On Thu, 2017-04-27 at 18:03 -0400, J. Bruce Fields wrote:
> On Thu, Apr 27, 2017 at 11:54:48AM +0800, Kinglong Mee wrote:
> >
> >
> > On 4/27/2017 11:24, Trond Myklebust wrote:
> > > On Thu, 2017-04-27 at 11:06 +0800, Kinglong Mee wrote:
> > > >
> > > > There is only one use of svc_set_num_threads that by nfsd,
> > > > I'd like to change svc_set_num_threads and update nfsd than add
> > > > a new
> > > > function.
> > >
> > > You can't really combine the two methods. Either you choose
> > > signals or
> > > you choose kthread_stop(). The problem is that signals require
> > > the
> > > thread to be able to persist past the nfsd_destroy() (which again
> > > forces things like nfsd() having to take nfsd_mutex), or you have
> > > to
> > > make things synchronous, in which case having nfsd() try to take
> > > nfsd_mutex causes deadlocks.
> > >
> > > IOW: if there is legacy behaviour here that requires the signal
> > > method,
> > > then knfsd cannot be converted.
> >
> > Got it.
> >
> > Tested-and-reviewed-by: Kinglong Mee <kinglongmee@xxxxxxxxx>
>
> So does what I have in git://linux-nfs.org/~bfields/linux.git nfsd-
> next
> look correct?
>
> (Alternatively, if Trond's taking this through his tree, that's fine
> too, feel free to add my ACK.)
>
If you've already got it queued up then I'm fine with that. I have it
in my "testing" branch (which passes ☺) but haven't yet moved it into
linux-next.
--
Trond Myklebust
Linux NFS client maintainer, PrimaryData
trond.myklebust@xxxxxxxxxxxxxxx
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
