mirror of
https://git.FreeBSD.org/src.git
synced 2024-12-15 10:17:20 +00:00
afc6f08629
from a 32-bit value to a 64-bit value. This commit does not actually change anything. It merely provides instructions, scripts, and a safety measure in Makefile.inc1 for people who want to make the change. The real change to 64-bit time_t's on sparc64 is scheduled to happen on March 10th, assuming that so major problems are found between now and then by early-adopters. Reviewed by: freebsd-sparc64
362 lines
18 KiB
Plaintext
362 lines
18 KiB
Plaintext
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
|
|
The FreeBSD/sparc64 port is going to change time_t from 32-bits to 64-bits.
|
|
This file explains the exact steps that users should follow to update their
|
|
sparc64 systems for this change. People running FreeBSD on other types of
|
|
hardware, such as CPU's from Intel or AMD, can ignore this file. For now,
|
|
this change is only happening for people running FreeBSD on Sparc hardware.
|
|
|
|
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
# Copyright (c) 2004 - Garance Alistair Drosehn <gad@FreeBSD.org>.
|
|
#
|
|
# All rights reserved.
|
|
#
|
|
# Redistribution, publication, translation and use, with or without
|
|
# modification, in full or in part, in any form or format of this
|
|
# document are permitted without further permission from the author.
|
|
#
|
|
# THIS DOCUMENT IS PROVIDED BY GARANCE DROSEHN ``AS IS'' AND ANY EXPRESS
|
|
# OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
|
# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
|
# DISCLAIMED. IN NO EVENT SHALL GARANCE DROSEHN BE LIABLE FOR ANY DIRECT,
|
|
# INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
|
# (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
|
|
# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
|
|
# STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
|
|
# IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
# POSSIBILITY OF SUCH DAMAGE.
|
|
#
|
|
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
# $FreeBSD$
|
|
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
|
|
If you are in too much of a hurry to read this file, then this is not the
|
|
time for you to upgrade to a 64-bit time_t. Period. Stick with a system
|
|
using 32-bit time_t until you have plenty of time to perform an upgrade.
|
|
|
|
This statement is true even if you have performed a thousand system upgrades
|
|
in the past, and you are certain that you know everything there is to know
|
|
about upgrades. This upgrade *will* take you more time than previous system
|
|
upgrades, simply because you must recompile at least some of your ports after
|
|
upgrading the base system.
|
|
|
|
Do not start this update unless you have the extra time.
|
|
|
|
* READ THIS ENTIRE DOCUMENT at least once before starting the upgrade. *
|
|
|
|
This is a major change. This change will *not* be backwards-compatible.
|
|
Any programs which call system-routines for handling time-values will
|
|
have to be recompiled after this change is made.
|
|
|
|
Because this change is not backwards-compatible, it is important that
|
|
the following steps be used when upgrading the system. "Shortcuts" that
|
|
have worked for EVERY SINGLE UPGRADE YOU HAVE EVER DONE IN YOUR LIFE are
|
|
probably irrelevant. This change is more disruptive than most of the
|
|
changes which are normally done on freebsd.
|
|
|
|
These steps are designed to minimize the chance of you running into any
|
|
trouble. We can not guarantee that these steps will avoid all possible
|
|
problems, but if you ignore these steps you are very likely to run into
|
|
some very painful and time-consuming headaches when upgrading.
|
|
|
|
Step Pre-1: Update to a recent snapshot of -current, keeping it as
|
|
a system with 32-bit time_t.
|
|
Step Pre-2: Install that system, using whatever steps you normally
|
|
use, and make sure that installation seems to work okay.
|
|
Step Pre-3: While still running that 32-bit time_t system, it would
|
|
probably be a good idea to cvsup your ports tree, and
|
|
then upgrade portupgrade (if you use it) and upgrade any
|
|
shells that you use. Eg:
|
|
portupgrade -Rr -f ruby portupgrade
|
|
portupgrade -Rr -f bash
|
|
That way you know you have the latest versions, and you
|
|
will also know you have the most-recent distfiles on
|
|
your machine.
|
|
|
|
Step Pre-4: For sparc64 machines which need DHCP:
|
|
The 'dhclient' in the base system is known to be unreliable
|
|
on a system which is upgraded to 64-bit time_t's. It may
|
|
work for you, but it probably will not.
|
|
As of March 3rd 2004, we have no fix for that.
|
|
However, the net/isc-dhcp3-client port does seem to work.
|
|
IF your machine needs DHCP, then you should probably install
|
|
that port and make sure you can get it working *before* you
|
|
make the change to use 64-bit time_t's.
|
|
|
|
<instructions for early-adopters>
|
|
edit the file /usr/src/sys/sparc64/include/_types.h
|
|
find the line:
|
|
typedef __int32_t __time_t; /* time()... */
|
|
and change '__int32_t' to '__int64_t'
|
|
|
|
For best results, do NOT make any other changes. Do NOT cvsup the
|
|
source tree trying to pick up any other changes. At this point you
|
|
know that you have a source tree that does work for your system, so
|
|
stick with that source tree (except for making the above 1-line
|
|
change, of course).
|
|
|
|
At one point in my testing, I did do a 'cvsup' which just happened
|
|
to pull in one bad commit that broke 'make buildworld', and a second
|
|
bad commit that broke 'make installworld'. Believe me, you REALLY
|
|
REALLY do *not* want to risk problems like that!
|
|
|
|
I am not suggesting that you have to do two whole buildworld/
|
|
installworld cycles in a single day. You could easily wait a few
|
|
days, or even a week between them. What I am suggesting is that
|
|
you should not 'cvsup' your sources inbetween the two buildworlds.
|
|
</instructions for early-adopters>
|
|
|
|
cd /usr/src #- 1.
|
|
make cleanworld #- 2. or 'rm -Rf /usr/obj/usr/src/*'
|
|
make buildworld #- 3.
|
|
make buildkernel #- 4. Add KERNCONF if you usually do.
|
|
NEWSPARC_TIMETYPE=__int64_t #- 5. (Used by a safety-check done
|
|
export NEWSPARC_TIMETYPE #- 5a. by installkernel)
|
|
make installkernel #- 6. Add KERNCONF if you usually do.
|
|
mergemaster -p #- 7.
|
|
|
|
# - - A section required for installs over NFS-mounts - - #
|
|
ifconfig -a #- NFS 8a. See note below.
|
|
shutdown now #- NFS 8b. NOT 'shutdown -r now'
|
|
cd /usr/src #- NFS 8c.
|
|
./installworld_oldk #- NFS 8d. See note below.
|
|
# - - End of this section for NFS-mounts - - #
|
|
|
|
reboot #- 9. MUST go into single-user mode
|
|
|
|
For many upgrades, it is true that you can "cheat" at this point, and
|
|
get away without actually going into single-user mode straight from
|
|
the reboot. But for this upgrade, you REALLY MUST start up straight
|
|
into single user mode. So, reboot the machine, type a space (or
|
|
anything other than 'Enter') when the boot-loader is counting down.
|
|
And then:
|
|
|
|
boot -s #- 10. (command to boot-loader)
|
|
|
|
The system will ask you if you want to use /bin/sh or some other shell.
|
|
For this upgrade, just hit enter, even if you usually prefer like some
|
|
other shell instead of /bin/sh.
|
|
|
|
fsck -p #- 11.
|
|
# - - A section required for installs over NFS-mounts - - #
|
|
PATH=/boot/kernel/bin:$PATH #- NFS 12.
|
|
# - - End of this section for NFS-mounts - - #
|
|
mount -a -t ufs #- 13.
|
|
swapon -a #- 14.
|
|
# - - A section required for installs over NFS-mounts - - #
|
|
ifconfig hme0 inet .... #- NFS 15a. See note below.
|
|
mount_nfs host:srcdir /usr/src #- NFS 15b. See note below.
|
|
mount_nfs host:objdir /usr/obj #- NFS 15c.
|
|
# - - End of this section for NFS-mounts - - #
|
|
cd /usr/src #- 16.
|
|
./installworld_newk #- 17. Might want to add -S
|
|
mergemaster #- 18.
|
|
rm -f /var/db/dhclient.leases #- 19. If this host uses DHCP
|
|
reboot #- 20.
|
|
|
|
At this point, you should be up-and-running on a system that has 64-bit
|
|
values for time_t. You will have to rebuild anything which depends on
|
|
time_t. Later in this file is a suggested order for upgrading ports.
|
|
|
|
If you have a lot of ports which start up daemons or do other processing
|
|
at system-startup, then you might want to have this reboot also go into
|
|
single-user mode for upgrading all of the ports. In my case, I've always
|
|
done a standard reboot at this point and did not run into problems, but
|
|
then I only have 25 ports installed on my SPARC64 system.
|
|
|
|
Aside: It is slightly more reasonable to use the 'reboot' command, although
|
|
you may be more familar with using 'shutdown -r now'. The shutdown command
|
|
just turns around and executes '/sbin/reboot', and with this upgrade it is
|
|
best to avoid such redirection.
|
|
|
|
# -------+---------+--------- Notes on the above -------+---------+---------+
|
|
|
|
General notes on NFS issues:
|
|
|
|
For this upgrade to 64-bit time_t's, the change is so disruptive that I
|
|
couldn't get NFS-mounts to work if I booted a "32-bit time_t system"
|
|
(ie: 32-bit versions of /bin, /sbin, /lib, ...) on a 64-bit kernel. So,
|
|
I added the installworld_oldk script. This script does two things:
|
|
1) Creates a mini-/bin inside /boot/kernel.
|
|
2) Does a minimal installworld (while still on the old kernel),
|
|
thus making it possible for NFS-mounts to work when you reboot.
|
|
|
|
The first half is a step that would be perfectly safe to do, for any
|
|
upgrade (including non-NFS ones), at any time. It is a generally safe
|
|
and interesting idea, although it really should be implemented as an
|
|
official target in /usr/src/Makefile to be done right.
|
|
|
|
The second half would USUALLY be a bad idea to do, but I think it's the
|
|
only way I can get this specific upgrade to work for people that install
|
|
from NFS-mounted directories. It is bad because you are clobbering parts
|
|
of your system even though (in the usual case) you would not know that
|
|
the new kernel actually works on your system. It also does not do a
|
|
full-install, so you end up booting into a system which is part old-
|
|
world, and part new-world. It looks like we can get away with that for
|
|
this upgrade, but the tactic would be too risky for "standard upgrades".
|
|
|
|
These instructions assume that you are already familiar with how to do
|
|
installations over NFS-mounted partitions. If you are not, you might
|
|
want to read other references, such as 'man development'.
|
|
|
|
Notes on step NFS 8a: ifconfig -a
|
|
|
|
This shows to the configuration of all your ethernet interfaces. Write
|
|
down the IP address and netmask of your main interface. This is
|
|
particularly important if the machine obtains its address via DHCP.
|
|
You will not be running dhclient after the reboot in step 8, so just
|
|
re-use the IP address that the machine is using for the present reboot.
|
|
|
|
Notes on step NFS 8b: shutdown now
|
|
|
|
This will drop you into single-user mode, without rebooting. It
|
|
will ask if you want to use /bin/sh for your shell. You do.
|
|
|
|
Notes on step NFS 8d: installworld_oldk
|
|
|
|
Note that this script only installs *part* of the new world. You will
|
|
still have to reboot into single-user mode and do the full installworld.
|
|
The installworld_oldk script will ask you if you want to build a
|
|
mini-/bin. For this upgrade, you should say "yes".
|
|
|
|
Notes on step NFS 15a:
|
|
On my Ultra-10, I have the 'hme0' device as my ethernet card. The output
|
|
of 'ifconfig -a' (from step 'NFS 7a') included the lines:
|
|
|
|
hme0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
|
|
inet 192.168.1.18 netmask 0xffffffe0 broadcast 192.168.1.31
|
|
|
|
So for this step, I typed in the command:
|
|
ifconfig hme0 inet 192.168.1.18 netmask 0xffffff00
|
|
|
|
Notes on step NFS 15b: mount_nfs
|
|
|
|
At this step, you may need to specify the host as an IP address instead
|
|
of a hostname, because the machine will only be able to resolve hostnames
|
|
that are in /etc/hosts.
|
|
|
|
In my case, I found it easier to create a source file ahead of time
|
|
which included the ifconfig and mount_nfs commands that I knew I would
|
|
need, and then I just sourced that file after rebooting into single user
|
|
mode. If you made such a source file and put it in your root partition,
|
|
perhaps under /boot, then that file could also include all of the steps
|
|
from 11 through 15c.
|
|
|
|
Also, it is best use the 'mount_nfs' command, instead of 'mount -t nfs'.
|
|
If you use the 'mount' command for NFS mounts, it will turn around and
|
|
directly execute /sbin/mount_nfs, and that is not desirable in this case.
|
|
|
|
Notes on step 17: ./installworld_newk
|
|
|
|
This script will do some setup work, and then ask you if want it to run
|
|
'make installworld'. Most people should just answer "y" (yes) to that
|
|
prompt. You can avoid the prompt by including "-y" or "-n" on the
|
|
command. If you say "n" (no), then it will tell you what commands
|
|
you must type to do the actual installworld.
|
|
|
|
The script also recognizes a "-S" parameter, which causes it to use
|
|
symlinks instead of making copies of programs used by the installation
|
|
process. This option will cause less filespace to be used up in /tmp,
|
|
but it might be slower in some cases (especially for installs using
|
|
an NFS-mounted directory for /usr/obj).
|
|
|
|
Both this script and the installworld_oldk script also recognize a "-M"
|
|
option. This option causes the script to use the absolute minimum PATH
|
|
setting that "should" be needed to complete an install. This option is
|
|
mainly just for debugging the scripts, though. If you request the
|
|
minimum PATH, and some important file was NOT properly copied, then the
|
|
installworld will immediately die at that point. This might be painful.
|
|
Without "-M", the same oversight would mean that you will run the wrong
|
|
*version* of the command, but that older version might actually work
|
|
perfectly fine. I did all my testing with "-M" to make sure I had
|
|
found all important programs, but there is probably no advantage for
|
|
using it for standard system upgrades. Also, if there are no important
|
|
files overlooked, then "-M" will not make any difference at all.
|
|
|
|
# -------+---------+---------+ Upgrading Ports +---------+---------+---------+
|
|
|
|
Similar to the recommendation for the upgrading the system, I suggest that
|
|
you do not 'cvsup' your local copy of the ports collection before trying to
|
|
rebuild everything for 64-bit time_t. For one thing, you will have a cvsup
|
|
compiled for 32-bTT (32-bit time_t's), and that will not work well on a
|
|
system which is using 64-bTT. You might find that you have to 'cvsup' for
|
|
some ports, but you will need to get a 64-bTT version of cvsup before you
|
|
can do that.
|
|
|
|
One tactic to use for upgrading ports is to rebuild your already-installed
|
|
ports one-at-a-time. If you want to do that, and if you use portupgrade
|
|
to upgrade your ports, then I suggest the first thing you should do is:
|
|
|
|
portupgrade -Rr -f ruby portupgrade #- Ports 1.
|
|
Aside: if you get an error about the "ruby-rdoc" port,
|
|
then enter: pkg_deinstall ruby-rdoc
|
|
and repeat the original command.
|
|
portupgrade -Rr -f bash #- Ports 2.
|
|
If you have 'bash' installed, or include any other shells
|
|
which you have installed from the ports collection. If
|
|
your session is *using* one of these shells, then logout
|
|
and log back in after recompiling that shell.
|
|
portupgrade -Rr -f ezm3 cvsup-without-gui #- Ports 3 (maybe).
|
|
If you want to rebuild a 64-bit time_t version of cvsup.
|
|
Note: ezm3 (modula-3) needs a patch to work correctly after
|
|
the change to 64-BTT. That fix has not been commited to the
|
|
port yet [as of Mar 3rd], but it should be commited soon.
|
|
|
|
There are pre-built packages available for ezm3 and cvsup-without-gui on
|
|
the new 64-bTT systems. This ezm3 package *does* include the necessarily
|
|
patch. These files are available on the standard ftp servers for FreeBSD.
|
|
If you have previous versions installed, then remove them with:
|
|
|
|
pkg_delete cvsup\*
|
|
pkg_delete ezm3\*
|
|
If you get warnings about "unable to completely remove" some
|
|
lib/m3 directories when deleting ezm3, then also enter:
|
|
rm -rf /usr/local/lib/m3
|
|
|
|
You can install the new packages with:
|
|
|
|
pkg_add ftp://ftp3.FreeBSD.org/pub/FreeBSD/ports/local-distfiles/gad/ez...
|
|
pkg_add ftp://ftp3.FreeBSD.org/pub/FreeBSD/ports/local-distfiles/gad/cv...
|
|
Replacing "ez..." with "ezm3-64btt-1.1_1.tbz" and "cv..."
|
|
with "cvsup-without-gui-64btt-16.1h.tbz". You can also use
|
|
some other standard ftp server, instead of ftp3.FreeBSD.org.
|
|
|
|
"Now look over all the other ports you have installed, and
|
|
re-compile everything that probably needs to be recompiled".
|
|
|
|
If you are going to do it piecemeal, the next ports to force-recompile
|
|
would probably be languages like perl and python, if you have them
|
|
installed. Or you might want to play it safe at this point, and simply
|
|
recompile *every* port that you have installed.
|
|
|
|
A different tactic to use for ports is to remove *all* ports before you
|
|
do the installkernel/installworld step (while you're still on a 32-bTT
|
|
system). Then, once you're up on the 64-bTT system, start making them
|
|
one-by-one. If you follow this tactic, you might want to save the output
|
|
of a 'pkg_info' command before you start removing ports.
|
|
|
|
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
|
|
If you run into problems when making this change, please report them to
|
|
the mailing list freebsd-sparc64@FreeBSD.org .
|
|
|
|
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
|
|
<Final notes for early-adopters>
|
|
For people who are helping out by testing these instructions, note
|
|
that once you make this change, you must remember to KEEP changing
|
|
__time_t in _types.h after every time you 'cvs update' or cvsup
|
|
your /usr/src tree. If you forget, and end up building a world
|
|
with 32-bit time_t's, you will probably have a very very bad day.
|
|
Once this change is committed for real (which is scheduled for
|
|
March 10th), you will not need to care about this issue as much.
|
|
|
|
Also, a change has been committed to /usr/src/Makefile.inc1 which
|
|
does try to protect you from making this mistake.
|
|
</final notes for early-adopters>
|
|
|
|
# -------+---------+---------+---------+---------+---------+---------+---------+
|
|
# Notice that the following command can be useful in some settings:
|
|
grep '#\- ' UPDATING.64BTT
|