Hello, I've been reading through the kernel source code trying to figure out when the final mmdrop() after a exit_mm() is called on a task that is having its virtual address deleted. I have discovered one instance in finish_task_switch(), the local variable mm is set (for a single access) to the last user task's active mm_struct, which is then mmdrop()'d if found not to be NULL. This implies that the user task that runs before a kernel thread has its mm_struct mmdrop()'d after said kernel thread is switched out because the kernel thread leeched off of its mm_struct. When kernel threads leeches off a user task's active_mm, it's logical for the kernel to increase the reference value to avoid race conditions, however, exit_mm() increases the mm_struct's count value in order to deffer the final mmdrop() to finish_task_switch() (according to the UTLK, end of chapter 9.5). I assume the final mmdrop() is deferred solely to have a valid pgd left in %cr3 after mmput() calls mmdrop(). This one pair of switching to a kernel thread then mmdrop()ing its active_mm once it's switched out makes sense to me. But what I don't understand is, how the kernel calls mmdrop() after exit_mm(). To clarify, the kernel executes the following code at finish_task_switch() (kernel/sched.c): struct mm_struct *mm = rq->prev_mm; ... if (mm) mmdrop(mm); I don't see how the kernel is supposed to execute mmdrop() on a exit_mm()'s behalf when exit_mm() does not set nor even touch rq->prev_mm.
Thank you, Robert G. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/