*BSD News Article 99543


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!news.mel.connect.com.au!munnari.OZ.AU!news.ecn.uoknor.edu!paladin.american.edu!02-newsfeed.univie.ac.at!01-newsfeed.univie.ac.at!news.radio.cz!news.shkoo.com!su-news-hub1.bbnplanet.com!cpk-news-hub1.bbnplanet.com!news.bbnplanet.com!news-peer.sprintlink.net!news-sea-19.sprintlink.net!news-in-west.sprintlink.net!news.sprintlink.net!Sprint!204.248.21.3!neonramp.com!cynjut.neonramp.com!cynjut.neonramp.com!not-for-mail
From: burgess@cynjut.neonramp.com (Dave Burgess)
Newsgroups: comp.unix.bsd.netbsd.announce,comp.unix.bsd.freebsd.announce,comp.answers,news.answers,comp.unix.openbsd.announce
Subject: [comp.unix.bsd] NetBSD, FreeBSD, and OpenBSD FAQ (Part 4 of 10)
Supersedes: <386bsd-faq-4-867394803@cynjut.neonramp.com>
Followup-To: comp.unix.bsd.netbsd.misc
Date: 13 Jul 1997 02:00:18 -0500
Organization: Dave's House in Omaha
Lines: 1513
Approved: news-answers-request@MIT.Edu,cgd@sun-lamp.cs.berkeley.edu
Expires: 07/31/97 02:00:04 CDT
Message-ID: <386bsd-faq-4-868777204@cynjut.neonramp.com>
References: <386bsd-faq-1-868777204@cynjut.neonramp.com>
Reply-To: burgess@cynjut.neonramp.com (BSD FAQ Maintainer)
NNTP-Posting-Host: cynjut.neonramp.com
Keywords: FAQ 386bsd NetBSD FreeBSD !Linux
X-Posting-Frequency: Posted on/about the 13th and the 27th of every month.
Xref: euryale.cc.adfa.oz.au comp.unix.bsd.netbsd.announce:474 comp.unix.bsd.freebsd.announce:620 comp.answers:25035 news.answers:18903

Posted-By: auto-faq 3.1.1.2
Archive-name: 386bsd-faq/part4

Section 3.	(Kernel Building and Maintenance)

3.0	System Internals
	
	One of the interesting aspects of *BSD is the fact that it comes 
	with the complete source.  This allows you to make changes to the 
	system, recompile, and test out your new ideas.  This section of 
	the FAQ describes many of the different aspects of this endeavor 
	and common problems and pitfalls that are encountered.  Kevin Lahey 
	provided the substantial portion of this section.  You can contact 
	him via E-Mail at (kml@rokkaku.atl.ga.us) or contact Dave Burgess 
	(burgess@cynjut.neonramp.com).

3.1	Kernel

3.1.1	How do I build a kernel?

	The kernel can be compiled in a variety of ways to support different 
	devices and configurations.  Compilation is controlled by a config 
	file that specifies the characteristics of the kernel.  A set of 
	different config files is located in /sys/i386/conf or 
	/sys/arch/i386/conf.  The configuration file names are in upper case.

	To build a particular kernel (in this example, we use the GENERICISA
	configuration file in NetBSD or FreeBSD):

	% cd /sys/i386/conf
	% config GENERICISA
	% cd /sys/compile/GENERICISA
	% make depend
	% make

	In NetBSD, since there are multiple architectures supported, there
	is an architecture line in the middle of the path to these files.
	See the build.kernel script in section 3.8 for more information.

	Remember, when structures in the kernel change, there are some
	programs (ps, top, etc.) that will cease to work correctly.  You
	will need to recompile these programs as well as the new kernel.
	You need to do the following to make sure that your programs get
	updated as well as the kernel:

	cd /usr/src/include
	make install
	cd /usr/src/lib/libkvm
	make clean && make && make install
	cd /usr/src/bin/ps
	make clean && make && make install
	cd /usr/src/...
	

3.1.1.1 Why does the kernel code for NetBSD still use the old K&R style
	declarations when the ANSI declarations are SO much better?
3.1.1.2	How do I port NetBSD to another platform?

	We still use the old style K&R definitions easier, because 
	bringing up a new port on a foreign machine with a brain-damaged 
	compiler can be impossible, or at least very difficult,
	if you don't do it this way.  Remember, NetBSD is multi-platform, 
	and tries to make it as easy as possible to port.  Which means 
	building pieces of the system with someone else's compiler.


3.1.2	I want to do one of the following things:
	* add a device not in the distributed kernel (third com
	  port, additional disk or tape, line printer driver, etc).
	* use a patch from the net or the patchkit to fix a kernel bug.
	* add another swap device.
	* recompile the kernel to remove extraneous devices so that
	  it takes up less space.
	* configure more pseudo-terminals to allow for more xterms
	  or network logins.
	
	You're going to have to recompile the kernel after you modify the 
	config file.  See section 3.2 below for more information about the 
	config file in general.
	

3.1.3	I want to build and profile a kernel.  What do I need to do?

	Step 1 is to build a profiled kernel:

		cd /sys/arch/<X>/conf
		config -p CONFIG.FILE.OF.YOUR.CHOICE
		cd /sys/arch/<X>/compile/CONFIG.FILE.OF.YOUR.CHOICE
		make depend
		make
		<install the kernel and reboot>

	Step 2 is to start the profiling process:

		log in
		su to root
		kgmon -r -h
		kgmon -b

	Step 3 is to run the application in question

	Step 4 is to stop the profile:

		kgmon -h
		kgmon -p  (this produces the gmon.out file, which is the
				profiles kernel information)

	Step 5 is to analyze the output:

		gprof /{kernel name} gmon.out > profile

	which produces a hierarchical call graph as well as flat
	profile.  The flat profile is easiest to use for beginners,
	although both are good information sources for kernel code
	performance.

	
3.1.4	Now that I have a kernel, how do I install it?

	Your kernel is called /kernel or /netbsd.  Copy the new kernel
	from /sys/compile/GENERICISA/(whatever) to /, assuming that it 
	is in that directory.  If you really screw up the new kernel, 
	you want to have something to fall back on, so be sure to 
	save /kernel to /kernel.old before copying in a new kernel.  


3.1.5	My system is complaining about stray interrupt 7.  Is my machine 
	going to explode or anything?

	No.  They are caused by lots of things.  They are, as far as
	anyone that should be expected to know about this stuff, harmless.
	There are ramifications on them being there, but for MOST users
	they do not pose a real threat to your operations.  For those of
	you that are doing REALLY interrupt intensive stuff, you may want
	to grab a technical reference and figure out why the 8259 is not
	getting reset correctly.

	In spite of the number of times this has come up (and people have
	even referenced this section) there are still at least two 
	questions on the net about this.  A memorable one was a guy who
	was just vehement that the stray int 7 was what was keeping his
	system from booting.  In fact, he went so far as to say that this
	document was practically worthless because I didn't tell him how
	to fix it.  Of course, once he configured his hard drive controller 
	so that it was on the right interrupt, his booting problem went 
	away.  I have said it before and I will say it again.  For MOST 
	users they do not pose a real threat to your operations.
	I have heard of three people (out of at least 2000) that have
	actually have problems so bad that they couldn't proceed.  They
	bought new computers, and the problem went away.

	These stray interrupts are caused by something in the PC.  
	I have yet to see a convincing explanation of precisely what,
	but they are definitely caused by something.  There are four
	ways to deal with this problem.

	1)  Ignore them.  They are spurious and do not effect the
	operation of your computer.

	2)  Implement the lpt driver.  This way, the driver traps 
	(the lpt driver expects IRQ 7) and then quietly discards them.  
	That is why when folks implement the lpt driver the 'problem' 
	goes away.  The computer is taught how to ignore them.

	3)  Do what the original 386bsd code did.  Comment out the
	diagnostic and associated code that tries to deal with them so
	you don't see the error message.

	4)  Buy a new computer that doesn't cause this problem.   It is a
	known hardware problem with the 8259 being reset incorrectly in
	hardware.

	Kalevi Suominen (jks@geom.helsinki.fi) offers this technical
	explanation of the 'stray interrupt 7' phenomenon.

	In the section of the Intel Peripheral Handbook dealing with
	the 8259A there is a description of the 6-step interrupt
	sequence for an 80x86 system (and 7-step for an MCS-80/85),
	and then the following paragraph:

	"If no interrupt request is present at step 4 of either sequence
	(i.e., the request was too short in duration) the 8259A will
	issue an interrupt level 7. Both the vectoring bytes and the CAS
	lines will look like an interrupt level 7 was requested."

	This explains how some transient disturbances or improperly
	functioning adapter cards could trigger a stray interrupt 7
	in a system operating in the *level* interrupt mode (such as
	a PS/2 with MCA): An interrupt request will disappear as soon
	as the interrupt line goes inactive.

	That such interrupts may also occur in a system operating in
	the *edge* triggered mode (such as an ordinary PC/AT with ISA)
	is a little harder to see. Yet it is possible - even without
	malfunctioning hardware - because masking an interrupt request
	will hide its presence from the 8259A as well:


	    1. The interrupt flag (IF) of the 80x86 is reset either 
	    directly (e.g., by a "cli" instruction) or because an 
	    interrupt handler is entered. In the latter case the 
	    corresponding in-service (IS) bit of the 8259A is set 
	    (effectively blocking interrupts of lower priority).

	    2. The 8259A receives an unmasked interrupt request (IRQn), 
	    and, in case an interrupt is being served and has higher 
	    priority than IRQn, the IS bit of the 8259A is reset by 
	    an end of interrupt (EOI) command. (These steps may occur 
	    in either order.) If IRQn has higher priority (e.g. IRQ0), 
	    no EOI is necessary.

	    3. The 8259A activates the interrupt (INT) line to the 80x86
   	    (which will ignore it - for the moment).

	    4. The interrupt mask (IM) bit of the 8259A for IRQn is set.
	    (A little late, though. The sequence has already started.)

	    5. The interrupt flag (IF) of the 80x86 is set (either 
	    directly, or as a consequence of e.g. an "iret" instruction).

	    6. The 80x86 will now acknowledge the INT request by activating
   	    the INTA line of the 8259A.

	    7. The 8259A will not see the masked IRQn and will continue
   	    by issuing a spurious interrupt of level 7 instead.


	The original interrupt request (IRQn) will not be lost, however.
	It is preserved by the associated edge sense latch of the 8259A,
	and will be acted on after the IM bit has been reset again.

	The net result is that a single interrupt request will be
	delivered *twice* to the 80x86: first as a stray interrupt 7
	and later as the proper interrupt. In particular, it is perfectly
	safe to ignore the stray interrupt (other than sending an EOI).
	It is just the ghost of an aborted interrupt sequence: the system
	was not prepared for it yet.

	Bill Roman provides this update, which is so technical I can't
	even figure out what to cut out or how to repackage it so it
	makes sense to people like me.  As an interesting aside, he is
	also a Linux user; thereby proving that I'll accept help the FAQ
	from everyone:

	First of all, Kalevi Suominen's explanation is correct, but 
	requires just a little more explanation.  Step 4 in the data 
	book concerns the 8259's detection of INTA (interrupt 
	acknowledge) asserted by the processor.  This means that the 
	processor has detected INT (interrupt request) asserted by the 
	8259, the previous instruction has ended, and the interrupt 
	enable flag is true.  The processor has begun its interrupt 
	processing, and the 8259 *must* supply a vector; there is no way 
	for it to tell the processor "never mind".

	INTA causes the 8259 to examine the currently asserted interrupt 
	requests and return the vector corresponding to the highest 
	current request.  If there is no longer any interrupt request 
	asserted, it supplies vector 7 as a default.  (Intel calls this 
	"DEFAULT IR7" later in the data book.)

	There is thus a race condition between deassertion of an interrupt 
	request and interrupt servicing by the processor.  An interrupt 
	request which is removed will not always cause DEFAULT IR7 -- 
	only if the request is deasserted after the processor has 
	detected INT and irrevocably decided to act on it, but before 
	the 8259 has detected INTA and prioritized its interrupts.

	It is easy to see how this could happen when the 8259 is in 
	level triggered mode, but it is counterintuitive that it should 
	happen in edge triggered mode.  To understand this, it is 
	necessary to look at Intel's "Priority Cell--Simplified Logic 
	Diagram" (figure 9 in a 1991 databook I have at hand).  The edge 
	sense latch output is gated by the request; if the request is 
	latched, then deasserted, the 8259 no longer sees it.  There 
	must be a reason Intel did it this way, but it's sure not 
	evident to me.

	The data book provides a little more information in a later 
	section titled "Edge and Level Triggered Modes".  The most 
	important point is that the corresponding bit in the In Service 
	Register is *not* set when the 8259 generates a DEFAULT IR7.  If 
	the IRQ 7 interrupt service routine sees this condition (i.e., 
	is entered and ISR bit 7 is zero) it should simply execute an 
	interrupt return instruction without sending an End of 
	Interrupt (EOI) command to the 8259.  Also, the IRQ 7 interrupt 
	service routine can be reentered due to this condition.  Intel 
	recommends that the interrupt service routine keep track of 
	whether it is currently active so this can be detected.

	I don't think that changing the interrupt mask bit can cause the 
	problem, especially if it is changed while the interrupt flag is 
	clear.  As I mentioned, the problem occurs only when the actual 
	interrupt acknowledge process has begun, which can't happen while 
	IF is clear.

	What can generate DEFAULT IR7 is a device driver that doesn't mask 
	off its device's interrupt (either in the 8259 or in the device 
	itself) while it is performing operations which can cause the 
	device to deassert its interrupt request.  For example: imagine 
	a hypothetical device with an interrupt status register.  This 
	register latches all the device's status which could cause an 
	interrupt, and clears this status when it is read.  If the 
	processor begins executing the instruction which reads this 
	register just as a status bit is set, the device will assert 
	and deassert the request within the space of that instruction.  
	On an x86 architecture processor I have worked with, this did 
	indeed generate a DEFAULT IR7.  All device drivers should make 
	sure that the device's interrupt request is disabled while the 
	device is being serviced.  A well-behaved device will never 
	deassert its request without explicit software action.

	This leaves only the poor folks who have had to buy new computers 
	to get rid of the problem.  My suspicion in this case is that 
	crosstalk on the mainboard is causing glitches on interrupt 
	request lines.  Remember that the f**king wizards at IBM made 
	the request lines active high instead of active low with a 
	pull-up (which would have allowed wire-or interrupt sharing).  
	Some devices (some serial port cards, I believe) use a 
	tri-state driver to drive the request high, but disable the 
	driver entirely when the request is deasserted, counting on a 
	weak pull-down.  This leaves interrupt request lines floating, 
	although the 8259 has the inputs enabled.  This is all 
	conjecture, though.  

	Provided by: Bill Roman  (roman@songdog.eskimo.com)


3.1.6	I keep getting "wd0c: extra interrupt".  What does it mean?

	It means that the drive was already processing a command 
	(active) when it received an interrupt from the system telling 
	it to see if it had anything to do.  This is mostly harmless 
	but could indicate that the drive/controller is having problems 
	if the message appears often.

	
3.1.7	I keep getting silo overflow messages, but the system doesn't
	seem to mind.  Is there a problem?

	Not exactly.  This simply means that the First in first out
	buffer is getting too full.  I markedly reduced the incidence 
	of silo overflows on my system by editing dev/isa/com.c to 
	change the FIFO threshold from 8 to 4 characters.  This way, the
	buffer gets more attention and reduces the chance of overflowing
	the buffer.

	Another possibility is caused when you are using internet over 
	a telephone line via SLIP or PPP.  You might have an unbuffered
	UART on your serial port, like the 8250 or the 16450. With the
	introduction of the IBM PS/2 systems the first 16550 UART's were
	shipped to provide a 16 byte buffer. But unfortunately the buffering
	of the original 16550 did not work. The problem was solved with
	the improved 16550A UART. If you run MSD (under MSDOS!), an
	application that comes with MS-Windows, you can determine the UART
	type of your serial ports (by choosing COMS). The UART type is also
	showed when your UNIX boots up.
	
	To solve the `silo overflow' problem you can by a Multi/IO card 
	with a `real' 16550A, or a card with an extra serial port, like 
	the HAYES ESP card. The Hayes card has a DOSSETUP utility to 
	configure its port address and IRQ. The port address can be 
	chosen between 180H and 380H, the IRQ address 3, 4, 5 or 9. 
	Normally COM1 (tty00) and COM2 (tty01) claim IRQ 3 and 4. So
	you can choose (for example) IRQ 9 for the Hayes ESP card.  
	Then you have to add the appropriate lines in your kernel 
	configuration:

   	    options COM_HAYESP
   
   	    device		com0	at isa? port "IO_COM1" irq 4
   	    device		com1	at isa? port "IO_COM2" irq 3
   	    # com2: Hayes ESP serial port
   	    device		com2    at isa? port "IO_COM3" irq 9
   
	For more information on com ports in general, try this URL:

	    http://comminfo.com/pages/ctsfaq01.htm#q6.


3.1.8	I found a bug in the kernel.  How do I report it?

	Both NetBSD and FreeBSD include a facility called 'bugfiler'.  
	While the instructions are included in both system, there is 
	still some apparent confusion about when to use (and when to
	NOT use) bugfiler.

	Jordan K. Hubbard (jkh@whisker.lotus.ie)  provides us with this
	short article for FreeBSD.

	To send bug reports, you want to use the sendbug(1) command.
	The entire package for sending and filing these bugs is known 
	as "the bugfiler", which is where the confusion stepped in, 
	but sendbug is definitely the command you want to use.

	Second, it doesn't take a "net connection" to use sendbug, 
	since all it does is package up your "bug report form" and mail 
	it to us; no direct Internet connectivity is required, just mail.

	So if you can send Internet mail you can use sendbug, or you can 
	also send mail to the `FreeBSD-bugs@freefall.cdrom.com' address 
	(do NOT send it to FreeBSD.cdrom.com since it will BOUNCE, this 
	is not the place to send bugs to, just to ftp stuff from!).

	NetBSD has a similar facility, but has a different program and
	host for bug reports.  The program for NetBSD is called send-pr
	and is slightly different in several respects.  It is part of
	the GNATS system, which the NetBSD core developers started using
	in February of 1994.  It is recommended that NetBSD users see the 
	man page on send-pr before filing bug reports.

	For getting information from GNATS, see the file doc/BUGS.

	There is a email interface to the NetBSD PR database.  To query 
	the database send mail to "query-pr@gnats.netbsd.org".  The mail 
	server will send a bug database listing back to the user.

	There are several flags that are useful to send to the mail server.
	The flags are entered in the "Subject" line:
	  --summary	Display an one-line entry for each bug
	  --full 	Display the full entry for each bug	
	  --help	Display a help string containing the rest of the flags.
	  PR		The Problem Report number of a particular bug

	For example, to send a query about all the bugs:
	  $ Mail -s "--summary" query-pr@gnats.netbsd.org < /dev/null

	If you want to know more about a particular bug, let's say bug 40:
	  $ Mail -s "--full 40" query-pr@gnats.netbsd.org < /dev/null


	John Conklin is trying to get a page set up at the NetBSD WWW site
	(www.netbsd.org) that will allow people to interactively query the 
	bug database.  It should be operational soon.


3.2	What exactly is this config file, anyway?  What are all of these 
	cryptic notations?

	The config file is the list of all of the optional (and settings
	for the mandatory) parts of the kernel.  If the system is made
	up of static object files which don't change, then all you
	should ever need to do is modify the config file, reconfigure
	the kernel objects, and relink.  Since both NetBSD and FreeBSD
	are distributed with source, these files are recompiled and a
	kernel is constructed.  Some of these have been deprecated, and
	may not be in use for a particular version of the system
	(i.e. ISO9660 and CD9660 are the same, CD9660 being the newer
	version). 
	

3.2.1	Okay, fine.  Why shouldn't I just add every device I can find to
	the kernel, so I'll never have to recompile this again?

	Because it takes up space.  The kernel is wired into memory, so 
	every byte it uses comes out of the pool of memory for everything 
	else.  It can't page out sections that aren't in use.  If your 
	kernel is larger than 640K, then it can't be loaded.  You'll need 
	to use Julian Elischer's bootblocks to put it in high memory, which 
	seem to be fairly complex.  Installing them (once they are 
	compiled) is as easy as using disklabel.

	Newer versions of the *BSD kith provide the capability to build
	a kernel that is larger than 640K.  Complete instructions are
	provided in the appropriate systems.


3.2.2	What should I remove from the kernel?

	What do you need?  If you only have an SCSI controller, you don't 
	need the wd0 device;  if you have another kind of disk controller, 
	you don't need sd0.  Unless you actually HAVE more than one Ethernet 
	controller, you should comment out all but one of them.  If you don't 
	have an ethernet controller, you don't need any of the controllers or 
	NFS compiled in.   Without a CD-ROM, ISOFS is kind of pointless.  Just 
	look at what you have and think about what you really need.


3.2.3	I can't get enough remote login sessions or xterm sessions.  I also
	can only get four sessions working at a time.  What can I do?

	Increase the count of pseudo-terminals --

	pseudo-device	pty	12  # or whatever

	Every pseudo terminal should have a /dev/pty* entry.  If you have 12
	pseudo terminals, you should also have at least 12 pty devices in the
	/dev directory.  The MAKEDEV script in /dev will create as many pseudo-
	terminals as you tell it to.


3.2.4	How do I get ddb, the kernel debugger, compiled into the kernel
	and running?

	If you are using older versions of the 386BSD family, you will
	need to add a line in your config file that looks like this:

	device-pseudo	ddb

	If you are using a more recent version (the division is pretty
	unclear about when the switch was made) and do not have any
	device-pseudo entries, you will need to add the line:

	options 	DDB

	to your config file.

	Build the kernel, then run dbsym on it:

	% dbsym ./kernel

	Install it and go for it.  Ctl-Alt-Esc drops you into the debugger.

	Note: DDB as shipped originally is a memory hog, and it is very
	difficult to get a kernel small enough with enough fun things in it
	to debug in 640K

	On the NetBSD-sparc system, the L1-A is used by the the DDB
	routines to drop you into the debugger.


3.2.5	I'm getting all kinds of errors when I try to build a new
	version of GCC.  How can I upgrade GCC to the most current version?

	Yes, this will happen on most architectures on the first compile
	of src/gnu/usr.bin/gcc/libgcc.  As was stated in the mailing
	list before, when you get to this error:
	  1) run a 'make' in the gcc directory.  It will error out (most
	     likely) on libgcc.
	  1) Do a 'make install' at this point to install at least gcc, 
	     cpp, and cc1.
	  2) go back and compile in src/gnu/usr.bin/gcc/ WITHOUT doing 
	     a "make clean"
	  3) install everything in src/gnu/usr.bin/gcc


3.2.6	Can I patch the current running OS image?

	In general, I think, the answer is no.  The prevailing philosophy 
	seems to be that one should use sysctl for such things, but that 
	requires that one has already compiled in the ability to change 
	the specific variable in question.  (I discovered this when I 
	wanted to patch tickadj at runtime; I added it to kernfs, and 
	when I offered the patches (which are quite small) I was told 
	sysctl was the `correct' way.  What's incorrect about /kern was 
	never quite explained; the closest anyone came was to invoke 
	internationalization concerns.  Of course, using /kern also 
	requires having compiled in support for tweaking the variable 
	you want to change.)

	Besides, unless you've patched securelevel, I don't think there 
	is any good way to twiddle variables in the running kernel.  
	/dev/{,k}mem are, I believe, read-only once init sets securelevel 
	to 1.

	If you need to know more about the sysctl command, try 
	"sysctl -a | more" to get a list of the parameters that can be
	modified while the system is running.


3.2.7	Can I have more than one config file?  Should I rename it to something
	else?  Any other hints?  

	You can create as many (or as few) config files as you desire.  The 
	system, once the patchkit is applied, will have between 10 and 15, 
	each of which implements certain functions or features.  In addition, 
	the normal place for the patchkit to make changes to the config files 
	is in the GENERICISA file.  Since this file should remain unchanged 
	and available, it is always a good idea to copy this file to a 
	meaningful name and modify that file.  In other words, change every 
	reference in 3.1.1 from GENERICISA to HAL (or whatever you call your 
	system).

	One final note.  Every /sys/compile directory takes up 800K or so;
	you might want to watch to see how big these all get.


3.2.8	I have been getting a lot of "virtual memory exhausted" errors when 
	I am compiling a program with a really big static array.  I have 
	128Meg of memory and 8Gig of swap.  How can this be happening?

	If you are using Csh, you can simply unlimit your processes in 
	your system level /etc/csh.cshrc file or your personal .cshrc 
	file.  You can also modify your kernel so that the
	amount of memory available is less limiting.  J"org Wunsch 
	(j@tcd-dresden.de) provides us with this brief description:

	From a recent posting regarding the problem how much virtual 
	memory one could get.

	| On the other hand, i've also changed the definitions you
	| mentioned. But i didn't like to modify the header files, and
	| actually, modifying the values is as easy as:
	| 
	| options		"DFLDSIZ='(16 * 1024 * 1024)'"
	| options		"MAXDSIZ='(64 * 1024 * 1024)'"
	| 
	| Include the above lines into your kernel's config file, reconfig
	| and rebuild it.
	| 

	This has been reported to work well for NetBSD, but causes
	problems in the mkdep step of the kernel compile in FreeBSD.
	Check the FreeBSD Web Site for a more definitive answer.
	<Insert address of pointer here>.

	This is just a hint for those people poking around with compiling
	large source files. Especially, due to some gcc problems with large
	static arrays, compiling X applications with huge bitmaps would
	cause virtual memory trouble. Increasing the limits (o'course, as
	long as the h/w resources suffice) could help there.

	The default definitions for the above parameters are found in
	/usr/include/machine/vmparam.h.


3.2.8.1	I am running NetBSD and really DO have 128 Meg of memory; but the 
	generic kernel only sees the first 64Meg.  How can I fix this?

	The EXTMEM_SIZE entry needs to be set for your system.  I'm
	completely clear on the concept, but it goes something like
	this:

	The value of EXTMEM_SIZE needs to be set to the number of bytes
	in the extended memory range for your system.  Since this
	excludes the first Meg of memory, the value for the machine
	above would be 127 * 1024, for a grand total of 130048.  Once
	you reset that, you will need to either get the -current sources
	(which will do the following for you) or you need to set NKPDE
	to something larger than 12.  For 128Meg, 16 seems to be a
	reasonable number; for 256Meg, 20 would probably work, with 24
	being adequate for 512Meg.

	Both of these can be set in the source, or you can use the
	"options" method listed in the last paragraph.


3.2.8.2	How do I dedicate 16Meg of memory to nothing but disk buffers?

	Set BUFPAGES to the number of 4K buffers you want to allocate.
	For 16Meg, you would use "options	BUFPAGES=4096".
	

3.2.9	Where can I learn more about all this?

	We've skipped over a lot of details here;  the straight dope comes 
	from "Building Berkeley UNIX Kernels with Config", by Samuel J. 
	Leffler and Michael J. Karels. 


3.3	Other kernel related kind of questions.
3.3.1	Has the method for system call changed in NetBSD?  

	Q.  Is there something special with the order I need to update 
	binaries and libraries in?  If I drop in the new libc, everything 
	gives me a bus error.  Both shared and static do this.

	A.  On the port-i386 list, Charles Hannum discussed changing the 
	system call mechanism (doing it via an INT instead of a call 
	gate).  Looking at src/lib/libc/arch/i386/sys/syscall.S, it looks 
	like this change is in.   Your binaries are (if you are using an 
	old kernel) probably crashing at each system call now.

	So.. first compiling a new kernel with COMPAT_10 in it should make 
	your newly linked binaries work, I guess (have not recompiled since 
	the update myself yet). Also don't forget that you need to use 
	config.new now.

	So, the answer is Yes, the mechanism for system calls has
	changed, but the old method (using a call gate) is still
	available by specifying COMPAT_10 in your configuration file.


3.3.2	Does anyone have a system building script that takes things like 
	building a new config and multiple config files into account?

	The program I use to rebuild my kernel is available from 
	http://cynjut.neonramp.com/build-kernel
	

3.3.3	How do I upgrade from my release version of NetBSD (and
	probably FreeBSD) to the '-current' development sources?
	
	Your best method _by far_ would to to pull down a snapshot
	(assuming one exists for your arch) & install all except the etc
	tarfile, then diff the etc tarfile contents with your setup to
	check for any updates you should make.

	If you really want to do it the hard way, here is your map.
	
	# Remember to make yourself a new config (not config.old) kernel 
	# config file.
	#
	# This means any old config file you may have had from
	# V1.1 needs to have the mainbus changes made. See GENERICADP and look
	# for the isa0,eisa0,pci0 at root add the mainbus0 stuff and attached
	# your buses to it.  For updates to V1.2, you will need to make
	# certain the COMPAT_* options are correct.
	#
	# Make sure you have COMPAT_11 as part of your kernel config 
	# options.
	#
	# This assumes that the -current source is in /usr/src

	(cd /usr/src/usr.sbin/config ; make && make install && make cleandir)

	# if you don't do this, config of your kernel config file will
	# fail with errors in files.i386

	(cd /usr/src/gnu/usr.bin/gas ; make && make install && make cleandir)

	# if you don't do this, you won't be able to build locore.s, with
	# errors about cpuid instruction not found

	(cd /sys/arch/i386/conf ; config BCC13)
	(cd /sys/arch/i386/compile/BCC13 ; make depend && make)

	# copy  new kernel to /, and boot off it
        
	(cd /usr/src/share/mk ; make install)

	# if you don't do this, you'll get errors building gcc, when it
	# doesn't know how to make the manual pages (don't know how to 
	# make gcc.0)
	#
	# Rebuild yacc first or the yacc from v1.1 dies on Error Code 1 when
	# it hits %expect 34 in /usr/src/gnu/gcc/common/c-parse.y. The
	# new yacc included with V1.1B handles it, build and install it
	# before it gets used in the gcc building.
	#
	# ***
	#
	(cd /usr/src/usr.bin/yacc; make && make install && make cleandir)
	#
	# The build now uses a new tsort option (-q) when it gets to the 
	# point of 'building standard cc1 library'. At this point if using 
	# the tsort that comes with V1.1 it dies on a Error 1 and no 
	# obvious explanation. Going to the common sub-directory under 
	# gcc (ie.  /usr/src/gnu/usr.bin/gcc/common)
	# and typing 'make' will reveal the full error.
	#
	(cd /usr/src/usr.bin/tsort; make && make install && make cleandir)
	#
	#
	cd /usr/src/gnu/usr.bin/gcc
	make
	#
	# Build will die here on libgcc
	#
	(cd cc; make install)
	(cd cc1; make install)
	(cd cc1obj; make install)
	(cd cc1plus; make install)
	(cd cpp; make install)
	(cd g++; make install)
	(cd libgcc; make; make install)
	(cd libobjc; make; make install)
	(cd /usr/src/gnu/usr.bin/gcc ; make && make install && make cleandir)
	#
	# Bernd Wiserner says that the ld.so that will be built next will
	# work only with libc.so.12.0, not with libc.so.12.3
	# His instructions to make a working ld.so follow:
	# Do NOT run ldconfig while doing the folowing 5 lines ...
	#
	(cd /usr/src/include ; make && make includes)
	#
	cp -p /usr/libexec/ld.so /usr/libexec/ld.so.good
	(cd /usr/src/gnu/usr.bin/ld ; make && make install && make cleandir)
	cp -p /usr/libexec/ld.so.good /usr/libexec/ld.so
	#
	# Then build ld.so again ...
	(cd /usr/src/gnu/usr.bin/ld ; make && make install && make cleandir)  
	# Thanks, Bernd...
	
	# And now the libraries
	(cd /usr/src/lib ; make && make install && make cleandir)
       	 
	(cd /usr/src/bin/sh ; make && make install &&  make cleandir)
       	 
	# and now back to the beginning and make the world
	#
	(cd /usr/src/bin ; make && make install && make cleandir)
	(cd /usr/src/sbin ; make && make install && make cleandir)
       	 
	mkdir /usr/share/doc/usd/13.viref
	# otherwise "make install" in /usr/src/usr.bin will fail because
	# the destination directory doesn't exist - from Tom Thai
    	 
	# if you're using the obj directory hierarchy, use the
	# initscan.c from the obj directory, otherwise use the initscan.c
	# that was created here.
	
	cd /usr/src/usr.bin/lex
	if test -d /usr/src/usr.bin/lex/obj ; then
       	    cp initscan.c obj/scan.c
	else
       	    cp initscan.c scan.c
	fi
	
	# if you don't, then lex won't be built
       	 
	(cd /usr/src/usr.bin ; make && make install && make cleandir)
	(cd /usr/src/usr.sbin ; make && make install && make cleandir)
	(cd /usr/src/libexec ; make && make install && make cleandir)
	(cd /usr/src/gnu ; make && make install && make cleandir)
	(cd /usr/src/share ; make && make install && make cleandir)
     	 
	mkdir /usr/share/doc/usd/30.rogue /usr/share/doc/usd/31.trek
	# otherwise "make install" in /usr/src/games might fail
	# if the dirs weren't already there
     	 
	(cd /usr/src/games ; make && make install && make cleandir)
	

3.3.4	Is there a Makefile that does all that happy world-building
	stuff?

	There is one in the /usr/src directory.  Unfortunately, it
	seldom gets sent down unless you are using sup to get -current.
	If you need it for NetBSD, you can FTP to ftp.netbsd.org and 
	get it from the src archive.

	ftp://ftp.netbsd.org/pub/NetBSD-current/src/Makefile

	The same can be said for FreeBSD and OpenBSD, but obviously you
	will need to get it from their FTP sites; not NetBSD's.
	

3.3.5	Can NetBSD do cross compilation?

	Sure.  Check out the cross-compiling howto for MacBSD for an
	example.  All of the NetBSD ports should be capable of doing
	'pretty much' the same thing.

	You can find Colin Wood's MacBSD cross compilation howto at 
	the following URL:

	http://www.macbsd.com/macbsd/howto/cc-HOWTO

3.3.6	My network memory seems to be leaking.  The numbers just keep
	increasing slowly over time.  Is there a problem I need to worry
	about?

	Probably not.  When the system starts, the kernel malloc pool is
	not backed by real memory.  As these pages are allocated, real
	memory is assigned to them, increasing your memory usage.  As
	these pages are released, they are returned to the malloc pool,
	but the memory is never returned to the system.  Eventually,
	your malloc area will reach a point of statis, where new pages
	are not needed from real memory; the releases to the malloc area
	should cover the new pages needed.


3.4	X11/XFree86/XS3
3.4.1	What options should I define to get the X extensions included?

	Once you have applied the patch kit, the only thing left to do is to
	modify the config file to include the following line:

	options	XSERVER, UCONSOLE
 
	recompile the kernel and the kernel should support X.

3.4.2	Where can I get the FAQ for 'X'?

	Steve Kotsopoulos' general 'X on Intel-based Unix' FAQ
	available by anonymous ftp from export.lcs.mit.edu in 
	/contrib/faq/Intel-Unix-X-faq. 


3.4.3	Why does X drop characters when using xdm?   When I run xdm 
	from the console, it keeps losing keystrokes and the shift keys 
	don't always work.  Why?

        You need to run xdm with the -nodaemon flag.  The reason is
        xdm normally detaches from the keyboard.  This allows other
        processes (like getty) to return to reading from the keyboard.
        A race condition results, where some keystrokes are sent to
        xdm and others are sent to other processes.  Using the
        -nodaemon flag causes xdm to stay attached to the keyboard
        so no other process can use it.  This answer comes from Michael 
        C. Newell (root@wanderer.nsi.nasa.gov)

	This bit of trivia is also covered in detail in the X FAQ and 
	the README that accompanies XFree86.


3.4.4	What can I do to figure out why 'X' doesn't work with NetBSD?

	The best way to debug problems with starting 'X' is to redirect
	the output of the 'startx' program to a file:

	% startx >& startx.log	   # csh
	% startx 2>&1 > startx.log # sh

	With the output from this, the problems should be fairly easy to
	identify and (hopefully) fix.


3.4.5	Under NetBSD and FreeBSD, xlock (or any other program that uses
	passwords) fails to validate user passwords. Anyone know why? 

	OK, first off, make sure you're using the latest version of 
	xlock. If you've pulled it out of the /ports/ distribution then 
	you've got version 1.14. This is woefully ancient, so get the 
	latest, which at the time of writing is 2.7 (just do an archie 
	search for 'xlockmore-2.7' to find it).

	Get this, compile it up and *make sure it's setuid root*. So, 
	after you've copied it into /usr/X11R6/bin or wherever, do the 
	following:

	    # chown root.wheel ./xlock
	    # chmod 4755 ./xlock

	After that, it should work fine. The problem is that without 
	being setuid root xlock cannot read the real system passwords. 
	If you look in /etc/passwd you'll see that the passwords are 
	all '*'d out, because FreeBSD and NetBSD use shadow passwords.

	That's what worked for me. A couple of other suggestions were 
	raised last time this problem cropped up:

	  o If you're using a pre-compiled xlock then it might have 
	  been linked against the USA encryption libraries. If you're 
	  outside the States then the encryption libraries are different, 
	  and a US xlock will not be able to read the passwords. The fix 
	  is to get the sources and recompile.

	  o If you've compiled it from source, made it setuid root, and 
	  it still doesn't work, someone suggested changing the size of 
	  the constant PASSLENGTH in xlock.c from '20' to '40'. I haven't 
	  had to do this, and in v2.7 it's '64' anyway, so it shouldn't 
	  be a problem.


3.5	I want to run 'XYZA'  which is  dynamically linked and from 'some 
	other operating system'.  What special things do I have to do to 
	get it working?

	For NetBSD:
	Assuming you are trying to simulate a SVR4 system, you want to 
	create the '/emul/svr4' directory.  You will also want to ensure
	the "COMPAT_SVR4" option is in your kernel config file.  This
	option will change in 1.3-current and later, so be sure to check
	out the config files included with those versions to ensure you
	have the right options.

	Note that there are known problems when trying to run a
	staically linked Linux ELF executable and you have SVR4
	emulation built into your kernel.  THe problem is the order in
	which ELF executables are processed through the kernel tables.
	The SVR4 emulation gets processed first, thus causing the LInux
	ELF executable to be improperly processed.  This may cause
	certain Linux static ELF executables to fail under the *BSD
	systems.  The way to fix this is to remove SVR4 emulation from
	the kernel or switch to a real SVR4 executable.

	Also note that newer versions of NetBSD do not make a
	distinction between Linux ELF executables and SVR4 ELF
	executables.  The only difference (from the 1.3 kernel's
	viewpoint) is whether they are 32 or 64 bit ELF executables.

	With this accomplished, you will need to copy several files into
	the emulation directoy.  A live example might be best at this
	point.  Most of this information is include in the
	'compat_linux(8)' manpage.

	First, set up the '/emul/linux' directory.  Within this
	directory (and virtually all of the emulation directories) you
	will need the following:

		etc/	(the emulated /etc directory)
		lib/	(the emulated /lib directory)
		usr/	(the emulated /usr directory)

	From there, the simplest way to populate these directories is to
	use a program like 'cpio' or 'tar' to build an archive.  Having
	a linux machine available will greatly simplify this.  Copy
	(basically everything from the Linux (or whatever) /etc and /lib
	directories.

	Any executables that you need from the Linux system should then
	be copied into an appropriate place in the usr/ directory.  For
	example, when creating the Linux Doom system on NetBSD, you need
	to have the following files (which should all come from the
	Linux Doom distribution).

	usr/X11R6/lib
		libX11.so.6.0

	usr/bin:
		as86
		ld86

	usr/games/doom
		README.linuxx
		doom1.wad
		linuxxdoom
		sndserver.old

	With Doom specifically, you may need to set 'DOOMWAD' (or
	whatever it is) to usr/games/doom for it to work correctly.  

	In addition to the 'X' version, the native version should also
	work (with recent versions of NetBSD (1.1+)

	It is generally accepted that NetBSD's emulation of the other
	systems is pretty close *except for* sound.  The audio drivers
	in NetBSD are not a robust (yet) as they probably should be.
	Don't be surprised if sound emulation works badly, if at all,
	for any of the other operating systems emulation works for.

	The good news is that work under way (early 1997) in the
	-current version of NetBSD (pre 1.3) should fix most of this.
	The sound subsystem has been modified so that machine
	independent components are seperated from the machine dependent
	stuff, and each of the machine dependent parts is getting a
	facelist.  Ultimately, the NetBSD sounds system should be 100%
	compatible with the one from FreeBSD (that was the design goal)
	and probably Linux as well. 


3.6	You promised to talk about timezones below.  Are you going to?

	>I've seen lots of stuff about timezone's being a bit dodgy,
	>especially with most European timezones changing over to DST on 
	>the 27th March.  I must say that that was NOT the case for me - 
	>pumpy (the author's system) is running off the 
	>/usr/share/zoneinfo/GB-Eire timezone file, (symbolically) linked 
	>to /etc/localtime, the CMOS clock is running off GMT, and the 
	>kernel is compiled with "timezone 0".

	For a full discussion of this problem, check out the
	'options(4)' manpage.  It describes the DST and TIMEZONE options
	in detail.

	In newer kernels, the problems are far less dramatic than they
	used to be (see below):

	The problems with timezones and real-time clocks is that most
	386 hardware is just stupid.  It doesn't recognize DST, even
	when told to.  It can't use network system time during DST
	because it makes all the PC clocks off by an hour.  

	For machines that dual boot DOS and *BSD, one of the simplest
	solutions involves using a program called 'clock372' from
	Simtelnet (the exact site is not available).  After you install
	it, you add a couple of lines to your DOS CONFIG.SYS and set the
	hardware clock to DST.  From there, you build a kernel with DST
	and TIMEZONE set to 0 and your whole system "just works".

	In older kernels the following works:

	I use /usr/share/zoneinfo/MET as /etc/localtime and have the 
	kernel configured as

		timezone        -1 dst 4

	(My wife is running DOS on this machine for doom sometimes ;-)

	I set this strange dst value after diging in some old ultrix(?) 
	man pages.  There were several dst-changing-method listed and 4 
	was the code for the central europe one.

	This gave me an idea... I use an Ultrix box every day, so why not...

	Now, I don't know how closely this applies to NetBSD since 
	Ultrix is based on a much older version of BSD, and this isn't 
	for the kernel config, but for an envar of timezone values, but 
	it's at least somewhat enlightening on possible meanings for 
	these things.  Could someone in the know shed light on how 
	accurately this models the timezone stuff in the kernel config?  
	When I did "man timezone" this is what I got (portion of this 
	quoted from the DEC MIPS Ultrix 4.3a timezone(3) manpage, 
	slightly hacked by me (Michael L. VanLoon (michaelv@iastate.edu))


		 STD offset [DST [offset][,start[/time],end[/time]]]

	     the components of the string have the following meaning:

	STD and DST	Three or more characters that are the 
			designation for the standard (STD) or 
			summer (DST) time zone. Only STD is required; 
			if DST is missing, then summer time does not apply 
			in this locale. Upper- and lowercase letters are 
			explicitly allowed.  Any characters except a 
			leading colon (:), digits, comma (,), minus (-), 
			plus (+), and ASCII NUL are allowed.

     	offset		Indicates the value to be added to the local 
			time to arrive at Coordinated Universal Time. The 
			offset has the form:

			hh[:mm[:ss]]

			The minutes (mm) and seconds (ss) are optional. 
			The hour (hh) is required and may be a single 
			digit.	The offset following STD is required. If 
			no offset follows DST, summer time is assumed to 
			be one hour ahead of standard time.  One or more 
			digits may be used; the value is always
			interpreted as a decimal number.  The hour must 
			be between 0 and 24, and the minutes (and
			seconds) - if present - between zero and 59. If 
			preceded by a "-", the time zone is east of the 
			Prime Meridian; otherwise it is west (which may
			be indicated by an optional preceding "+").

     start and end	Indicates when to change to and back from summer
     			time. Start describes the date when the change
			from standard to summer time occurs and end 
			describes the date when the change back
			happens.  The format of start and end must be
			one of the following:

			Jn	The Julian day n (1 < n < 365). Leap
				days are not counted.  That is, in all
				years, including leap years, February
				28 is day 59 and March 1 is day 60.  It
				is impossible to explicitly refer to
				the occasional February 29.

			n	The zero-based Julian day (0 < n <
				365).  Leap days are counted, and it is	
				possible to refer to February 29.

			Mm.n.d	The nth d day of month m (1 < n < 5, 
				0 < d < 6, 1 < m < 12).  When n is 5 it 
				refers to the last d day of month m.
				Day 0 is Sunday.

     			time	The time field describes the time when,
				in current time, the change to or from
				summer time occurs. Time has the same
				format as offset except that no leading
				sign (a minus (-) or a plus (+) sign)
				is allowed. The default, if time is not
				given, is 02:00:00.

	As an example of the previous format, if the TZ environment
	variable had the value EST5EDT4,M4.1.0,M10.5.0 it would	describe 
	the rule, which went into effect in 1987, for the Eastern time
	zone in the USA.  Specifically, EST would be the designation for
	standard time, which is 5 hours behind GMT.  EDT would be the 
	designation for DST, which is 4 hours behind GMT.  DST starts
	on the first Sunday in April and ends on the last Sunday in
	October.  In both cases, since the time was not specified, the
	change to and from DST would occur at the default time of 2:00 AM.

	The timezone call remains for compatibility reasons only; it is 
	impossible to reliably map timezone's arguments (zone, a
	`minutes west of GMT' value and DST, a `daylight saving time in
	effect' flag) to a time zone abbreviation.


3.6.1	How do you change the timezone on NetBSD (FreeBSD also?)?

	Relink /etc/localtime.  This will correct the difference from
	GMT (or its trendy equivelant) to your local timezone.  In
	addition, the kernel needs to be modified to take the clock
	time in your CMOS into account.  Since most folks that run DOS
	prefer to have their clocks set to local time, the timezone
	hack was introduced to allow the kernel to adjust the CMOS
	clock time to GMT.  Once GMT has been computed, the
	/etc/localtime file can be referenced to determine the
	corrected local time.

	All generic kernels are built using the offset from California
	(why is anyone's guess:-) so just about everyone's clock will
	be off by their timezone offset from Berkeley.

	Also, it may pay to actually copy the correct timezone file
	rather than link it.  That way, you clock will be correct even
	in single users mode (when the /usr partition is not even
	mounted.  The disadvantage of this is that anytime the timezone
	file gets updated, you will need to make certain that the file
	is copied into the /etc directory.


3.6.2	The translation between seconds-since-the-epoch and date
	differs by about 18 seconds between BSD and other Unixes when
	running ntp; why?

	See ntp FAQ. Apparently, the time correction takes leap seconds
	into account twice.  The timezone files in our system take the 
	leap seconds into account in the kernel, and nntp takes the
	same 18 leap seconds into account when syncing the time.
	Because of that, the time will appear to be off by eighteen
	seconds.  (Henning Schulzrinne)


3.7	How can I implement CVS to track MY changes to the kernel source
	tree, yet still follow the -current development tree?

	I'll append the scripts I use, but be warned, they may bite you if 
	you are careless...

	The main reason I use cvs import is to handle updates from the 
	``vendor''.  The best way I've found is to import _exactly_ what 
	was shipped.  This means unconfigured, and I put config.h, etc, 
	in .cvsignore.  If I have to modify configure.in then obviously 
	I commit them :-)

#!/bin/sh
# This is a shell archive.
# remove everything above the "#!/bin/sh" line
# and feed to /bin/sh
# Use -c option to overwrite existing files
#
# Contents: 
#	README.import
#	import.sh
#	prune.sh
#
# packed by: <sjg@zen.void.oz.au> on Sat Jun 17 20:00:34 EST 1995
#
PATH=/bin:/usr/bin:/usr/ucb ; export PATH
if test -f README.import -a x$1 != x-c ; then
  echo shar: Will not over-write existing file \"README.import\"
else
  echo shar: Extracting \"README.import\" \(2902 characters\)
  sed 's/^X//' >README.import << '!EOF'
XThe following may be of use to others wanting to use CVS to merge
XNetBSD sources with local changes but are not confident that they have
Xinterpreted the documentation accurately.
X
XMuch thanks to Chris Demetriou (cgd) for taking the time to spell out
Xthe steps he used when merging NetBSD sources without which I might
Xnot have taken the plunge myself :-) The following is based on Chris'
Xtips, though of course any errors are mine.
X
XOk.  My NetBSD sources are kept in usr.src, if NetBSD is all
Xyou use CVS for you might want to simply call it src.
X
XI unpack tar files and/or sup into a directory /d2/current.
X
XTo import the entire tree I:
X
Xcd /d2/current/src
X
Xcvs import "-I! " -m "from netbsd-current as of 950508" usr.src NetBSD \
XNetBSD-950508 > /tmp/cvs.out 2>&1
X
XWhere:
Xusr.src		is the repository where the imported data goes (so set it
X		according to your own needs), 
XNetBSD		is a vendor tag.
XNetBSD-950508	is a release tag (there can be multiple release tags given).
X
XI use "-I! " as otherise some files that you need (like
Xbin/csh/USD.doc/csh.a) will be ignored..  The space after the ! is
Xneeded.
X
XIt takes quite a while.  It is a good idea to save the output to a file.
X
XAt the end you may well get a message like:
X
X	cvs checkout -jNetBSD:yesterday -jNetBSD usr.src
X
XThis means there were some conflicts between your local sources and
Xthe import.  So I do what it says - but not in my working tree.
X
Xcd /d2/tmp
Xcvs checkout -jNetBSD:yesterday -jNetBSD usr.src > /tmp/merge.out 2>&1
X
XYou can then go find all the files with conflicts.
XEither grep '^C' /tmp/merge.out or find usr.src -name '.#*' -print
XGo edit them to resolce the conflicts.  This is usually obvious.
X
XWhen happy. 
X
Xcd /d2/tmp/usr.src
Xcvs commit -m"merged local changes with NetBSD-950508"
Xcd ..
Xrm -rf usr.src
X
XOk. Now if you are brave you can:
X
Xcd /usr.src (or whereever your working sources are)
Xcvs update
X
XFinally, you should occasionally make sure you remove old files.
X
XI use a script to do this.  It does a diff between files on the NetBSD
Xbranch looking for the latest release tag (eg. NetBSD-950508). 
XIf cvs diff remports that a file does not have that tag, it is because
Xit was not present in the import (ie removed). 
X
XThe command sequence is:
X
Xcvs diff -s -r NetBSD -r NetBSD-950508 > /tmp/prune.out 2>&1
X
X# check that all went well...
Xcat /tmp/prune.out | awk '/Diffing/ { dir=$NF } 
X/NetBSD-/ { file=$NF; print dir "/" file }'  > /tmp/pruned
X
Xcat /tmp/pruned | xargs cvs tag -d NetBSD
Xcat /tmp/pruned | xargs rm -f
Xcat /tmp/pruned | xargs cvs delete
X
XNote that this is a slow process on a 486DX33!  So don't plan on
Xmerging everything very often.  Folk who mainly hack the kernel can do
Xsrc/sys more frequently.  The sequence is analogous eg.
X
Xcd /d2/current/src/sys
X
Xcvs import "-I! " -m "from netbsd-current as of 950508" usr.src/sys NetBSD \
XNetBSD-950508 > /tmp/cvs.out 2>&1
X
Xetc.
X
XHope this helps.
X
X--sjg
!EOF
  if test 2902 -ne `wc -c < README.import`; then
    echo shar: \"README.import\" unpacked with wrong size!
  fi
  
fi
if test -f import.sh -a x$1 != x-c ; then
  echo shar: Will not over-write existing file \"import.sh\"
else
  echo shar: Extracting \"import.sh\" \(290 characters\)
  sed 's/^X//' >import.sh << '!EOF'
X:
Xtoday=`date '+%y%m%d'`
X
Xrep=${1:-usr.src}
X
X# -I! doesn't work, it needs a space after the !
Xcvs import "-I! " -m "from netbsd-current as of $today" $rep NetBSD NetBSD-$today
X
X# cd somewhere
X# cvs checkout -jNetBSD:yesterday -jNetBSD usr.src > /tmp/cvs.out 2>&1
X# merge changes and commit
!EOF
  if test 290 -ne `wc -c < import.sh`; then
    echo shar: \"import.sh\" unpacked with wrong size!
  fi
  chmod +x import.sh
fi
if test -f prune.sh -a x$1 != x-c ; then
  echo shar: Will not over-write existing file \"prune.sh\"
else
  echo shar: Extracting \"prune.sh\" \(491 characters\)
  sed 's/^X//' >prune.sh << '!EOF'
X:
Xthen=${1:-`date '+%y%m%d'`}
XTF=/tmp/prune.$$
XTF2=/tmp/prune2.$$
X#S=-s
XS=
X
Xcase `echo -n .` in -n*) N=; C="\c";; *) N=-n; C=;; esac
X
Xask () { echo $N "${2:-$1?} "$C; read $1; }
X  
Xcvs diff $S -r NetBSD -r NetBSD-$then > $TF 2>&1 || cat $TF >&2
X
Xcat $TF | awk '/Diffing/ { dir=$NF } /NetBSD-/ { file=$NF; print dir "/" file }' > $TF2
X
Xcat $TF2
Xask proceed
Xcase "$proceed" in
X[yY]*)
Xcat $TF2 | xargs cvs tag -d NetBSD
Xcat $TF2 | xargs rm -f
Xcat $TF2 | xargs cvs delete
X;;
Xesac
Xrm -f $TF $TF2
!EOF
  if test 491 -ne `wc -c < prune.sh`; then
    echo shar: \"prune.sh\" unpacked with wrong size!
  fi
  chmod +x prune.sh
fi
exit 0



3.8	Optional Op-codes for NetBSD, FreeBSD, and other systems.

	MNEMONIC	INSTRUCTION
	----------------------------------
	AAC		Alter All Commands
	AAR		Alter At Random
	AB		Add Backwards
	AFVC		Add Finagle's Variable Constant
	AIB		Attack Innocent Bystander
	AWTT		Assemble With Tinker Toys
	BAC		Branch to Alpha Centauri
	BAF		Blow All Fuses
	BAFL		Branch And Flush
	BBIL		Branch on Blown Indicator Light
	BBT		Branch on Binary Tree
	BBW		Branch Both Ways
	BCF		Branch and Catch Fire
	BCIL		Branch Creating Infinite Loop
	BDC		Break Down and Cry
	BDT		Burn Data Tree
	BEW		Branch Either Way
	BF		Belch Fire
	BH		Branch and Hang
	BOB		Branch On Bug
	BOD		Beat On the Disk
	BOI		Bite Operator Immediately
	BPDI		Be Polite, Don't Interrupt
	BPO		Branch on Power Off
	BRSS		Branch on Sunspot
	BST		Backspace and Stretch Tape
	BW		Branch on Whim
	CDC		Close Disk Cover
	CDIOOAZ		Calm Down, It's Only Ones and Zeros
	CEMU		Close Eyes and Monkey with User space
	CH		Create Havoc
	CLBR		Clobber Register
	CM		Circulate Memory
	CML		Compute Meaning of Life
	COLB		Crash for Operators Lunch Break
	CPPR		Crumple Printer Paper and Rip
	CRASH		Continue Running After Stop or Halt
	CRB		Crash and Burn
	CRN		Convert to Roman Numerals
	CS		Crash System
	CSL		Curse and Swear Loudly
	CU		Convert to Unary
	CVG		Convert to Garbage
	CWOM		Complement Write-Only Memory
	CZZC		Convert Zone to Zip Code
	DBZ		Divide By Zero
	DC		Divide and Conquer
	DMNS		Do what I Mean, Not what I Say
	DMPK		Destroy Memory Protect Key
	DPMI		Declare Programmer Mentally Incompetent
	DPR		Destroy Program
	DTC		Destroy This Command
	DTE		Decrement Telephone Extension
	DTVFL		Destroy Third Variable From Left
	DW		Destroy World
	ECO		Electrocute Computer Operator
	EFD		Emulate Frisbee Using Disk Pack
	EIAO		Execute In Any Order
	EIOC		Execute Invalid Opcode
	ENF		Emit Noxious Fumes
	EO		Execute Operator
	EROS		Erase Read-Only Storage
	FLI		Flash Lights Impressively
	FSM		Fold, Spindle and Mutilate
	GCAR		Get Correct Answer Regardless
	GDP		Grin Defiantly at Programmer
	GFM		Go Forth and Multiply
	IAE		Ignore All Exceptions
	IBP		Insert Bug and Proceed
	ISC		Insert Sarcastic Comments
	JTZ		Jump to Twilight Zone
	LCC		Load and Clear Core
	MAZ		Multiply Answer by Zero
	MLR		Move and Lose Record
	MWAG		Make Wild-Assed Guess
	MWT		Malfunction Without Telling
	OML		Obey Murphy's Laws
	PD		Play Dead
	PDSK		Punch Disk
	PEHC		Punch Extra Holes on Cards
	POCL		Punch Out Console Lights
	POPI		Punch Operator Immediately
	RA		Randomize Answer
	RASC		Read And Shred Card
	RCB		Read Command Backwards
	RD		Reverse Directions
	RDA		Refuse to Disclose Answer
	RDB		Run Disk Backwards
	RIRG		Read Inter-Record Gap
	RLI		Rotate Left Indefinitely
	ROC		Randomize Opcodes
	RPB		Read, Print and Blush
	RPM		Read Programmer's Mind
	RSD		On Read Error Self-Destruct
	RWCR		Rewind Card Reader
	SAI		Skip All Instructions
	SAS		Sit and Spin
	SCCA		Short Circuit on Correct Answer
	SFH		Set Flags to Half mast
	SLMTU=x		SLIP MTU size
	SLP		Sharpen Light Pen
	SPS		Set Panel Switches
	SPSW		Scramble Program Status Word
	SQPC		Sit Quietly and Play with your Crayons
	SRDR		Shift Right Double Ridiculous
	STA		Store Anywhere
	TARC		Take Arithmetic Review Course
	TPF		Turn Power Off
	TPN		Turn Power On
	UCB		Uncouple CPU and Branch
	ULDA		Unload Accumulator
	UP		Understand Program
	WBT		Water Binary Tree
	WHFO		Wait Until Hell Freezes Over
	WI		Write Illegibly
	WSWW		Work in Strange and Wondrous Ways
	XSP		Execute Systems Programmer
	ZAR		Zero Any Register

	If you have gotten this far, you deserved some humor.

-- 
Dave Burgess                   Network Engineer - Nebraska On-Ramp, Inc.
*bsd FAQ Maintainer / SysAdmin for the NetBSD system in my spare bedroom
"Just because something is stupid doesn't mean there isn't someone that 
doesn't want to do it...."