简体繁体 English

从动态库调用jni时，JVM堆的内存不足

[英]Not enough memory for JVM heap when calling jni from dynamic library

原文 2015-10-23 12:58:25 3 1 java/ c++/ dll/ jvm/ java-native-interface

I have strange situation: we have .dll library (lets call it 'sdk_wrapper'), which is wrapper for java classes, which are opening JRE jvm.dll and use some other .dlls (PKCS11 implementations, to be more specific - but I think it doesn't matter). 我有一个奇怪的情况：我们有.dll库（我们称其为“ sdk_wrapper”），该库是Java类的包装器，这些类正在打开JRE jvm.dll并使用其他一些.dll（PKCS11实现，具体而言，但我认为没关系）。

Everything works fine when I use sdk_wrapper directly in my c++ program (alias 'run_dll'). 当我直接在我的C ++程序中使用sdk_wrapper时，一切工作正常（别名为“ run_dll”）。 But when I pack it into another .dll (to prepare configurations file, initialize library, etc) and export one function which is dong everything the same as program 'run_dll', calling this causes jvm initialization error: Could not reserve enough space for object heap . 但是，当我将其打包到另一个.dll（以准备配置文件，初始化库等）并导出与程序“ run_dll”相同的一个函数时，调用此函数将导致jvm初始化错误： Could not reserve enough space for object heap JVM.dll initialization is doing with -Xmx512m . JVM.dll初始化使用-Xmx512m 。

Have you any idea what's going on? 你知道发生了什么吗？ I readed that JVM needs continous part of memory to initialize, but what's the difference between calling 'sdk_wrapper' directly from program and from another dll? 我读到JVM需要连续的一部分内存来初始化，但是直接从程序和另一个dll调用'sdk_wrapper'有什么区别？ They are at the same place (I mean in the same directory). 它们在同一位置（我的意思是在同一目录中）。

1 个解决方案

As you load libraries into a memory, you end of with memory fragmentation of the address space. 当您将库加载到内存中时，您将以地址空间的内存碎片结束。 Win32 is particularly bad for this. Win32对此特别糟糕。

Note: if you use a 64-bit process this isn't an issue as you will have plenty of virtual memory. 注意：如果您使用64位进程，这不是问题，因为您将拥有大量的虚拟内存。