diff --git a/CMakeLists.txt b/CMakeLists.txt
index 5d09dca..61359b7 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -54,16 +54,18 @@ if(USE_SBRK)
   add_definitions(-DUSE_SBRK)
 endif()
 
-if(NOT MSVC)
-  add_library(snmallocshim SHARED src/override/malloc.cc)
-  target_link_libraries(snmallocshim -pthread)
-  target_include_directories(snmallocshim PRIVATE src)
-endif()
+macro(add_shim name)
+  add_library(${name} SHARED src/override/malloc.cc)
+  target_link_libraries(${name} -pthread)
+  target_include_directories(${name} PRIVATE src)
+  target_compile_definitions(${name} PRIVATE "SNMALLOC_EXPORT=__attribute__((visibility(\"default\")))")
+  set_target_properties(${name} PROPERTIES CXX_VISIBILITY_PRESET hidden)
+endmacro()
 
 if(NOT MSVC)
-  add_library(snmallocshim-1mib SHARED src/override/malloc.cc)
-  target_link_libraries(snmallocshim-1mib -pthread)
-  target_include_directories(snmallocshim-1mib PRIVATE src)
+  add_shim(snmallocshim)
+
+  add_shim(snmallocshim-1mib)
   target_compile_definitions(snmallocshim-1mib PRIVATE IS_ADDRESS_SPACE_CONSTRAINED)
 endif()
 
diff --git a/src/mem/largealloc.h b/src/mem/largealloc.h
index 3c8dd71..11aa6e0 100644
--- a/src/mem/largealloc.h
+++ b/src/mem/largealloc.h
@@ -196,13 +196,15 @@ namespace snmalloc
             memory_provider.template zero<true>(p, OS_PAGE_SIZE);
 
           memory_provider.template notify_using<zero_mem>(
-            (void*)((size_t)p + OS_PAGE_SIZE), size - OS_PAGE_SIZE);
+            (void*)((size_t)p + OS_PAGE_SIZE),
+            bits::align_up(size, OS_PAGE_SIZE) - OS_PAGE_SIZE);
         }
         else
         {
           // This is a superslab that has not been decommitted.
           if (zero_mem == YesZero)
-            memory_provider.template zero<true>(p, size);
+            memory_provider.template zero<true>(
+              p, bits::align_up(size, OS_PAGE_SIZE));
         }
       }
 
diff --git a/src/test/func/memory/memory.cc b/src/test/func/memory/memory.cc
index 5ae4759..a9bdec1 100644
--- a/src/test/func/memory/memory.cc
+++ b/src/test/func/memory/memory.cc
@@ -277,6 +277,17 @@ void test_alloc_16M()
   alloc->dealloc(p1);
 }
 
+void test_calloc_16M()
+{
+  auto* alloc = ThreadAlloc::get();
+  // sizes >= 16M use large_alloc
+  const size_t size = 16'000'000;
+
+  void* p1 = alloc->alloc<YesZero>(size);
+  assert(Alloc::alloc_size(Alloc::external_pointer(p1)) >= size);
+  alloc->dealloc(p1);
+}
+
 int main(int argc, char** argv)
 {
 #ifdef USE_SYSTEMATIC_TESTING
@@ -296,6 +307,7 @@ int main(int argc, char** argv)
   test_double_alloc();
   test_external_pointer();
   test_alloc_16M();
+  test_calloc_16M();
 
   return 0;
 }