From owner-cvs-sys  Sun Feb 16 05:32:28 1997
Return-Path: <owner-cvs-sys>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id FAA04607
          for cvs-sys-outgoing; Sun, 16 Feb 1997 05:32:28 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id FAA04592;
          Sun, 16 Feb 1997 05:32:15 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.3/8.6.9) id AAA19640; Mon, 17 Feb 1997 00:28:19 +1100
Date: Mon, 17 Feb 1997 00:28:19 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199702161328.AAA19640@godzilla.zeta.org.au>
To: mpp@freefall.freebsd.org, peter@spinner.dialix.com
Subject: Re: cvs commit: src/sys/sys types.h
Cc: ache@freefall.freebsd.org, bde@zeta.org.au, cvs-all@freefall.freebsd.org,
        CVS-committers@freefall.freebsd.org, cvs-sys@freefall.freebsd.org
Sender: owner-cvs-sys@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>> Sounds like it is time for us to implement the poll system
>> call.  I noticed that OpenBSD has already done so.
>
>So have I..   But then again, you're all probably sick of hearing 
>that by now :-]

How much slower would it be? :-)

>Also, aren't you misreading the roundup?
>
>        /* The amount of space we need to allocate */
>        ni = howmany(roundup2 (uap->nd, FD_SETSIZE), NFDBITS) *
>                sizeof(fd_mask);
>        if (ni > p->p_selbits_size) {
>		/* realloc p_selbits */
>	}
> ...later...
>        /* The amount of space we need to copyin/copyout */
>        ni = howmany(uap->nd, NFDBITS) * sizeof(fd_mask);

Oops.  I had forgotten that select() doesn't actually initialize the
entire descriptor set like its man page says.  It only zeros a convenient
number of bits above `nfds'.  We considered only rounding to a byte
boundary but decided to be bug for bug compatible.  The first `ni' seems
to be left over from a version that wanted to round to an fd_set boundary.
It doesn't hurt to allocate a few extra bits, but bzeroing them is just
wasteful.

>Note the reuse of 'ni'.  The application is only exposed to sizeof(fd_mask)
>(fd_mask == unsigned long) rounding, not FD_SETSIZE.  The FD_SETSIZE
>rounding is only to size the initial workspace to save excess reallocation
>calls to the kernel allocator.  Even then it expands by 32 bytes (256
>bits), not FD_SETSIZE.

Yes, the change for user space was OK except for performance problems.
Sorry for the false alarm.

>Also, it seems to me that the code is not very optimal.  I'm probably 
>about to embarress myself here, but wouldn't it be better something like 
>this?:
>...

It should just clear 6 * (minimum (rounded) number of bits) and not use
the bits at the end.

Bruce