for (unsigned int i = 0; i < btl_proc->proc_addr_count; i++) {
if (MCA_BTL_TCP_AF_INET == btl_proc->proc_addrs[i].addr_family) {
btl_proc->proc_addrs[i].addr_family = AF_INET;
}
#if OPAL_ENABLE_IPV6
if (MCA_BTL_TCP_AF_INET6 == btl_proc->proc_addrs[i].addr_family) {
btl_proc->proc_addrs[i].addr_family = AF_INET6;
}
#endif
}
} while (0);
if (OPAL_SUCCESS == rc) {
/* add to hash table of all proc instance. */
opal_proc_table_set_value(&mca_btl_tcp_component.tcp_procs,
proc->proc_name, btl_proc);
} else {
if (btl_proc) {
OBJ_RELEASE(btl_proc);
btl_proc = NULL;
}
}
OPAL_THREAD_UNLOCK(&mca_btl_tcp_component.tcp_lock);
return btl_proc;
}
int
ompi_mtl_portals4_add_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs,
struct ompi_proc_t** procs)
{
int ret, me;
size_t i;
bool new_found = false;
ptl_process_t *maptable;
if (ompi_mtl_portals4.use_logical) {
maptable = malloc(sizeof(ptl_process_t) * nprocs);
if (NULL == maptable) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: malloc failed\n",
__FILE__, __LINE__);
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
/* Get the list of ptl_process_id_t from the runtime and copy into structure */
for (i = 0 ; i < nprocs ; ++i) {
ptl_process_t *modex_id;
size_t size;
if( procs[i] == ompi_proc_local_proc ) {
me = i;
}
if (procs[i]->super.proc_arch != ompi_proc_local()->super.proc_arch) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"Portals 4 MTL does not support heterogeneous operations.");
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"Proc %s architecture %x, mine %x.",
OMPI_NAME_PRINT(&procs[i]->super.proc_name),
procs[i]->super.proc_arch, ompi_proc_local()->super.proc_arch);
return OMPI_ERR_NOT_SUPPORTED;
}
OPAL_MODEX_RECV(ret, &mca_mtl_portals4_component.mtl_version,
&procs[i]->super.proc_name, (uint8_t**)&modex_id, &size);
if (OMPI_SUCCESS != ret) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: ompi_modex_recv failed: %d\n",
__FILE__, __LINE__, ret);
return ret;
} else if (sizeof(ptl_process_t) != size) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: ompi_modex_recv failed: %d\n",
__FILE__, __LINE__, ret);
return OMPI_ERR_BAD_PARAM;
}
if (NULL == procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_PORTALS4]) {
ptl_process_t *peer_id;
peer_id = malloc(sizeof(ptl_process_t));
if (NULL == peer_id) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: malloc failed: %d\n",
__FILE__, __LINE__, ret);
return OMPI_ERR_OUT_OF_RESOURCE;
}
if (ompi_mtl_portals4.use_logical) {
peer_id->rank = i;
maptable[i].phys.pid = modex_id->phys.pid;
maptable[i].phys.nid = modex_id->phys.nid;
opal_output_verbose(50, ompi_mtl_base_framework.framework_output,
"logical: global rank=%d pid=%d nid=%d\n",
(int)i, maptable[i].phys.pid, maptable[i].phys.nid);
} else {
*peer_id = *modex_id;
}
procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_PORTALS4] = peer_id;
new_found = true;
} else {
ptl_process_t *proc = (ptl_process_t*) procs[i]->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_PORTALS4];
if (ompi_mtl_portals4.use_logical) {
if ((size_t)proc->rank != i) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: existing peer and rank don't match\n",
__FILE__, __LINE__);
return OMPI_ERROR;
}
maptable[i].phys.pid = modex_id->phys.pid;
maptable[i].phys.nid = modex_id->phys.nid;
}
else if (proc->phys.nid != modex_id->phys.nid ||
proc->phys.pid != modex_id->phys.pid) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: existing peer and modex peer don't match\n",
__FILE__, __LINE__);
return OMPI_ERROR;
}
}
}
if (ompi_mtl_portals4.use_logical) {
ret = PtlSetMap(ompi_mtl_portals4.ni_h, nprocs, maptable);
if (OMPI_SUCCESS != ret) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: logical mapping failed: %d\n",
__FILE__, __LINE__, ret);
return ret;
}
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"logical mapping OK\n");
free(maptable);
}
portals4_init_interface();
/* activate progress callback */
ret = opal_progress_register(ompi_mtl_portals4_progress);
if (OMPI_SUCCESS != ret) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: opal_progress_register failed: %d\n",
__FILE__, __LINE__, ret);
return ret;
}
#if OMPI_MTL_PORTALS4_FLOW_CONTROL
if (new_found) {
ret = ompi_mtl_portals4_flowctl_add_procs(me, nprocs, procs);
if (OMPI_SUCCESS != ret) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: flowctl_add_procs failed: %d\n",
__FILE__, __LINE__, ret);
return ret;
}
}
#endif
return OMPI_SUCCESS;
}
