Merge pull request #609 from bgoglin/pid-children

combine hwloc-ps and lstopo for more customizable lstopo --top

Author: bgoglin

contrib/completion/bash/hwloc

@@ -1,5 +1,5 @@
 #
-# Copyright © 2018-2022 Inria.  All rights reserved.
+# Copyright © 2018-2023 Inria.  All rights reserved.
 # See COPYING in top-level directory.
 #
 
@@ -80,6 +80,7 @@ _lstopo() {
 		   --export-xml-flags
 		   --export-synthetic-flags
 		   --ps --top
+		   --misc-from
 		   --version
 		   -h --help
 		  )
@@ -480,6 +481,7 @@ complete -F _hwloc_distrib hwloc-distrib
 _hwloc_ps(){
     local OPTIONS=(-a
 		   --pid
+		   --children-of-pid
 		   --name
 		   --uid
 		   -l --logical
@@ -491,6 +493,7 @@ _hwloc_ps(){
 		   --pid-cmd
 		   --short-name
 		   --disallowed --whole-system
+		   --lstopo-misc
 		   --json-server
 		   --json-port
 		   -v --verbose
@@ -510,7 +513,7 @@ _hwloc_ps(){
 	--uid)
 	    COMPREPLY=( "<uid>" "all" "" )
 	    ;;
-	--pid)
+	--pid|--children-of-pid)
 	    COMPREPLY=( "<pid>" "" )
 	    ;;
 	--pid-cmd)

utils/hwloc/common-ps.c

@@ -393,3 +393,63 @@ int hwloc_ps_foreach_process(hwloc_topology_t topology, hwloc_const_bitmap_t top
   return -1;
 #endif /* HAVE_DIRENT_H */
 }
+
+int hwloc_ps_foreach_child(hwloc_topology_t topology, hwloc_const_bitmap_t topocpuset,
+                           long pid,
+                           void (*callback)(hwloc_topology_t topology, struct hwloc_ps_process *proc, void *cbdata),
+                           void *cbdata,
+                           unsigned long flags, const char *only_name, long uid)
+{
+#ifdef HAVE_DIRENT_H
+  struct hwloc_ps_process proc;
+  DIR *taskdir;
+  char path[512];
+
+  proc.pid = pid;
+  proc.cpuset = NULL;
+  proc.nthreads = 0;
+  proc.nboundthreads = 0;
+  proc.threads = NULL;
+  if (hwloc_ps_read_process(topology, topocpuset, &proc, flags) < 0)
+    goto next;
+  if (only_name && !strstr(proc.name, only_name))
+    goto next;
+  if (uid != HWLOC_PS_ALL_UIDS && proc.uid != HWLOC_PS_ALL_UIDS && proc.uid != uid)
+    goto next;
+  callback(topology, &proc, cbdata);
+ next:
+  hwloc_ps_free_process(&proc);
+
+  snprintf(path, sizeof(path), "/proc/%ld/task", proc.pid);
+  taskdir = opendir(path); /* should be enough for the vast majority of cases */
+  if (taskdir) {
+    struct dirent *taskdirent;
+    while ((taskdirent = readdir(taskdir))) {
+      char pidline[4096];
+      FILE *file;
+      char *begin;
+      size_t len;
+      snprintf(path, sizeof(path), "/proc/%ld/task/%s/children", proc.pid, taskdirent->d_name);
+      file = fopen(path, "r");
+      if (!file)
+        continue;
+      len = fread(pidline, 1, sizeof(pidline)-1, file);
+      fclose(file);
+      pidline[len] = '\0';
+      begin = pidline;
+      while (1) {
+        char *end;
+        long childpid = strtoul(begin, &end, 10);
+        if (end == begin)
+          break;
+        hwloc_ps_foreach_child(topology, topocpuset, childpid, callback, cbdata, flags, only_name, uid);
+        begin = end;
+      }
+    }
+    closedir(taskdir);
+  }
+  return 0;
+#else /* HAVE_DIRENT_H */
+  return -1;
+#endif /* HAVE_DIRENT_H */
+}

utils/hwloc/common-ps.h

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2009-2021 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -45,6 +45,12 @@ int hwloc_ps_foreach_process(hwloc_topology_t topology, hwloc_const_bitmap_t top
 			     void *cbdata,
 			     unsigned long flags, const char *only_name, long only_uid);
 
+int hwloc_ps_foreach_child(hwloc_topology_t topology, hwloc_const_bitmap_t topocpuset,
+                           long pid,
+                           void (*callback)(hwloc_topology_t topology, struct hwloc_ps_process *proc, void *cbdata),
+			     void *cbdata,
+			     unsigned long flags, const char *only_name, long only_uid);
+
 void hwloc_ps_pidcmd(struct hwloc_ps_process *proc, const char *pidcmd);
 
 void hwloc_ps_free_process(struct hwloc_ps_process *proc);

utils/hwloc/hwloc-ps.1in

@@ -1,5 +1,5 @@
 .\" -*- nroff -*-
-.\" Copyright © 2010-2021 Inria.  All rights reserved.
+.\" Copyright © 2010-2023 Inria.  All rights reserved.
 .\" Copyright © 2009-2010 Cisco Systems, Inc.  All rights reserved.
 .\" See COPYING in top-level directory.
 .TH HWLOC-PS "1" "%HWLOC_DATE%" "%PACKAGE_VERSION%" "%PACKAGE_NAME%"
@@ -28,6 +28,10 @@ specific part of the machine.
 Only show process of PID \fI<pid>\fR,
 even if it is not bound to any specific part of the machine.
 .TP
+\fB\-\-children\-of\-pid <pid>\fR
+Only show process of PID \fI<pid>\fR and its hierarchy of children,
+even if they are not bound to any specific part of the machine.
+.TP
 \fB\-\-name <name>\fR
 Only show processes whose name contains \fI<name>\fR,
 even if they are not bound to any specific part of the machine.
@@ -82,6 +86,10 @@ in each process and display it at the end of the line.
 On Linux, try to find the process MPI rank (by querying some widespread
 environment variables) and display it at the end of the line.
 
+.TP
+\fB\-\-lstopo\-misc\fR <file>
+Output a file that may be given to \fBlstopo \-\-misc\-from\fR for displaying
+processes/threads as Misc objects. See EXAMPLES below.
 .TP
 \fB\-\-json\-server\fR
 Run the tool as a JSON server that waits for other process' requests
@@ -171,6 +179,27 @@ it only appears in the thread-aware output (or if explicitly selected):
     $ hwloc-ps --pid 4759
     4759	Machine:0		myprogram
 
+The output may be a file that lstopo uses for adding Misc objects
+(more flexible version of lstopo --top):
+
+    $ hwloc-ps --misc-from foo
+    $ cat foo
+    name=12444 myprogram
+    cpuset=0x000000f0
+    subtype=Process
+
+    name=12444 mythread1
+    cpuset=0x00000050
+    subtype=Thread
+
+    name=12444 mythread2
+    cpuset=0x000000a0
+    subtype=Thread
+
+This may be directly given to lstopo:
+
+    $ hwloc-ps --misc-from - | lstopo --misc-from -
+
 On Linux, hwloc-ps may also display some process specific environment
 variable at the end of the line. This is for instance useful
 for identify MPI ranks among processes:

utils/hwloc/hwloc-ps.c

@@ -1,5 +1,5 @@
 /*
- * Copyright © 2009-2022 Inria.  All rights reserved.
+ * Copyright © 2009-2023 Inria.  All rights reserved.
  * Copyright © 2009-2012 Université Bordeaux
  * Copyright © 2009-2011 Cisco Systems, Inc.  All rights reserved.
  * See COPYING in top-level directory.
@@ -35,18 +35,21 @@ static int logical = 1;
 static int single_ancestor = 0;
 #define NO_ONLY_PID -1
 static long only_pid = NO_ONLY_PID;
+static long children_of_pid = NO_ONLY_PID;
 static long only_uid;
 static int json_server = 0;
 static int json_port = JSON_PORT;
 static FILE *json_output = NULL;
 static int verbose = 0;
+static FILE *lstopo_misc_output = NULL;
 
 void usage(const char *name, FILE *where)
 {
   fprintf (where, "Usage: %s [ options ] ...\n", name);
   fprintf (where, "Options:\n");
   fprintf (where, "  -a                 Show all processes, including those that are not bound\n");
   fprintf (where, "  --pid <pid>        Only show process of pid number <pid>\n");
+  fprintf (where, "  --children-of-pid <pid> Only show process of pid number <pid> and its children\n");
   fprintf (where, "  --name <name>      Only show processes whose name contains <name>\n");
 #ifdef HWLOC_LINUX_SYS
   fprintf (where, "  --uid <uid>        Only show processes of the user with the given uid\n");
@@ -64,6 +67,7 @@ void usage(const char *name, FILE *where)
   fprintf (where, "  --pid-cmd <cmd>    Append the output of <cmd> <pid> to each PID line\n");
   fprintf (where, "  --short-name       Show only the process short name instead of the path\n");
   fprintf (where, "  --disallowed       Include objects disallowed by administrative limitations\n");
+  fprintf (where, "  --lstopo-misc <file>  Output Misc object to be given to lstopo --misc-from <file>\n");
   fprintf (where, "  --json-server      Run as a JSON server\n");
   fprintf (where, "  --json-port <n>    Use port <n> for JSON server (default is %d)\n", JSON_PORT);
   fprintf (where, "  -v --verbose       Increase verbosity\n");
@@ -133,6 +137,45 @@ static void print_process(hwloc_topology_t topology,
 	print_task(topology, proc->threads[i].tid, proc->threads[i].name, proc->threads[i].cpuset, NULL, 1);
 }
 
+static void print_process_lstopo_misc(hwloc_topology_t topology __hwloc_attribute_unused,
+                                      struct hwloc_ps_process *proc)
+{
+  /* sort of similar to foreach_process_cb() in lstopo.c */
+  char name[100];
+  char *s;
+  unsigned i;
+
+  snprintf(name, sizeof(name), "%ld", proc->pid);
+  if (*proc->name)
+    snprintf(name, sizeof(name), "%ld %s", proc->pid, proc->name);
+  hwloc_bitmap_asprintf(&s, proc->cpuset);
+  fprintf(lstopo_misc_output,
+          "name=%s\n"
+          "cpuset=%s\n"
+          "subtype=Process\n"
+          "\n",
+          name, s);
+  free(s);
+
+  if (proc->nthreads)
+    for(i=0; i<proc->nthreads; i++)
+      if (proc->threads[i].cpuset) {
+	char task_name[150];
+	if (*proc->threads[i].name)
+	  snprintf(task_name, sizeof(task_name), "%s %li %s", name, proc->threads[i].tid, proc->threads[i].name);
+	else
+	  snprintf(task_name, sizeof(task_name), "%s %li", name, proc->threads[i].tid);
+        hwloc_bitmap_asprintf(&s, proc->threads[i].cpuset);
+        fprintf(lstopo_misc_output,
+                "name=%s\n"
+                "cpuset=%s\n"
+                "subtype=Thread\n"
+                "\n",
+                task_name, s);
+        free(s);
+      }
+}
+
 static void print_process_json(hwloc_topology_t topology,
 			       struct hwloc_ps_process *proc)
 {
@@ -197,22 +240,28 @@ static void foreach_process_cb(hwloc_topology_t topology,
   const char *pidcmd = cbdata;
 
   /* don't print anything if the process isn't bound and if no threads are bound and if not showing all */
-  if (!proc->bound && (!proc->nthreads || !proc->nboundthreads) && !show_all && !only_name)
+  if (!proc->bound && (!proc->nthreads || !proc->nboundthreads) && !show_all && !only_name && children_of_pid == NO_ONLY_PID)
     return;
 
   if (pidcmd)
     hwloc_ps_pidcmd(proc, pidcmd);
 
   if (json_output)
     print_process_json(topology, proc);
+  else if (lstopo_misc_output)
+    print_process_lstopo_misc(topology, proc);
   else
     print_process(topology, proc);
 }
 
 static int run(hwloc_topology_t topology, hwloc_const_bitmap_t topocpuset,
 	       unsigned long psflags, char *pidcmd)
 {
-  if (only_pid == NO_ONLY_PID) {
+  if (children_of_pid != NO_ONLY_PID) {
+    /* show children */
+    return hwloc_ps_foreach_child(topology, topocpuset, children_of_pid, foreach_process_cb, pidcmd, psflags, only_name, only_uid);
+
+  } else if (only_pid == NO_ONLY_PID) {
     /* show all */
     return hwloc_ps_foreach_process(topology, topocpuset, foreach_process_cb, pidcmd, psflags, only_name, only_uid);
 
@@ -235,6 +284,8 @@ static int run(hwloc_topology_t topology, hwloc_const_bitmap_t topocpuset,
 
       if (json_output)
 	print_process_json(topology, &proc);
+      else if (lstopo_misc_output)
+        print_process_lstopo_misc(topology, &proc);
       else
 	print_process(topology, &proc);
     }
@@ -317,6 +368,7 @@ run_json_server(hwloc_topology_t topology, hwloc_const_bitmap_t topocpuset)
 
       only_name = NULL;
       only_pid = NO_ONLY_PID;
+      children_of_pid = NO_ONLY_PID;
       current = req;
       while (*current) {
 	if (!strncmp(current, "lastcpulocation ", 16)) {
@@ -338,6 +390,10 @@ run_json_server(hwloc_topology_t topology, hwloc_const_bitmap_t topocpuset)
 	  psflags |= HWLOC_PS_FLAG_THREADS;
 	  show_all = 1;
 	  break;
+	} else if (!strncmp(current, "childrenofpid=", 14)) {
+	  children_of_pid = atoi(current+14);
+	  show_all = 1;
+	  break;
 	} else if (!strncmp(current, "name=", 5)) {
 	  only_name = current+5;
 	  show_all = 1;
@@ -418,6 +474,13 @@ int main(int argc, char *argv[])
       }
       only_pid = strtol(argv[1], NULL, 10);
       opt = 1;
+    } else if (!strcmp(argv[0], "--children-of-pid")) {
+      if (argc < 2) {
+	usage(callname, stderr);
+	exit(EXIT_FAILURE);
+      }
+      children_of_pid = strtol(argv[1], NULL, 10);
+      opt = 1;
     } else if (!strcmp(argv[0], "--name")) {
       if (argc < 2) {
 	usage(callname, stderr);
@@ -449,6 +512,21 @@ int main(int argc, char *argv[])
       pidcmd = argv[1];
       opt = 1;
 
+    } else if (!strcmp (argv[0], "--lstopo-misc")) {
+      if (argc < 2) {
+	usage(callname, stderr);
+	exit(EXIT_FAILURE);
+      }
+      if (!strcmp(argv[1], "-"))
+        lstopo_misc_output = stdout;
+      else
+        lstopo_misc_output = fopen(argv[1], "w");
+      if (!lstopo_misc_output) {
+        fprintf(stderr, "Failed to open --lstopo-misc output `%s' for writing (%s)\n", argv[1], strerror(errno));
+        exit(EXIT_FAILURE);
+      }
+      opt = 1;
+
     } else if (!strcmp (argv[0], "--json-server")) {
       json_server = 1;
     } else if (!strcmp (argv[0], "--json-port")) {
@@ -522,5 +600,7 @@ int main(int argc, char *argv[])
  out_with_topology:
   hwloc_topology_destroy(topology);
  out:
+  if (lstopo_misc_output && lstopo_misc_output != stdout)
+    fclose(lstopo_misc_output);
   return err;
 }

utils/lstopo/lstopo-draw.c

@@ -1139,7 +1139,8 @@ lstopo_set_object_color(struct lstopo_output *loutput,
   }
 
   case HWLOC_OBJ_MISC:
-    if (loutput->show_process_color && obj->subtype && !strcmp(obj->subtype, "Process"))
+    if (loutput->show_process_color && obj->subtype &&
+        (!strcmp(obj->subtype, "Process") || !strcmp(obj->subtype, "Thread")))
       s->bg = &loutput->palette->process;
     else
       s->bg = &loutput->palette->misc;