I'm currently studying Linux. I know the thread is a kind of lightweight process on Linux. But I wonder to know where the thread stack space comes from.
The stack of the thread is private. It is independent of the process stack. Based on my search, some people said thread stack was created by mmap(). And also, some people said mmap() space is not heap. It is between stack and heap. So thread stack comes from the memory mapping segment of the process. Is that correct?
As far as the Linux kernel is concerned, threads are processes with some more sharing than usual (e.g. their address space, their signal handling, and their process id, which is really their thread group id).
When a process starts, it has a single thread, with a stack etc. When that thread starts another thread, it’s up to the creating thread to provide a stack for the new thread; that is typically done using mmap, because mmap supports various flags which help ensure that the allocated memory is suitable for use as a stack. See the example program in man 2 clone. However there is no requirement to use mmap, any allocated block of memory meeting the requirements for a stack can be used.
Stacks set up for threads aren’t private: they are visible to other threads sharing the same address space. However, they must be reserved for a single thread’s use as a stack (multiple threads sharing a single stack won’t work well, to say the least).
See Are threads implemented as processes on Linux? for more context and history.
