Merge pull request #176 from igchor/hwloc_valgrind

Fix Valgrind runs

Author: lukaszstolarczuk

.github/workflows/nightly.yml

@@ -22,7 +22,7 @@ jobs:
         - name: Install apt packages
           run: |
             sudo apt-get update
-            sudo apt-get install -y cmake libhwloc-dev libjemalloc-dev libnuma-dev libtbb-dev valgrind
+            sudo apt-get install -y cmake hwloc libhwloc-dev libjemalloc-dev libnuma-dev libtbb-dev valgrind
 
         - name: Configure CMake
           run: >

src/memory_targets/memory_target_numa.c

@@ -41,6 +41,7 @@ static umf_result_t numa_initialize(void *params, void **memTarget) {
 
 static void numa_finalize(void *memTarget) { free(memTarget); }
 
+// sets maxnode and allocates and initializes mask based on provided memory targets
 static umf_result_t
 numa_targets_create_nodemask(struct numa_memory_target_t **targets,
                              size_t numTargets, unsigned long **mask,
@@ -61,13 +62,26 @@ numa_targets_create_nodemask(struct numa_memory_target_t **targets,
         }
     }
 
-    int nrUlongs = hwloc_bitmap_nr_ulongs(bitmap);
-    if (nrUlongs == -1) {
+    int lastBit = hwloc_bitmap_last(bitmap);
+    if (lastBit == -1) {
+        // no node is set
         hwloc_bitmap_free(bitmap);
-        return UMF_RESULT_ERROR_UNKNOWN;
+        return UMF_RESULT_ERROR_INVALID_ARGUMENT;
     }
 
+    *maxnode = lastBit + 1;
+
+    // Do not use hwloc_bitmap_nr_ulongs due to:
+    // https://github.com/open-mpi/hwloc/issues/429
+    unsigned bits_per_long = sizeof(unsigned long) * 8;
+    int nrUlongs = (lastBit + bits_per_long) / bits_per_long;
+
     unsigned long *nodemask = malloc(sizeof(unsigned long) * nrUlongs);
+    if (!nodemask) {
+        hwloc_bitmap_free(bitmap);
+        return UMF_RESULT_ERROR_OUT_OF_HOST_MEMORY;
+    }
+
     int ret = hwloc_bitmap_to_ulongs(bitmap, nrUlongs, nodemask);
     hwloc_bitmap_free(bitmap);
 
@@ -77,7 +91,6 @@ numa_targets_create_nodemask(struct numa_memory_target_t **targets,
     }
 
     *mask = nodemask;
-    *maxnode = nrUlongs * sizeof(unsigned long) * 8;
 
     return UMF_RESULT_SUCCESS;
 }

src/provider/provider_os_memory.c

@@ -111,7 +111,7 @@ static umf_result_t nodemask_to_hwloc_nodeset(const unsigned long *nodemask,
 
     unsigned bits_per_mask = sizeof(unsigned long) * 8;
     hwloc_bitmap_from_ulongs(
-        *out_nodeset, (maxnode + bits_per_mask) / bits_per_mask, nodemask);
+        *out_nodeset, (maxnode + bits_per_mask - 1) / bits_per_mask, nodemask);
 
     return UMF_RESULT_SUCCESS;
 }

test/test_valgrind.sh

@@ -49,13 +49,18 @@ esac
 
 FAIL=0
 
+mkdir -p cpuid
+
+echo "Gathering data for hwloc so it can be run under valgrind:"
+hwloc-gather-cpuid ./cpuid
+
 echo "Running: \"valgrind $OPTION\" for the following tests:"
 
 for tf in $(ls -1 ./test/umf_test-*); do
 	[ ! -x $tf ] && continue
 	echo -n "$tf "
 	LOG=${tf}.log
-	valgrind $OPTION $tf >$LOG 2>&1 || echo -n "(valgrind failed) "
+	HWLOC_CPUID_PATH=./cpuid valgrind $OPTION $tf >$LOG 2>&1 || echo -n "(valgrind failed) "
 	if grep -q -e "ERROR SUMMARY: 0 errors from 0 contexts" $LOG; then
 		echo "- OK"
 		rm $LOG