BOOT_ADMIN"> PALO"> via"> Windows"> [user@machine ~/dir]>"> ]> &PA;/Linux Boot HOWTO Thomas Marteau TuxFamily
marteaut@tuxfamily.org
Deb Richardson The Puffin Group
deb@thepuffingroup.com
Thibaut Varène PA/Team
T-Bone@parisc-linux.org
1999 The Puffin Group and Deb Richardson. 2001 2002 2003 Thomas Marteau. 2002 2003 2006 Thibaut Varène. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 as published by the Free Software Foundation; with no Invariant Sections, with no Front-Cover Texts, and with no Back-Cover Texts. A copy of the license can be found at http://www.gnu.org/copyleft/fdl.html. No liability for the contents of this document can be accepted. Use the concepts, examples and information at your own risk. There may be errors and inaccuracies, that could be damaging to your system. Proceed with caution, and although it is highly unlikely that accidents will happen because of following advice or procedures described in this document, the author(s) do not take any responsibility for any damage claimed to be caused by doing so. All copyrights are held by their by their respective owners, unless specifically noted otherwise. Use of a term in this document should not be regarded as affecting the validity of any trademark or service mark. Naming of particular products or brands should not be seen as endorsements. PA-RISC Linux boot This document outlines the procedures to get the &PA;/&L; kernel to boot on your &PA; system. It also explains the usage of &PO;, the kernel loader for &PL;. You will find much information on how to compile a kernel from the source available at http://cvs.parisc-linux.org/. Please note that this HOWTO version is newer than Deb Richardson's and includes more accurate information because of the progress of the port. Nevertheless, it's worth mentioning that this document kept parts of Deb's original work and unveiled some remarkable information. If you are looking for information related to HP hardware but not directly to &PA;, please read Bruno Cornec's HP-HOWTO. Note: by the time this HOWTO was started, &D; was the only &L; distribution available for the &PA; platform, hence the "&D; color" of this document. Some times, &D; specific commands will have to be replaced by their equivalent, if any. 2006-06-06 version 1.5 for DocBook V4.2 1.5 2006-06-06 tv XML cleanup, improved and reorganized contents. Updated to palo 1.5+ and kernel 2.6. 1.1 2003-11-01 tm Added Jeremy Drake's &W; server boot howto. 1.0 2002-10-04 tm & tv Content done by Thibaut. Ready for Woody release. Added Glossary and bibliography. XML conversion. 0.9 2002-01-15 tm This version brings you some useful advices for compiling your own kernel on hppa systems. 0.8 2001-10-17 tm This version takes care of the change of name of the official FTP and CVS sites and modify the license. 0.7 2001-10-13 tm This version adds some updates due to the progress of &PL;. 0.6 draft 2001-09-26 tm This version contains some minor changes and complete the "obtaining bootp/tftpd" section. 0.5 draft 2001-07-03 tm This version is a large update from Deb's work. 0.3 draft 1999-11-24 dlr The initial and published version of this HOWTO. Introduction You just received this &PA; box you bought online, or maybe you got it from your company scrapyard. Anyway, here comes the question of the operating system you are going to use. The &PL; project consists in porting &L; to the &PA; architecture, and we hope that if you decide to use it on your box, this HOWTO will help you in the process of setting it up. In addition to porting the &L; kernel, the development team is working on porting the &D; project to &PA;. In fact, by the time we wrote this document, over 97% of the package pool is available for the stable &D; release (3.1, aka Sarge) on hppa (see the buildd stats for detailed data). Some &D; developers and users reported that the port was one of the easiest to install, as it feels like installing an i386 system. For more information about the &PA;/&L; porting project, please see http://www.parisc-linux.org/, or a mirror site like http://www.fr.parisc-linux.org/. This site deals with kernel development and improvement. For user-space troubles, please refer to &D; hppa port pages. In a few words, this HOWTO is aimed at anyone looking for some help and information about using &L; on a &PA; system. No particular prior knowledge is necessary but bases about how &D; packages work, and general background about &L; can be helpful. It is also worth mentioning that some sections of the present document aren't Linux-centric, and may be of use to people dealing with other OSes, such as BSDs or even HP-UX. After listing supported hardware, this HOWTO explains some commands of the Boot Console Handler (BCH) available at boot time. Then, the features of the &PL; kernel loader are introduced in another chapter, and finally many ways to get your system up and running are detailed. At the end, the text goes deep in the kernel compilation and configuration, as well as a few appendices giving some extra hints. Supported Hardware With the release of &PA; architecture in &D; 3.0 (aka Woody), a major improvement was made in term of quantity and quality of hardware support. Since 0.9.3 releasedBefore the first release of Debian on hppa, there has been unofficial releases of Woody snapshots, entitled "&PL; releases", and numbered 0.x. Immediately after the release of Woody, the &PL; port switched to the normal Debian numbering scheme. In other words, &D; 3.0 is consecutive to &PL; 0.9.3., the kernel has been greatly improved, so that much unsupported hardware by the time 0.9.3 went out is now handled. That's why even if your model is not listed here, you might give it a try and report your result to the mailing list: parisc-linux@lists.parisc-linux.org. Mind that as of this document's publication date, the 2.4 series of the Linux kernel are deemed obsolete, and no development happens in that branch. Any new comer to the PA-RISC/Linux port should look at the 2.6 kernel series, which supports much more hardware than the 2.4 does. Accordingly, one will not try to use the now aging Woody release and will instead focus on its successor: Sarge, or even better, the upcoming Etch. The following &PA; machines should work just fine, provided that you follow the instructions of the present document. Please note that this list can change at any time. The best way to get an up to date version is to look at http://www.pateam.org/list.html. This is the place to find relevant information about a specific model, including special boot procedures. More hardware information can be found on OpenPA. SMP machines should work with SMP kernels (and UP ones, of course) unless stated otherwise. All 712 models. All 715 models including Strider series. All 705, 710, 720, 730, 750 models should work. Some 725, 735 (no HVD SCSI), 755 models are now working. But since there was not a lot of feedback about these machines, we can not be more explicit. The VME-like systems are supported. This includes 742 and 743. A180 and similar. A500, and similar (rp2400 series). BXXX models like B132, B160 and B180. These boxes can be used in the framebuffer mode through the Standard Text Interface. BXXXX models like B1000, B2000 and B2600. These boxes can be used with STI_CONSOLE, but framebuffer only works with VIS-EG cards. FX adapters are not supported. CXXX models like C100, C110, C160, C180, C200, C240, C360. CXXXX models. Indeed, BXXXX and CXXXX are the same kind of machines, based on Astro/Elroy (aka SBA/LBA) chipsets with varying CPU speeds, number of memory/PCI slots. C3000, C3600, C3750 are reported to be working. D class works unless you have a Remote Management Card installed. Even then, it still kind of works, it's just that ttyS0 gets assigned to the second serial port and you have to switch cables around. J class is quite well supported. It has the same split as C class, i.e. JXXX and J2240 are U2/Uturn based and JXXXX are Astro/Elroy. It is the SMP version of CXXXX models. K class is supported if you are using recent ISO images (e.g. Sarge ones). L class and similar (rp5400 series). N class: N4000 (some rp7400 series are reported to work). R class is basically the same as D class. These are not really working yet but work is being carried on. The current 8-way (and bigger) machines using SX1000 chipset and pa8800 (or pa8900) processors do not work yet. Similarly, smaller 2-socket and 4-socket platforms using ZX1 chipset and pa8800 (or pa8900) such as rp3400 series do not work correctly yet either. No plan to get the following hardware completely supported in the near future: E class: E35 and E55 are known to work diskless. The SCSI support is not expected to work anytime soon. F,G,H,I classes: Currently not supported. SuperDome: It boots "single-cell", multi-IOMMU doesn't work. T 5XX and V class: Nobody is working on it at the moment. rp8400: these are cell based and probably don't work yet. The following hardware might never work: T600. Vis-FX graphic adapters. Preparing to boot Like any other system, machines based on &PA; processors have to go through several steps in order to get &L; up and running. The next section introduces you to the early boot management of your &PA; computer. (By the way, to be a bit less awkward, we might from time to time call it a 'PA' box). This chapter will give you some key concepts like &BA;. &BA; First of all, you must learn what is and how to use &BA; on your &PA; box, before thinking about doing anything with it. &BA; is a firmware application, used to manage a &PA; machine at an early boot stage, i.e. when the box has not yet started its Operating System. It is also known as the . Those familiar with the x86 world will probably see it as a kind of BIOS, whilst PPC fans might think of it as an equivalent for Open Firmware. We named it &BA; throughout this document since it is the name it is the most common prompt it will display on most &PA; machines. You will see through this HOWTO that there are many references to it, therefore it's worth saying that minimalistic &BA; skills are mandatory! Entering the &BA; interface Entering the &BA; management tool isn't that awful: Turn your &PA; box on. During the boot process, the following message will appear on the current console (see ): Searching for Potential Boot Devices. To terminate search, press and hold the ESCAPE key. When this message is displayed, press and hold the Esc key until an option menu appears. This can take a while, be patient. On recent machines, pressing any key interrupts the boot process as well. By default, you should enter the &BA; console. Though on some 715s and 725s, an option menu looking like this may be shown: b) Boot from specified device s) Search for bootable devices a) Enter Boot Administration mode x) Exit and continue boot sequence ?) Help Select from menu: Type 'a' and hit Enter to enter Boot Administration mode. This will bring up a 'BOOT_ADMIN>' prompt. Once you have the 'BOOT_ADMIN>' prompt, you can pat yourself on the back: you are in &BA; mode! As it has been said before, the prompt can differ between machines. On recent ones, for instance, it looks like that: Main Menu: Enter command or menu > &BA; commands &BA; is an early boot subsystem (a Boot Console Handler, as said before) where you can execute a limited set of commands. You should find here everything you need to know about them. All HP-PA systems have a BCH. The display can be different but the idea remains the same. That's why the following list is not complete but consistent enough. Another important thing is that for each command, you have a shorter way to invoke it. You can see the shortcut shown as uppercase letters in the command name. Full names will be used in this section. Be cautious when dealing with the BCH, you can harm your system. Some commands may appear in several different menus, and all commands listed here may not necessarily be available on your particular system, this is normal. The <literal>main</literal> commands These commands are the basic ones. boot may be followed by an argument which indicates the path you want to boot. The path should be the definition of a device like for example FWSCSI.6.0 or PRI if you have set this variable correctly. Usually defaults to PRI. path displays or sets the current paths. Invoked with only one argument it will display the current path of the entity passed as argument: path alt will display the current alternative boot path. path pri fwscsi.6.0 will setup the primary boot path as the device attached to Fast and Wide SCSI controller with ID 6 and LUN 0. You can also set and display the paths of console (graphics/serial) and keyboard (ps2/hil/usb). search is a very useful command. It automagically checks all possible boot devices and displays these bootable paths. In several firmware versions, it links them to a shortcut (like P0). It can even search the LAN, if the box is able to boot it. Some firmware revisions allow you to restrain the search path like: search lan or search disk. display redisplays the current menu. help gives you an overview of the available commands and their action. help name will give you details on command name. By default, you can list all main commands by typing help main. main will bring you back to the main menu, whatever menu you might be currently browsing. On almost every systems, there is a reset instruction. It restarts the machine with the latest parameters you have set. The <literal>configuration</literal> commands These commands are available in the configuration menu. So, in order to use them, you must enter this menu by typing configuration at the 'BOOT_ADMIN>' prompt. auto will tell you if the box will automatically start booting when switched on, or will do a search for boot devices, depending on the first argument passed to the command (boot, search, start). You can modify this parameter with the keywords ON and OFF. default sets back the factory defaults. monitor (only in graphic console) sets your display configuration by typing monitor <path> <type> which indicates your console path and type. You can list the available modes by typing monitor list. fastboot displays or sets the boot tests execution. The <literal>information</literal> commands They give you access to global information about your system. Going into this menu is done by asking for information. all should display everything. bootinfo lists all the boot parameters of the system. fwrversion gives your firmware revision. You can check if your firmware is up-to-date at this webpage. lanaddress shows the MAC (Ethernet) address of the system. On some boxes (especially 712s), two different addresses may appear. The one you are looking for is the first. The <literal>service</literal> commands It is a &PA; guru menu. You will find nothing really interesting for an end-user here. We recommend you not to play with it unless you really know what you are doing. pim [<proc>] [HPMC|LPMC|TOC] displays the content of a and Error Log. It is very useful after a to collect debugging information. clearpim clears data. scroll enables or disables the scrolling mode in &BA; on recent boxes. Consoles Throughout your PA-RISC experience, you will be often told about consoles. This section aims at de-obfuscating what this word means and how to use said consoles. In order to boot your &PA; system with the &PL; kernel, you must first set up a console. A console is basically the device where the kernel (and the firmware) will display its output, and where input can by sent to control the system at an early boot stage. You can use either graphic console, which requires having a monitor and a keyboard attached to the system, or serial console, which allows serial line communication between the system and another Linux machine, or any VT system. Please note that the firmware console and kernel console are not necessarily the same. For instance, it is possible to interact with &BA; using keyboard & monitor, and once &PL; is up, to have kernel output sent to serial port only. By default, will try to use firmware console as the kernel one. Workstations usually boot in graphic mode, whereas servers boot in serial mode. Some boxes will also automatically switch to serial if no keyboard is connected, or if you hold down TOC switch while powering the system on. If you don't know what the actual console of your box is, it's quite simple: find the place where first output is sent when the box is turned on (serial line or monitor output, if any), that is the console. If you are trying to setup a &PA; workstation and have a monitor handy, the easiest method is to use graphic console. If you get into troubles, or are trying to configure a server, choose serial console. Using graphic console To use the graphic console, you must first ensure that the Linux kernel supports your system's graphic card. There are two ways to deal with the graphic console. If you think about bug-reporting any trouble, you must know how to differentiate both. First, the STI console is the classical video text console, like VGA on a common PC for example. This name is due to the fact that each &PA; box with graphical capabilities features the which defines some standardized ways to access the video memory. The other graphic console is the well known framebuffer console (which on HP-PA uses STI in a special manner, hence the name STIfb). In this case, when booting, you will see a characteristic little penguin appearing on the top-left corner. This is the easiest way to differentiate the two graphic modes. Obviously, if you can use graphic console, it is the easiest way to proceed. Nevertheless, you must be sure that your hardware is supported. All HP-provided graphics cards can deal with , but not all of them are supported in Linux. This is especially true for Vis-FX cards that can only be used through STIcon. Using serial console The serial console is a good way to obtain all console messages, including the BCH ones. It is very useful for bug reports, as its output can be easily dumped. Moreover, most of the &PA; servers can only be managed with serial console. The only cases where serial console HAS TO be used is either if you don't have a monitor handy, or if the machine doesn't support graphics. It is also possible that the kernel can NOT handle some specific graphics hardware present in the machine, but that is pretty rare (STIcon should work everywhere). Here is the procedure to setup serial console support. Serial Cable To connect a &PA; machine to a PC's RS232 port, you need a 9-pin-to-9-pin female plugs null-modem cable. You should be able to obtain such a cable at your local computer hardware reseller. Obviously, you can also choose to connect the other end of the cable to a terminal (in this case it will probably need a 25-pin male plug). Anyway, the most practical method is to connect it to another box running minicom or cu, which makes all output easily available for further usage (dump report, session log, and so on). Configuring <command>minicom</command> on &L; In order to communicate with a &PA; machine on a serial line, you have to set it up in serial console mode (see below), and configure a serial communication program. We recommend minicom, which can be found in most &L; distributions. If you don't have minicom on your system, you can find the latest package on any major &L; software website. Most of the minicom configuration is machine dependent. However, you must ensure that: The baud rate is set to 9600 Protocol is set to 8-N-1 (8bit data, No parity check, 1 stop bit) Don't worry too much as these are the default values on &PL;. If you are running minicom on a PC, you will probably need to change the baud rate. Switching consoles It might prove useful that you learn how to manage the console mode on a &PA; box. The following section will explain the various operations regarding console modes. Checking current console mode Type: path console to see the current console mode. If it's graphic console mode, it will return something like: 'Console path = graphic_1'. If it's serial console, it will return: 'Console path = rs232_a.9600.8.none' or something similar. On some models, there can be slight differences in the naming, but the idea stays the same. If you want to see more descriptions here, please send us a message describing the box you use and what you get. Changing to serial console mode To change to serial console mode, type the following command at the 'BOOT_ADMIN>' command prompt: path console rs232_a.9600.8.none or, like on B132L+ path console serial_1 Anyway, on most boxes if you try to setup an invalid path for the console, you will be warned and prompted again for a valid path. To verify that the console path has been correctly set, type path console. This should return something like 'Console path = rs232_a.9600.8.none', indicating that the system is now set up to boot using serial console, on RS232 port 'A'. If your machine has only one, this is OK, if not, take care to use the right one. reset will reboot your system with the new parameters. How can I change the boot console to serial on a 712? Unfortunately, it is *normally* not possible. Although 712s are configured for in-house HP development to use serial, this cannot be set in &BA;. You will have to use graphic console on 712s. And why the hell would we use this beautiful 712 with serial console when we can have X on it?! Anyway, if you feel like trying bleeding edge solutions, there is a tip at the &PL; mailing list archive. This explains how to change the console from an HP/UX ISL prompt. Since you actually need HP/UX to be able to do the serial trick, you can find a small HP/UX lifimage here: http://www.pateam.org/archive/uxbootlf. (See further to learn how to netboot a lifimage). In fact, serial console on 712 is especially useful if you want to boot the box without having a keyboard attached to it, which is otherwise not possible. The following takedown is highly unofficial, unsupported and in a general way a bad idea, as you can make your 712 unbootable, needing intervention from a HP-techie, if something goes wrong. Beware! Here is the procedure: Turn the box on and when in &BA;, boot into HP/UX ISL. For example: BOOT_ADMIN> boot lan isl Once you get the 'ISL>' prompt, type the following: Switching to serial: conspath 2/0/4.0x283 Switching to graphic: conspath 1/0/0.0 Still at the 'ISL>' prompt, type disp, and check that console path is either '(hex) 2/0/4.283.0.0.0.0.0' for serial, or '(hex) 1/0/0.0.0.0.0.0.0' for graphic. Power cycle the system to bring it up on the new console. Changing to graphic console mode This is the reverse of the previous operation. By checking your console path, you should see 'Console path = rs232_a.9600.8.none'. Now, you can switch to the graphic mode by issuing the following command at 'BOOT_ADMIN>' prompt: path console graphic_1 The actual switch will happen after a reset. If the monitor does not seem to work properly, try to press the Tab key (on the keyboard attached to the box of course) at the beginning of the boot sequence to change the resolution of the display. By pressing this key, the monitor resolution cycles from one to another. Keep that in mind when changing monitors. &PO;, the &PL; kernel loader What is &PO;? &PO; is a set of two programs, a boot loader, which is loaded by the &PA; firmware into memory and then executed, and a boot media management tool, which prepares and updates bootable media such as hard disk drives. The &PO; boot loader executable is stored in a file called iplboot. 'IPL' is HP jargon for Initial Program Loader (See the glossary). The boot media management tool is called &PO;, which stands for &PL; LOader, just as on x86 the boot media management tool is called LILO. Even though &PO; is much alike LILO (both have a userland application and a boot loader executable), it's worth mentioning that &PO; doesn't usually need to be called every time you build and install a new kernel, as LILO doesFor the knowledge addict: &PO; can actually access and read ext2/ext3 filesystem, and therefore follow symlinks, whereas LILO bootloader will only know the physical disk address to access the kernel. See this for further details.. &PO; is strongly related to &PL; development. Thus, several versions have been released. Some changes in the way make palo operates are explained by the author of &PO;, Paul Bame, in this mail. What does &PO;? The main idea is to boot a kernel, passing it all needed parameters. This is what the boot loader part of &PO; does (see ). Once it has been called by the firmware, it will load the Linux Kernel in memory, passing to it the given arguments, and tell the processor to branch to its entry point. This will begin the execution of the kernel on the &PA; computer. The &PO; management tool can transform the usual vmlinux into a &PA; bootable lifimage, including or not RAMDISK or NFSROOT support. However, it can also make a hard disk drive bootable, specifying the console output and the root device. We are going to see all these points precisely. What must be kept in mind is that vmlinux is the kernel alone, which is not bootable as is. It needs &PO; to be turned into a bootable lifimage for CD or network boot, or to be loaded at boot time from a prepared hard disk drive. Have a look at about these words. Quoting a well known &PL; hacker:
Richard Hirst People often try to put a lifimage in /boot, or a vmlinux on the network boot server.
Which is obviously wrong. &PO; management tool usage Here we will show what can be done with the &PO; boot media management tool. For in-depth information about palo usage, we strongly advise you to take a look at &PO;'s README file, which can be found in palo/ directory on http://cvs.parisc-linux.org/. For the next two steps, you will need a compiler toolchain, see . Making a <filename>lifimage</filename> with <literal>RAMDISK</literal> First things first: when should you walk this way? At an earlier stage of the &PL; project, the lifimage was very useful. In fact, simply putting this file in a boot server tree allows you to boot your PA box using the boot lan instruction without any further involvement (see ). The main advantage of a RAMDISK is that it unpacks its own file system in RAM, and therefore is completely independent of the machine I/O capabilities (hard drives, etc). The main drawback is that you have to build your own RAMDISK if you have memory constraints or need some customized files. Now, let's see how to obtain a lifimage with RAMDISK. We assume you got the latest source of the &PL; kernel tree (to which we'll refer below as the "linux/ directory"), and that you are somewhat familiar with kernel configuration. Check for &PA; specific options. Mainly, you will need a (cross-)compiler, the linux/ directory and the &PO; package installed. If you do not have it, run as root apt-get install palo. Everything can also be found at http://www.parisc-linux.org/. Go through the kernel configuration step. Then, run make palo and if &PO; is installed, the following message should appear at the end of the compilation: A generic palo config file (./palo.conf) has been created for you. You should check it and re-run "make palo". WARNING: the "lifimage" file is now placed in this directory by default! So, edit the palo.conf file: # This a generic Palo configuration file. For more information about how # it works try 'palo -?'. # # Most people using 'make palo' want a bootable file, usable for # network or tape booting for example. --init-tape=lifimage --recoverykernel=vmlinux ########## Pick your ROOT here! ########## # You need at least one 'root='! # # If you want a root ramdisk, use the next 2 lines # (Edit the ramdisk image name!!!!) --ramdisk=ram-disk-image-file --commandline=0/vmlinux HOME=/ root=/dev/ram initrd=0/ramdisk # If you want NFS root, use the following command line (Edit the HOSTNAME!!!) #--commandline=0/vmlinux HOME=/ root=/dev/nfs nfsroot=HOSTNAME ip=bootp # If you have root on a disk partition, use this (Edit the partition name!!!) #--commandline=0/vmlinux HOME=/ root=/dev/sda1 As you can see, the RAMDISK mode is the default. The string ram-disk-image-file should give to &PO; the path of your RAMDISK file. You shouldn't change anything else to this file. After editing palo.conf, you can run make palo again. The result, a lifimage file, is waiting for you in the linux/ directory. Making a <filename>lifimage</filename> with <literal>NFSROOT</literal> This method is widely used because the kernel and the file system are directly accessible on your boot server. It is also very easy to test a new kernel. You just have to generate the kernel and put it in the correct directory. When starting up, the PA box will boot the new kernel by typing boot lan in &BA;. Finally, it is the only way to go for systems which I/O devices are not supported (such as E class, by the writing of this document). Enabling NFSROOT support is easier than RAMDISK. You have to edit palo.conf to specify the boot server IP address instead of the string HOSTNAME. For instance, if your server has 10.10.10.2 as its IP address, then the palo.conf file should contain: # This a generic Palo configuration file. For more information about how # it works try 'palo -?'. # # Most people using 'make palo' want a bootable file, usable for # network or tape booting for example. --init-tape=lifimage --recoverykernel=vmlinux ########## Pick your ROOT here! ########## # You need at least one 'root='! # # If you want a root ramdisk, use the next 2 lines # (Edit the ramdisk image name!!!!) #--ramdisk=ram-disk-image-file #--commandline=0/vmlinux HOME=/ root=/dev/ram initrd=0/ramdisk # If you want NFS root, use the following command line (Edit the HOSTNAME!!!) --commandline=0/vmlinux HOME=/ root=/dev/nfs nfsroot=10.10.10.2 ip=bootp # If you have root on a disk partition, use this (Edit the partition name!!!) #--commandline=0/vmlinux HOME=/ root=/dev/sda1 If you have another IP, this field must be filled in with the correct data. You shouldn't change anything else to this file. After having properly configured palo.conf, you can go into the linux/ directory and issue a make palo. The result, a lifimage file, is as usual waiting for you in the linux/ directory. For advanced details on NFSROOT management, take a look at for the appropriate HOWTOs. Making a bootable partition In this part, &PO; can be seen as a LILO clone. palo is mainly a program that enables a PA box to boot a kernel present on its hard disk drive. This section is going to explain how to make it work. After installing the &PO; package, a copy of the default palo.conf can be found at /usr/share/doc/palo/palo.conf. We will explain here how to customize it to fit your needs. To setup a bootable hard disk, you have to partition it properly (that is, if you want to use it as your primary boot device). This implies that this step can only be achieved either if you have already booted a minimal system on your &PA; box (from CD or network, see ), or if you intend to prepare your hard disk using another computer than the target one (which can be useful to unpack and setup a downloaded file system for a slow box, for example). The point of this HOWTO is not to teach you how to use fdisk and friends, so here are the few things you HAVE TO know: A partition entirely contained within the first 2GB of your target device has to be of partition type 'f0', which is the reserved partition type for &PO; boot loader. There are two ways to use &PO;: the old scheme (available on all versions), in which that partition will only store configuration and recovery kernel; and the new scheme (available since &PO; 1.5), in which that partition will be formatted as ext2 or ext3 and mounted in /boot. In the first case, the partition does not need to be huge. This is were &PO; will save its configuration, recovery kernel(s) - about 5MB each - and optional recovery ramdisk, so 32MB seem far sufficient. Alternatively, in the second case, since you will use it as /boot, you should size it decently. 100MB is a good cut. Beware! The vmlinux file that will be actually booted has also to be located within the first 2GB of the hard disk. We strongly recommend to either (in the old scheme) create a separate /boot partition at the beginning of the disk (unless you plan to boot recovery kernels every time), or use the new scheme and mount the &PO; partition as /boot, because if ever your vmlinux binary gets physically stored past the first 2GB of the disk (like when filling up '/' with data), the box won't boot anymore. In fact, this third usage of &PO; is the most common as the default /etc/palo.conf makes it easy to configure. The old scheme: hidden partition The hidden partition is deprecated. Don't use this for a new setup. Use the new scheme instead (see ). The hidden partition method is documented for the sake of posterity. Here is the output of fdisk which represents the hard drive of a box with 16MB &PO; space, 128MB swap space and about 1GB '/' partition: bash# fdisk -l /dev/sda Disk /dev/sda: 133 heads, 62 sectors, 1017 cylinders Units = cylinders of 8246 * 512 bytes Device Boot Start End Blocks Id System /dev/sda1 * 1 4 16461 f0 Linux/&PA; boot /dev/sda2 5 34 123690 82 Linux swap /dev/sda3 35 277 1001889 83 Linux Now let's deal with &PO; configuration. Here are the various parameters you can change: recoverykernel is (as the name suggests) the path to the kernel you want to boot within a failsafe session, it will be stored in the 'f0-type' partition. bootloader is the path to the iplboot boot loader utility which is created by &PO; when you issue a make iplboot. Usually you don't want to change or even specify this. init-partitioned is used to indicate the pre-partionned device where palo will write its boot parameters. The effect is immediate. It means that &PO; is going to write on (and erase the content of) the 'f0' partition of this device, which has to exist. commandline: the first digit is the number of your ext2/ext3 partition where the kernel file is located, as reported by fdisk. Logically, the next string is the absolute path to the kernel from the root of THIS partition Example: /boot is mounted from a separate partition, which number is, say, 4 according to fdisk. From a &L; point of view, the absolute path of the file is /boot/vmlinux, but from a partition point of view, it is /vmlinux. Therefore, the commandline will start with "4/vmlinux". We hope that's clear enough!. The following space separated parameters (do NOT use any quotation mark!) will be passed to the kernel as its arguments. e.g.: HOME= and TERM= are environmental parameters passed to init when booting. They are not compulsory but they can be useful. root= tells the kernel which partition it must mount as the root file system while booting. It can be tricky when you have more than one disk, and is a mandatory argument. Maximum length for the commandline is 127 characters. You can also add console=, to force the designation of the output console. You should remember that console=ttyS0 is for a serial console and console=tty0 is for a STI (graphic) console. Support for the MUX console (if enabled in the kernel, see ) has been added, using console=ttyB0. Recent versions of &PO; auto detect the right console path (except for MUX), and can figure out whether a 32bit or 64bit kernel should be used. If not, please mail to the mailing list. Last but not least, if you are using &D; 2.6 kernels, you will also need to add initrd=X/path/to/initrd, following the same rules as for the kernel path, see above. According the above fdisk example, we want to use /dev/sda3 as our root partition. Thus, the configuration file should look like that: # The following arguments are set up for booting from /dev/sda, specifically # mounting partition 3 as root, and using /boot/vmlinux as both the # recovery kernel, and the default dynamically-booted kernel. --recoverykernel=/boot/vmlinux --init-partitioned=/dev/sda --commandline=3/boot/vmlinux root=/dev/sda3 The new scheme: mounted partition Following is a practical example using the new way of doing things, by using a formatted &PO; partition. That scheme should be the preferred one for new installations. Looking at the previous example (), very little things need to be changed. Essentially, if you had the need for a separate /boot partition, it will be gone in the scheme detailed below. The configuration for &PO; will be a little bit different as well, but that's about it. Here is the output of fdisk which represents the hard drive of a box with 100MB &PO; space (which will be mounted as /boot), 128MB swap space and about 1GB '/' partition (bear in mind that the f0 partition must still be wholy contained within the first 2GB of the disk): bash# fdisk -l /dev/sda Disk /dev/sda: 133 heads, 62 sectors, 1017 cylinders Units = cylinders of 8246 * 512 bytes Device Boot Start End Blocks Id System /dev/sda1 * 1 26 100461 f0 Linux/&PA; boot /dev/sda2 27 36 123690 82 Linux swap /dev/sda3 37 277 917889 83 Linux Now let's deal with &PO; configuration. Contrary to the previous example, there are far less options to put in the configuration file, and the setup is a single step operation: the initialization step, which needs only to be done once. To initialize for the first time the &PO; partition as something the Linux system can understand (ext2 or ext3 filesystem), you need to run the following command (as root): &prompt; palo --format-as=N --init-partitioned=target_disk Where N is either 2 for ext2 or 3 for ext3, and target_disk is the device which contains the &PO; partition, /dev/sda in the current example. Continuing with this example, assuming we want an ext3 partition we would use: &prompt; palo --format-as=3 --init-partitioned=/dev/sda Do NOT use mkfs or mke2fs to generate the filesystem on this partition. &PO; marks blocks as used where the boot loader portion of &PO; is stored on disk. These tools don't know about &PO; boot loader! This needs to be run only once for it will erase any existing data on the f0 partition. Finally, we need to tell &PO; about the partition, how we want it to be used, and how we don't want it to be erased everytime &PO; is being run. Thus, keeping in sync with our current example, the configuration file should look like that: # The following arguments are set up for booting from /dev/sda, specifically # mounting partition 3 as root and booting the vmlinux file in /dev/sda1, palo # partition formatted as ext3. --update-partitioned=/dev/sda --format-as=3 --commandline=1/vmlinux root=/dev/sda3 As one can see, since we will be storing our kernels directly on the f0 partition, we tell &PO; to load them from it (hence the 1 in the commandline argument. In the previous section () we dealt with the init-partitioned parameter. Here, we use update-partitioned instead, which, contrary to the former, tells &PO; to not erase the content of the partition when runThe format-as switch is a bit misleading. When used with init-partitioned it is meant to tell which filesystem to format the new partition, but with update-partitioned, it is meant to tell &PO; which filesystem is used on the already formatted partition.. How to use &PO; at early boot stage? The theory You have setup everything, rebooted your box, and suddenly you want to change something to the kernel boot arguments, or even boot another kernel. Damn it! How could you, now that the box is booting? Well, stay calm and relax, have a deep breath, we have the solution! Unfortunately, by the time you'll learn about it, your box will have finished booting ;o) First, you must learn how to interact with &PO; during the startup sequence. You have to enter &BA;, as explained in . For some old models (up to 712 or so), you must add the ipl (or isl) string to your boot command in the &BA; console: BOOT_ADMIN> boot pri ipl On most &PA; boxes, the system will ask you if you want to interact with IPL anyway. You just have to answer "y" and hit Enter. You will then end up to &PO; configuration display, with the list of all parameters and their corresponding numbers. You just have to enter the number corresponding to the parameter you want to change. Hit Enter, modify it and validate the changes by hitting Enter again. The system will redisplay the new list. This modification is not permanent To save your changes, you will have to run /sbin/palo when your system will be up and running, and it will write on the disk all the parameters contained in the configuration file, (/etc/palo.conf), which you will have properly modified if needed.! If you want to add a supplementary parameter, select any one and write yours on the editing line, beginning with a space: <#> edit the numbered field 'b' boot with this command line 'r' restore command line 'l' list dir ? 0 3/boot/vmlinux initrd=3/initrd.img After validation, the list will count one more parameter. If you want to delete one, select it and erase the complete entry. You will see that the list counts one less parameter. For more informations about &PO;, please take look at the &PO; README. You can find a copy of this file after having installed the palo package in /usr/share/doc/palo/README.html. This HOWTO section is mostly inspired from the above file, written by Paul Bame. A complete example This example has been suggested by Michael Damaschke. We will use notions explained in and , and refer to concept such as console, seen in . So, let's go for the story of the happy &PL; user booting a kernel, also called "I don't know how to configure my workstation to use the kernel I want during boot sequence!". After switching your workstation on, a message on the console will tell you that the workstation is about to start automatically the boot sequence, except if you hold the Esc key to stop the auto-booting process. This is a very difficult step: you must hold the Esc key down ;o) Depending on your model, you might need to press this key during a quite long time. In some cases when using graphic console, the monitor can be too slow to trigger on, and won't allow you to see the warning message. A good workaround is to keep a close eye on the keyboard's lights: when they all blink at once, this is the right time to press and hold the Esc key. If you still have troubles, please refer to . There are a few different ways to get access to &BA; (see ). If you have an old box, you will see an information message displayed, where the workstation's firmware tells you that it will start searching for all bootable devices, or that you can break this by holding down the Esc key. This is the same procedure as just mentioned, you must press the Esc key. As usual, on some machines you might then get a menu where you should press the a key followed by Enter. You are now facing the deadly 'BOOT_ADMIN>' prompt :^). First, we will turn off auto boot process by entering the following lines: BOOT_ADMIN> auto boot off then hit Enter to validate. This will prevent the box from further attempts at auto-booting. In other words, you won't have to stop the boot process with Esc, it will stop on its own on subsequent reboots and wait for your instructions. Now, you must tell the system from which boot device you would like to boot. If it's a hard drive, it must have a 'f0' partition at the beginning (see ). In this example, the old kernel is vmlinux and the new one is vmlinux-2.4.17-pa3. The chosen SCSI boot device is designed by: SCSI.X.0, where X is the SCSI-ID of the disk you want to boot fromFor those who wonder what the final 0 means, it's the device LUN. Since most SCSI devices have only one LUN (especially disks), you can safely use 0 as in this example.. e.g.: BOOT_ADMIN> boot SCSI.5.0 At the end of the previous command line, you must add the IPL token if you have a HP 9000/7xx system to specify that you want to interact with IPL. If you have a more recent hardware, the system will ask if you want to interact with IPL anyway: Interact with IPL (Y or N)?> Say Y and hit Enter. Now, you can manually configure the &PO; boot parameters. A new menu is displayed, where you can configure on line '0' (selected by default) the boot partition number, and the path of your boot kernel. Here is the complete session log of a A500 serial console output, taken from &PO; version 1.5. You can find in a session log with an older version of palo, such as the one that can be found on &D; 3.0 install disks. Main Menu: Enter command or menu > bo scsi.5.0 Interact with IPL (Y, N, or Cancel)?> y Booting... Boot IO Dependent Code (IODC) revision 1 HARD Booted. palo ipl 1.5 root@c3k Fri May 14 16:17:38 MDT 2004 Skipping extended partition 6 - beyond reach of IPL Partition Start(MB) End(MB) Id Type 1 1 31 f0 Palo 2 32 153 83 ext2 3 154 1107 82 swap 5 1108 5875 83 ext2 PALO(F0) partition contains: 0/vmlinux64 5279419 bytes @ 0x44000 Information: No console specified on kernel command line. This is normal. PALO will choose the console currently used by firmware (serial). Current command line: 2/vmlinux root=/dev/sdb5 HOME=/ console=ttyS0 TERM=vt102 0: 2/vmlinux 1: root=/dev/sdb5 2: HOME=/ 3: console=ttyS0 4: TERM=vt102 <#> edit the numbered field 'b' boot with this command line 'r' restore command line 'l' list dir ? 0 2/vmlinux-2.6-cvs initrd=2/initrd.img-cvs Current command line: 2/vmlinux-2.6-cvs initrd=2/initrd.img-cvs root=/dev/sdb5 HOME=/ console=ttyS0 TERM=vt102 0: 2/vmlinux-2.6-cvs 1: initrd=2/initrd.img-cvs 2: root=/dev/sdb5 3: HOME=/ 4: console=ttyS0 5: TERM=vt102 <#> edit the numbered field 'b' boot with this command line 'r' restore command line 'l' list dir ? 1 Current command line: 2/vmlinux-2.6-cvs root=/dev/sdb5 HOME=/ console=ttyS0 TERM=vt102 0: 2/vmlinux-2.6-cvs 1: root=/dev/sdb5 2: HOME=/ 3: console=ttyS0 4: TERM=vt102 <#> edit the numbered field 'b' boot with this command line 'r' restore command line 'l' list dir ? b &PO; was first setup to boot the kernel file vmlinux located on the second partition of the SCSI device ID 5 LUN 0. (We know this since we have asked &BA; to boot on this device). But we wanted another kernel this time. We have pressed the Enter key (to validate the default choice '0') and modified the text to match our needs, here vmlinux-2.6-cvs. We have also added an initrd=2/initrd.img-cvs argument to the command line. We have validated our changes by hitting the Enter key. Finally we have decided that we didn't want this additional argument, so we have selected it and erased it. At the end it asked again which field we wanted to edit, we just typed 'b' instead of any number and hit Enter to boot our new kernel. Please don't change any other parameter unless you really know what you are doing! That's it! &PO; has no more secrets for you :-) As you might have noticed, the &BA; interface can take several aspects, so don't be disappointed if yours does not exactly match our examples. Available boot solutions Booting from CD Booting from CD is one of the easiest way to start and install your &PA; machine; assuming you have a CD drive handy and a bootable CD. You can download official &D; ISOs as well as Net Install ISO (see ) from the &D; Installer pages, or from the &PA;/Linux official website. Start the box and enter the &BA; mode. () Place your bootable CD on the CD tray and close it. Sounds obvious, but we know guys who missed that step :) There are two options from there: either you know the full PATH to your CD device, then you can jump to next step, or you don't. In this last case, issue a search ipl to list all available bootable devices. You can also specify search [PATH], which is fastest. For instance if you want to search the SCSI bus: search SCSI On recent boxes, search disk is quite helpful. Take a look at help search for details specific to your box. Once you know the full PATH to your CD drive, you can issue a boot <PATH>. That's all. If everything goes fine, it will start booting the CD present in the CD reader. Real life example: boot ide Booting from hard drive Booting from hard drive is not really more difficult that booting from CD. The only thing really important is that your hard drive has to be correctly prepared. Take a look at to learn how to prepare it. Start the box and enter the &BA; mode. () There are two options from there: either you know the full PATH to your hard disk device, then you can jump to next step, or you don't. In this last case, issue a search ipl to list all available bootable devices. You can also specify search [PATH]. For instance if you want to search the Single Ended SCSI bus: search SESCSI Take a look at help search for details specific to your box. Once you know the full PATH to your hard drive, you can issue a boot <PATH>. That's all. If everything goes fine, it will start booting the kernel as setup by &PO; (see ). Real life example: boot scsi.6 Booting from network Booting from the network is only required in certain cases. Booting from the network is very usefull when you have unsupported I/O devices, diskless systems, or systems with broken hardware. Network booting is detailed below. Preparing to boot from network Booting from the network involves two machines: the boot server and the boot client, the latter being the &PA; system you are trying to start up, and the former, the machine that will serve over the network the files which the client needs. The rest of this section will extensively deal with setting up the boot server since this is probably the trickiest part. You will need a lifimage to perform a network boot. See to learn how to create one. You can also use the one from &D; Installer. Needless to say, all server-side setup is meant to be performed by the super-user, also known as root. rboot or bootp? All 'recent' machines can boot using BOOTP, starting from 715/100, 715/120, and 712s. Older ones, mostly early 715s, 710s and 725s need RBOOT. To use BOOTP you have to enable the IP: Kernel level autoconfiguration IP: BOOTP support within the 'Networking options' section of the kernel configuration, if you want to use a home-made kernel. See for details. Please note that though deals with RBOOT only, two different implementations of the BOOTP protocol are detailed in and . We detail these two because we can, but if you need to use the BOOTP protocol, you will have to choose one. If you don't know which BOOTP implementation to use, go for the dhcp one, it is much easier to deal with. Using rboot Obtaining rbootd If you have an old machine that requires rboot to boot over network, use the following procedure to set up and configure a boot server, and boot using the &PA;/Linux kernel. Old machines, including the Scorpio 715s, use the RBOOT protocol. You need rbootd to handle their boot requests. Look for it in your favorite distribution archive (assuming you will be servicing boot requests from a Linux box). Here are two ways of getting the rboot daemon: If you are using a &D;-powered server (which you really should be doing ;o), you're almost done. Run from a command shell: &prompt; apt-get install rbootd If you can't find any rbootd package for your system (which is very possible since it is a very old netboot protocol), you can find its source in the &D; archive: rbootd. You will have to build it from source. Configuring rbootd As we already said, to boot a RBOOT-aware system, you need a separate machine with rbootd installed (this is the 'boot server') on which you will store the &PA;/Linux kernel lifimage that you want to use to boot your &PA; system with. Once the rbootd server software is installed, read the following to configure it to work with your &PA; system: In /etc/rbootd.conf you will have to add a line like: ethernet_addr bootfile Replace bootfile with the name of your &PA;/Linux kernel image, usually 'lifimage'. Now get the Ethernet address of your &PA; system by typing lanaddress at the 'BOOT_ADMIN>' prompt (see ). It will return a number like 080009-7004b6. Take note of this number. In /etc/rbootd.conf on your boot server, the Ethernet address has to be colon-delimited. That means you will have to modify the number you just obtained so that every set of two characters (after removing the '-') is separated by a colon. For example: 080009-7004b6 becomes 08:00:09:70:04:b6. Add the colon delimited Ethernet address to /etc/rbootd.conf on your boot server. The resulting file will look something like this: # ethernet addr boot file comments 08:00:09:87:e4:8f lifimage_715 # PA/Linux kernel for 715/33 08:00:09:70:04:b6 lifimage_720 # PA/Linux kernel for 720 This rbootd.conf example contains the Ethernet addresses and boot file names for two different machines. Once you have changed the configuration file, restart rbootd. By default, rbootd assumes that bootfiles are located in /var/lib/rbootd/. Therefore, you will have to put your bootable kernel image in that directory, or, if you really hate that directory for some reason, you can rebuild rbootd to use a different directory. The easiest thing, of course, is just to drop your kernel images in the default directory! Using dhcp/tftp We will see here how to setup a DHCP server to handle BOOTP requests (since &PA; box use BOOTP, unless they need RBOOT, as mentioned above). &W; users might want to look at . Obtaining dhcp/tftp &D; users will just have to install the packages using the following commands, executed as root: &prompt; apt-get install dhcp tftpd If you need rpm packages (for the ISC dhcp server), the best way is to go to http://rpmfind.net/. The dhcp package can do much more than a simple bootp daemon. Nevertheless, it is also known to be much easier to configure. If you really want to try regular bootp, skip this and go to . Configuring dhcp/tftp Here are the instructions to set up dhcp on your boot server. To keep this explanation simple, we will assume that you want to assign a fixed IP to your box, without DNS update. Your subnet will be 192.168.1.0/24, with optional: gateway at 192.168.1.1, domain name foo.com and DNS at 192.168.1.4. Feel free to replace these values with those which would suit your needs in the next sections. This section is dedicated to &D; users. For others distributions, it should be similar though there may be some differences like default directories. Edit /etc/inetd.conf on your boot server to add the following line, if it doesn't already exist: tftp dgram udp wait nobody /usr/sbin/tcpd \ /usr/sbin/in.tftpd /tftpboot Here, /tftpboot/ is being used as tftpd server's root (this is where you will put the lifimage file). You can choose another directory if you want. According to man tftpd, this is the usual default directory. When this is done, reload inetd with: /etc/init.d/inetd reload. Non-&D; users can also issue a killall -HUP inetd. According to man 5 dhcpd.conf, edit the /etc/dhcpd.conf file to contain something like: allow bootp; default-lease-time 600; max-lease-time 7200; # This will tell the box its hostname while booting: use-host-decl-names on; subnet 192.168.1.0 netmask 255.255.255.0 { option routers 192.168.1.1; option domain-name "foo.com"; option domain-name-server 192.168.1.4; } host [hostname] { hardware ethernet [mac address]; fixed-address [ip address]; filename "[boot filename]"; option root-path "[root path]"; } You have to fill in the [hostname], [mac address], [ip address], [boot filename] and [root path] fields with the appropriate information, where: [hostname] is the name of the &PA; system. [mac address] is the colon-delimited ethernet address of the &PA; system, which can be obtained by typing lanaddress at the 'BOOT_ADMIN>' prompt (see ). [ip address] is the IP address you wish to assign to the &PA; system. [boot filename] is the name of the bootable kernel image you want to boot your system with. [root path] is the path to the NFS root filesystem exported by the server. Additionally, if the tftp server you want to use is not the same as the one running the dhcp server, you can add next-server [ip address];, replacing [ip address] with the actual IP of the tftp server, to the dhcp configuration. You'll end up with something like this for each box you want to netboot: host tatooine { hardware ethernet 00:40:05:18:0c:dd; fixed-address 192.168.1.22; filename "lifimage-tatooine"; option root-path "/exports/tatooineroot"; } Using bootp/tftp Obtaining bootp/tftp For &D; users, you just have to install the packages by typing these commands as user root: &prompt; apt-get install bootp tftpd If you need rpm packages, the best way is to go to http://rpmfind.net/. You'll have been warned! This daemon is far more obfuscated in its configuration. Configuring bootp/tftp Follow these instructions to use the bootp daemon on your boot server: This section is dedicated to &D; users. For others distributions, it should be similar though there may be some differences like default directories. Edit /etc/inetd.conf on your boot server to add the following lines, if they don't already exist: tftp dgram udp wait nobody /usr/sbin/tcpd \ /usr/sbin/in.tftpd /tftpboot bootps dgram udp wait root /usr/sbin/bootpd \ bootpd -i -t 120 Here, /tftpboot/ is being used as tftpd server's root (this is where you will put the lifimage file). You can choose another directory if you want. According to man tftpd, this is the usual default directory. When this is done, reload inetd with: /etc/init.d/inetd reload. Non-&D; users can also issue a killall -HUP inetd. According to man 5 bootptab, edit the /etc/bootptab file to contain: [hostname]:hd=/tftpboot:\ :rp=[root path]:\ :ht=ethernet:\ :ha=[mac address]:\ :ip=[ip address]:\ :bf=[boot filename]:\ :sm=255.255.255.0:\ :to=7200: You have to fill in the [hostname], [mac address], [ip address] and [root path] fields with the appropriate information, where: [hostname] is the name of the &PA; system. [mac address] is the NOT-delimited ethernet address of the &PA; system, which can be obtained by typing lanaddress at the 'BOOT_ADMIN>' prompt (see ). [ip address] is the IP address you wish to assign to the &PA; system. [boot filename] is the name of the bootable kernel image you want to boot your system with. [root path] is the path to the NFS root filesystem exported by the server. You'll end up with something like this: vodka:hd=/tftpboot:\ :rp=/usr/src/parisc/:\ :ht=ethernet:\ :ha=080069088717:\ :ip=140.244.9.208:\ :bf=lifimage:\ :sm=255.255.255.0:\ :to=7200: Effectively booting from network To conclude with the developers' way to boot the kernel, this section will tell you how to actually boot your system from a network server. But it tends to be less and less used. Most users will prefer to stick to once their system is properly setup. Here we are. We assume that you've done everything outlined above, your network boot server is on the same physical subnet as your &PA; machine, you've got a bootable &PL; kernel lifimage on your boot server, and you're willing to give it a try. If everything is ready (including you!), the following procedure will introduce you to the joy of network booting your PA box into &L;. Fire up your &PA; system. Watch your &PA; box starting up. When the following message appears during the &PA; machine's boot process, press and hold the Esc key: Searching for Potential Boot Devices. To terminate search, press and hold the ESCAPE key. If needed, select 'a) Enter Boot Administration mode' from the menu. This brings up the 'BOOT_ADMIN>' prompt. Type the following at the prompt: boot lan. Watch your &PA; system magically becoming a &PL; system. Ta dah! Of course your are supposed to run only one boot server at a time on your network, in order to avoid conflicts... Building and installing a custom kernel To build a Linux kernel, you need a compiler and the kernel source. The first element is not a trivial thing to find because it depends on how you want to build your kernel. The second is easier since it can be found at the official CVS site. First, we will discuss about GCC compiler. Then, the preparation of the build will be explained. The last paragraph will deal with the installation of this new kernel. We will deal only with a kernel built without modules, to simplify the explanations. GCC compiler You can build the kernel directly on your own &PA; box (self-hosted or native build). But on old systems, you may prefer to use another - faster - non &PA; computer to compile your kernel (cross-compilation). We will see the two possibilities. By the time version 1.0 of this howto was released, only gcc-3.0.X was able to build working kernels. There was a bug in more recent versions that made the box crash when network activity occurs. It should be fixed by now, so using the latest version of gcc should be fine. If ever the above mentioned bug occurs, you'll know what's wrong. Anyway, if you want to build any kernel after 2.6.12-rc3, you will need at least gcc-3.3. Native build Since &D; was the first distribution to support &PA; architecture, if you want to use the Super Cow powers, you need to have some basic knowledge about the &D; packaging system. We will explain here how to quickly get a gcc compiler ready on your &PA; box. If you are not using &D;, well, we're afraid we can't do much for you: you will have to transpose what is said below to your distribution. We will assume you know how to use If you are using your own &PA; box, you only need the good old GCC compiler. You can install the required tools to build a kernel by issuing: &prompt; apt-get install build-essential libncurses5-dev Essentially, this will install everything you need to build a kernel (and even a bit more). This boils down to binutils, gcc, libc-dev, make, fileutils and libncurses5-dev. When this is done, you can proceed to the kernel settings. Cross compiled build In this kernel build method, everything depends on the architecture of your building machine. If you want to compile your own toolchain, there is a slightly out-of-date HOWTO (). Otherwise, we assume you can either find a cross-compiler package for your build host, or make one by yourself. As there is not yet a 64bit userspace on HP-PA, you have to cross-compile 64bit kernel even if you are building on a 64bit &PA; box. You can get unofficial debs for hppa64 compilers and binutils by running for instance: &prompt; apt-get install gcc-3.3-hppa64 binutils-hppa64 See the &PA; Linux Website for details. Kernel configuration If you want to take advantage of the latest kernel improvements, we suggest you retrieve it from the official &PA;/Linux CVS. Please mind that the vanilla kernel that can be found at http://www.kernel.org/ is generally out of sync with the above mentioned CVS kernel, and that snapshots of this kernel are available too, check the download area. In the following, we will focus on a fresh CVS tree. The best way to obtain appreciable performances is to get a well configured kernel. For the &PA; platform, make oldconfig is a kind of default setup. If you want to make your own kernel, the first step is to know what hardware you have. The best way to grab useful info is to look at your box and find a maximum of data (model name, partnumber, chipsets, and so on). If you have already booted your box, you can take a look at dmesg output. Then, go to the official hardware database or to the HP partsurfer website. Once you know what is inside your box and what you want to do with it, just run make menuconfig or another config command. Configuring 2.4 kernels Here is a brief list of architecture dependent menus for 2.4 kernels. You should take a look at them, to see if the values set match your hardware. Mind that 2.4 kernels are now considered deprecated anyway: you will not get community support for them. Remember that make oldconfig is a good base to start with, since it works for almost any machine. Processor type - indicates your CPU model General options - tells you what is going to be enabled in your kernel (U2/Uturn, USC/GSC/HSC, Lasi, Wax, Dino, LBA/Elroy, SuperIO) Parallel port support - enables/disables the Lasi/ASP parport SCSI support - check there for your SCSI chipset (Lasi, Zalon, NCR/SYM53C8XX or other) Network device support - is used to set your network card (Lasi, Tulip...) Character devices - defines your I/O capabilities (Lasi, Dino, MUX see ) HIL Support - useful if you have a HIL controller. See below . Console drivers - is directly related to your console mode (STI console or STI framebuffer) Sound - enables/disables the Harmony driver As you can see, menus specifically concerned by &PA; hardware are not that numerous, but there are lots of dependencies between them. Now, you must configure the kernel accordingly to what you plan to use this box for. Here is a list of some menus you should be going through to configure additional functionalities you might want: General setup - is responsible for binary formats handled by the kernel. You need ELF, and can try SOM (support for HP/UX binaries. It *might* work with some static executables). Block devices - sets the ramdisk and loopback support. You probably won't use them. ATA/IDE/MFM/RLL support - You will need to check this to enable IDE. See . File Systems/Network File Systems - is where to set EXT3 or NFS support. USB support - If you have enabled SuperIO and want USB, look here: . By the time this HOWTO was written, there was no floppy drive support; and what's more, it is not expected to ever be supported. When you're done with it, save your kernel configuration. Everything is written in the .config file. You should back it up because make distclean will remove it. At this stage, you can do make dep vmlinux and if everything goes fine, you will have a new kernel in a couple of minutes. Here follows brief information about specific hardware configurations. <acronym>HIL</acronym> Support in 2.4 Since kernel-2.4.18-pa45, there is a full HIL support, for mice, tablets and keyboards. It is based on the Linux Input Driver model. See the &PA;/Linux FAQ and the mail posted on the mailing list by Helge Deller. Here is how to configure it: Make sure you have a 2.4.18-pa45 or higher kernel source. Look at your kernel configuration for the following options: CONFIG_INPUT=y CONFIG_INPUT_KEYBDEV=y CONFIG_INPUT_MOUSEDEV=y CONFIG_INPUT_MOUSEDEV_SCREEN_X=1024 CONFIG_INPUT_MOUSEDEV_SCREEN_Y=768 CONFIG_INPUT_EVDEV=y CONFIG_INPUT_SERIO=y CONFIG_HIL=y CONFIG_HP_SDC=y CONFIG_HIL_MLC=y CONFIG_HP_SDC_MLC=y CONFIG_HIL_KBD=y CONFIG_HIL_PTR=y There is no more CONFIG_HIL_KBD_BASIC. On your target system, check that the following devices are available: /dev/input/mice /dev/input/mouseX /dev/input/eventX If they are not yet present, create them as root by running: &prompt; cd /dev; MAKEDEV input Configure gpm with the following options in /etc/gpm.conf: device=/dev/input/mice type=imps2 Here is a sample /etc/X11/XF86Config-4: Section "InputDevice" Identifier "HIL Keyboard" Driver "keyboard" Option "CoreKeyboard" EndSection Section "InputDevice" Identifier "HIL Mouse" Driver "mouse" Option "CorePointer" Option "Device" "/dev/input/mice" Option "Protocol" "ImPS/2" Option "ZAxisMapping" "4 5" EndSection Section "ServerLayout" Identifier "Default Layout" Screen "Default Screen" InputDevice "HIL Keyboard" InputDevice "HIL Mouse" EndSection You can also download a sample XF86Config-4 here: ftp://ftp.parisc-linux.org/XFree86/XF86Config-4, adjust color depth and resolution, and put it in your /etc/X11/. <acronym>USB</acronym> Support in 2.4 USB support on HP-PA is still experimental, therefore it is only configured as modules in default kernel configuration. We have tried to install a B2000 with builtin USB support, both 32 and 64bit, and it worked fine, despite some keyboard problems. Don't worry, nothing critical: the range of keys located between the main part of the keyboard (the letters, backspace, enter...) and the numeric pad are broken. They do not behave at all as expected. You can use the numeric pad as arrow keys: when NumLock is not activated, it behaves as a navigation pad. e.g. 8 is Up Arrow, 4 is Left Arrow and so on. Make sure you have a 2.4.18 or higher kernel source. Look at your kernel configuration for the following options: CONFIG_SUPERIO=y CONFIG_HOTPLUG=y CONFIG_INPUT=y CONFIG_INPUT_KEYBDEV=y CONFIG_INPUT_MOUSEDEV=y CONFIG_INPUT_MOUSEDEV_SCREEN_X=1024 CONFIG_INPUT_MOUSEDEV_SCREEN_Y=768 CONFIG_INPUT_EVDEV=y CONFIG_USB=y CONFIG_USB_DEVICEFS=y CONFIG_USB_OHCI=y CONFIG_HID=y On your target system, check that the following devices are available: /dev/input/mice /dev/input/mouseX /dev/input/eventX If they are not yet present, create them as root by running: &prompt; cd /dev; MAKEDEV input Configure gpm with the following options in /etc/gpm.conf: device=/dev/input/mice type=imps2 The XF86-Config-4 is similar to the HIL one, as it is also using the Linux Input Driver. <acronym>MUX</acronym> Console Support in 2.4 MUX Console has been improved by Richard Hirst in 2.4.18-pa37 kernel, though it is still a very experimental feature. It is expected to provide adequate MUX Console support to E- and K-Class machines. Feedback would be really appreciated. Now follow these steps to get it to work: Make sure you have a 2.4.18-pa37 or higher kernel source. Look at your kernel configuration for the following options: CONFIG_SERIAL_CONSOLE=y CONFIG_SERIAL_GSC=y CONFIG_SERIAL_NONSTANDARD=y CONFIG_SERIAL_MUX=y On your target system, check that the following devices are available: /dev/ttyB0 If they are not yet present, create them as root by running: &prompt; cd /dev; MAKEDEV ttyB0 It needs a recent MAKEDEV package to be created this way. Now you can boot your system, taking care that &PO; uses console=ttyB0. <acronym>IDE</acronym> Devices Support in 2.4 There is nothing really special about IDE support. You have to check that the IDE Chipset in use in your box is supported by the kernel. A common chipset found on &PA; hardware is NS87415. You can find it on B2000, J5000 and C3000 for instance. You will need IDE support to use some CD-ROM devices. Here is an example to get IDE to work with this chipset: Make sure you have a recent kernel source. Look at your kernel configuration for the following options: CONFIG_IOMMU_CCIO=y CONFIG_PCI=y CONFIG_PCI_LBA=y CONFIG_IOSAPIC=y CONFIG_IOMMU_SBA=y CONFIG_SUPERIO=y CONFIG_IDE=y CONFIG_BLK_DEV_IDE=y CONFIG_BLK_DEV_IDEPCI=y CONFIG_BLK_DEV_IDEDMA=y CONFIG_BLK_DEV_ADMA=y CONFIG_BLK_DEV_IDEDMA=y CONFIG_BLK_DEV_NS87415=y On your target system, check that the following devices are available: /dev/hd* If they are not yet present, create them as root by running: &prompt; cd /dev; MAKEDEV hda hdb hdc hdd hde Of course we didn't mention much of the architecture independent options. Moreover, the above settings may vary depending on your hardware. This is just an example. Configuring 2.6 kernels Here is a brief list of architecture dependent menus for 2.6 kernels. You should take a look at them, to see if the values set match your hardware: Processor type and features - indicates your CPU model and some specific features such as SMP or Discontigmem support Bus options - tells you what bus support is going to be enabled in your kernel (U2/Uturn, USC/GSC/HSC, Lasi, Wax, Dino, LBA/Elroy, SuperIO) PA-RISC specific drivers - enables/disables some &PA; specific drivers, such as LED support, GSP and Stable Storage support. As you can see, menus specifically concerned by &PA; hardware are not that numerous, and everything else is much generic by now. Still, you must configure the kernel accordingly to what you plan to use this box for and what features you want supported. Many other drivers are found in their respective submenus, such as SCSI, with the Zalon, Lasi SCSI and SYM2 drivers being there, or the Framebuffer devices (STI) in the Graphics Support menu, or the sound drivers (Harmony and AD1889) in the Sound menu. Help is often provided, feel free to look at it. Most of what was said for 2.4 is somewhat still applicable to 2.6. Kernel installation If you have made a native build on the box you wish to install, you can setup the new kernel as follows: within the kernel source tree linux/, as root execute: &prompt; cp vmlinux /boot/vmlinux-[kernelversion] &prompt; cp System.map /boot/System.map-[kernelversion] &prompt; cp .config /boot/config-[kernelversion] Though it is not mandatory, we suggest you to replace [kernelversion] by the version of the kernel you built, e.g.: vmlinux-2.4.18-pa44. This will help you dealing with multiple kernel versions on the same machine. The same applies to .config. It is not needed to have a working kernel, though it might be very helpful when configuring a new one. Now, do cd /boot, make sure that vmlinux is a symbolic link to another file, as in the following example: &prompt; ls -l vmlinux lrwxrwxrwx 1 root root 35 Jun 23 01:38 vmlinux -> vmlinux-2.4.18-64-SMP Make sure to remember the name of the kernel actually running on your box if ever the new one won't work properly. You are now able to ask &PO; to boot on it if needed (see for more information). Now do the following: &prompt; rm -f vmlinux &prompt; ln -s vmlinux-[kernelversion] vmlinux &prompt; sync If you want to boot from network you can forget all this, as you will need to set &PO; as explained in the , and run make palo to create the bootable lifimage. If you have made a cross-compiled build or built a kernel on a PA box which is not the one you wish to install, you have to find a way to put vmlinux, System.map and eventually .config in /boot/ as mentioned before. You can use the network (like ftp) or a CD to do so, or even direct copy to the hard disk drive. &W; 2003 boot server howto This appendix has been greatly contributed by Jeremy Drake. It describes the process of setting up a &W; 2003 Server to serve boot requests for a &PL; box. Setup the DHCP service As for the UNIX/Linux based approach (discussed in ), you need to collect some information and data before setting everything up. First of all, you need the MAC address of your &PA; box. Please check rboot preparation for details. You are going to need a lifimage file. Please read . Then, you have to enable DHCP service on your &W; box. You can do that by going into the Control Panel, open Add/Remove Programs, select Windows Components and finally Networking Services. There, enable Dynamic Host Configuration Protocol (DHCP). You need to configure the DHCP service now. Launch the DHCP admin tool by going into the Control Panel, open Admin Tools and select DHCP. Expand the server tree. Right click on Reservations New Reservation... In Reservation name, put the workstation's host name. Enter an unused IP address. Enter the &PA; box' MAC address (no delimiters, just the hex number). Select Both for whether it should be bootp or dhcp. Click Ok to close this window. Look for your newly created reservation at the bottom of the list under Reservations and click it. Right click on Configure Options... It should have inherited your server's default options, so we won't cover setting router, dns, wins and lease length. Scroll down the list of options to 066: Boot Server Host Name. Check the box next to option 066. Enter your tftp server's ip address, because IPL can't resolve hostnames. Check option 067: Bootfile Name and enter the name of the lifimage. Generally, lifimage is a good choice here. Click Ok and your dhcp server is ready! Get & setup the TFTP server To get the network boot process operational, you need the TFTP service that provides the basic file system at boot time. Get Tftpd from http://tftpd32.jounin.net/. You must download the latest version in zip format. Unzip it and store it in your favorite place. Then, you must setup the monster. Run tftpd32. Click the Browse button. Browse to where you put your lifimage, highlight it and click Ok. Make sure the IP address below the directory is the one you gave to your &PA; box. Let tftpd32 open. The tftp server only runs when the GUI is displayed. If you want to run it as a NT service, you have to download a Microsoft program. Please refer to the \ Tftpd32 FAQ. Attempt to netboot Now, you are fully set up to try to boot your &PA; box via network. You can follow the instructions in . If you have any trouble, start by looking at those points and then ask the &PL; mailing list. Settings on the DHCP server (verify the &PA; MAC address is correct). Your dhcp server is on the same physical network segment as the &PA; box. The state of the network connection of the 2 boxes. Try to tcpdump while you are booting the &PA; box over the lan. Older &PO; dumps A500 Session dump using &PO; 0.97 Main Menu: Enter command or menu > bo scsi.5.0 ipl Interact with IPL (Y, N, or Cancel)?> y Booting... Boot IO Dependent Code (IODC) revision 1 HARD Booted. palo ipl 0.97 root@c3k Tue Nov 27 14:51:48 MST 2001 Information: Boot device can't seek past 2Gb (ignore next error). byteio_read: seekread() returned -1 expected 2048 Partition Start(MB) End(MB) Id Type 1 1 15 f0 Palo 2 16 503 82 swap 3 504 2887 83 ext2 PALO(F0) partition contains: 0/vmlinux64 39