Large Disk mini-HOWTO
  Andries Brouwer, aeb@cwi.nl
  v1.0, 960626

  All about disk geometry and the 1024 cylinder limit for disks.

  1.  The problem

  Suppose you have a disk with more than 1024 cylinders.  Suppose
  moreover that you have an operating system that uses the BIOS.  Then
  you have a problem, because the usual INT13 BIOS interface to disk I/O
  uses a 10-bit field for the cylinder on which the I/O is done, so that
  cylinders 1024 and past are inaccessible.

  Fortunately, Linux does not use the BIOS, so there is no problem.

  Well, except for two things:

  (1) When you boot your system, Linux isn't running yet and cannot save
  you from BIOS problems.  This has some consequences for LILO and
  similar boot loaders.

  (2) It is necessary for all operating systems that use one disk to
  agree on where the partitions are.  In other words, if you use both
  Linux and, say, DOS on one disk, then both must interpret the
  partition table in the same way.  This has some consequences for the
  Linux kernel and for fdisk.

  Below a rather detailed description of all relevant details.  Note
  that I used kernel version 2.0.8 source as a reference.  Other
  versions may differ a bit.

  2.  Booting

  When the system is booted, the BIOS reads sector 0 (known as the MBR -
  the Master Boot Record) from the first disk (or from floppy), and
  jumps to the code found there - usually some bootstrap loader.  These
  small bootstrap programs found there typically have no own disk
  drivers and use BIOS services.  This means that a Linux kernel can
  only be booted when it is entirely located within the first 1024
  cylinders.

  This problem is very easily solved: make sure that the kernel (and
  perhaps other files used during bootup, such as LILO map files) are
  located on a partition that is entirely contained in the first 1024
  cylinders of a disk that the BIOS can access - probably this means the
  first or second disk.

  Another point is that the boot loader and the BIOS must agree as to
  the disk geometry.  It may help to give LILO the `linear' option.
  More details below.

  3.  Disk geometry and partitions

  If you have several operating systems on your disks, then each uses
  one or more disk partitions.  A disagreement on where these partitions
  are may have catastrophic consequences.

  The MBR contains a partition table describing where the (primary)
  partitions are.  There are 4 table entries, for 4 primary partitions,
  and each looks like

  struct partition {
          char active;    /* 0x80: bootable, 0: not bootable */
          char begin[3];  /* CHS for first sector */
          char type;
          char end[3];    /* CHS for last sector */
          int start;      /* 32 bit sector number (counting from 0) */
          int length;     /* 32 bit number of sectors */
  };

  (where CHS stands for Cylinder/Head/Sector).

  Thus, this information is redundant: the location of a partition is
  given both by the 24-bit begin and end fields, and by the 32-bit start
  and length fields.

  Linux only uses the start and length fields, and can therefore handle
  partitions of not more than 2^32 sectors, that is, partitions of at
  most 2 TB.  That is two hundred times larger than the disks available
  today, so maybe it will be enough for the next ten years or so.

  Unfortunately, the BIOS INT13 call uses CHS coded in three bytes, with
  10 bits for the cylinder number, 8 bits for the head number, and 6
  bits for the track sector number.  Possible cylinder numbers are
  0-1023, possible head numbers are 0-255, and possible track sector
  numbers are 1-63 (yes, sectors on a track are counted from 1, not 0).
  With these 24 bits one can address 8455716864 bytes (7.875 GB), two
  hundred times larger than the disks available in 1983.

  Even more unfortunately, the standard IDE interface allows 256
  sectors/track, 65536 cylinders and 16 heads.  This in itself allows
  access to 2^37 = 137438953472 bytes (128 GB), but combined with the
  BIOS restriction to 63 sectors and 1024 cylinders only 528482304 bytes
  (504 MB) remain addressable.

  This is not enough for present-day disks, and people resort to all
  kinds of trickery, both in hardware and in software.

  4.  Translation and Disk Managers

  Nobody is interested in what the `real' geometry of a disk is.
  Indeed, the number of sectors per track often is variable - there are
  more sectors per track close to the outer rim of the disk - so there
  is no `real' number of sectors per track.  For the user it is best to
  regard a disk as just a linear array of sectors numbered 0, 1, ...,
  and leave it to the controller to find out where a given sector lives
  on the disk.

  This linear numbering is known as LBA.  The linear address belonging
  to (c,h,s) for a disk with geometry (C,H,S) is c*H*S + h*S + (s-1).
  All SCSI controllers speak LBA, and some IDE controllers do.

  If the BIOS converts the 24-bit (c,h,s) to LBA and feeds that to a
  controller that understands LBA, then again 7.875 GB is addressable.
  Not enough for all disks, but still an improvement.  Note that here
  CHS, as used by the BIOS, no longer has any relation to `reality'.

  Something similar works when the controller doesn't speak LBA but the
  BIOS knows about translation.  (In the setup this is often indicated
  as `Large'.)  Now the BIOS will present a geometry (C',H',S') to the
  operating system, and use (C,H,S) while talking to the disk
  controller.  Usually S = S', C' = C/N and H' = H*N, where N is the
  smallest power of two that will ensure C' <= 1024 (so that least
  capacity is wasted by the rounding down in C' = C/N).  Again, this
  allows access of up to 7.875 GB.

  If a BIOS does not know about `Large' or `LBA', then there are
  software solutions around.  Disk Managers like OnTrack or EZ-Drive
  replace the BIOS disk handling routines by their own.  Often this is
  accomplished by having the disk manager code live in the MBR and
  subsequent sectors (OnTrack calls this code DDO: Dynamic Drive
  Overlay), so that it is booted before any other operating system.
  That is why one may have problems when booting from a floppy when a
  Disk Manager has been installed.

  The effect is more or less the same as with a translating BIOS - but
  especially when running several different operating systems on the
  same disk, disk managers can cause a lot of trouble.

  Linux does support OnTrack Disk Manager since version 1.3.14, and EZ-
  Drive since version 1.3.29.  Some more details are given below.

  5.  Kernel disk translation for IDE disks

  If the Linux kernel detects the presence of some disk manager on an
  IDE disk, it will try to remap the disk in the same way this disk
  manager would have done, so that Linux sees the same disk partitioning
  as for example DOS with OnTrack or EZ-Drive.  However, NO remapping is
  done when a geometry was specified on the command line - so a
  `hd=cyls,heads,secs' command line option might well kill compatibility
  with a disk manager.

  The remapping is done by trying 4, 8, 16, 32, 64, 128, 255 heads
  (keeping H*C constant) until either C <= 1024 or H = 255.

  The details are as follows - subsection headers are the strings
  appearing in the corresponding boot messages.  Here and everywhere
  else in this text partition types are given in hexadecimal.

  5.1.  EZD

  EZ-Drive is detected by the fact that the first primary partition has
  type 55.  The geometry is remapped as described above, and the
  partition table from sector 0 is discarded - instead the partition
  table is read from sector 1.  Disk block numbers are not changed, but
  writes to sector 0 are redirected to sector 1.  This behaviour can be
  changed by recompiling the kernel with
   #define FAKE_FDISK_FOR_EZDRIVE  0 in ide.c.

  5.2.  DM6:DDO

  OnTrack DiskManager (on the first disk) is detected by the fact that
  the first primary partition has type 54.  The geometry is remapped as
  described above and the entire disk is shifted by 63 sectors (so that
  the old sector 63 becomes sector 0).  Afterwards a new MBR (with
  partition table) is read from the new sector 0.  Of course this shift
  is to make room for the DDO - that is why there is no shift on other
  disks.

  5.3.  DM6:AUX

  OnTrack DiskManager (on other disks) is detected by the fact that the
  first primary partition has type 51 or 53.  The geometry is remapped
  as described above.

  5.4.  DM6:MBR

  An older version of OnTrack DiskManager is detected not by partition
  type, but by signature.  (Test whether the offset found in bytes 2 and
  3 of the MBR is not more than 430, and the short found at this offset
  equals 0x55AA, and is followed by an odd byte.) Again the geometry is
  remapped as above.

  5.5.  PTBL

  Finally, there is a test that tries to deduce a translation from the
  start and end values of the primary partitions: If some partition has
  start and end cylinder less than 256, and start and end sector number
  1 and 63, respectively, and end heads 31, 63 or 127, then, since it is
  customary to end partitions on a cylinder boundary, and since moreover
  the IDE interface uses at most 16 heads, it is conjectured that a BIOS
  translation is active, and the geometry is remapped to use 32, 64 or
  128 heads, respectively.  (Maybe there is a flaw here, and genhd.c
  should not have tested the high order two bits of the cylinder
  number?)  However, no remapping is done when the current idea of the
  geometry already has 63 sectors per track and at least as many heads
  (since this probably means that a remapping was done already).

  6.  Consequences

  What does all of this mean?  For Linux users only one thing: that they
  must make sure that LILO and fdisk use the right geometry where
  `right' is defined for fdisk as the geometry used by the other
  operating systems on the same disk, and for LILO as the geometry that
  will enable successful interaction with the BIOS at boot time.
  (Usually these two coincide.)

  How does fdisk know about the geometry?  It asks the kernel, using the
  HDIO_GETGEO ioctl.  But the user can override the geometry
  interactively or on the command line.

  How does LILO know about the geometry?  It asks the kernel, using the
  HDIO_GETGEO ioctl.  But the user can override the geometry using the
  `disk=' option.  One may also give the linear option to LILO, and it
  will store LBA addresses instead of CHS addresses in its map file, and
  find out of the geometry to use at boot time (by using INT 13 Function
  8 to ask for the drive geometry).

  How does the kernel know what to answer?  Well, first of all, the user
  may have specified an explicit geometry with a `hd=cyls,heads,secs'
  command line option.  And otherwise the kernel will ask the hardware.

  6.1.  IDE details

  Let me elaborate.  The IDE driver has four sources for information
  about the geometry.  The first (G_user) is the one specified by the
  user on the command line.  The second (G_bios) is the BIOS Fixed Disk
  Parameter Table (for first and second disk only) that is read on
  system startup, before the switch to 32-bit mode.  The third (G_phys)
  and fourth (G_log) are returned by the IDE controller as a response to
  the IDENTIFY command - they are the `physical' and `current logical'
  geometries.

  On the other hand, the driver needs two values for the geometry: on
  the one hand G_fdisk, returned by a HDIO_GETGEO ioctl, and on the
  other hand G_used, which is actually used for doing I/O.  Both G_fdisk
  and G_used are initialized to G_user if given, to G_bios when this
  information is present according to CMOS, and to to G_phys otherwise.
  If G_log looks reasonable then G_used is set to that.  Otherwise, if
  G_used is unreasonable and G_phys looks reasonable then G_used is set
  to G_phys.  Here `reasonable' means that the number of heads is in the
  range 1-16.

  To say this in other words: the command line overrides the BIOS, and
  will determine what fdisk sees, but if it specifies a translated
  geometry (with more than 16 heads), then for kernel I/O it will be
  overridden by output of the IDENTIFY command.

  6.2.  SCSI details

  The situation for SCSI is slightly different, as the SCSI commands
  already use logical block numbers, so a `geometry' is entirely
  irrelevant for actual I/O.  However, the format of the partition table
  is still the same, so fdisk has to invent some geometry, and also uses
  HDIO_GETGEO here - indeed, fdisk does not distinguish between IDE and
  SCSI disks.  As one can see from the detailed description below, the
  various drivers each invent a somewhat different geometry.  Indeed,
  one big mess.

  If you are not using DOS or so, then avoid all extended translation
  settings, and just use 64 heads, 32 sectors per track (for a nice,
  convenient 1 MB per cylinder), if possible, so that no problems arise
  when you move the disk from one controller to another.  Some SCSI disk
  drivers (aha152x, pas16, ppa, qlogicfas, qlogicisp) are so nervous
  about DOS compatibility that they will not allow a Linux-only system
  to use more than about 8 GB.  This is a bug.

  What is the real geometry?  The easiest answer is that there is no
  such thing.  And if there were, you wouldn't want to know, and
  certainly NEVER, EVER tell fdisk or LILO or the kernel about it.  It
  is strictly a business between the SCSI controller and the disk.  Let
  me repeat that: only silly people tell fdisk/LILO/kernel about the
  true SCSI disk geometry.

  But if you are curious and insist, you might ask the disk itself.
  There is the important command READ CAPACITY that will give the total
  size of the disk, and there is the MODE SENSE command, that in the
  Rigid Disk Drive Geometry Page (page 04) gives the number of cylinders
  and heads (this is information that cannot be changed), and in the
  Format Page (page 03) gives the number of bytes per sector, and
  sectors per track.  This latter number is typically dependent upon the
  notch, and the number of sectors per track varies - the outer tracks
  have more sectors than the inner tracks.  The Linux program scsiinfo
  will give this information.  There are many details and complications,
  and it is clear that nobody (probably not even the operating system)
  wants to use this information.  Moreover, as long as we are only
  concerned about fdisk and LILO, one typically gets answers like
  C/H/S=4476/27/171 - values that cannot be used by fdisk because the
  partition table reserves only 10 resp. 8 resp. 6 bits for C/H/S.

  Then where does the kernel HDIO_GETGEO get its information from?
  Well, either from the SCSI controller, or by making an educated guess.
  Some drivers seem to think that we want to know `reality', but of
  course we only want to know what the DOS or OS/2 FDISK (or Adaptec
  AFDISK, etc) will use.

  Note that Linux fdisk needs the numbers H and S of heads and sectors
  per track to convert LBA sector numbers into c/h/s addresses, but the
  number C of cylinders does not play a role in this conversion.  Some
  drivers use (C,H,S) = (1023,255,63) to signal that the drive capacity
  is at least 1023*255*63 sectors.  This is unfortunate, since it does
  not reveal the actual size, and will limit the users of most fdisk
  versions to about 8 GB of their disks - a real limitation in these
  days.

  In the description below, M denotes the total disk capacity, and C, H,
  S the number of cylinders, heads and sectors per track.  It suffices
  to give H, S if we regard C as defined by M / (H*S).

  By default, H=64, S=32.

     aha1740, dtc, g_NCR5380, t128, wd7000:
        H=64, S=32.

     aha152x, pas16, ppa, qlogicfas, qlogicisp:
        H=64, S=32 unless C > 1024, in which case H=255, S=63, C =
        min(1023, M/(H*S)).  (Thus C is truncated, and H*S*C is not an
        approximation to the disk capacity M.  This will confuse most
        versions of fdisk.)  The ppa.c code uses M+1 instead of M and
        says that due to a bug in sd.c M is off by 1.

     advansys:
        H=64, S=32 unless C > 1024 and moreover the `> 1 GB' option in
        the BIOS is enabled, in which case H=255, S=63.

     aha1542:
        Ask the controller which of two possible translation schemes is
        in use, and use either H=255, S=63 or H=64, S=32.  In the former
        case there is a boot message "aha1542.c: Using extended bios
        translation".

     aic7xxx:
        H=64, S=32 unless C > 1024, and moreover either the "extended"
        boot parameter was given, or the `extended' bit was set in the
        SEEPROM or BIOS, in which case H=255, S=63.

     buslogic:
        H=64, S=32 unless C >= 1024, and moreover extended translation
        was enabled on the controller, in which case if M < 2^22 then
        H=128, S=32; otherwise H=255, S=63.  However, after making this
        choice for (C,H,S), the partition table is read, and if for one
        of the three possibilities (H,S) = (64,32), (128,32), (255,63)
        the value endH=H-1 is seen somewhere then that pair (H,S) is
        used, and a boot message is printed "Adopting Geometry from
        Partition Table".

     fdomain:
        Find the geometry information in the BIOS Drive Parameter Table,
        or read the partition table and use H=endH+1, S=endS for the
        first partition, provided it is nonempty, or use H=64, S=32 for
        M < 2^21 (1 GB), H=128, S=63 for M < 63*2^17 (3.9 GB) and H=255,
        S=63 otherwise.

     in2000:
        Use the first of (H,S) = (64,32), (64,63), (128,63), (255,63)
        that will make C <= 1024.  In the last case, truncate C at 1023.

     seagate:
        Read C,H,S from the disk.  (Horrors!)  If C or S is too large,
        then put S=17, H=2 and double H until C <= 1024.  This means
        that H will be set to 0 if M > 128*1024*17 (1.1 GB).  This is a
        bug.

     ultrastor and u14_34f:
        One of three mappings ((H,S) = (16,63), (64,32), (64,63)) is
        used depending on the controller mapping mode.

  If the driver does not specify the geometry, we fall back on an edu­
  cated guess using the partition table, or using the total disk capac­
  ity.

  Look at the partition table.  Since by convention partitions end on a
  cylinder boundary, we can, given end = (endC,endH,endS) for any
  partition, just put H = endH+1 and S = endS.  (Recall that sectors are
  counted from 1.)  More precisely, the following is done.  If there is
  a nonempty partition, pick the partition with the largest beginC.  For
  that partition, look at end+1, computed both by adding start and
  length and by assuming that this partition ends on a cylinder
  boundary.  If both values agree, or if endC = 1023 and start+length is
  an integral multiple of (endH+1)*endS, then assume that this partition
  really was aligned on a cylinder boundary, and put H = endH+1 and S =
  endS.  If this fails, either because there are no partitions, or
  because they have strange sizes, then look only at the disk capacity
  M.  Algorithm: put H = M/(62*1024) (rounded up), S = M/(1024*H)
  (rounded up), C = M/(H*S) (rounded down).  This has the effect of
  producing a (C,H,S) with C at most 1024 and S at most 62.