cvs commit: src/sys/netinet sctp_bsd_addr.c

[M. Warner Losh]

In message: <477A2D6D.9030403 at cisco.com>
            Randall Stewart <rrs at cisco.com> writes:
: M. Warner Losh wrote:
: > In message: <20080101161858.A10345 at delplex.bde.org>
: >             Bruce Evans <brde at optusnet.com.au> writes:
: > : On Mon, 31 Dec 2007, M. Warner Losh wrote:
: > : 
: > : > In message: <200712311219.08286.jhb at freebsd.org>
: > : >            John Baldwin <jhb at FreeBSD.org> writes:
: > : 
: > : > : The more correct fix though is to do a 'sched_prio()' at the start of the
: > : > : thread's main loop to set the priority and then not adjust it via msleep().
: > : > : Kernel threads really should never pass a priority to msleep() but always '0'
: > : > : (which means "don't change my priority").
: > : >
: > : > Not PZERO?  When should one use PZERO and when should one use a bare
: > : > '0'?  Can this information be added to the man page?
: > : 
: > : PZERO is compatibility cruft which should never be used.  Just a few
: > : places in kern still use it to invent a priority when no suitable
: > : priority (like PSOCK or PRIBIO) is already #defined.  It isn't clear
: > : where these invented priorities are suitable.
: > 
: > Do we want to document the other Pxxxx priorities?
: > 
: > : Otherwise, PZERO has a completely different meaning from either priority
: > : 0 (maximal) or the bare 0 arg to msleep.  It means some middle priority,
: > : or the bias from priority 0 to get to that middle priority, so that
: > : after subtracting it, 0 becomes the middle priority.  The bare 0 is
: > : actualy priority 0 (maximal) overloaded to mean "don't change the
: > : priority".  This overloading doesn't lose anything except clarity since
: > : nothing is permitted to wake up at maximal priority after a sleep.
: > : (Maximal priority is reserved for realtime priority ithreads and even
: > : much lower priority ithreads are not permitted to sleep, and non-interrupt
: > : threads aren't permitted to run at ithread priorities except temporarily
: > : for priority propagation.)
: > 
: > So would the following be a reasonable change to sleep.9?
: > 
: > Index: sleep.9
: > ===================================================================
: > RCS file: /home/ncvs/src/share/man/man9/sleep.9,v
: > retrieving revision 1.61
: > diff -u -r1.61 sleep.9
: > --- sleep.9     30 Mar 2007 18:07:26 -0000      1.61
: > +++ sleep.9     1 Jan 2008 09:44:01 -0000
: > @@ -93,6 +93,10 @@
: >  runnable with the specified
: >  .Fa priority
: >  when it resumes.
: > +.Dv PZERO 
: > +should never be used, as it is for compatibility only.
: > +A new priority of 0 means to use the thread's current priority when
: > +it is made runnable again.
: >  If
: >  .Fa priority
: >  includes the
: > 
: > Warner
: > 
: That sure would have been nice to read in the manual page .. I would
: have not changed this at all if I would have had that to read :-(
: 
: I was pinged on xchat over an issue where the sctp worker thread
: was hanging... and in the process I was told that 0 was max
: priority.. which really concerned me... 0 having a special
: meaning of.. "don't change my priority" means the old code
: was fine :-0
: 
: Not that having a mid-range lower priority is a bad thing
: for this task...

I had someone point out that I wasn't using PZERO in the code I wrote
at work, so I went through and changed all the drivers there to use
PZERO always.  The fact it is compatibility shimming wasn't clear at
all, hence my desire to document it :0

Warner