Re: Fix nagging make_dev() warnings

["Frank W. Josellis" <frank@xxxxxxxxxxxxxxxxxxxxx>]

On Thu, 7 Sep 2006, Simon 'corecode' Schubert wrote:

> [...]
> 
> i'm not sure if dev->si_ops->head.flags is the right place to add D_NAGGED,
> but something like this seems to be the right thing.
> 

Let's recall that in kernel mode dev_t is a pointer to a struct specinfo 
(defined in <sys/conf.h>), where the former struct cdevsw member from 
FreeBSD has been replaced by a struct dev_ops for Dragonfly. With that in 
mind it's not difficult to see where these flags should go. However, we 
still need an explanation why the warning is issued at all, and meanwhile 
I think that this becomes clear from inspecting the function spec_open() 
in spec_vnops.c: 

Here we deal with a dev_t dev obtained via udev2dev(), in which respect 
Dragonfly behaves significantly different from FreeBSD. Indeed, from 
sys/kern/kern_conf.c we see that udev2dev() calls hashdev(), and that for 
the returned dev_t the member dev->si_name is *not* set as it could be 
expected for FreeBSD 4.x. In particular, for the dev in question we expect 
devtoname(dev) to return "consolectl" with FreeBSD 4.x, whereas we get 
"#sc/255" with Dragonfly. But that's not a bug, that's a feature resulting 
from using udev2dev(). You may wish to turn on some debugging output for 
hashdev() to see what devtoname() actually returns for the various 
devices.

So the problem appears to be solved by just dropping that obsolete 
warning, it's completely useless now (patch below).

Regards,
Frank Josellis


--- patch begins here ---
--- sys/vfs/specfs/spec_vnops.c.orig	2006-08-12 02:25:58.000000000 +0200
+++ sys/vfs/specfs/spec_vnops.c	2006-09-08 08:21:33.000000000 +0200
@@ -266,13 +266,6 @@
 			dev->si_bsize_phys = DEV_BSIZE;
 		vinitvmio(vp, IDX_TO_OFF(INT_MAX));
 	}
-	if ((dev_dflags(dev) & D_DISK) == 0) {
-		cp = devtoname(dev);
-		if (*cp == '#') {
-			printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
-			    dev_dname(dev), cp);
-		}
-	}
 
 	/*
 	 * If we were handed a file pointer we may be able to install a
--- patch ends here ---