OS: does the process scheduler runs in separate process

Question

I have few doubts about how operating system works.

Scheduler: Does the scheduler runs in a separate process(like any other process). What exactly happens at the time of swapping in new process(i know the processor registers and memory tables are updated, my question is how they are updated. Can we write a program to update the registers(sc, pc) to point to a different process).

Answer 1

The process schedule could feasibly run in a separate process, but such a design would be very inefficient since you would have to swap from one process to the scheduling process (which would then have to make several system calls to the kernel) and then back to the new process, as opposed to just placing the scheduler in the kernel where you will not need system calls nor need to swap contexts more than once. Therefore, the scheduler is generally in the exclusive realm of the kernel.

Here are the steps that occur:

1. The scheduler determines which process will run in the next time slot (through various different algorithms).

2. The scheduler tells the Memory Managing Unit (MMU) to use the page table for the next process to run (this is done by setting a register to point to the table).

3. The scheduler programs the Programmable Interrupt Timer (PIT) to generate an interrupt after N clock cycles.

4. The scheduler restores the state of the registers from when the process was last running (or sets them to default values for new processes)

5. The scheduler jumps to the address of the last instruction that was not executed in the process.

6. After N clock cycles, an interrupt occurs and the operating system recognizes it as caused by the PIT, which is registered to be handled by the scheduler.

7. The scheduler saves the state of the registers (including stack pointer, etc) and grabs the program counter of where the interrupt occured (and saves it as the address to jump to next time around) and then goes back to step 1.

This is just one example of how it can be done, and many of the low level details are architecture specific. Essentially all the registers (the program state) can be saved to any place in RAM (say a linked list of structures that represent processes each having space for the registers, etc) and the virtual address space (defined by page tables) can be arbitrarily swapped out.

So essentially your question:

"Can we write a program to update the registers to point to a different process?"

is simply stated, yet the answer is correct. We sure can.