运行中的JVM程序包含哪些类型的栈？

Great question—digging into the JVM spec's stack details is a fantastic way to build a solid foundation of how Java executes under the hood. You've already got the Java Virtual Machine Stack (per-thread, holds method frames, handles method calls/returns) down, so let's walk through the other stack types you'll encounter in the JVM:

• Native Method Stack
  As defined in §2.5.5 of the JVM Specification, this is a per-thread private stack just like the Java Virtual Machine Stack, but it's designed for executing native methods (code written in C/C++ via JNI). It manages the call stack, local variables, and return flow for native code. Some JVM implementations (like HotSpot) optimize by merging this with the Java Virtual Machine Stack, but the spec explicitly defines it as a separate component.

• Operand Stack
  This is a core part of every method frame (§2.6.1) within the Java Virtual Machine Stack. It's a LIFO stack that acts as the workspace for bytecode execution. For example, when running an iadd instruction, the JVM pops two integers from the operand stack, computes their sum, then pushes the result back. It also handles passing method arguments and receiving return values. Think of it as the "calculator" for individual method execution.

• JIT Compiler Internal Stacks
  While not mandated by the official JVM spec, all modern JVMs with Just-In-Time (JIT) compilers (such as HotSpot's C1 and C2) use internal stacks during the compilation process. These stacks store intermediate code representations, compilation context, and state as the JVM translates bytecode into optimized machine code. They're implementation-specific but critical for delivering the performance we expect from modern Java applications.

• Signal Handler Stacks
  Many JVM implementations allocate dedicated stacks for signal handling threads. Operating systems send signals to the JVM process (e.g., SIGSEGV for invalid memory access, SIGINT for user interrupts), and using a separate stack ensures the handler can run safely without overlapping with an application thread's stack. This avoids potential stack overflow issues that could crash the entire JVM during signal processing.

内容的提问来源于stack exchange，提问作者Adelin