SRM Firmware Howto
  David Mosberger <mailto:davidm@azstarnet.com>
  v0.5, 17 August 1996

  This document describes how to boot Linux/Alpha using the SRM
  firmware, which is the firmware normally used to boot DEC Unix.  Gen�
  erally, it is preferable to use MILO instead of aboot since MILO is
  perfectly adapted to the needs of Linux.  However, MILO is not always
  available for a particular system and MILO does not presently have the
  ability to boot over the network.  In either case, using the SRM con�
  sole may be  the right solution.

  Unless you're interested in technical details, you may want to skip
  right to Section ``''.

  1.  How Does SRM Boot an OS?

  All versions of SRM can boot from SCSI disks and the versions for
  recent platforms, such as the Noname or AlphaStations can boot from
  floppy disks as well.  Network booting via bootp is supported.  Note
  that older SRM versions (notably the one for the Jensen) cannot boot
  from floppy disks.  Also, booting from IDE disk drives is unsupported.

  Booting Linux with SRM is a two step process: first, SRM loads and
  transfers control to the secondary bootstrap loader.  Then the
  secondary bootstrap loader sets up the environment for Linux, reads
  the kernel image from a disk filesystem and finally transfers control
  to Linux.

  Currently, there are two secondary bootstrap loaders for Linux: the
  raw loader that comes with the Linux kernel and aboot which is
  distributed separately.  These two loaders are described in more
  detail below.

  1.1.  Loading The Secondary Bootstrap Loader

  SRM knows nothing about filesystems or disk-partitions.  It simply
  expects that the secondary bootstrap loader occupies a consecutive
  range of physical disk sector, starting from a given offset.  The
  information on the size of the secondary bootstrap loader and the
  offset of its first disk sector is stored in the first 512 byte
  sector.  Specifically, the long integer at offset 480 stores the size
  of the secondary bootstrap loader (in 512-byte blocks) and the long at
  offset 488 gives the sector number at which the secondary bootstrap
  loader starts.  The first sector also stores a flag-word at offset 496
  which is always 0 and a checksum at offset 504.  The checksum is
  simply the sum of the first 63 long integers in the first sector.

  If the checksum in the first sector is correct, SRM goes ahead and
  reads the size sectors starting from the sector given in the sector
  number field and places them in virtual memory at address 0x20000000.
  If the reading completes successfully, SRM performs a jump to address
  0x20000000.

  2.  The Raw Loader

  The sources for this loader can be found in directory

               linux/arch/alpha/boot

  of the Linux kernel source distribution.  It loads the Linux kernel by
  reading START_SIZE bytes starting at disk offset BOOT_SIZE+512 (also
  in bytes).  The constants START_SIZE and BOOT_SIZE are defined in
  linux/include/asm-alpha/system.h.  START_SIZE must be at least as big
  as the kernel image (i.e., the size of the BOOT_SIZE must be at least
  as big as the image of the raw bootstrap loader.  Both constants
  should be an integer multiple of the sector size, which is 512 bytes.
  The default values are currently 2MB for START_SIZE and 16KB for
  BOOT_SIZE.  Note that if you want to boot from a 1.44MB floppy disk,
  you have to reduce START_SIZE to 1400KB and make sure that the kernel
  you want to boot is no bigger than that.

  To build a raw loader, simply type make rawboot in /usr/src/linux.
  This should produce the following files in arch/alpha/boot:

     tools/lxboot:
        The first sector on the disk.  It contains the offset and size
        of the next file in the format described above.

     tools/bootlx:
        The raw boot loader that will load the file below.

     vmlinux.nh:
        The raw kernel image consisting of the .text, .data, and .bss
        segments of the object file in /usr/src/linux/vmlinux.  The
        extension .nh indicates that this file has no object-file
        header.

  The concatenation of these three files should be written to the disk
  from which you want to boot.  For example, to boot from a floppy,
  insert an empty floppy disk in, say, /dev/fd0 and then type:

       cat tools/lxboot tools/bootlx vmlinux >/dev/fd0

  You can then shutdown the system and boot from the floppy by issueing
  the command boot dva0.

  3.  The aboot Loader

  When using the SRM firmware, aboot is the preferred way of booting
  Linux.  It supports:

  �  direct booting from various filesystems (ext2, ISO9660, and UFS,
     the DEC Unix filesystem)

  �  booting of executable object files (both ELF and ECOFF)

  �  booting compressed kernels

  �  network booting (using bootp)

  �  partition tables in DEC Unix format (which is compatible with BSD
     Unix partition tables)

  �  interactive booting and default configurations for SRM consoles
     that cannot pass long option strings

  3.1.  Getting and Building aboot

  The latest sources for aboot are available in this ftp directory
  <ftp://ftp.azstarnet.com/pub/linux/axp/aboot>.  The description in
  this manual applies to aboot version 0.5 or newer.

  Once you downloaded and extracted the latest tar file, take a look at
  the README and INSTALL files for installation hints.  In particular,
  be sure to adjust the variables in Makefile and in include/config.h to
  match your environment.  Normally, you won't need to change anything
  when building under Linux, but it is always a good idea to double
  check.  If you're satisfied with the configuration, simply type make
  to build it (if you're not building under Linux, be advised that aboot
  requires GNU make).

  After running make, the aboot directory should contain the following
  files:

     aboot
        This is the actual aboot executable (either an ECOFF or ELF
        object file).

     bootlx
        Same as above, but it contains only the text, data and bss
        segments---that is, this file is not an object file.

     sdisklabel/writeboot
        Utility to install aboot on a hard disk.

     tools/e2writeboot
        Utility to install aboot on an ext2 filesystem (usually used for
        floppies only).

     tools/isomarkboot
        Utility to install aboot on a iso9660 filesystem (used by CD-ROM
        distributors).

     tools/abootconf
        Utility to configure an installed aboot.

  3.2.  Floppy Installation

  The bootloader can be installed on a floppy using the e2writeboot
  command (note: this can't be done on a Jensen since its firmware does
  not support booting from floppy).  This command requires that the disk
  is not overly fragmented as it needs to find enough contiguous file
  blocks to store the entire aboot image (currently about 90KB).  If
  e2writeboot fails because of this, reformat the floppy and try again
  (e.g., with fdformat(1)).  For example, the following steps install
  aboot on floppy disk assuming the floppy is in drive /dev/fd0:

  fdformat /dev/fd0
  mke2fs /dev/fd0
  e2writeboot /dev/fd0 bootlx

  3.3.  Harddisk Installation

  Since the e2writeboot command may fail on highly fragmented disks and
  since reformatting a harddisk is not without pain, it is generally
  safer to install aboot on a harddisk using the swriteboot command.
  swriteboot requires that the first few sectors are reserved for
  booting purposes.  We suggest that the disk be partitioned such that
  the first partition starts at an offset of 2048 sectors.  This leaves
  1MB of space for storing aboot.  On a properly partitioned disk, it is
  then possible to install aboot as follows (assuming the disk is
  /dev/sda):

       swriteboot /dev/sda bootlx

  On a Jensen, you will want to leave some more space, since you need to
  write a kernel to this place, too---2MB should be sufficient when
  using compressed kernels.  Use swriteboot as described in Section ``''
  to write bootlx together with the Linux kernel.

  3.4.  CD-ROM Installation

  To make a CD-ROM bootable by SRM, simply build aboot as described
  above.  Then, make sure that the bootlx file is present on the iso9660
  filesystem (e.g., copy bootlx to the directory that is the filesystem
  master, then run mkisofs on that directory).  After that, all that
  remains to be done is to mark the filesystem as SRM bootable.  This is
  achieved with a command of the form:

       isomarkboot filesystem bootlx

  The command above assumes that filesystem is a file containing the
  iso9660 filesystem and that bootlx has been copied into the root
  directory of that filesystem.  That's it!

  3.5.  Building the Linux Kernel

  A bootable Linux kernel can be built with the following steps.  During
  the make config, be sure to answer "yes" to the question whether you
  want to boot the kernel via SRM.

  cd /usr/src/linux
  make config
  make dep
  make boot

  The last command will build the file arch/alpha/boot/vmlinux.gz which
  can then be copied to the disk from which you want to boot from.  In
  our floppy disk example above, this would entail:

       mount /dev/fd0 /mnt
       cp arch/alpha/boot/vmlinux.gz /mnt
       umount /mnt

  3.6.  Booting Linux

  With the SRM firmware and aboot installed, Linux is generally booted
  with a command of the form:

       boot devicename -fi filename -fl flags

  The filename and flags arguments are optional.  If they are not
  specified, SRM uses the default values stored in environment variables
  BOOT_OSFILE and BOOT_OSFLAGS.  The syntax and meaning of these two
  arguments is described in more detail below.

  3.6.1.  Boot Filename

  The filename argument takes the form:

       [n/]filename

  n is a single digit in the range 1..8 that gives the partition number
  from which to boot from.  filename is the path of the file you want
  boot.  For example to boot from the second partition of SCSI device 6,
  you would enter:

       boot dka600 -file 2/vmlinux.gz

  Or to boot from floppy drive 0, you'd enter:

       boot dva0 -file vmlinux.gz

  If a disk has no partition table , aboot pretends the disk contains
  one ext2 partition starting at the first diskblock.  This allows
  booting from floppy disks.

  As a special case, partition number 0 is used to request booting from
  a disk that does not (yet) contain a file system.  When specifying
  "partition" number 0, aboot assumes that the Linux kernel is stored
  right behind the aboot image.  Such a layout can be achieved with the
  swriteboot command.  For example, to setup a filesystem-less boot from
  /dev/sda, one could use the command:

       swriteboot /dev/sda bootlx vmlinux.gz

  Booting a system in this way is not normally necessary.  The reason
  this feature exists is to make it possible to get Linux installed on a
  systems that can't boot from a floppy disk (e.g., the Jensen).

  3.6.2.  Boot Flags

  A number of bootflags can be specified.  The syntax is:

       -flags "options..."

  Where "options..." is any combination the following options (separated
  by blanks).  There are many more bootoptions, depending on what
  drivers your kernel has installed.  The options listed below are
  therefore just examples to illustrate the general idea:

     load_ramdisk=1
        Copy root file system from a (floppy) disk to the RAM disk
        before starting the system.  The RAM disk will be used in lieu
        of the root device.  This is useful to bootstrap Linux on a
        system with only one floppy drive.

     floppy=str
        Sets floppy configuration to str.

     root=dev
        Select device dev as the root-file system. The device can be
        specified as a major/minor hex number (e.g., 0x802 for
        /dev/sda2) or one of a few canonical names (e.g., /dev/fd0,
        /dev/sda2).

     single
        Boot system in single user mode.

     kgdb
        Enable kernel-gdb (works only if CONFIG_KGDB is enabled; a
        second Alpha system needs to be connected over the serial port
        in order to make this work)

  Some SRM implementations (e.g., the one for the Jensen) are
  handicapped and allow only short option strings (e.g., at most 8
  characters).  In such a case, aboot can be booted with the single-
  character boot flag "i".  With this flag, aboot will prompt the user
  to interacively enter a boot option string of up to 256 characters.
  For example:

       boot dka0 -fl i
       aboot> 3/vmlinux.gz root=/dev/sda3 single

  Since booting in that manner quickly becomes tedious, aboot allows to
  define short-hands for frequently used commandlines.  In particular, a
  single digit option (0-9) requests that aboot uses the corresponding
  option string stored in file /etc/aboot.conf.  A sample aboot.conf is
  shown below:

       #
       # aboot default configurations
       #
       0:3/vmlinux.gz root=/dev/sda3
       1:3/vmlinux.gz root=/dev/sda3 single
       2:3/vmlinux.new.gz root=/dev/sda3
       3:3/vmlinux root=/dev/sda3
       8:- root=/dev/sda3            # fs-less boot of raw kernel
       9:0/vmlinux.gz root=/dev/sda3 # fs-less boot of (compressed) ECOFF kernel
       -

  With this configuration file, the command

       boot dka0 -fl 1

  corresponds exactly to the boot command shown above.  It is quite easy
  to forget what number corresponds to what option string.  To alleviate
  this problem, boot with option "h" and aboot will print the contents
  of /etc/aboot.conf before issueing the prompt for the full option
  string.

  Finally, whenever aboot prompts for an option string, it is possible
  to enter one of the single character flags ("i", "h", or "0"-"9") to
  get the same effect as if that flag had been specified in the boot
  command line.  For example, you could boot with flag "i" and then type
  "h" (followed by return) to remind yourself of the contents of
  /etc/aboot.conf

  3.6.2.1.  Selecting the Partition of /etc/aboot.conf

  When installed on a harddisk, aboot needs to know what partition to
  search for the /etc/aboot.conf file.  A newly compiled aboot will
  search the second partition (e.g., /dev/sda2).  Since it would be
  inconvenient to have to recompile aboot just to change the partition
  number, abootconf allows to directly modify an installed aboot.
  Specifically, if you want to change aboot to use the third partition
  on disk /dev/sda, you'd use the command:

       abootconf /dev/sda 3

  You can verify the current setting by simply omitting the partition
  number.  That is: abootconf /dev/sda will print the currently selected
  partition number.  Note that aboot does have to be installed already
  for this command to succeed.  Also, when installing a new aboot, the
  partition number will fall back to the default (i.e., it will be
  necessary to rerun abootconf).

  Since aboot version 0.5, it is also possible to select the aboot.conf
  partition via the boot command line. This can be done with a command
  line of the form a:b where a is the partition that holds
  /etc/aboot.conf and b is a single-letter option as described above
  (0-9, i, or h). For example, if you type boot -fl "3:h" dka100 the
  system boots from SCSI ID 1, loads /etc/aboot.conf from the third
  partition, prints its contents on the screen and waits for you to
  enter the boot options.

  3.7.  Booting Over the Network

  Two prelimenary steps are necessary before Linux can be booted via a
  network.  First, you need to set the SRM environment variables to
  enable booting via the bootp protocol and second you need to setup
  another machine as the your boot server.  Please refer to the SRM
  documentation that came with your machine for information on how to
  enable bootp.  Setting up the boot server is obviously dependent on
  what operating system that machine is running, but typically it
  involves starting the program bootpd in the background after
  configuring the /etc/bootptab file.  The bootptab file has one entry
  describing each client that is allowed to boot from the server.  For
  example, if you want to boot the machine myhost.cs.arizona.edu, then
  an entry of the following form would be needed:

       myhost.cs.arizona.edu:\
               :hd=/remote/:bf=vmlinux.bootp:\
               :ht=ethernet:ha=08012B1C51F8:hn:vm=rfc1048:\
               :ip=192.12.69.254:bs=auto:

  This entry assumes that the machine's Ethernet address is 08012B1C51F8
  and that its IP address is 192.12.69.254.  The Ethernet address can be
  found with the show device command of the SRM console or, if Linux is
  running, with the ifconfig command.  The entry also defines that if
  the client does not specify otherwise, the file that will be booted is
  vmlinux.bootp in directory /remote.  For more information on
  configuring bootpd, please refer to its man page.

  Next, build aboot with with the command make netboot.  Make sure the
  kernel that you want to boot has been built already.  By default, the
  aboot Makefile uses the kernel in
  /usr/src/linux/arch/alpha/boot/vmlinux.gz (edit the Makefile if you
  want to use a different path).  The result of make netboot is a file
  called vmlinux.bootp which contains aboot and the Linux kernel, ready
  for network booting.

  Finally, copy vmlinux.bootp to the bootsever's directory.  In the
  example above, you'd copy it into /remote/vmlinux.bootp.  Next, power
  up the client machine and boot it, specifying the Ethernet adapter as
  the boot device.  Typically, SRM calls the first Ethernet adapter
  ewa0, so to boot from that device, you'd use the command:

       boot ewa0

  The -fi and -fl options can be used as usual.  In particular, you can
  ask aboot to prompt for Linux kernel arguments by specifying the
  option -fl i.

  4.  Sharing a Disk With DEC Unix

  Unfortunately, DEC Unix doesn't know anything about Linux, so sharing
  a single disk between the two OSes is not entirely trivial.  However,
  it is not a difficult task if you heed the tips in this section.  The
  section assumes you are using aboot version 0.5 or newer.

  4.1.  Partitioning the disk

  First and foremost: never use any of the Linux partitioning programs
  (minlabel or fdisk) on a disk that is also used by DEC Unix.  The
  Linux minlabel program uses the same partition table format as DEC
  Unix disklabel, but there are some incompatibilities in the data that
  minlabel fills in, so DEC Unix will simply refuse to accept a
  partition table generated by minlabel.  To setup a Linux ext2
  partition under DEC Unix, you'll have to change the disktab entry for
  your disk.  For the purpose of this discussion, let's assume that you
  have an rz26 disk (a common 1GB drive) on which you want to install
  Linux.  The disktab entry under DEC Unix v3.2 looks like this (see
  file /etc/disktab):

       rz26|RZ26|DEC RZ26 Winchester:\
               :ty=winchester:dt=SCSI:ns#57:nt#14:nc#2570:\
               :oa#0:pa#131072:ba#8192:fa#1024:\
               :ob#131072:pb#262144:bb#8192:fb#1024:\
               :oc#0:pc#2050860:bc#8192:fc#1024:\
               :od#393216:pd#552548:bd#8192:fd#1024:\
               :oe#945764:pe#552548:be#8192:fe#1024:\
               :of#1498312:pf#552548:bf#8192:ff#1024:\
               :og#393216:pg#819200:bg#8192:fg#1024:\
               :oh#1212416:ph#838444:bh#8192:fh#1024:

  The interesting fields here are o?, and p?, where ? is a letter in the
  range a-h (first through 8-th partition).  The o value gives the
  starting offset of the partition (in sectors) and the p value gives
  the size of the partition (also in sectors).  See disktab(4) for more
  info.  Note that DEC Unix likes to define overlapping partitions.  For
  the entry above, the partition layout looks like this (you can verify
  this by adding up the various o and p values):

         a     b         d           e           f
       |---|-------|-----------|-----------|-----------|

                               c
       |-----------------------------------------------|

                            g                 h
                   |-----------------|-----------------|

  DEC Unix insists that partition a starts at offset 0 and that
  partition c spans the entire disk.  Other than that, you can setup the
  partition table any way you like.

  Let's suppose you have DEC Unix using partition g and want to install
  Linux on partition h with partition b being a (largish) swap
  partition.  To get this layout without destroying the existing DEC
  Unix partition, you need to set the partition types explicitly.  You
  can do this by adding a t field for each partition.  In our case, we
  add the following line to the above disktab entry.

               :ta=unused:tb=swap:tg=4.2BSD:th=resrvd8:

  Now why do we mark partition h as "reservd8" instead of "ext2"?  Well,
  DEC Unix doesn't know about Linux.  It so happens that partition type
  "ext2" corresponds to a numeric value of 8, and DEC Unix uses the
  string "reservd8" for that value.  Thus, in DEC Unix speak, "reservd8"
  means "ext2".  OK, this was the hard part.  Now we just need to
  install the updated disktab entry on the disk.  Let's assume the disk
  has SCSI id 5.  In this case, we'd do:

       disklabel -rw /dev/rrz5c rz26

  You can verify that everything is all right by reading back the
  disklabel with disklabel -r /dev/rrz5c.  At this point, you may want
  to reboot DEC Unix and make sure the existing DEC Unix partition is
  still alive and well.  If that is the case, you can shut down the
  machine and start with the Linux installation.  Be sure to skip the
  disk partitioning step during the install.  Since we already installed
  a good partition table, you should be able to proceed and select the
  8th partition as the Linux root partition and the 2nd partition as the
  swap partition.  If the disk is, say, the second SCSI disk in the
  machine, then the device name for these partitions would be /dev/sdb8
  and /dev/sdb2, respectively (note that Linux uses letters to name the
  drives and numbers to name the partitions, which is exactly reversed
  from what DEC Unix does; the Linux scheme makes more sense, of course
  ;-).

  4.2.  Installing aboot

  First big caveat: with the SRM firmware, you can boot one and only one
  operating system per disk.  For this reason, it is generally best to
  have at least two SCSI disks in a machine that you want to dualboot
  between Linux and DEC Unix.  Of course, you could also boot Linux from
  a floppy if speed doesn't matter or over the network, if you have a
  bootp-capable server.  But in this section we assume you want to boot
  Linux from a disk that contains one or more DEC Unix partitions.

  Second big caveat: installing aboot on a disk shared with DEC Unix
  renders the first and third partition unusable (since those must have
  a starting offset of 0).  For this reason, we recommend that you
  change the size of partition a to something that is just big enough to
  hold aboot (1MB should be plenty).

  Once these two caveats are taken care of, installing aboot is almost
  as easy as usual: since partition a and c will overlap with aboot, we
  need to tell swriteboot that this is indeed OK.  We can do this under
  Linux with a command line of the following form (again, assuming we're
  trying to install aboot on the second SCSI disk):

       swriteboot -f1 -f3 /dev/sdb bootlx

  The -f1 means that we want to force writing bootlx even though it
  overlaps with partition 1.  The corresponding applies for partition 3.

  This is it.  You should now be able to shutdown the system and boot
  Linux from the harddisk.  In our example, the SRM command line to do
  this would be:

       boot dka5 -fi 8/vmlinux.gz -fl root=/dev/sdb8