diff --git a/include/mimalloc-internal.h b/include/mimalloc-internal.h
index 74a5a9cc..d3336e87 100644
--- a/include/mimalloc-internal.h
+++ b/include/mimalloc-internal.h
@@ -104,6 +104,7 @@ bool       _mi_arena_memid_is_suitable(size_t memid, mi_arena_id_t req_arena_id)
 void*      _mi_segment_cache_pop(size_t size, mi_commit_mask_t* commit_mask, mi_commit_mask_t* decommit_mask, bool* large, bool* is_pinned, bool* is_zero, mi_arena_id_t req_arena_id, size_t* memid, mi_os_tld_t* tld);
 bool       _mi_segment_cache_push(void* start, size_t size, size_t memid, const mi_commit_mask_t* commit_mask, const mi_commit_mask_t* decommit_mask, bool is_large, bool is_pinned, mi_os_tld_t* tld);
 void       _mi_segment_cache_collect(bool force, mi_os_tld_t* tld);
+void       _mi_segment_cache_free_all(mi_os_tld_t* tld);
 void       _mi_segment_map_allocated_at(const mi_segment_t* segment);
 void       _mi_segment_map_freed_at(const mi_segment_t* segment);
 
@@ -149,6 +150,7 @@ void       _mi_heap_destroy_pages(mi_heap_t* heap);
 void       _mi_heap_collect_abandon(mi_heap_t* heap);
 void       _mi_heap_set_default_direct(mi_heap_t* heap);
 bool       _mi_heap_memid_is_suitable(mi_heap_t* heap, size_t memid);
+void       _mi_heap_destroy_all(void);
 
 // "stats.c"
 void       _mi_stats_done(mi_stats_t* stats);
diff --git a/include/mimalloc.h b/include/mimalloc.h
index 7f1ce80a..38524b8d 100644
--- a/include/mimalloc.h
+++ b/include/mimalloc.h
@@ -342,6 +342,7 @@ typedef enum mi_option_e {
   mi_option_allow_decommit,
   mi_option_segment_decommit_delay,  
   mi_option_decommit_extend_delay,
+  mi_option_destroy_on_exit,          
   _mi_option_last
 } mi_option_t;
 
diff --git a/src/heap.c b/src/heap.c
index 15ca3603..83a0a29e 100644
--- a/src/heap.c
+++ b/src/heap.c
@@ -347,7 +347,20 @@ void mi_heap_destroy(mi_heap_t* heap) {
   }
 }
 
-
+void _mi_heap_destroy_all(void) {
+  mi_heap_t* bheap = mi_heap_get_backing();
+  mi_heap_t* curr = bheap->tld->heaps;
+  while (curr != NULL) {
+    mi_heap_t* next = curr->next;
+    if (curr->no_reclaim) {
+      mi_heap_destroy(curr);
+    }
+    else {
+      _mi_heap_destroy_pages(curr);
+    }
+    curr = next;
+  }
+}
 
 /* -----------------------------------------------------------
   Safe Heap delete
diff --git a/src/init.c b/src/init.c
index e27d72b2..18eadf1f 100644
--- a/src/init.c
+++ b/src/init.c
@@ -627,7 +627,7 @@ static void mi_cdecl mi_process_done(void) {
   #if defined(_WIN32) && !defined(MI_SHARED_LIB)
   FlsFree(mi_fls_key);  // call thread-done on all threads (except the main thread) to prevent dangling callback pointer if statically linked with a DLL; Issue #208
   #endif
-  
+
   #ifndef MI_SKIP_COLLECT_ON_EXIT
     #if (MI_DEBUG != 0) || !defined(MI_SHARED_LIB)  
     // free all memory if possible on process exit. This is not needed for a stand-alone process
@@ -637,6 +637,14 @@ static void mi_cdecl mi_process_done(void) {
     #endif
   #endif
 
+  // Forcefully release all retained memory; this can be dangerous in general if overriding regular malloc/free
+  // since after process_done there might still be other code running that calls `free` (like at_exit routines,
+  // or C-runtime termination code.
+  if (mi_option_is_enabled(mi_option_destroy_on_exit)) {
+    _mi_heap_destroy_all();                          // forcefully release all memory held by all heaps (of this thread only!)
+    _mi_segment_cache_free_all(&_mi_heap_main_get()->tld->os);  // release all cached segments
+  }
+
   if (mi_option_is_enabled(mi_option_show_stats) || mi_option_is_enabled(mi_option_verbose)) {
     mi_stats_print(NULL);
   }
diff --git a/src/options.c b/src/options.c
index 367bc0d2..6c6f8f2f 100644
--- a/src/options.c
+++ b/src/options.c
@@ -94,7 +94,8 @@ static mi_option_desc_t options[_mi_option_last] =
   { 8,    UNINIT, MI_OPTION(max_segment_reclaim)},// max. number of segment reclaims from the abandoned segments per try.  
   { 1,    UNINIT, MI_OPTION(allow_decommit) },    // decommit slices when no longer used (after decommit_delay milli-seconds)
   { 500,  UNINIT, MI_OPTION(segment_decommit_delay) }, // decommit delay in milli-seconds for freed segments
-  { 2,    UNINIT, MI_OPTION(decommit_extend_delay) }
+  { 2,    UNINIT, MI_OPTION(decommit_extend_delay) },
+  { 0,    UNINIT, MI_OPTION(destroy_on_exit)}     // release all OS memory on process exit; careful with dangling pointer or after-exit frees!
 };
 
 static void mi_option_init(mi_option_desc_t* desc);
diff --git a/src/segment-cache.c b/src/segment-cache.c
index 436ce2bf..26786c92 100644
--- a/src/segment-cache.c
+++ b/src/segment-cache.c
@@ -45,7 +45,11 @@ static bool mi_cdecl mi_segment_cache_is_suitable(mi_bitmap_index_t bitidx, void
   return _mi_arena_memid_is_suitable(slot->memid, req_arena_id);
 }
 
-mi_decl_noinline void* _mi_segment_cache_pop(size_t size, mi_commit_mask_t* commit_mask, mi_commit_mask_t* decommit_mask, bool* large, bool* is_pinned, bool* is_zero, mi_arena_id_t _req_arena_id, size_t* memid, mi_os_tld_t* tld)
+mi_decl_noinline static void* mi_segment_cache_pop_ex(
+                              bool all_suitable,
+                              size_t size, mi_commit_mask_t* commit_mask, 
+                              mi_commit_mask_t* decommit_mask, bool* large, bool* is_pinned, bool* is_zero, 
+                              mi_arena_id_t _req_arena_id, size_t* memid, mi_os_tld_t* tld)
 {
 #ifdef MI_CACHE_DISABLE
   return NULL;
@@ -66,8 +70,8 @@ mi_decl_noinline void* _mi_segment_cache_pop(size_t size, mi_commit_mask_t* comm
   mi_bitmap_index_t bitidx = 0;
   bool claimed = false;
   mi_arena_id_t req_arena_id = _req_arena_id;
-  mi_bitmap_pred_fun_t pred_fun = &mi_segment_cache_is_suitable;  // cannot pass NULL as the arena may be exclusive itself; todo: do not put exclusive arenas in the cache?
-  
+  mi_bitmap_pred_fun_t pred_fun = (all_suitable ? NULL : &mi_segment_cache_is_suitable);  // cannot pass NULL as the arena may be exclusive itself; todo: do not put exclusive arenas in the cache?
+
   if (*large) {  // large allowed?
     claimed = _mi_bitmap_try_find_from_claim_pred(cache_available_large, MI_CACHE_FIELDS, start_field, 1, pred_fun, &req_arena_id, &bitidx);
     if (claimed) *large = true;
@@ -97,6 +101,12 @@ mi_decl_noinline void* _mi_segment_cache_pop(size_t size, mi_commit_mask_t* comm
 #endif
 }
 
+
+mi_decl_noinline void* _mi_segment_cache_pop(size_t size, mi_commit_mask_t* commit_mask, mi_commit_mask_t* decommit_mask, bool* large, bool* is_pinned, bool* is_zero, mi_arena_id_t _req_arena_id, size_t* memid, mi_os_tld_t* tld)
+{
+  return mi_segment_cache_pop_ex(false, size, commit_mask, decommit_mask, large, is_pinned, is_zero, _req_arena_id, memid, tld);
+}
+
 static mi_decl_noinline void mi_commit_mask_decommit(mi_commit_mask_t* cmask, void* p, size_t total, mi_stats_t* stats)
 {
   if (mi_commit_mask_is_empty(cmask)) {
@@ -163,6 +173,30 @@ void _mi_segment_cache_collect(bool force, mi_os_tld_t* tld) {
   mi_segment_cache_purge(force, tld );
 }
 
+void _mi_segment_cache_free_all(mi_os_tld_t* tld) {
+  mi_commit_mask_t commit_mask;
+  mi_commit_mask_t decommit_mask;
+  bool is_pinned;
+  bool is_zero;
+  size_t memid;
+  const size_t size = MI_SEGMENT_SIZE;
+  // iterate twice: first large pages, then regular memory 
+  for (int i = 0; i < 2; i++) {
+    void* p;
+    do {
+      // keep popping and freeing the memory
+      bool large = (i == 0);  
+      p = mi_segment_cache_pop_ex(true /* all */, size, &commit_mask, &decommit_mask,
+                                  &large, &is_pinned, &is_zero, _mi_arena_id_none(), &memid, tld);
+      if (p != NULL) {
+        size_t csize = _mi_commit_mask_committed_size(&commit_mask, size);
+        if (csize > 0 && !is_pinned) _mi_stat_decrease(&_mi_stats_main.committed, csize);
+        _mi_arena_free(p, size, MI_SEGMENT_ALIGN, 0, memid, is_pinned /* pretend not committed to not double count decommits */, tld->stats);
+      }
+    } while (p != NULL);
+  }
+}
+
 mi_decl_noinline bool _mi_segment_cache_push(void* start, size_t size, size_t memid, const mi_commit_mask_t* commit_mask, const mi_commit_mask_t* decommit_mask, bool is_large, bool is_pinned, mi_os_tld_t* tld)
 {
 #ifdef MI_CACHE_DISABLE
diff --git a/src/segment.c b/src/segment.c
index 0d0f98a7..3423bd53 100644
--- a/src/segment.c
+++ b/src/segment.c
@@ -803,7 +803,7 @@ static mi_segment_t* mi_segment_init(mi_segment_t* segment, size_t required, siz
     bool is_pinned = false;
     size_t memid = 0;
     size_t align_offset = 0;
-    size_t alignment = MI_SEGMENT_SIZE;
+    size_t alignment = MI_SEGMENT_ALIGN;
     
     if (page_alignment > 0) {
       mi_assert_internal(huge_page != NULL);
diff --git a/test/main-override-static.c b/test/main-override-static.c
index 70b6293c..6c994bca 100644
--- a/test/main-override-static.c
+++ b/test/main-override-static.c
@@ -33,9 +33,10 @@ int main() {
   // invalid_free();
   // test_reserved();
   // negative_stat();
+  // test_heap_walk();
   // alloc_huge();
   // test_heap_walk();
-  test_heap_arena();
+  // test_heap_arena();
   
   void* p1 = malloc(78);
   void* p2 = malloc(24);