jronald said:
The kernel schedules only processes, include processes, and it doesn't know about user thread.
Well, this is not exactly right.
The kernel schedules only
scheduling entities which can be far smaller than an entire process.
To make short, let's call them
kernel-threads.
In a 1:1 scheduling model such as the one priviledged under freeBSD 7, A single
user-thread is associated to a single
kernel-thread.
which means that, although the kernel knows nothing about user-threads, by scheduling a
kernel-thread, a single user thread (i.e. not an entire process if this one is multi-threaded) will be scheduled.
I understand the part of the book you mention as dealing with a N:M scheduling model. In such a model, there can be more than one
user-thread associated to a single
kernel-thread.
"
The kernel simply schedules the underlying process or LWP, [i.e kernel-thread]
which in turn uses library functions to schedule its threads. When the process or LWP is preempted,
so are its threads.
In other words, all the user-threads associated with the kernel-thread will be scheduled when the kernel-thread is scheduled.
However, this does not mean that the whole user process will be scheduled. It will depend on how the userland part of the scheduler will have made the associations.
Anyway, this is no longer the priviledged approach under freeBSD since V7.
