C - newlib中的malloc():在一次大的失败分配后是否会浪费内存

Question

我正在为STM32F7编写嵌入式软件，我的libc是newlib-2.4.0.20160527。

我已经实现了_sbrk()如下:

extern intptr_t g_bss_end; /* value after the last byte in .bss */
extern intptr_t g_msp_lim; /* stack buffer starts at this address */
intptr_t _sbrk(ptrdiff_t heap_incr)
{
    static intptr_t heap_end = 0;
    intptr_t prev_heap_end;
    intptr_t new_heap_end;
    if(heap_end == 0) {
        heap_end = (intptr_t)&g_bss_end;
    }
    prev_heap_end = heap_end;
    new_heap_end = prev_heap_end + heap_incr;
    if(new_heap_end >= g_msp_lim) {
        errno = ENOMEM;
        return -1;
    }
    heap_end = new_heap_end;
    return prev_heap_end;
}

然后，当我执行以下操作时:

/* total capacity of my heap is 0x40000 */
void * mem = malloc(0x40000);
free(mem); mem = 0;
mem = malloc(0x40000);

一切正常(即malloc两次返回非零)。

但是当我做以下操作时(用于测试目的):

for(int32_t sz = 0x50000; sz >= 0; sz--) {
    void * mem = malloc(sz);
    if(mem != 0) {
        __BKPT();
        free(mem);
        break;
    }
}

每个malloc()都失败，甚至malloc(0)(即__BKPT() 从未到达)。所以，实际上堆上没有分配的内存(我没有得到任何mem != 0，所以我甚至不能free()什么的)，也没有可用的内存。

我期望malloc()在每个sz > 0x40000中失败，并在每个sz <= 0x40000中成功(假设free()在每个malloc()之后工作良好)。

我错过了什么，或者这是一个错误或预期的行为在newlib?

Answer 1

由于newlib/libc/stdlib/mallocr.c:2137中的malloc_extend_top()例程不好，newlib的malloc()在分配整个堆内存时不能正常工作。呼叫_sbrk()成功后

  brk = (char*)(MORECORE (sbrk_size)); /* MORECORE = _sbrk */
  /* Fail if sbrk failed or if a foreign sbrk call killed our space */
  if (brk == (char*)(MORECORE_FAILURE) || 
      (brk < old_end && old_top != initial_top))
    return;

它尝试计算校正以适合页面对齐:

/* Guarantee alignment of first new chunk made from this space */
front_misalign = (POINTER_UINT)chunk2mem(brk) & MALLOC_ALIGN_MASK;
if (front_misalign > 0) 
{
  correction = (MALLOC_ALIGNMENT) - front_misalign;
  brk += correction;
}
else
  correction = 0;
/* Guarantee the next brk will be at a page boundary */
correction += pagesz - ((POINTER_UINT)(brk + sbrk_size) & (pagesz - 1));

更正总是正的，因为即使分配非常合适，它也会尝试分配下一整个页面。例如，如果页面大小为4096和brk + sbrk_size = 4096*n，表达式4096 - ((brk + sbrk_size) & 4095)将给出4096，因此需要下一个空页，但没有空间。

例程没有正确处理这种情况，只留下已分配的数据(brk值)，导致永久的"不可释放"的整个堆分配。这样的浪费:-)

Answer 2

经过测试，工作正常

From 1473e08d2a16ad448afedb7036a476231a785643 Mon Sep 17 00:00:00 2001
From: Jeff Johnston <[e-mail]>
Date: Thu, 24 May 2018 23:53:15 -0400
Subject: [PATCH] Fix issue with malloc_extend_top
- when calculating a correction to align next brk to page boundary,
  ensure that the correction is less than a page size
- if allocating the correction fails, ensure that the top size is
  set to brk + sbrk_size (minus any front alignment made)
Signed-off-by: Jeff Johnston <[e-mail]>
---
 newlib/libc/stdlib/mallocr.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/newlib/libc/stdlib/mallocr.c b/newlib/libc/stdlib/mallocr.c
index ecc445f..26d1c89 100644
--- a/newlib/libc/stdlib/mallocr.c
+++ b/newlib/libc/stdlib/mallocr.c
@@ -2198,13 +2198,18 @@ static void malloc_extend_top(RARG nb) RDECL INTERNAL_SIZE_T nb;
     /* Guarantee the next brk will be at a page boundary */
     correction += pagesz - ((POINTER_UINT)(brk + sbrk_size) & (pagesz - 1));
+    /* To guarantee page boundary, correction should be less than pagesz */
+    correction &= (pagesz - 1);
+
     /* Allocate correction */
     new_brk = (char*)(MORECORE (correction));
     if (new_brk == (char*)(MORECORE_FAILURE))
       {
    correction = 0;
    correction_failed = 1;
-   new_brk = brk;
+   new_brk = brk + sbrk_size;
+   if (front_misalign > 0)
+     new_brk -= (MALLOC_ALIGNMENT) - front_misalign;
       }
     sbrked_mem += correction;
-- 
1.8.3.1