MMAP learning

MMAP

Virtual process address space (vm_area_struct) And virtual kernel address space (vm_struct) All mapped to the same physical memory space . When Client End and Server When the client sends data ,Client（ As a data sender ） Start with your own process space IPC Communication data copy_from_user Copy to kernel space , and Server End （ As a data receiver ） Share data with the kernel , No longer need to copy data , It's through the offset of the memory address space , You can get the memory address , The whole process only takes place one memory copy . In general , need Client Copy the end process space to the kernel space , Then copy from kernel space to Server Process space , There will be two copies .

For process and kernel virtual address mapping to the same physical memory operation occurs at the data receiving end , The data sender still needs to copy the data in user mode to kernel mode . Here we are , Why not map the sender and receiver directly to the same physical space , That way, even one copy operation is not needed ,0 The second copy operation is related to Linux Shared memory of standard kernel IPC There's no difference in mechanism , For shared memory, though efficient , However, the problem of multi process synchronization is complex , And the pipes / Message queuing, etc IPC It needs to be copied 2 two , Low efficiency .
Here is a rough flow chart ：