From owner-freebsd-current  Thu May 29 14:18:35 1997
Return-Path: <owner-current>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id OAA09946
          for current-outgoing; Thu, 29 May 1997 14:18:35 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id OAA09941
          for <current@freebsd.org>; Thu, 29 May 1997 14:18:30 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.5/8.6.9) id HAA29423; Fri, 30 May 1997 07:16:00 +1000
Date: Fri, 30 May 1997 07:16:00 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199705292116.HAA29423@godzilla.zeta.org.au>
To: bde@zeta.org.au, dfr@nlsystems.com
Subject: Re: disk cache challenged by small block sizes
Cc: current@freebsd.org
Sender: owner-current@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

>This is caused by my recent changes to vfs_bio to make NFS mmap work
>properly.  I didn't test on a filesystem with small block sizes.  This
>patch should fix it:

Thanks.

>...
> static void
>-vfs_page_set_valid(struct buf *bp, vm_offset_t off, vm_page_t m)
>+vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, int pageno, vm_page_t m)
> {
> 	struct vnode *vp = bp->b_vp;
> 	vm_offset_t soff, eoff;
> 
> 	soff = off;
>-	eoff = min(off + PAGE_SIZE, bp->b_bufsize);
>+	eoff = off + min(PAGE_SIZE, bp->b_bufsize);
>+	vm_page_set_invalid(m, soff, eoff);

This sometimes traps.  It only works on my test system.  I think soff needs
to be reduced (mod PAGE_SIZE)?

> 	if (vp->v_tag == VT_NFS) {
>-		soff = max((bp->b_validoff + DEV_BSIZE - 1) & -DEV_BSIZE, soff);
>-		eoff = min(bp->b_validend & -DEV_BSIZE, eoff);
>+		vm_offset_t sv, ev;
>+		sv = off + (bp->b_validoff + DEV_BSIZE - 1) & -DEV_BSIZE
>+			- pageno * PAGE_SIZE;
>+		ev = off + bp->b_validend & -DEV_BSIZE - pageno * PAGE_SIZE;

This has too few parentheses even for me :-).  Please use `& ~(BSIZE - 1)'
instead of -DEV_BSIZE.  The latter assumes 2's complement and may be less
idiomatic.

Bruce