diff options
Diffstat (limited to 'UPDATING.64BTT')
-rw-r--r-- | UPDATING.64BTT | 368 |
1 files changed, 0 insertions, 368 deletions
diff --git a/UPDATING.64BTT b/UPDATING.64BTT deleted file mode 100644 index 6b04d25..0000000 --- a/UPDATING.64BTT +++ /dev/null @@ -1,368 +0,0 @@ -# -------+---------+---------+---------+---------+---------+---------+---------+ - - The FreeBSD/sparc64 port has now changed 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. - -As of March 10th, the official value for time_t on sparc64 has changed to -be 64-bits. If you really must build system with 32-bit time_t's, then it -would be best to stick with a snapshot of current from before March 10th. -With src snapshots after that point, there will soon be ports which assume -you have a 64-bit time_t based on the value of __FreeBSD_version. - -If you are ready to upgrade, then *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, and first build - that as a system with 32-bit time_t. To do this, edit - the file /usr/src/sys/sparc64/include/_types.h - find the line: - typedef __int64_t __time_t; /* time()... */ - and change '__int64_t' to '__int32_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 10th 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. - -Step Pre-4: For people using database-related ports: - Ports like PostgreSQL may change how they store data after - they are recompiled for 64-bTT. So, you may have to do a - full dump of your data while you still have a 32-bTT system, - and do a matching initdb/restore of your data after the port - has been recompiled on the upgraded 64-bTT system. - -After you have built and installed that src-snapshot with 32-bit time_t's, - edit the file /usr/src/sys/sparc64/include/_types.h - find the line: - typedef __int32_t __time_t; /* time()... */ - and change '__int32_t' back 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. - -And then follow these steps to build and install the 64-bit time_t system: - - 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. - sh 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. - sh 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: sh 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: sh 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 10th], but hopefully it will 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-rebuild would -probably be languages like perl and python, if you have them installed. -After that, force-rebuild the ports like autoconf and automake, 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 . - -# -------+---------+---------+---------+---------+---------+---------+---------+ -# Notice that the following command can be useful in some settings: - grep '#\- ' UPDATING.64BTT |