about platform_execute in usb_drv

[Norman Feske]

Hello Jaeyong,

>> This function is used to switch the stack and execute a given function.
> 
> Why do we have to switch the stack?
> For instance, what would happen if we do not switch the stack?

in contrast to normal function calls, user-level threads are executed
without the need to return from a function. For example, let us assume
that a user-level thread executes the following code:

  void thread_entry()
  {
    int signal_count = 0;
    for (;;) {
      wait_for_signal();
      signal_count++;
      printf("signal_count is %d", signal_count);
      ...
    }
  }

Here, the variable 'signal_count' is allocated on the thread's stack
because it is a local variable. The function 'wait_for_signal()' blocks
the user-level thread for an incoming event and invokes scheduler of
other user-level threads. So here, a user-level context switch might happen.

Now, imagine that you have two of such threads running. Let's call them
A and B. If they would share one stack, and the threads were scheduled
in the order A, then B, the resulting stack would look as follows:

  ------------------------------------------ <- stack top
   context of B (i.e., 'signal_count' of B)
  ------------------------------------------
   context of A (i.e., 'signal_count' of A)
  ------------------------------------------

How would the stack look after the user-level scheduler would now pick
A? If we just switched to the last stack pointer used by A, the picture
would look like this:

  ------------------------------------------
   context of B (i.e., 'signal_count' of B)
  ------------------------------------------ <- stack top
   context of A (i.e., 'signal_count' of A)
  ------------------------------------------

Now, thread A executes the 'printf' function, which uses the stack and
thereby overwrites the context of 'B'. After returning from 'printf',
the stack would look like this:

  ------------------------------------------
   left-overs of the printf function
  ------------------------------------------ <- stack top
   context of A (i.e., 'signal_count' of A)
  ------------------------------------------

The context of B has been overwritten by the function call frames used
by the 'printf' function and its subroutines.

By hosting the stack of each user-level thread in a distinct memory
area, such situations cannot happen. The threads can be executed in any
order without overwriting each other's context.

Cheers
Norman

-- 
Dr.-Ing. Norman Feske
Genode Labs

http://www.genode-labs.com · http://genode.org

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth