Hi,
> What is the difference between mm_count, mm_users and
> mm_realusers?
> Also what is active_mm and when is it set/cleared and to
> what??
A process address spaces is described by it's memory descriptor
'mm_struct'; mm field in task_struct points to the address of
this descriptor (task_struct->mm). Only one mm_struct exists for
each process, and (task_struct->active_mm == task_struct->mm) is
always true.
Well, I am not 'the' expert in linux memory management, and if
there is an error in some aspects of my explanation I appreciate
any corrections... but I think what is really used by the Linux
memory management subsystem is the active_mm field, and not the
mm field.
Kernel threads have no user space context/ no own memory areas /
page tables... and so the task_struct->mm field is always NULL.
Anyway, no matter if the current task is a kernel thread or an user
space process, Linux expects and requires a memory descriptor for
the current context (e.g. to map it's virtual memory to physical
memory it needs the address of the pgd). Kernel threads borrow the
mm of the previous task and place it in the task_struct->active_mm
field. (Results in better performance, e.g. just borrowing mm_struct
instead of carrying out an expensive TLB flush etc...).
Have a look at the source code in kernel/sched.c:
...
673 prepare_to_switch();
674 {
675 struct mm_struct *mm = next->mm;
676 struct mm_struct *oldmm = prev->active_mm;
677 if (!mm) {
678 if (next->active_mm) BUG();
679 next->active_mm = oldmm;
680 atomic_inc(&oldmm->mm_count);
681 enter_lazy_tlb(oldmm, next, this_cpu);
682 } else {
683 if (next->active_mm != mm) BUG();
684 switch_mm(oldmm, mm, next, this_cpu);
685 }
686
687 if (!prev->mm) {
688 prev->active_mm = NULL;
689 mmdrop(oldmm);
690 }
691 }
As you see, every task has a memory descriptor, but mem. descriptors
can be shared between tasks.
mm_count is a reference counter to the mm. This is important for lazy
TLB
switches where a task may be using one mm_struct TEMPORARILY.
mm_users counts of the number of threads accessing an mm. A cloned
thread
will up this count to make sure an mm_struct() is not destroyed early.
The
swap_out code will increment this count when swapping out portions of
the
mm
Hope this throws a little more light on the usage of memory descriptors.
Thomas
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
