Fix psz/pind edge cases.

Add an "over-size" extent heap in which to store extents which exceed
the maximum size class (plus cache-oblivious padding, if enabled).
Remove psz2ind_clamp() and use psz2ind() instead so that trying to
allocate the maximum size class can in principle succeed.  In practice,
this allows assertions to hold so that OOM errors can be successfully
generated.

include/jemalloc/internal/arena.h

@@ -212,8 +212,8 @@ struct arena_s {
 	 * Heaps of extents that were previously allocated.  These are used when
 	 * allocating extents, in an attempt to re-use address space.
 	 */
-	extent_heap_t		extents_cached[NPSIZES];
-	extent_heap_t		extents_retained[NPSIZES];
+	extent_heap_t		extents_cached[NPSIZES+1];
+	extent_heap_t		extents_retained[NPSIZES+1];
 	/*
 	 * Ring sentinel used to track unused dirty memory.  Dirty memory is
 	 * managed as an LRU of cached extents.

include/jemalloc/internal/jemalloc_internal.h.in

@@ -434,7 +434,7 @@ extern arena_t	**arenas;
  * pind2sz_tab encodes the same information as could be computed by
  * pind2sz_compute().
  */
-extern size_t const	pind2sz_tab[NPSIZES];
+extern size_t const	pind2sz_tab[NPSIZES+1];
 /*
  * index2size_tab encodes the same information as could be computed (at
  * unacceptable cost in some code paths) by index2size_compute().
@@ -516,9 +516,7 @@ void	jemalloc_postfork_child(void);
 #include "jemalloc/internal/large.h"
 
 #ifndef JEMALLOC_ENABLE_INLINE
-pszind_t	psz2ind_impl(size_t psz, bool clamp);
 pszind_t	psz2ind(size_t psz);
-pszind_t	psz2ind_clamp(size_t psz);
 size_t	pind2sz_compute(pszind_t pind);
 size_t	pind2sz_lookup(pszind_t pind);
 size_t	pind2sz(pszind_t pind);
@@ -544,11 +542,11 @@ ticker_t	*decay_ticker_get(tsd_t *tsd, unsigned ind);
 
 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_C_))
 JEMALLOC_ALWAYS_INLINE pszind_t
-psz2ind_impl(size_t psz, bool clamp)
+psz2ind(size_t psz)
 {
 
 	if (unlikely(psz > LARGE_MAXCLASS))
-		return (clamp ? NPSIZES-1 : NPSIZES);
+		return (NPSIZES);
 	{
 		pszind_t x = lg_floor((psz<<1)-1);
 		pszind_t shift = (x < LG_SIZE_CLASS_GROUP + LG_PAGE) ? 0 : x -
@@ -567,24 +565,12 @@ psz2ind_impl(size_t psz, bool clamp)
 	}
 }
 
-JEMALLOC_INLINE pszind_t
-psz2ind(size_t psz)
-{
-
-	return (psz2ind_impl(psz, false));
-}
-
-JEMALLOC_INLINE pszind_t
-psz2ind_clamp(size_t psz)
-{
-
-	return (psz2ind_impl(psz, true));
-}
-
 JEMALLOC_INLINE size_t
 pind2sz_compute(pszind_t pind)
 {
 
+	if (unlikely(pind == NPSIZES))
+		return (LARGE_MAXCLASS + PAGE);
 	{
 		size_t grp = pind >> LG_SIZE_CLASS_GROUP;
 		size_t mod = pind & ((ZU(1) << LG_SIZE_CLASS_GROUP) - 1);
@@ -614,7 +600,7 @@ JEMALLOC_INLINE size_t
 pind2sz(pszind_t pind)
 {
 
-	assert(pind < NPSIZES);
+	assert(pind < NPSIZES+1);
 	return (pind2sz_lookup(pind));
 }
 
@@ -623,7 +609,7 @@ psz2u(size_t psz)
 {
 
 	if (unlikely(psz > LARGE_MAXCLASS))
-		return (0);
+		return (LARGE_MAXCLASS + PAGE);
 	{
 		size_t x = lg_floor((psz<<1)-1);
 		size_t lg_delta = (x < LG_SIZE_CLASS_GROUP + LG_PAGE + 1) ?

include/jemalloc/internal/private_symbols.txt

@@ -380,8 +380,6 @@ prof_thread_active_set
 prof_thread_name_get
 prof_thread_name_set
 psz2ind
-psz2ind_clamp
-psz2ind_impl
 psz2u
 rtree_child_read
 rtree_child_read_hard