Merge branch 'dev3' into dev3-bin

include/mimalloc.h

@@ -394,8 +394,8 @@ typedef enum mi_option_e {
   mi_option_guarded_precise,            // disregard minimal alignment requirement to always place guarded blocks exactly in front of a guard page (=0)
   mi_option_guarded_sample_rate,        // 1 out of N allocations in the min/max range will be guarded (=1000)
   mi_option_guarded_sample_seed,        // can be set to allow for a (more) deterministic re-execution when a guard page is triggered (=0)
-  mi_option_page_reclaim_on_free,       // allow to reclaim an abandoned segment on a free (=1)
-  mi_option_page_full_retain,           // retain N full pages per size class (=2)
+  mi_option_page_reclaim_on_free,       // reclaim abandoned pages on a free (=0). -1 disallowr always, 0 allows if the page originated from the current heap, 1 allow always
+  mi_option_page_full_retain,           // retain N full (small) pages per size class (=2)
   mi_option_page_max_candidates,        // max candidate pages to consider for allocation (=4)
   mi_option_max_vabits,                 // max user space virtual address bits to consider (=48)
   mi_option_pagemap_commit,             // commit the full pagemap (to always catch invalid pointer uses) (=0)

include/mimalloc/internal.h

@@ -492,7 +492,7 @@ static inline mi_page_t* _mi_unchecked_ptr_page(const void* p) {
 // 2-level page map:
 // double indirection, but low commit and low virtual reserve.
 //
-// the page-map is usually 4 MiB and points to sub maps of 64 KiB.
+// the page-map is usually 4 MiB (for 48 bits virtual addresses) and points to sub maps of 64 KiB.
 // the page-map is committed on-demand (in 64 KiB parts) (and sub-maps are committed on-demand as well)
 // one sub page-map = 64 KiB => covers 2^(16-3) * 2^16 = 2^29 = 512 MiB address space
 // the page-map needs 48-(16+13) = 19 bits => 2^19 sub map pointers = 4 MiB size.
@@ -519,7 +519,7 @@ static inline mi_page_t* _mi_checked_ptr_page(const void* p) {
   size_t sub_idx;
   const size_t idx = _mi_page_map_index(p, &sub_idx);
   mi_page_t** const sub = _mi_page_map[idx];
-  if mi_unlikely(sub == NULL) return NULL;
+  if mi_unlikely(sub == NULL) return (mi_page_t*)&_mi_page_empty;
   return sub[sub_idx];
 }
 

src/free.c

@@ -123,6 +123,10 @@ static void mi_decl_noinline mi_free_generic_local(mi_page_t* page, void* p) mi_
 // free a pointer owned by another thread (page parameter comes first for better codegen)
 static void mi_decl_noinline mi_free_generic_mt(mi_page_t* page, void* p) mi_attr_noexcept {
   if (p==NULL) return;  // a NULL pointer is seen as abandoned (tid==0) with a full flag set
+  #if !MI_PAGE_MAP_FLAT
+  if (page==&_mi_page_empty) return;  // an invalid pointer may lead to using the empty page
+  #endif
+  mi_assert_internal(p!=NULL && page != NULL && page != &_mi_page_empty);
   mi_block_t* const block = _mi_page_ptr_unalign(page, p); // don't check `has_aligned` flag to avoid a race (issue #865)
   mi_block_check_unguard(page, block, p);
   mi_free_block_mt(page, block);
@@ -135,10 +139,9 @@ void mi_decl_noinline _mi_free_generic(mi_page_t* page, bool is_local, void* p)
 }
 
 
-// Get the segment data belonging to a pointer
-// This is just a single `and` in release mode but does further checks in debug mode
-// (and secure mode) to see if this was a valid pointer.
-static inline mi_page_t* mi_checked_ptr_page(const void* p, const char* msg)
+// Get the page belonging to a pointer
+// Does further checks in debug mode to see if this was a valid pointer.
+static inline mi_page_t* mi_validate_ptr_page(const void* p, const char* msg)
 {
   MI_UNUSED_RELEASE(msg);
   #if MI_DEBUG
@@ -146,9 +149,14 @@ static inline mi_page_t* mi_checked_ptr_page(const void* p, const char* msg)
     _mi_error_message(EINVAL, "%s: invalid (unaligned) pointer: %p\n", msg, p);
     return NULL;
   }
-  mi_page_t* const page = _mi_safe_ptr_page(p);
-  if (page == NULL && p != NULL) {
-    _mi_error_message(EINVAL, "%s: invalid pointer: %p\n", msg, p);
+  mi_page_t* page = _mi_safe_ptr_page(p);
+  if (page == NULL) {
+    if (p != NULL) {
+      _mi_error_message(EINVAL, "%s: invalid pointer: %p\n", msg, p);
+    }
+    #if !MI_PAGE_MAP_FLAT
+    page = (mi_page_t*)&_mi_page_empty;
+    #endif
   }
   return page;
   #else
@@ -160,12 +168,13 @@ static inline mi_page_t* mi_checked_ptr_page(const void* p, const char* msg)
 // Fast path written carefully to prevent register spilling on the stack
 void mi_free(void* p) mi_attr_noexcept
 {
-  mi_page_t* const page = mi_checked_ptr_page(p,"mi_free");
+  mi_page_t* const page = mi_validate_ptr_page(p,"mi_free");
 
-  #if MI_PAGE_MAP_FLAT                  // if not flat, NULL will point to `_mi_page_empty` and get to `mi_free_generic_mt`
+  #if MI_PAGE_MAP_FLAT  // if not flat, p==NULL leads to `_mi_page_empty` which leads to `mi_free_generic_mt`
   if mi_unlikely(page==NULL) return;
   #endif
-
+  mi_assert_internal(page!=NULL);
+  
   const mi_threadid_t xtid = (_mi_prim_thread_id() ^ mi_page_xthread_id(page));
   if mi_likely(xtid == 0) {                        // `tid == mi_page_thread_id(page) && mi_page_flags(page) == 0`
     // thread-local, aligned, and not a full page
@@ -211,7 +220,7 @@ static void mi_decl_noinline mi_free_try_collect_mt(mi_page_t* page, mi_block_t*
   if (mi_page_all_free(page))
   {
     // first remove it from the abandoned pages in the arena (if mapped, this waits for any readers to finish)
-      _mi_arenas_page_unabandon(page);
+    _mi_arenas_page_unabandon(page);
     // we can free the page directly
     _mi_arenas_page_free(page);
     return;
@@ -235,8 +244,9 @@ static void mi_decl_noinline mi_free_try_collect_mt(mi_page_t* page, mi_block_t*
     // can we reclaim?
     if (heap != NULL && heap->allow_page_reclaim) {
       if (heap == page->heap ||                  // only reclaim if we were the originating heap,
-          (reclaim_on_free == 1 &&               // OR if the reclaim option across heaps is enabled
+          (reclaim_on_free == 1 &&               // OR if the reclaim across heaps is allowed
            !mi_page_is_used_at_frac(page, 8) &&  //    and the page is not too full
+           !heap->tld->is_in_threadpool &&       //    and not part of a threadpool
            _mi_arena_memid_is_suitable(page->memid, heap->exclusive_arena))  // and the memory is suitable    
          )
       {
@@ -283,7 +293,7 @@ static size_t mi_decl_noinline mi_page_usable_aligned_size_of(const mi_page_t* p
 }
 
 static inline size_t _mi_usable_size(const void* p, const char* msg) mi_attr_noexcept {
-  const mi_page_t* const page = mi_checked_ptr_page(p,msg);
+  const mi_page_t* const page = mi_validate_ptr_page(p,msg);
   if mi_unlikely(page==NULL) return 0;
   if mi_likely(!mi_page_has_aligned(page)) {
     const mi_block_t* block = (const mi_block_t*)p;

src/heap.c

@@ -181,10 +181,10 @@ void _mi_heap_init(mi_heap_t* heap, mi_arena_id_t arena_id, bool allow_destroy,
   heap->tag        = heap_tag;
   if (heap->tld->is_in_threadpool) {
     // if we run as part of a thread pool it is better to not arbitrarily reclaim abandoned pages into our heap.
-    heap->allow_page_reclaim = false;
-    // .. but abandoning is good in this case: quarter the full page retain (possibly to 0)
+    // this is checked in `free.c:mi_free_try_collect_mt`
+    // .. but abandoning is good in this case: halve the full page retain (possibly to 0)
     // (so blocked threads do not hold on to too much memory)
-    if (heap->page_full_retain >= 0) {
+    if (heap->page_full_retain > 0) {
       heap->page_full_retain = heap->page_full_retain / 4;
     }
   }

src/options.c

@@ -168,7 +168,7 @@ static mi_option_desc_t options[_mi_option_last] =
   { MI_DEFAULT_GUARDED_SAMPLE_RATE,
          UNINIT, MI_OPTION(guarded_sample_rate)},       // 1 out of N allocations in the min/max range will be guarded (=4000)
   { 0,   UNINIT, MI_OPTION(guarded_sample_seed)},
-  { 0,   UNINIT, MI_OPTION_LEGACY(page_reclaim_on_free, abandoned_reclaim_on_free) },// reclaim an abandoned segment on a free: -1 = disable completely, 0 = only reclaim into the originating heap, 1 = reclaim on free across heaps
+  { 0,   UNINIT, MI_OPTION_LEGACY(page_reclaim_on_free, abandoned_reclaim_on_free) },// reclaim abandoned pages on a free: -1 = disable completely, 0 = only reclaim into the originating heap, 1 = reclaim on free across heaps
   { 2,   UNINIT, MI_OPTION(page_full_retain) },         // number of (small) pages to retain in the free page queues
   { 4,   UNINIT, MI_OPTION(page_max_candidates) },      // max search to find a best page candidate
   { 0,   UNINIT, MI_OPTION(max_vabits) },               // max virtual address space bits

src/page-map.c

@@ -206,7 +206,7 @@ bool _mi_page_map_init(void) {
   if (!mi_page_map_memid.initially_committed) {
     _mi_os_commit(&_mi_page_map[0], os_page_size, NULL);  // commit first part of the map
   }
-  _mi_page_map[0] = (mi_page_t**)((uint8_t*)_mi_page_map + page_map_size);  // we reserved 2 subs at the end already
+  _mi_page_map[0] = (mi_page_t**)((uint8_t*)_mi_page_map + page_map_size);  // we reserved 2 sub maps at the end already
   if (!mi_page_map_memid.initially_committed) {
     _mi_os_commit(_mi_page_map[0], os_page_size, NULL);   // only first OS page
   }
@@ -315,10 +315,10 @@ void _mi_page_map_unregister_range(void* start, size_t size) {
   mi_page_map_set_range(NULL, idx, sub_idx, slice_count);  // todo: avoid committing if not already committed?
 }
 
-// Return the empty page for the NULL pointer to match the behaviour of `_mi_ptr_page`
+// Return NULL for invalid pointers
 mi_page_t* _mi_safe_ptr_page(const void* p) {
+  if (p==NULL) return NULL;
   if mi_unlikely(p >= mi_page_map_max_address) return NULL;
-  if (p == NULL) return (mi_page_t*)&_mi_page_empty; // to match `_mi_ptr_page` (see `mi_free` as well)
   size_t sub_idx;
   const size_t idx = _mi_page_map_index(p,&sub_idx);
   if mi_unlikely(!mi_page_map_is_committed(idx,NULL)) return NULL;
@@ -328,7 +328,7 @@ mi_page_t* _mi_safe_ptr_page(const void* p) {
 }
 
 mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept {
-  return (p != NULL && _mi_safe_ptr_page(p) != NULL);
+  return (_mi_safe_ptr_page(p) != NULL);
 }
 
 #endif

src/page.c

@@ -358,11 +358,11 @@ static void mi_page_to_full(mi_page_t* page, mi_page_queue_t* pq) {
 
   mi_heap_t* heap = mi_page_heap(page);
   if (heap->allow_page_abandon) {
-    // abandon full pages
+    // abandon full pages (this is the usual case in order to allow for sharing of memory between heaps)
     _mi_page_abandon(page, pq);
   }
   else if (!mi_page_is_in_full(page)) {
-    // put full pages in a heap local queue
+    // put full pages in a heap local queue (this is for heaps that cannot abandon, for example, if the heap can be destroyed)
     mi_page_queue_enqueue_from(&mi_page_heap(page)->pages[MI_BIN_FULL], pq, page);
     _mi_page_free_collect(page, false);  // try to collect right away in case another thread freed just before MI_USE_DELAYED_FREE was set
   }

src/prim/osx/alloc-override-zone.c

@@ -64,7 +64,8 @@ static void* zone_valloc(malloc_zone_t* zone, size_t size) {
 
 static void zone_free(malloc_zone_t* zone, void* p) {
   MI_UNUSED(zone);
-  mi_cfree(p);
+  // mi_cfree(p);  // checked free as `zone_free` may be called with invalid pointers
+  mi_free(p); // with the page_map and pagemap_commit=1 we can use the regular free
 }
 
 static void* zone_realloc(malloc_zone_t* zone, void* p, size_t newsize) {