adding code to figure out which processes are on the same

host, and limit the number of procs that can use a given
shared memory region.

This commit was SVN r1230.

src/mca/ptl/sm/src/ptl_sm.c

@@ -45,7 +45,41 @@ int mca_ptl_sm_add_procs(
     struct mca_ptl_base_peer_t **peers,
     ompi_bitmap_t* reachability)
 {
+    int proc,rc;
+    size_t size;
+    mca_ptl_sm_exchange_t **sm_proc_info;
+
+    /* allocate array to hold setup shared memory from all
+     * other procs */
+    sm_proc_info=(mca_ptl_sm_exchange_t **)
+        malloc(nprocs*sizeof(mca_ptl_sm_exchange_t *));
+    if( NULL == sm_proc_info ){
+        rc=OMPI_ERR_OUT_OF_RESOURCE;
+        goto CLEANUP;
+    }
+
+    /* get unique host identifier for each process in the list */
+    for( proc=0 ; proc < nprocs; proc++ ) {
+        rc = mca_base_modex_recv(
+                &mca_ptl_sm_module.super.ptlm_version, proc,
+                (void**)(sm_proc_info+proc), &size);
+        if(rc != OMPI_SUCCESS) {
+            ompi_output(0, "mca_ptl_sm_add_procs: mca_base_modex_recv: failed with return value=%d", rc);
+            goto CLEANUP;
+        }
+    }
+
+    /* free local memory */
+    if(sm_proc_info){
+        free(sm_proc_info);
+    }
+
     return OMPI_SUCCESS;
+
+CLEANUP:
+    if(sm_proc_info){
+        free(sm_proc_info);
+    }
 }
 
 

src/mca/ptl/sm/src/ptl_sm.h

@@ -27,6 +27,7 @@ struct mca_ptl_sm_module_1_0_0_t {
     int sm_free_list_num;                /**< initial size of free lists */
     int sm_free_list_max;                /**< maximum size of free lists */
     int sm_free_list_inc;                /**< number of elements to alloc when growing free lists */
+    int sm_max_procs;                    /**< upper limit on the number of processes using the shared memory pool */
     void* sm_base_addr;                  /**< base address of mmaped region */
     ompi_free_list_t sm_send_requests;    /**< free list of sm send requests -- sendreq + sendfrag */
     ompi_free_list_t sm_send_frags;       /**< free list of sm send fragments */
@@ -194,5 +195,15 @@ extern int mca_ptl_sm_send(
     int flags
 );
 
+/**
+ * Data structure used to hold information that will be exchanged with
+ * all other procs at startup.  !!!!! This is only temporary, until the
+ * registry is complete
+ */
+#define MCA_PTL_SM_MAX_HOSTNAME_LEN  128
+typedef struct mca_ptl_sm_exchange{
+    char host_name[MCA_PTL_SM_MAX_HOSTNAME_LEN];
+}mca_ptl_sm_exchange_t;
+
 #endif
 

src/mca/ptl/sm/src/ptl_sm_mmap.c

@@ -30,7 +30,7 @@ static mca_ptl_sm_mmap_t* mca_ptl_sm_mmap_open(size_t size)
 
     while(fd < 0) {
         struct timespec ts;
-        fd = shm_open(mca_ptl_sm_module.sm_mmap_file, O_CREAT|O_RDWR, 0000);
+        /*fd = shm_open(mca_ptl_sm_module.sm_mmap_file, O_CREAT|O_RDWR, 0000); */
         if(fd < 0 && errno != EACCES) {
             ompi_output(0, "mca_ptl_sm_mmap_open: open failed with errno=%d\n", errno);
             return NULL;
@@ -61,6 +61,9 @@ mca_ptl_sm_mmap_t* mca_ptl_sm_mmap_init(size_t size)
 {
     static int segnum = 0;
 
+    /* debug !!!!! */
+    return OMPI_SUCCESS;
+
     ompi_job_handle_t job_handle = mca_pcm.pcm_handle_get();
     char hostname[64];
     int fd;
@@ -69,7 +72,7 @@ mca_ptl_sm_mmap_t* mca_ptl_sm_mmap_init(size_t size)
 
     gethostname(hostname, sizeof(hostname));
     sprintf(mca_ptl_sm_module.sm_mmap_file, "/%s.%s.%d", hostname, job_handle, segnum++);
-    fd = shm_open(mca_ptl_sm_module.sm_mmap_file, O_CREAT|O_RDWR, 0000);
+    /*fd = shm_open(mca_ptl_sm_module.sm_mmap_file, O_CREAT|O_RDWR, 0000); */
     if(fd < 0) {
         if(errno == EACCES)
             return mca_ptl_sm_mmap_open(size);

src/mca/ptl/sm/src/ptl_sm_module.c

@@ -13,11 +13,13 @@
 #include "util/if.h"
 #include "util/argv.h"
 #include "util/output.h"
+#include "util/sys_info.h"
 #include "mca/pml/pml.h"
 #include "mca/ptl/ptl.h"
 #include "mca/ptl/base/ptl_base_sendreq.h"
 #include "mca/base/mca_base_param.h"
 #include "mca/base/mca_base_module_exchange.h"
+#include "mca/ptl/sm/src/ptl_sm.h"
 #include "ptl_sm.h"
 #include "ptl_sm_sendreq.h"
 #include "ptl_sm_sendfrag.h"
@@ -114,6 +116,8 @@ int mca_ptl_sm_module_open(void)
         mca_ptl_sm_param_register_int("free_list_max", -1);
     mca_ptl_sm_module.sm_free_list_inc =
         mca_ptl_sm_param_register_int("free_list_inc", 256);
+    mca_ptl_sm_module.sm_max_procs =
+        mca_ptl_sm_param_register_int("max_procs", -1);
 
     /* initialize objects */
     OBJ_CONSTRUCT(&mca_ptl_sm_module.sm_lock, ompi_mutex_t);
@@ -226,7 +230,40 @@ int mca_ptl_sm_module_progress(mca_ptl_tstamp_t tstamp)
 
 static int mca_ptl_sm_module_exchange()
 {
+    /*
+     *  !!!!  This is temporary, and will be removed when the
+     *  registry is implemented
+     */
+    mca_ptl_sm_exchange_t mca_ptl_sm_setup_info;
+    size_t len,size;
+    char *ptr;
+    int rc;
+
+    /* determine length of host name */
+    len=strlen(ompi_system_info.nodename);
+    /* check if string is zero length or there is an error */
+    if( 0 >= len) {
+        return OMPI_ERROR;
+    }
+    /* check if string is too long */
+    if( MCA_PTL_SM_MAX_HOSTNAME_LEN < (len+1) ){
+        return OMPI_ERROR;
+    }
+
+    /* copy string into structure that will be used to send data around */
+    ptr=NULL;
+    ptr=strncpy(&(mca_ptl_sm_setup_info.host_name[0]),
+            ompi_system_info.nodename, len);
+    if( NULL == ptr ) {
+        return OMPI_ERROR;
+    }
+    mca_ptl_sm_setup_info.host_name[len]='\0';
+
+    /* exchange setup information */
+    size=sizeof(mca_ptl_sm_exchange_t);
+    rc =  mca_base_modex_send(&mca_ptl_sm_module.super.ptlm_version, 
+            &mca_ptl_sm_setup_info, size);
+    
     return OMPI_SUCCESS;
 }
 
-