/**
* opal_carto_base_get_nodes_distance - returns the distance of
* all the nodes from the reference node.
*
* @param graph
* @param reference_node
* @param node_type the type of the nodes in the returned array
* @param dist_array
*
* @return int number of nodes in the returned array.
*/
int opal_carto_base_get_nodes_distance_fn(opal_carto_graph_t *graph, opal_carto_base_node_t *reference_node,
const char *node_type, opal_value_array_t *dist_array)
{
opal_value_array_t *distance_array;
vertex_distance_from_t *vertex_distance;
opal_carto_base_node_t *node;
uint32_t i, graph_order;
int distance_array_size;
opal_carto_node_distance_t node_distance;
distance_array = OBJ_NEW(opal_value_array_t);
opal_value_array_init(distance_array, sizeof(vertex_distance_from_t));
opal_value_array_reserve(distance_array,50);
/* use dijkstra algorithm to receive the distance of all the nodes from the referenced node */
graph_order = opal_graph_dijkstra(graph, reference_node->vertex, distance_array);
/* for all the nodes in the dijkstra array */
for (i = 0, distance_array_size = 0; i < graph_order; i++) {
vertex_distance = opal_value_array_get_item(distance_array, i);
node = vertex_distance->vertex->vertex_data;
/* check if the node is in the correct type */
if (NULL == node_type || 0 == strcmp(node->node_type, node_type)) {
/* assigne the result distance array */
node_distance.node = vertex_distance->vertex->vertex_data;
node_distance.node_distance = vertex_distance->weight;
opal_value_array_append_item(dist_array, (void *)&node_distance);
}
}
/* return the result distance array */
return distance_array_size;
}
/**
* This graph API returns all the adjacents of a vertex and the
* distance (weight) of those adjacents and the vertex.
*
* @param graph
* @param vertex The reference vertex
* @param adjacents An allocated pointer array of vertices and
* their distance from the reference vertex.
* Note that this pointer should be free after
* usage by the user
*
* @return int the number of adjacents in the list.
*/
int opal_graph_get_adjacent_vertices(opal_graph_t *graph, opal_graph_vertex_t *vertex, opal_value_array_t *adjacents)
{
opal_adjacency_list_t *adj_list;
opal_graph_edge_t *edge;
int adjacents_number;
opal_list_item_t *item;
vertex_distance_from_t distance_from;
int i;
/**
* Verify that the vertex belongs to the graph.
*/
if (graph != vertex->in_graph) {
OPAL_OUTPUT((0,"Vertex %p not in the graph %p\n", (void *)vertex, (void *)graph));
return 0;
}
/**
* find the adjacency list that this vertex belongs to
*/
adj_list = (opal_adjacency_list_t *) vertex->in_adj_list;
/* find the number of adjcents of this vertex */
adjacents_number = opal_list_get_size(adj_list->edges);
/* Run on all the edges from this vertex */
for (item = opal_list_get_first(adj_list->edges), i = 0;
item != opal_list_get_end(adj_list->edges);
item = opal_list_get_next(item), i++) {
edge = (opal_graph_edge_t *)item;
/* assign vertices and their weight in the adjcents list */
distance_from.vertex = edge->end;
distance_from.weight = edge->weight;
opal_value_array_append_item(adjacents, &distance_from);
}
/* return the number of the adjacents in the list */
return adjacents_number;
}
static void start_sequence(orte_jobid_t jobid, orte_node_t *node,
orte_regex_node_t *ndreg, char suffix, int32_t nodenum)
{
int32_t j, ppn;
orte_vpid_t start_vpid, end_vpid;
orte_node_rank_t nrank;
opal_value_array_append_item(&ndreg->suffix, &suffix);
opal_value_array_append_item(&ndreg->nodes, &nodenum);
j = 0;
opal_value_array_append_item(&ndreg->cnt, &j);
compute_vpids(node, jobid, &start_vpid, &end_vpid, &ppn, &nrank);
opal_value_array_append_item(&ndreg->starting_vpid, &start_vpid);
opal_value_array_append_item(&ndreg->ppn, &ppn);
opal_value_array_append_item(&ndreg->nrank, &nrank);
}
/* ////////////////////////////////////////////////////////////////////////// */
static int
verbs_runtime_query(mca_base_module_t **module,
int *priority,
const char *hint)
{
int rc = OSHMEM_SUCCESS;
openib_device_t my_device;
openib_device_t *device = &my_device;
int num_devs = 0;
int i = 0;
*priority = 0;
*module = NULL;
memset(device, 0, sizeof(*device));
#ifdef HAVE_IBV_GET_DEVICE_LIST
device->ib_devs = ibv_get_device_list(&num_devs);
#else
#error unsupported ibv_get_device_list in infiniband/verbs.h
#endif
if (num_devs == 0 || !device->ib_devs) {
return OSHMEM_ERR_NOT_SUPPORTED;
}
/* Open device */
if (NULL != mca_sshmem_verbs_component.hca_name) {
for (i = 0; i < num_devs; i++) {
if (0 == strcmp(mca_sshmem_verbs_component.hca_name, ibv_get_device_name(device->ib_devs[i]))) {
device->ib_dev = device->ib_devs[i];
break;
}
}
} else {
device->ib_dev = device->ib_devs[0];
}
if (NULL == device->ib_dev) {
rc = OSHMEM_ERR_NOT_FOUND;
goto out;
}
if (NULL == (device->ib_dev_context = ibv_open_device(device->ib_dev))) {
rc = OSHMEM_ERR_RESOURCE_BUSY;
goto out;
}
/* Obtain device attributes */
if (ibv_query_device(device->ib_dev_context, &device->ib_dev_attr)) {
rc = OSHMEM_ERR_RESOURCE_BUSY;
goto out;
}
/* Allocate the protection domain for the device */
device->ib_pd = ibv_alloc_pd(device->ib_dev_context);
if (NULL == device->ib_pd) {
rc = OSHMEM_ERR_RESOURCE_BUSY;
goto out;
}
/* Allocate memory */
if (!rc) {
void *addr = NULL;
size_t size = getpagesize();
struct ibv_mr *ib_mr = NULL;
uint64_t access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ;
uint64_t exp_access_flag = 0;
OBJ_CONSTRUCT(&device->ib_mr_array, opal_value_array_t);
opal_value_array_init(&device->ib_mr_array, sizeof(struct ibv_mr *));
#if defined(MPAGE_ENABLE) && (MPAGE_ENABLE > 0)
exp_access_flag = IBV_EXP_ACCESS_ALLOCATE_MR |
IBV_EXP_ACCESS_SHARED_MR_USER_READ |
IBV_EXP_ACCESS_SHARED_MR_USER_WRITE;
#endif /* MPAGE_ENABLE */
struct ibv_exp_reg_mr_in in = {device->ib_pd, addr, size, access_flag|exp_access_flag, 0};
ib_mr = ibv_exp_reg_mr(&in);
if (NULL == ib_mr) {
rc = OSHMEM_ERR_OUT_OF_RESOURCE;
} else {
device->ib_mr_shared = ib_mr;
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
}
#if defined(MPAGE_ENABLE) && (MPAGE_ENABLE > 0)
if (!rc) {
struct ibv_exp_reg_shared_mr_in in_smr;
access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ|
IBV_EXP_ACCESS_NO_RDMA;
addr = (void *)mca_sshmem_base_start_address;
mca_sshmem_verbs_fill_shared_mr(&in_smr, device->ib_pd, device->ib_mr_shared->handle, addr, access_flag);
ib_mr = ibv_exp_reg_shared_mr(&in_smr);
if (NULL == ib_mr) {
mca_sshmem_verbs_component.has_shared_mr = 0;
} else {
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
mca_sshmem_verbs_component.has_shared_mr = 1;
}
}
#endif /* MPAGE_ENABLE */
}
/* all is well - rainbows and butterflies */
if (!rc) {
*priority = mca_sshmem_verbs_component.priority;
*module = (mca_base_module_t *)&mca_sshmem_verbs_module.super;
}
out:
if (device) {
if (opal_value_array_get_size(&device->ib_mr_array)) {
struct ibv_mr** array;
struct ibv_mr* ib_mr = NULL;
array = OPAL_VALUE_ARRAY_GET_BASE(&device->ib_mr_array, struct ibv_mr *);
while (opal_value_array_get_size(&device->ib_mr_array) > 0) {
ib_mr = array[0];
ibv_dereg_mr(ib_mr);
opal_value_array_remove_item(&device->ib_mr_array, 0);
}
if (device->ib_mr_shared) {
device->ib_mr_shared = NULL;
}
OBJ_DESTRUCT(&device->ib_mr_array);
}
if (device->ib_pd) {
ibv_dealloc_pd(device->ib_pd);
device->ib_pd = NULL;
}
if(device->ib_dev_context) {
ibv_close_device(device->ib_dev_context);
device->ib_dev_context = NULL;
}
if(device->ib_devs) {
ibv_free_device_list(device->ib_devs);
device->ib_devs = NULL;
}
}
return rc;
}
/**
* segment_attach can only be called after a successful call to segment_create
*/
static void *
segment_attach(map_segment_t *ds_buf, sshmem_mkey_t *mkey)
{
openib_device_t *device = &memheap_device;
static int mr_count = 0;
void *addr = NULL;
assert(ds_buf);
assert(mkey->va_base == 0);
if (MAP_SEGMENT_SHM_INVALID == (int)(mkey->u.key)) {
return (mkey->va_base);
}
/* workaround mtt problem - request aligned addresses */
++mr_count;
addr = (void *)((uintptr_t)mca_sshmem_base_start_address +
mca_sshmem_verbs_component.mr_interleave_factor * 1024ULL * 1024ULL * 1024ULL * mr_count);
{
struct ibv_mr *ib_mr = NULL;
uint64_t access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ |
IBV_EXP_ACCESS_NO_RDMA;
struct ibv_exp_reg_shared_mr_in in;
mca_sshmem_verbs_fill_shared_mr(&in, device->ib_pd, mkey->u.key, addr, access_flag);
ib_mr = ibv_exp_reg_shared_mr(&in);
if (NULL == ib_mr) {
mkey->va_base = (void *)-1;
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error to ibv_reg_shared_mr() %llu bytes errno says %d: %s",
(unsigned long long)ds_buf->seg_size, errno, strerror(errno))
);
} else {
if (ib_mr->addr != addr) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"Failed to map shared region to address %p got addr %p. Try to increase 'memheap_mr_interleave_factor' from %d",
addr, ib_mr->addr, mca_sshmem_verbs_component.mr_interleave_factor)
);
}
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
mkey->va_base = ib_mr->addr;
}
}
OPAL_OUTPUT_VERBOSE(
(70, oshmem_sshmem_base_framework.framework_output,
"%s: %s: attach successful "
"(id: %d, addr: %p size: %lu, name: %s | va_base: 0x%p len: %d key %llx)\n",
mca_sshmem_verbs_component.super.base_version.mca_type_name,
mca_sshmem_verbs_component.super.base_version.mca_component_name,
ds_buf->seg_id, ds_buf->super.va_base, (unsigned long)ds_buf->seg_size, ds_buf->seg_name,
mkey->va_base, mkey->len, (unsigned long long)mkey->u.key)
);
/* update returned base pointer with an offset that hides our stuff */
return (mkey->va_base);
}
/* ////////////////////////////////////////////////////////////////////////// */
static int
segment_create(map_segment_t *ds_buf,
const char *file_name,
size_t size)
{
int rc = OSHMEM_SUCCESS;
openib_device_t *device = &memheap_device;
int num_devs = 0;
int i = 0;
assert(ds_buf);
/* init the contents of map_segment_t */
shmem_ds_reset(ds_buf);
memset(device, 0, sizeof(*device));
#ifdef HAVE_IBV_GET_DEVICE_LIST
device->ib_devs = ibv_get_device_list(&num_devs);
#else
#error unsupported ibv_get_device_list in infiniband/verbs.h
#endif
if (num_devs == 0 || !device->ib_devs) {
return OSHMEM_ERR_NOT_SUPPORTED;
}
/* Open device */
if (NULL != mca_sshmem_verbs_component.hca_name) {
for (i = 0; i < num_devs; i++) {
if (0 == strcmp(mca_sshmem_verbs_component.hca_name, ibv_get_device_name(device->ib_devs[i]))) {
device->ib_dev = device->ib_devs[i];
break;
}
}
} else {
device->ib_dev = device->ib_devs[0];
}
if (NULL == device->ib_dev) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error getting device says %d: %s",
errno, strerror(errno))
);
return OSHMEM_ERR_NOT_FOUND;
}
if (NULL == (device->ib_dev_context = ibv_open_device(device->ib_dev))) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error obtaining device context for %s errno says %d: %s",
ibv_get_device_name(device->ib_dev), errno, strerror(errno))
);
return OSHMEM_ERR_RESOURCE_BUSY;
}
/* Obtain device attributes */
if (ibv_query_device(device->ib_dev_context, &device->ib_dev_attr)) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error obtaining device attributes for %s errno says %d: %s",
ibv_get_device_name(device->ib_dev), errno, strerror(errno))
);
return OSHMEM_ERR_RESOURCE_BUSY;
}
/* Allocate the protection domain for the device */
device->ib_pd = ibv_alloc_pd(device->ib_dev_context);
if (NULL == device->ib_pd) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error allocating protection domain for %s errno says %d: %s",
ibv_get_device_name(device->ib_dev), errno, strerror(errno))
);
return OSHMEM_ERR_RESOURCE_BUSY;
}
/* Allocate memory */
if (!rc) {
void *addr = NULL;
struct ibv_mr *ib_mr = NULL;
uint64_t access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ;
uint64_t exp_access_flag = 0;
OBJ_CONSTRUCT(&device->ib_mr_array, opal_value_array_t);
opal_value_array_init(&device->ib_mr_array, sizeof(struct ibv_mr *));
#if (MPAGE_ENABLE > 0)
exp_access_flag = IBV_EXP_ACCESS_ALLOCATE_MR |
IBV_EXP_ACCESS_SHARED_MR_USER_READ |
IBV_EXP_ACCESS_SHARED_MR_USER_WRITE;
#endif /* MPAGE_ENABLE */
struct ibv_exp_reg_mr_in in = {device->ib_pd, addr, size, access_flag|exp_access_flag, 0};
#if MPAGE_HAVE_IBV_EXP_REG_MR_CREATE_FLAGS
if (0 == mca_sshmem_verbs_component.has_shared_mr) {
in.addr = (void *)mca_sshmem_base_start_address;
in.comp_mask = IBV_EXP_REG_MR_CREATE_FLAGS;
in.create_flags = IBV_EXP_REG_MR_CREATE_CONTIG;
in.exp_access = access_flag;
}
#endif
ib_mr = ibv_exp_reg_mr(&in);
if (NULL == ib_mr) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error to ibv_exp_reg_mr() %llu bytes errno says %d: %s",
(unsigned long long)size, errno, strerror(errno))
);
rc = OSHMEM_ERR_OUT_OF_RESOURCE;
} else {
device->ib_mr_shared = ib_mr;
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
}
#if (MPAGE_ENABLE > 0)
if (!rc && mca_sshmem_verbs_component.has_shared_mr) {
void *addr = NULL;
access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ|
IBV_EXP_ACCESS_NO_RDMA;
addr = (void *)mca_sshmem_base_start_address;
struct ibv_exp_reg_shared_mr_in in;
mca_sshmem_verbs_fill_shared_mr(&in, device->ib_pd, device->ib_mr_shared->handle, addr, access_flag);
ib_mr = ibv_exp_reg_shared_mr(&in);
if (NULL == ib_mr) {
OPAL_OUTPUT_VERBOSE(
(5, oshmem_sshmem_base_framework.framework_output,
"error to ibv_reg_shared_mr() %llu bytes errno says %d: %s has_shared_mr: %d",
(unsigned long long)size, errno, strerror(errno),
mca_sshmem_verbs_component.has_shared_mr
)
);
rc = OSHMEM_ERR_OUT_OF_RESOURCE;
} else {
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
}
}
#endif /* MPAGE_ENABLE */
if (!rc) {
OPAL_OUTPUT_VERBOSE(
(70, oshmem_sshmem_base_framework.framework_output,
"ibv device %s shared_mr: %d",
ibv_get_device_name(device->ib_dev),
mca_sshmem_verbs_component.has_shared_mr)
);
if (mca_sshmem_verbs_component.has_shared_mr) {
assert(size == device->ib_mr_shared->length);
ds_buf->type = MAP_SEGMENT_ALLOC_IBV;
ds_buf->seg_id = device->ib_mr_shared->handle;
} else {
ds_buf->type = MAP_SEGMENT_ALLOC_IBV_NOSHMR;
ds_buf->seg_id = MAP_SEGMENT_SHM_INVALID;
}
ds_buf->super.va_base = ib_mr->addr;
ds_buf->seg_size = size;
ds_buf->super.va_end = (void*)((uintptr_t)ds_buf->super.va_base + ds_buf->seg_size);
}
}
OPAL_OUTPUT_VERBOSE(
(70, oshmem_sshmem_base_framework.framework_output,
"%s: %s: create %s "
"(id: %d, addr: %p size: %lu, name: %s)\n",
mca_sshmem_verbs_component.super.base_version.mca_type_name,
mca_sshmem_verbs_component.super.base_version.mca_component_name,
(rc ? "failure" : "successful"),
ds_buf->seg_id, ds_buf->super.va_base, (unsigned long)ds_buf->seg_size, ds_buf->seg_name)
);
return rc;
}
/**
* This graph API returns the distance (weight) from a reference
* vertex to all other vertices in the graph using the Dijkstra
* algorithm
*
* @param graph
* @param vertex The reference vertex.
* @param distance_array An array of vertices and
* their distance from the reference vertex.
*
* @return uint32_t the size of the distance array
*/
uint32_t opal_graph_dijkstra(opal_graph_t *graph, opal_graph_vertex_t *vertex, opal_value_array_t *distance_array)
{
int graph_order;
vertex_distance_from_t *Q, *q_start, *current_vertex;
opal_list_item_t *adj_list_item;
opal_adjacency_list_t *adj_list;
int number_of_items_in_q;
int i;
uint32_t weight;
/**
* Verify that the reference vertex belongs to the graph.
*/
if (graph != vertex->in_graph) {
OPAL_OUTPUT((0,"opal:graph:dijkstra: vertex %p not in the graph %p\n",(void *)vertex,(void *)graph));
return 0;
}
/* get the order of the graph and allocate a working queue accordingly */
graph_order = opal_graph_get_order(graph);
Q = (vertex_distance_from_t *)malloc(graph_order * sizeof(vertex_distance_from_t));
/* assign a pointer to the start of the queue */
q_start = Q;
/* run on all the vertices of the graph */
for (adj_list_item = opal_list_get_first(graph->adjacency_list), i=0;
adj_list_item != opal_list_get_end(graph->adjacency_list);
adj_list_item = opal_list_get_next(adj_list_item), i++) {
adj_list = (opal_adjacency_list_t *)adj_list_item;
/* insert the vertices pointes to the working queue */
Q[i].vertex = adj_list->vertex;
/**
* assign an infinity distance to all the vertices in the queue
* except the reference vertex which its distance should be 0.
*/
if (Q[i].vertex == vertex) {
Q[i].weight = 0;
}
else {
Q[i].weight = DISTANCE_INFINITY;
}
}
number_of_items_in_q = i;
/* sort the working queue according the distance from the reference vertex */
qsort(q_start, number_of_items_in_q, sizeof(vertex_distance_from_t), compare_vertex_distance);
/* while the working queue is not empty */
while (number_of_items_in_q > 0) {
/* start to work with the first vertex in the working queue */
current_vertex = q_start;
/* remove the first vertex from the queue */
q_start++;
/* decrees the number of vertices in the queue */
number_of_items_in_q--;
/* find the distance of all other vertices in the queue from the first vertex in the queue */
for (i = 0; i < number_of_items_in_q; i++) {
weight = opal_graph_adjacent(graph, current_vertex->vertex, q_start[i].vertex);
/**
* if the distance from the first vertex in the queue to the I
* vertex in the queue plus the distance of the first vertex in
* the queue from the referenced vertex is smaller than the
* distance of the I vertex from the referenced vertex, assign
* the lower distance to the I vertex.
*/
if (current_vertex->weight + weight < q_start[i].weight) {
q_start[i].weight = weight + current_vertex->weight;
}
}
/* sort again the working queue */
qsort(q_start, number_of_items_in_q, sizeof(vertex_distance_from_t), compare_vertex_distance);
}
/* copy the working queue the the returned distance array */
for (i = 0; i < graph_order-1; i++) {
opal_value_array_append_item(distance_array, (void *)&(Q[i+1]));
}
/* free the working queue */
free(Q);
/* assign the distance array size. */
return graph_order - 1;
}
static int _ibv_attach(map_segment_t *s, size_t size)
{
int rc = OSHMEM_SUCCESS;
static openib_device_t memheap_device;
openib_device_t *device = &memheap_device;
int num_devs = 0;
assert(s);
memset(device, 0, sizeof(*device));
#ifdef HAVE_IBV_GET_DEVICE_LIST
device->ib_devs = ibv_get_device_list(&num_devs);
#else
#error unsupported ibv_get_device_list in infiniband/verbs.h
#endif
if (num_devs == 0 || !device->ib_devs)
{
rc = OSHMEM_ERR_NOT_SUPPORTED;
}
/* Open device */
if (!rc)
{
int i = 0;
if (num_devs > 1)
{
if (NULL == mca_memheap_base_param_hca_name)
{
MEMHEAP_VERBOSE(5, "found %d HCAs, choosing the first", num_devs);
}
else
{
MEMHEAP_VERBOSE(5, "found %d HCAs, searching for %s", num_devs, mca_memheap_base_param_hca_name);
}
}
for (i = 0; i < num_devs; i++)
{
device->ib_dev = device->ib_devs[i];
device->ib_dev_context = ibv_open_device(device->ib_dev);
if (NULL == device->ib_dev_context)
{
MEMHEAP_ERROR("error obtaining device context for %s errno says %d: %s",
ibv_get_device_name(device->ib_dev), errno, strerror(errno));
rc = OSHMEM_ERR_RESOURCE_BUSY;
}
else
{
if (NULL != mca_memheap_base_param_hca_name)
{
if (0 == strcmp(mca_memheap_base_param_hca_name,ibv_get_device_name(device->ib_dev)))
{
MEMHEAP_VERBOSE(5, "mca_memheap_base_param_hca_name = %s, selected %s as %d of %d", mca_memheap_base_param_hca_name, ibv_get_device_name(device->ib_dev), i, num_devs);
rc = OSHMEM_SUCCESS;
break;
}
}
else
{
MEMHEAP_VERBOSE(5, "mca_memheap_base_param_hca_name = %s, selected %s as %d of %d", mca_memheap_base_param_hca_name, ibv_get_device_name(device->ib_dev), i, num_devs);
rc = OSHMEM_SUCCESS;
break;
}
}
}
}
/* Obtain device attributes */
if (!rc)
{
if (ibv_query_device(device->ib_dev_context, &device->ib_dev_attr))
{
MEMHEAP_ERROR("error obtaining device attributes for %s errno says %d: %s",
ibv_get_device_name(device->ib_dev), errno, strerror(errno));
rc = OSHMEM_ERR_RESOURCE_BUSY;
}
else
{
MEMHEAP_VERBOSE(5, "ibv device %s",
ibv_get_device_name(device->ib_dev));
}
}
/* Allocate the protection domain for the device */
if (!rc)
{
device->ib_pd = ibv_alloc_pd(device->ib_dev_context);
if (NULL == device->ib_pd)
{
MEMHEAP_ERROR("error allocating protection domain for %s errno says %d: %s",
ibv_get_device_name(device->ib_dev), errno, strerror(errno));
rc = OSHMEM_ERR_RESOURCE_BUSY;
}
}
/* Allocate memory */
if (!rc)
{
void *addr = NULL;
struct ibv_mr *ib_mr = NULL;
int access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ;
OBJ_CONSTRUCT(&device->ib_mr_array, opal_value_array_t);
opal_value_array_init(&device->ib_mr_array, sizeof(struct ibv_mr *));
#if defined(MPAGE_ENABLE) && (MPAGE_ENABLE > 0)
access_flag |= IBV_ACCESS_ALLOCATE_MR |
IBV_ACCESS_SHARED_MR_USER_READ |
IBV_ACCESS_SHARED_MR_USER_WRITE;
#endif /* MPAGE_ENABLE */
ib_mr = ibv_reg_mr(device->ib_pd, addr, size, access_flag);
if (NULL == ib_mr)
{
MEMHEAP_ERROR("error to ibv_reg_mr() %llu bytes errno says %d: %s",
(unsigned long long)size, errno, strerror(errno));
rc = OSHMEM_ERR_OUT_OF_RESOURCE;
}
else
{
device->ib_mr_shared = ib_mr;
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
}
#if defined(MPAGE_ENABLE) && (MPAGE_ENABLE > 0)
if (!rc)
{
access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ|
IBV_ACCESS_NO_RDMA;
addr = (void *)mca_memheap_base_start_address;
ib_mr = ibv_reg_shared_mr(device->ib_mr_shared->handle,
device->ib_pd, addr, access_flag);
if (NULL == ib_mr)
{
MEMHEAP_ERROR("error to ibv_reg_shared_mr() %llu bytes errno says %d: %s",
(unsigned long long)size, errno, strerror(errno));
rc = OSHMEM_ERR_OUT_OF_RESOURCE;
}
else
{
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
}
}
#endif /* MPAGE_ENABLE */
if (!rc)
{
assert(size == device->ib_mr_shared->length);
s->type = MAP_SEGMENT_ALLOC_IBV;
s->shmid = device->ib_mr_shared->handle;
s->start = ib_mr->addr;
s->size = size;
s->end = (void*)((uintptr_t)s->start + s->size);
s->context = &memheap_device;
}
}
return rc;
}
static void memheap_attach_segment(mca_spml_mkey_t *mkey, int tr_id)
{
/* process special case when va was got using shmget(IPC_PRIVATE)
* this case is notable for:
* - key is set as (type|shmid);
* - va_base is set as 0;
*/
if (!mkey->va_base
&& ((int) MEMHEAP_SHM_GET_ID(mkey->key) != MEMHEAP_SHM_INVALID)) {
MEMHEAP_VERBOSE(5,
"shared memory usage tr_id: %d key %llx base_va %p shmid 0x%X|0x%X",
tr_id,
(unsigned long long)mkey->key,
mkey->va_base,
MEMHEAP_SHM_GET_TYPE(mkey->key),
MEMHEAP_SHM_GET_ID(mkey->key));
if (MEMHEAP_SHM_GET_TYPE(mkey->key) == MAP_SEGMENT_ALLOC_SHM) {
mkey->va_base = shmat(MEMHEAP_SHM_GET_ID(mkey->key),
0,
0);
} else if (MEMHEAP_SHM_GET_TYPE(mkey->key) == MAP_SEGMENT_ALLOC_IBV) {
#if defined(MPAGE_ENABLE) && (MPAGE_ENABLE > 0)
openib_device_t *device = NULL;
struct ibv_mr *ib_mr;
void *addr;
static int mr_count;
int access_flag = IBV_ACCESS_LOCAL_WRITE |
IBV_ACCESS_REMOTE_WRITE |
IBV_ACCESS_REMOTE_READ |
IBV_ACCESS_NO_RDMA;
device = (openib_device_t *)memheap_map->mem_segs[HEAP_SEG_INDEX].context;
assert(device);
/* workaround mtt problem - request aligned addresses */
++mr_count;
addr = (void *)(mca_memheap_base_start_address + mca_memheap_base_mr_interleave_factor*1024ULL*1024ULL*1024ULL*mr_count);
ib_mr = ibv_reg_shared_mr(MEMHEAP_SHM_GET_ID(mkey->key),
device->ib_pd, addr, access_flag);
if (NULL == ib_mr)
{
mkey->va_base = (void*)-1;
MEMHEAP_ERROR("error to ibv_reg_shared_mr() errno says %d: %s",
errno, strerror(errno));
}
else
{
if (ib_mr->addr != addr) {
MEMHEAP_WARN("Failed to map shared region to address %p got addr %p. Try to increase 'memheap_mr_interleave_factor' from %d", addr, ib_mr->addr, mca_memheap_base_mr_interleave_factor);
}
opal_value_array_append_item(&device->ib_mr_array, &ib_mr);
mkey->va_base = ib_mr->addr;
}
#endif /* MPAGE_ENABLE */
} else {
MEMHEAP_ERROR("tr_id: %d key %llx attach failed: incorrect shmid 0x%X|0x%X",
tr_id,
(unsigned long long)mkey->key,
MEMHEAP_SHM_GET_TYPE(mkey->key),
MEMHEAP_SHM_GET_ID(mkey->key));
oshmem_shmem_abort(-1);
}
if ((void *) -1 == (void *) mkey->va_base) {
MEMHEAP_ERROR("tr_id: %d key %llx attach failed: errno = %d",
tr_id, (unsigned long long)mkey->key, errno);
oshmem_shmem_abort(-1);
}
}
}