Date: Sun, 14 Jul 1996 19:30:46 +0200 (MET DST) From: J Wunsch <j@uriah.heep.sax.de> To: freebsd-hackers@freebsd.org (FreeBSD hackers) Cc: pjchilds@imforei.apana.org.au, jonny@gaia.coppe.ufrj.br (Joao Carlos Mendes Luis) Subject: Re: Preach it (was Some recent changes to GENERIC) Message-ID: <199607141730.TAA11956@uriah.heep.sax.de> In-Reply-To: <199607141548.MAA19290@mailhost.coppe.ufrj.br> from Joao Carlos Mendes Luis at "Jul 14, 96 12:48:16 pm"
index | next in thread | previous in thread | raw e-mail
As Joao Carlos Mendes Luis wrote:
(Moved to -hackers again, this is becoming technical.)
> I was thinking about a completely different opioin this week. What if the
> config program could make the kernel something like a microkernel ? For
> example, suppose I want to build an embedded BSD (in EPROM) just to make
> some kind of industrial process control. I would not need lots of things.
> For example: SWAP, FS support (MFS and/or NFS only !), networking, console,
> and maybe even some system calls could disappear.
>
> I have a friend making some tests on this and he cannot make something less
> than 300K for the kernel. It' still too much for such an application.
Uaah!
I've just tried to configurate a `minimal' kernel, with just the
following devices:
isa0, fdc0, fd0, wdc0, wd0, sc0, npx0, sio0, sio1, sio2, sio3, lpt0
Only FFS configured, and no other pseudo-devices.
First bummer: the kernel doesn't link. _netisr_set is undefined.
After fiddling around, the following patch fixes this:
Index: sys/i386/i386/machdep.c
===================================================================
RCS file: /usr/home/cvs/src/sys/i386/i386/machdep.c,v
retrieving revision 1.192
diff -u -u -r1.192 machdep.c
--- machdep.c 1996/06/08 11:03:01 1.192
+++ machdep.c 1996/07/14 17:13:35
@@ -40,6 +40,7 @@
#include "npx.h"
#include "isa.h"
+#include "ether.h"
#include "opt_sysvipc.h"
#include "opt_ddb.h"
#include "opt_bounce.h"
@@ -208,12 +209,14 @@
int cpu_class = CPUCLASS_386; /* smallest common denominator */
static void dumpsys __P((void));
+#if NETHER > 0
static void setup_netisrs __P((struct linker_set *)); /* XXX declare elsewhere */
+extern struct linker_set netisr_set;
+#endif
static vm_offset_t buffer_sva, buffer_eva;
vm_offset_t clean_sva, clean_eva;
static vm_offset_t pager_sva, pager_eva;
-extern struct linker_set netisr_set;
#define offsetof(type, member) ((size_t)(&((type *)0)->member))
@@ -271,10 +274,12 @@
}
}
+#if NETHER > 0
/*
* Quickly wire in netisrs.
*/
setup_netisrs(&netisr_set);
+#endif
/*
#ifdef ISDN
@@ -460,6 +465,7 @@
}
}
+#if NETHER > 0
int
register_netisr(num, handler)
int num;
@@ -486,6 +492,7 @@
register_netisr(nit->nit_num, nit->nit_isr);
}
}
+#endif
static struct cpu_nameclass i386_cpus[] = {
{ "Intel 80286", CPUCLASS_286 }, /* CPU_286 */
The fix is arguably bogus, but i found _netisr_set defined (for some
obfuscated reason i couldn't examine) in if_ether.o in another kernel.
So i assume it depends on NETHER. If somebody knows it better (and
understands all the magic behind these linker sets), he should feel
free to correct the above patch, and commit a fix (or submit it for
committing).
Still, the resulting kernel is:
j@uriah 195% size kernel
text data bss dec hex
434176 40960 25828 500964 7a4e4
The prime single offending object files by size are syscons.o, pmap.o,
and userconfig.o. Umm, the latter is not even an option right now --
we bloat all our kernels with it, regardless of whether it's needed or
not.
Next attempt, use MFS (but no FFS), and isa0, npx0, sio0, sio1, sio2,
sio3, lpt0.
text data bss dec hex
393216 32768 22776 448760 6d8f8
Still pompous.
--
cheers, J"org
joerg_wunsch@uriah.heep.sax.de -- http://www.sax.de/~joerg/ -- NIC: JW11-RIPE
Never trust an operating system you don't have sources for. ;-)
help
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199607141730.TAA11956>