summaryrefslogtreecommitdiffstats
path: root/UPDATING.64BTT
diff options
context:
space:
mode:
Diffstat (limited to 'UPDATING.64BTT')
-rw-r--r--UPDATING.64BTT361
1 files changed, 361 insertions, 0 deletions
diff --git a/UPDATING.64BTT b/UPDATING.64BTT
new file mode 100644
index 0000000..bad7cea
--- /dev/null
+++ b/UPDATING.64BTT
@@ -0,0 +1,361 @@
+# -------+---------+---------+---------+---------+---------+---------+---------+
+
+ 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
OpenPOWER on IntegriCloud