Are Split Stacks Necessary on x86_64 Architectures?

Split Stacks Unnecessary on x86_64?

The topic of split stacks in runtime models has raised questions about their necessity on 64-bit architectures like x86_64. Some opinions suggest that split stacks are no longer required due to the expanded address space available.

Answer to Question 1: Interpretation of Opinions

Yes, the opinions quoted in the question imply that split stacks are unnecessary on x86_64 architectures. The references to "millions of threads" and the "narrow use case" on 32-bit systems indicate that the large address space of 64-bit systems can accommodate a sufficient number of stack ranges.

Answer to Question 2: Rationale for Unnecessity

Split stacks are no longer necessary on 64-bit architectures due to the significantly larger virtual address space. This allows for each stack to have a dedicated and sizable address range that is comparable to the entire 32-bit address space.

In modern flat memory models, the hardware Memory Management Unit (MMU) translates virtual addresses to physical memory locations. On x86_64, allocating large chunks of virtual address space for each stack and mapping only the initial page to actual RAM is found to be more efficient.

This approach allows the stack to grow and shrink as needed, using contiguous virtual addresses. The OS can re-configure the MMU to map additional pages of virtual memory to real memory as the stack grows or shrinks past specified thresholds.

By carefully selecting these thresholds, stacks can behave like dynamic arrays with O(1) average stack operation complexity. They retain the benefit of supporting millions of stacks that can grow to arbitrary sizes while consuming only the memory they utilize.

The above is the detailed content of Are Split Stacks Necessary on x86_64 Architectures?. For more information, please follow other related articles on the PHP Chinese website!