hwloc_get_last_cpu_location fails in chroot when /proc not mounted

[bgoglin]

(moved from @dgloe second comment of #213 to here)

When I try again with debugging information turned on I get a different error:

hwloc_get_last_cpu_location: hwloc_get_last_cpu_location.c:28: check: Assertion `!ret' failed.
./wrapper.sh: line 14: 20654 Aborted                 (core dumped) "$@"

[bgoglin]

What kind of debug did you enable? Compiler options? Or hwloc --enable-debug? I don't see why debugging would change anything here. This function shouldn't check. Can you printf errno?

[dgloe]

By enabling debugging I just mean adding -g to CFLAGS so I can get information out of the core file.

[dgloe]

The errno value is 38.

[bgoglin]

Try this. If it fails to build, we need to see if SYS_gettid was moved somewhere else, or is just missing for real. If it builds, you'll get a print, hopfully.

diff --git a/src/topology-linux.c b/src/topology-linux.c
index 26bde8e..bea4c9b 100644
--- a/src/topology-linux.c
+++ b/src/topology-linux.c
@@ -1014,6 +1014,7 @@ hwloc_linux_get_tid_last_cpu_location(hwloc_topology_t topology __hwloc_attribut
 #ifdef SYS_gettid
     tid = syscall(SYS_gettid);
 #else
+#error no SYS_gettid
     errno = ENOSYS;
     return -1;
 #endif
@@ -1022,18 +1023,21 @@ hwloc_linux_get_tid_last_cpu_location(hwloc_topology_t topology __hwloc_attribut
   snprintf(name, sizeof(name), "/proc/%lu/stat", (unsigned long) tid);
   file = fopen(name, "r");
   if (!file) {
+printf("fopen %s %s\n", name, strerror(errno));
     errno = ENOSYS;
     return -1;
   }
   tmp = fgets(buf, sizeof(buf), file);
   fclose(file);
   if (!tmp) {
+printf("fgets %s\n", strerror(errno));
     errno = ENOSYS;
     return -1;
   }

   tmp = strrchr(buf, ')');
   if (!tmp) {
+printf("strrchr failed\n");
     errno = ENOSYS;
     return -1;
   }
@@ -1044,6 +1048,7 @@ hwloc_linux_get_tid_last_cpu_location(hwloc_topology_t topology __hwloc_attribut
   for(i=0; i<36; i++) {
     tmp = strchr(tmp, ' ');
     if (!tmp) {
+printf("strchr %u failed\n", i);
       errno = ENOSYS;
       return -1;
     }
@@ -1053,6 +1058,7 @@ hwloc_linux_get_tid_last_cpu_location(hwloc_topology_t topology __hwloc_attribut

   /* read the last cpu in the 38th field now */
   if (sscanf(tmp, "%d ", &i) != 1) {
+printf("sscanf %s failed\n", tmp);
     errno = ENOSYS;
     return -1;
   }

[dgloe]

fopen /proc/25645/stat No such file or directory

Looks like the chroot didn't have the /proc filesystem set up.
Executing mount -t proc proc proc/ fixed the issue.

[bgoglin]

Thanks, I retitled for documentation and I am closing now.