Add mallctl to set and get ncached_max of each cache_bin.

1. `thread_tcache_ncached_max_read_sizeclass` allows users to get the
    ncached_max of the bin with the input sizeclass, passed in through
    oldp (will be upper casted if not an exact bin size is given).
2. `thread_tcache_ncached_max_write` takes in a char array
    representing the settings for bins in the tcache.

include/jemalloc/internal/arena_inlines_b.h

@@ -198,7 +198,8 @@ arena_malloc(tsdn_t *tsdn, arena_t *arena, size_t size, szind_t ind, bool zero,
 			assert(sz_can_use_slab(size));
 			return tcache_alloc_small(tsdn_tsd(tsdn), arena,
 			    tcache, size, ind, zero, slow_path);
-		} else if (likely(ind < TCACHE_NBINS_MAX &&
+		} else if (likely(
+		    ind < tcache_nbins_get(tcache->tcache_slow) &&
 		    !tcache_bin_disabled(ind, &tcache->bins[ind],
 		    tcache->tcache_slow))) {
 			return tcache_alloc_large(tsdn_tsd(tsdn), arena,

include/jemalloc/internal/cache_bin.h

@@ -210,6 +210,11 @@ cache_bin_info_ncached_max_get(cache_bin_t *bin, cache_bin_info_t *info) {
 	return info->ncached_max;
 }
 
+/* Gets ncached_max without asserting that the bin is enabled. */
+static inline cache_bin_sz_t
+cache_bin_ncached_max_get_unsafe(cache_bin_t *bin) {
+	return bin->bin_info.ncached_max;
+}
 /*
  * Internal.
  *
@@ -229,7 +234,7 @@ cache_bin_assert_earlier(cache_bin_t *bin, uint16_t earlier, uint16_t later) {
  * Does difference calculations that handle wraparound correctly.  Earlier must
  * be associated with the position earlier in memory.
  */
-static inline uint16_t
+static inline cache_bin_sz_t
 cache_bin_diff(cache_bin_t *bin, uint16_t earlier, uint16_t later) {
 	cache_bin_assert_earlier(bin, earlier, later);
 	return later - earlier;
@@ -584,19 +589,17 @@ cache_bin_nitems_get_remote(cache_bin_t *bin, cache_bin_info_t *info,
 	cache_bin_sz_t diff = bin->low_bits_empty -
 	    (uint16_t)(uintptr_t)bin->stack_head;
 	cache_bin_sz_t n = diff / sizeof(void *);
-
-	cache_bin_sz_t ncached_max = cache_bin_info_ncached_max_get(bin, info);
-	assert(n <= ncached_max);
 	*ncached = n;
 
 	/* Racy version of cache_bin_nstashed_get_internal. */
 	uint16_t low_bits_low_bound = cache_bin_low_bits_low_bound_get(bin,
 	    info);
 	n = (bin->low_bits_full - low_bits_low_bound) / sizeof(void *);
-
-	assert(n <= ncached_max);
 	*nstashed = n;
-	/* Note that cannot assert ncached + nstashed <= ncached_max (racy). */
+	/*
+	 * Note that cannot assert anything regarding ncached_max because
+	 * it can be configured on the fly and is thus racy.
+	 */
 }
 
 /*

include/jemalloc/internal/ctl.h

@@ -14,6 +14,7 @@
 
 /* Maximum ctl tree depth. */
 #define CTL_MAX_DEPTH	7
+#define CTL_MULTI_SETTING_MAX_LEN 1000
 
 typedef struct ctl_node_s {
 	bool named;

include/jemalloc/internal/jemalloc_internal_macros.h

@@ -37,8 +37,10 @@
 /* Various function pointers are static and immutable except during testing. */
 #ifdef JEMALLOC_JET
 #  define JET_MUTABLE
+#  define JET_EXTERN extern
 #else
 #  define JET_MUTABLE const
+#  define JET_EXTERN static
 #endif
 
 #define JEMALLOC_VA_ARGS_HEAD(head, ...) head

include/jemalloc/internal/tcache_externs.h

@@ -26,7 +26,7 @@ extern unsigned opt_lg_tcache_flush_large_div;
  * it should not be changed on the fly.  To change the number of tcache bins
  * in use, refer to tcache_nbins of each tcache.
  */
-extern unsigned	global_do_not_change_nbins;
+extern unsigned	global_do_not_change_tcache_nbins;
 
 /*
  * Maximum cached size class.  Same as above, this is only used during threads
@@ -55,6 +55,9 @@ void tcache_bin_flush_large(tsd_t *tsd, tcache_t *tcache,
     cache_bin_t *cache_bin, szind_t binind, unsigned rem);
 void tcache_bin_flush_stashed(tsd_t *tsd, tcache_t *tcache,
     cache_bin_t *cache_bin, szind_t binind, bool is_small);
+bool tcache_bins_ncached_max_write(tsd_t *tsd, char *settings, size_t len);
+bool tcache_bin_ncached_max_read(tsd_t *tsd, size_t bin_size,
+    cache_bin_sz_t *ncached_max);
 void tcache_arena_reassociate(tsdn_t *tsdn, tcache_slow_t *tcache_slow,
     tcache_t *tcache, arena_t *arena);
 tcache_t *tcache_create_explicit(tsd_t *tsd);

include/jemalloc/internal/tcache_inlines.h

@@ -46,7 +46,7 @@ tcache_bin_settings_backup(tcache_t *tcache,
     cache_bin_info_t tcache_bin_info[TCACHE_NBINS_MAX]) {
 	for (unsigned i = 0; i < TCACHE_NBINS_MAX; i++) {
 		cache_bin_info_init(&tcache_bin_info[i],
-		    tcache->bins[i].bin_info.ncached_max);
+		    cache_bin_ncached_max_get_unsafe(&tcache->bins[i]));
 	}
 }
 
@@ -54,6 +54,7 @@ JEMALLOC_ALWAYS_INLINE bool
 tcache_bin_disabled(szind_t ind, cache_bin_t *bin,
     tcache_slow_t *tcache_slow) {
 	assert(bin != NULL);
+	assert(ind < TCACHE_NBINS_MAX);
 	bool disabled = cache_bin_disabled(bin);
 
 	/*
@@ -66,7 +67,7 @@ tcache_bin_disabled(szind_t ind, cache_bin_t *bin,
 	 * ind < nbins and ncached_max > 0.
 	 */
 	unsigned nbins = tcache_nbins_get(tcache_slow);
-	cache_bin_sz_t ncached_max = bin->bin_info.ncached_max;
+	cache_bin_sz_t ncached_max = cache_bin_ncached_max_get_unsafe(bin);
 	if (ind >= nbins) {
 		assert(disabled);
 	} else {
@@ -215,6 +216,8 @@ tcache_dalloc_large(tsd_t *tsd, tcache_t *tcache, void *ptr, szind_t binind,
 	assert(tcache_salloc(tsd_tsdn(tsd), ptr) > SC_SMALL_MAXCLASS);
 	assert(tcache_salloc(tsd_tsdn(tsd), ptr) <=
 	    tcache_max_get(tcache->tcache_slow));
+	assert(!tcache_bin_disabled(binind, &tcache->bins[binind],
+	    tcache->tcache_slow));
 
 	cache_bin_t *bin = &tcache->bins[binind];
 	if (unlikely(!cache_bin_dalloc_easy(bin, ptr))) {

include/jemalloc/internal/util.h

@@ -130,4 +130,12 @@ util_prefetch_write_range(void *ptr, size_t sz) {
 
 #undef UTIL_INLINE
 
+/*
+ * Reads the settings in the following format:
+ * key1-key2:value|key3-key4:value|...
+ * Note it does not handle the ending '\0'.
+ */
+bool
+multi_setting_parse_next(const char **setting_segment_cur, size_t *len_left,
+    size_t *key_start, size_t *key_end, size_t *value);
 #endif /* JEMALLOC_INTERNAL_UTIL_H */