workspace/tags/glibc/

Glibc

  • 2.4x 的 tcache 机制分析

    简述

    在 2.39 到 2.40 中，malloc.c 并未发生变化，因此我们这里主要讲一些 2.41 到 2.43 的 tcache 相关的一些小细节 在这三个版本中，主要变化可以简单概括为

    • 2.41:
      • calloc 可以分配 tcache 块
      • 小 chunk 被 free 后会直接进入 small bin
    • 2.42
      • large_tcache 的引入
    • 2.43
      • tcache 扩容以补偿 fastbin 删除的性能问题
      • large_tcache 分配改为需要精确 chunk_size
      • mmap chunk 可以进入 tcache

    因为 tcache 初始化时机这几个版本一直改来改去的很乱，只做简述 以下为细分的分析，我会粘贴一些源码辅助理解

    tcache 基本机制

    tcache 结构体的变化

    在 glibc 2.42版本中，tcache_perthread_struct 结构进行了调整

    • 2.41
    /* We overlay this structure on the user-data portion of a chunk when
   the chunk is stored in the per-thread cache.  */
typedef struct tcache_entry
{
  struct tcache_entry *next;
  /* This field exists to detect double frees.  */
  uintptr_t key;
} tcache_entry;

/* There is one of these for each thread, which contains the
   per-thread cache (hence "tcache_perthread_struct").  Keeping
   overall size low is mildly important.  Note that COUNTS and ENTRIES
   are redundant (we could have just counted the linked list each
   time), this is for performance reasons.  */
typedef struct tcache_perthread_struct
{
  uint16_t counts[TCACHE_MAX_BINS];
  tcache_entry *entries[TCACHE_MAX_BINS];
} tcache_perthread_struct;

static __thread bool tcache_shutting_down = false;
static __thread tcache_perthread_struct *tcache = NULL;
    • 2.42
    /* We overlay this structure on the user-data portion of a chunk when
   the chunk is stored in the per-thread cache.  */
typedef struct tcache_entry
{
  struct tcache_entry *next;
  /* This field exists to detect double frees.  */
  uintptr_t key;
} tcache_entry;

/* There is one of these for each thread, which contains the
   per-thread cache (hence "tcache_perthread_struct").  Keeping
   overall size low is mildly important.  The 'entries' field is linked list of
   free blocks, while 'num_slots' contains the number of free blocks that can
   be added.  Each bin may allow a different maximum number of free blocks,
   and can be disabled by initializing 'num_slots' to zero.  */
typedef struct tcache_perthread_struct
{
  uint16_t num_slots[TCACHE_MAX_BINS];
  tcache_entry *entries[TCACHE_MAX_BINS];
} tcache_perthread_struct;

    entry 的部分基本没有变化，这里主要变化还是原本的 counts 数组变成了 num_slots 数组 原本的 counts 数组表示对应索引下 tcache 链表的长度，现在 num_slots 数组中变为了对应索引下设剩下的空槽位

Terminal

C0nvR3 Lab terminal ready. Type help for commands.