KSE, libpthread & libthr: almost newbie question
Julian Elischer
julian at elischer.org
Fri Oct 27 19:34:27 UTC 2006
Ruslan Ermilov wrote:
> On Fri, Oct 27, 2006 at 10:56:21AM +0200, Ivan Voras wrote:
>> Lev Serebryakov wrote:
>>> Hello ,
>>>
>>> I've was sure, that both libpthread and libthr use KSE to make
>>> multithreading. They use KSE in different ways: libpthread uses N:M
>>> model and libthr uses 1:1 model, but bot use KSE to work.
>>> How will be possible to use these libraries (read: multithreaded
>>> programs) when KSE will be optional, on kernel without KSE?!
>> Yes, isn't KSE by definition "that thing that is scheduled in the kernel"?
>>
> KSE == N:M threading
>
> A 1:1 threading (libthr) is much simpler than N:M threading (libpthread),
> and thus doesn't require KSE support in the kernel; see kse(2) manpage
> for details. Without the KSE option in the kernel, all kse(2) syscalls
> will return EOPNOTSUPP, and a lot of code becomes redundant.
KSE is a misnomer that I abandoned long ago..
mostly it is the thread fairness code that is independent of what
threading library is running (see the other email I just sent)
(or should be)
>
> : /*
> : * Initialize global thread allocation resources.
> : */
> : void
> : threadinit(void)
> : {
> :
> : mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
> : tid_unrhdr = new_unrhdr(PID_MAX + 1, INT_MAX, &tid_lock);
> :
> : thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
> : thread_ctor, thread_dtor, thread_init, thread_fini,
> : UMA_ALIGN_CACHE, 0);
> : #ifdef KSE
> : ksegrp_zone = uma_zcreate("KSEGRP", sched_sizeof_ksegrp(),
> : ksegrp_ctor, NULL, NULL, NULL,
> : UMA_ALIGN_CACHE, 0);
> : kseinit(); /* set up kse specific stuff e.g. upcall zone*/
> : #endif
The KSEGRP is a part of the fairness code in general and independent
of M:N and 1:1
> : }
>
>
> Cheers,
More information about the freebsd-current
mailing list