aeb@cwi.nl
Suppose you have a disk with more than 1024 cylinders. Suppose moreover that you have an operating system that uses the old INT13 BIOS interface to disk I/O. Then you have a problem, because this interface 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.
You got a new large disk. What to do? Well, on the software side:
use fdisk
(or, better, cfdisk
) to create partitions,
and then mke2fs
to create a filesystem, and then mount
to attach the new filesystem to the big file hierarchy.
A year ago or so I could write:
You need not read this HOWTO since there are no problems
with large hard disks these days. The great majority of
apparent problems is caused by people who think there might
be a problem and install a disk manager, or go into fdisk
expert mode, or specify explicit disk geometries to LILO
or on the kernel command line.
However, typical problem areas are: (i) ancient hardware, (ii) several operating systems on the same disk, and sometimes (iii) booting.
These days the situation is a bit worse. Maybe 2.3.21 and later will be good for all disks again.
Advice:
For large SCSI disks: Linux has supported them from very early on. No action required.
For large IDE disks (over 8.4 GB): get a recent stable kernel (2.0.34 or later). Usually, all will be fine now, especially if you were wise enough not to ask the BIOS for disk translations like LBA and the like.
For very large IDE disks (over 33.8 GB): see IDE problems with 34+ GB disks below.
If LILO hangs at boot time, also specify
linear
in the
configuration file /etc/lilo.conf
.
(And if you did have linear
, try without it.)
There may be geometry problems that can be solved by giving an explicit geometry to kernel/LILO/fdisk.
If you have an old fdisk
and it warns about
overlapping partitions:
ignore the warnings, or check using cfdisk
that really all is well.
If you think something is wrong with the size of your disk,
make sure that you are not confusing binary and decimal
units
,
and realize that the free space that df
reports on an empty disk
is a few percent smaller than the partition size, because there
is administrative overhead.
Now, if you still think there are problems, or just are curious, read on.
A kilobyte (kB) is 1000 bytes. A megabyte (MB) is 1000 kB. A gigabyte (GB) is 1000 MB. A terabyte (TB) is 1000 GB. This is the SI norm. However, there are people that use 1 MB=1024000 bytes and talk about 1.44 MB floppies, and people who think that 1 MB=1048576 bytes. Here I follow the recent standard and write Ki, Mi, Gi, Ti for the binary units, so that these floppies are 1440 KiB (1.47 MB, 1.41 MiB), 1 MiB is 1048576 bytes (1.05 MB), 1 GiB is 1073741824 bytes (1.07 GB) and 1 TiB is 1099511627776 bytes (1.1 TB).
Quite correctly, the disk drive manufacturers follow the SI norm and use the decimal units. However, Linux boot messages and some fdisk-type programs use the symbols MB and GB for binary, or mixed binary-decimal units. So, before you think your disk is smaller than was promised when you bought it, compute first the actual size in decimal units (or just in bytes).
Concerning terminology and abbreviation for binary units, Knuth has an alternative proposal, namely to use KKB, MMB, GGB, TTB, PPB, EEB, ZZB, YYB and to call these large kilobyte, large megabyte, ... large yottabyte. He writes: `Notice that doubling the letter connotes both binary-ness and large-ness.' This is a good proposal - `large gigabyte' sounds better than `gibibyte'. For our purposes however the only important thing is to stress that a megabyte has precisely 1000000 bytes, and that some other term and abbreviation is required if you mean something else.
In the present text a sector has 512 bytes. This is almost always
true, but for example certain MO disks use a sectorsize of 2048 bytes,
and all capacities given below must be multiplied by four.
(When using fdisk
on such disks, make sure you have version
2.9i or later, and give the `-b 2048' option.)
A disk with C cylinders, H heads and S sectors per track
has C*
H*
S sectors in all, and can store
C*
H*
S*
512 bytes.
For example, if the disk label says C/H/S=4092/16/63
then the disk has 4092*
16*
63=4124736 sectors, and can hold
4124736*
512=2111864832 bytes (2.11 GB).
There is an industry convention to give C/H/S=16383/16/63
for disks larger than 8.4 GB, and the disk size can no longer
be read off from the C/H/S values reported by the disk.
In order to read or write something from or to the disk, we have to specify a position on the disk, for example by giving a sector or block number. If the disk is a SCSI disk, then this sector number goes directly into the SCSI command and is understood by the disk. If the disk is an IDE disk using LBA, then precisely the same holds. But if the disk is old, RLL or MFM or IDE from before the LBA times, then the disk hardware expects a triple (cylinder,head,sector) to designate the desired spot on the disk.
The correspondence between the linear numbering and this 3D notation
is as follows: for a disk with C cylinders, H heads and S sectors/track
position (c,h,s) in 3D or CHS notation is the same as position
c*
H*
S + h*
S + (s-1) in linear or LBA notation.
(The minus one is because traditionally sectors are counted from 1,
not 0, in this 3D notation.)
Consequently, in order to access a very old non-SCSI disk, we need to know its geometry, that is, the values of C, H and S.
Linux does not use the BIOS, but some other systems do.
The BIOS, which predates LBA times, offers with INT13
disk I/O routines that have (c,h,s) as input.
(More precisely: AH
selects the function to perform,
CH
is the low 8 bits of the cylinder number, CL
has in bits 7-6 the high two bits of the cylinder number
and in bits 5-0 the sector number, DH
is the head number,
and DL
is the drive number (80h or 81h).
This explains part of the layout of the partition table.)
Thus, we have CHS encoded in three bytes, with 10 bits for the cylinder number, 8 bits for the head number, and 6 bits for the track sector number (numbered 1-63). It follows that cylinder numbers can range from 0 to 1023 and that no more than 1024 cylinders are BIOS addressable.
DOS and Windows software did not change when IDE disks with LBA support were introduced, so DOS and Windows continued needing a disk geometry, even when this was no longer needed for the actual disk I/O, but only for talking to the BIOS. This again means that Linux needs the geometry in those places where communication with the BIOS or with other operating systems is required, even on a modern disk.
This state of affairs lasted for four years or so, and then disks appeared on the market that could not be addressed with the INT13 functions (because the 10+8+6=24 bits for (c,h,s) can address not more than 8.5 GB) and a new BIOS interface was designed: the so-called Extended INT13 functions, where DS:SI points at a 16-byte Disk Address Packet that contains an 8-byte starting absolute block number.
Very slowly the Microsoft world is moving towards using these Extended INT13 functions. Probably a few years from now no modern system on modern hardware will need the concept of `disk geometry' anymore.
At most 65536 cylinders (numbered 0-65535), 16 heads (numbered 0-15), 255 sectors/track (numbered 1-255), for a maximum total capacity of 267386880 sectors (of 512 bytes each), that is, 136902082560 bytes (137 GB). This is not yet a problem (in 1999), but will be a few years from now.
At most 1024 cylinders (numbered 0-1023), 256 heads (numbered 0-255), 63 sectors/track (numbered 1-63) for a maximum total capacity of 8455716864 bytes (8.5 GB). This is a serious limitation today. It means that DOS cannot use present day large disks.
If the same values for c,h,s are used for the BIOS Int 13 call and for the IDE disk I/O, then both limitations combine, and one can use at most 1024 cylinders, 16 heads, 63 sectors/track, for a maximum total capacity of 528482304 bytes (528MB), the infamous 504 MiB limit for DOS with an old BIOS. This started being a problem around 1993, and people resorted to all kinds of trickery, both in hardware (LBA), in firmware (translating BIOS), and in software (disk managers). The concept of `translation' was invented (1994): a BIOS could use one geometry while talking to the drive, and another, fake, geometry while talking to DOS, and translate between the two.
Some older BIOSes only allocate 12 bits for the field in CMOS RAM that
gives the number of cylinders. Consequently, this number can be at most
4095, and only 4095*
16*
63*
512=2113413120 bytes are accessible.
The effect of having a larger disk would be a hang at boot time.
This made disks with geometry 4092/16/63 rather popular. And still today
many large disk drives come with a jumper to make them appear 4092/16/63.
See also
over2gb.htm.
Other BIOSes
would not hang but just detect a much smaller disk, like 429 MB instead of 2.5 GB.
There was a bug in the Phoenix 4.03 and 4.04 BIOS firmware that would cause the system to lock up in the CMOS setup for drives with a capacity over 3277 MB. See over3gb.htm.
Simple BIOS translation (ECHS=Extended CHS, sometimes called `Large
disk support' or just `Large')
works by repeatedly doubling the number of heads and halving the number
of cylinders shown to DOS, until the number of cylinders is at most 1024.
Now DOS and Windows 95 cannot handle 256 heads,
and in the common case that the disk reports 16 heads, this means that
this simple mechanism only works up to 8192*
16*
63*
512=4227858432
bytes (with a fake geometry with 1024 cylinders, 128 heads, 63 sectors/track).
Note that ECHS does not change the number of sectors per track, so if
that is not 63, the limit will be lower.
See
over4gb.htm.
Slightly smarter BIOSes avoid the previous problem by first adjusting the
number of heads to 15 (`revised ECHS'), so that a fake geometry with
240 heads can be obtained, good for
1024*
240*
63*
512=7927234560 bytes.
Finally, if the BIOS does all it can to make this translation a success,
and uses 255 heads and 63 sectors/track (`assisted LBA' or just `LBA')
it may reach 1024*
255*
63*
512=8422686720 bytes, slightly less
than the earlier 8.5 GB limit because the geometries with 256 heads must be
avoided.
(This translation will use for the number of heads the first value H
in the sequence 16, 32, 64, 128, 255 for which the total disk capacity
fits in 1024*
H*
63*
512, and then computes the number of
cylinders C as total capacity divided by (H*
63*
512).)
The next hurdle comes with a size over 33.8 GB. The problem is that with the default 16 heads and 63 sectors/track this corresponds to a number of cylinders of more than 65535, which does not fit into a short. Most BIOSes in existence today can't handle such disks. (See, e.g., Asus upgrades for new flash images that work.) Linux kernels older than 2.2.14 / 2.3.21 need a patch. See IDE problems with 34+ GB disks below.
For another discussion of this topic, see Breaking the Barriers, and, with more details, IDE Hard Drive Capacity Barriers.
Hard drives over 8.4 GB are supposed to report their geometry as 16383/16/63. This in effect means that the `geometry' is obsolete, and the total disk size can no longer be computed from the geometry.
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 or CDROM), 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.
Thus: create a small partition, say 10 MB large, so that there
is room for a handful of kernels, making sure that it is entirely
contained within the first 1024 cylinders of the first or second
disk. Mount it on /boot
so that LILO will put its stuff there.
Another point is that the boot loader and the BIOS must agree
as to the disk geometry. LILO asks the kernel for the geometry,
but more and more authors of disk drivers follow the bad habit
of deriving a geometry from the partition table, instead of
telling LILO what the BIOS will use. Thus, often the geometry
supplied by the kernel is worthless. In such cases it helps
to give LILO the `linear
' option. The effect of this is that
LILO does not need geometry information at boot loader install
time (it stores linear addresses in the maps) but does the conversion
of linear addresses at boot time. Why is this not the default?
Well, there is one disadvantage: with the `linear' option, LILO
no longer knows about cylinder numbers, and hence cannot warn you
when part of the kernel was stored above the 1024 cylinder limit,
and you may end up with a system that does not boot.
With LILO versions below v21 there is another disadvantage: the address conversion done at boot time has a bug: when c*H is 65536 or more, overflow occurs in the computation. For H larger than 64 this causes a stricter limit on c than the well-known c < 1024; for example, with H=255 and an old LILO one must have c < 258. (c=cylinder where kernel image lives, H=number of heads of disk)
Tim Williams writes: `I had my Linux partition within the first 1024 cylinders and still it wouldnt boot. First when I moved it below 1 GB did things work.' How can that be? Well, this was a SCSI disk with AHA2940UW controller which uses either H=64, S=32 (that is, cylinders of 1 MiB = 1.05 MB), or H=255, S=63 (that is, cylinders of 8.2 MB), depending on setup options in firmware and BIOS. No doubt the BIOS assumed the former, so that the 1024 cylinder limit was found at 1 GiB, while Linux used the latter and LILO thought that this limit was at 8.4 GB.
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).
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 TiB. That is sixty times
larger than the disks available today, so maybe it will be
enough for the next seven years or so.
(So, partitions can be very large, but there is a serious
restriction in that a file in an ext2 filesystem on hardware
with 32-bit integers cannot be larger than 2 GiB.)
DOS uses the begin
and end
fields, and uses the
BIOS INT13 call to access the disk, and can therefore only
handle disks of not more than 8.4 GB, even with a translating
BIOS. (Partitions cannot be larger than 2.1 GB because of
restrictions of the FAT16 file system.) The same holds for
Windows 3.11 and WfWG and Windows NT 3.*.
Windows 95 has support for the Extended INT13 interface, and
uses special partition types (c, e, f instead of b, 6, 5)
to indicate that a partition should be accessed in this way.
When these partition types are used, the begin
and end
fields
contain dummy information (1023/255/63).
Windows 95 OSR2 introduces the FAT32 file system (partition type
b or c), that allows partitions of size at most 2 TiB.
What is this nonsense you get from fdisk
about `overlapping'
partitions, when in fact nothing is wrong?
Well - there is something `wrong': if you look at the begin
and end
fields of such partitions, as DOS does, they overlap.
(And that cannot be corrected, because these fields cannot store
cylinder numbers above 1024 - there will always be `overlap'
as soon as you have more than 1024 cylinders.)
However, if you look at the start
and length
fields,
as Linux does, and as Windows 95 does in the case of partitions
with partition type c, e or f, then all is well.
So, ignore these warnings when cfdisk
is satisfied and you
have a Linux-only disk. Be careful when the disk is shared with DOS.
Use the commands cfdisk -Ps /dev/hdx
and cfdisk -Pt /dev/hdx
to look at the partition table of /dev/hdx
.
Disk geometry (with heads, cylinders and tracks) is something from the age of MFM and RLL. In those days it corresponded to a physical reality. Nowadays, with IDE or SCSI, nobody is interested in what the `real' geometry of a disk is. Indeed, the number of sectors per track 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. Quite the contrary: the IDE command INITIALIZE DRIVE PARAMETERS (91h) serves to tell the disk how many heads and sectors per track it is supposed to have today. It is quite normal to see a large modern disk that has 2 heads report 15 or 16 heads to the BIOS, while the BIOS may again report 255 heads to user software.
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 firmware to find out where a given sector lives on the disk. This linear numbering is called LBA.
So now the conceptual picture is the following. DOS, or some boot loader, talks to the BIOS, using (c,h,s) notation. The BIOS converts (c,h,s) to LBA notation using the fake geometry that the user is using. If the disk accepts LBA then this value is used for disk I/O. Otherwise, it is converted back to (c',h',s') using the geometry that the disk uses today, and that is used for disk I/O.
Note that there is a bit of confusion in the use of the expression `LBA': As a term describing disk capabilities it means `Linear Block Addressing' (as opposed to CHS Addressing). As a term in the BIOS Setup, it describes a translation scheme sometimes called `assisted LBA' - see above under ` The 8.4 GB limit '.
Something similar works when the firmware 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 8.4 GB (7.8 GiB).
(The third setup option usually is `Normal', where no translation is involved.)
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.
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.
If you are hit by this, and know someone who can compile a new
kernel for you, find the file linux/drivers/block/ide.c
and remove in the routine ide_xlate_1024()
the test
if (drive->forced_geom) { ...; return 0; }
.
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.
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
.
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.
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.
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.
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 sector number 1 and 63, respectively,
and end heads 31, 63, 127 or 254, 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, 128 or 255 heads, respectively.
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).
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
in /etc/lilo.conf
(see lilo.conf(5)).
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 `hda=
cyls,
heads,
secs'
kernel command line option (see bootparam(7)), perhaps by hand, or by
asking the boot loader to supply such an option to the kernel.
For example, one can tell LILO to supply such an option by adding
an `append = "hda=
cyls,
heads,
secs"
'
line in /etc/lilo.conf
(see lilo.conf(5)).
And otherwise the kernel will guess, possibly using values
obtained from the BIOS or the hardware.
It is possible (since Linux 2.1.79) to change the kernel's ideas
about the geometry by using the /proc
filesystem.
For example
# sfdisk -g /dev/hdc
/dev/hdc: 4441 cylinders, 255 heads, 63 sectors/track
# cd /proc/ide/ide1/hdc
# echo bios_cyl:17418 bios_head:128 bios_sect:32 > settings
# sfdisk -g /dev/hdc
/dev/hdc: 17418 cylinders, 128 heads, 32 sectors/track
#
Sometimes it is useful to force a certain geometry
by adding `hda=
cyls,
heads,
secs'
on the kernel command line. Almost always one wants secs=63,
and the purpose of adding this is to specify heads.
(Reasonable values today are heads=16 and heads=255.)
What should one specify for cyls? Precisely that number
that will give the right total capacity of C*H*S sectors.
For example, for a drive with 71346240 sectors (36529274880 bytes)
one would compute C as 71346240/(255*63)=4441 (for example using
the program bc
), and give boot parameter hdc=4441,255,63
.
How does one know the right total capacity? For example,
# hdparm -g /dev/hdc | grep sectors
geometry = 4441/255/63, sectors = 71346240, start = 0
# hdparm -i /dev/hdc | grep LBAsects
CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=71346240
gives two ways of finding the total number of sectors 71346240.
The kernel output
# dmesg | grep hdc
...
hdc: Maxtor 93652U8, 34837MB w/2048kB Cache, CHS=70780/16/63
hdc: [PTBL] [4441/255/63] hdc1 hdc2 hdc3! hdc4 < hdc5 > ...
tells us about (at least) 34837*2048=71346176 and about (at least)
70780*16*63=71346240 sectors. In this case the second value happens
to be precisely correct, but in general both may be rounded down.
This is a good way to approximate the disk size when hdparm
is unavailable. Never give a too large value for cyls!
In the case of SCSI disks the precise number of sectors is given
in the kernel boot messages:
SCSI device sda: hdwr sector= 512 bytes. Sectors= 17755792 [8669 MB] [8.7 GB]
(and MB, GB are rounded, not rounded down, and `binary').
The IDE driver has five sources of 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 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.
Note that G_bios is rather unreliable: for systems booting from SCSI the first and second disk may well be SCSI disks, and the geometry that the BIOS reported for sda is used by the kernel for hda. Moreover, disks that are not mentioned in the BIOS Setup are not seen by the BIOS. This means that, e.g., in an IDE-only system where hdb is not given in the Setup, the geometries reported by the BIOS for the first and second disk will apply to hda and hdc.
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 MiB 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 GiB. 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 GiB 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.
H=64, S=32.
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.
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.
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".
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. In Linux 2.0.36 this extended translation would always be set in case no SEEPROM was found, but in Linux 2.2.6 if no SEEPROM is found extended translation is set only when the user asked for it using this boot parameter (while when a SEEPROM is found, the boot parameter is ignored). This means that a setup that works under 2.0.36 may fail to boot with 2.2.6 (and require the `linear' keyword for LILO, or the `aic7xxx=extended' kernel boot parameter).
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".
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 GiB),
H=128, S=63 for M < 63*
2^17 (3.9 GiB) and H=255, S=63 otherwise.
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.
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 GiB). This is a bug.
One of three mappings ((H,S) = (16,63), (64,32), (64,63)) is used depending on the controller mapping mode.
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)
,
then assume that this partition really was aligned on a cylinder
boundary, and put H = *
endSendH+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.
The Linux IDE driver gets the geometry and capacity of a disk
(and lots of other stuff) by using an ATA IDENTIFY request.
Until recently the driver would not believe the returned value
of lba_capacity if it was more than 10% larger than the capacity
computed by C*
H*
S. However, by industry agreement
large IDE disks (with more than 16514064 sectors)
return C=16383, H=16, S=63, for a total of 16514064 sectors (7.8 GB)
independent of their actual size, but give their actual size in
lba_capacity.
Recent Linux kernels (2.0.34, 2.1.90) know about this
and do the right thing. If you have an older Linux kernel and do
not want to upgrade, and this kernel only sees 8 GiB of a much larger disk,
then try changing the routine lba_capacity_is_ok
in
/usr/src/linux/drivers/block/ide.c
into something like
static int lba_capacity_is_ok (struct hd_driveid *id) {
id->cyls = id->lba_capacity / (id->heads * id->sectors);
return 1;
}
For a more cautious patch, see 2.1.90.
As just mentioned, large disks return the geometry C=16383, H=16, S=63 independent of the actual size, while the actual size is returned in the value of LBAcapacity. Some BIOSes do not recognize this, and translate this 16383/16/63 into something with fewer cylinders and more heads, for example 1024/255/63 or 1027/255/63. So, the kernel must not only recognize the single geometry 16383/16/63, but also all BIOS-mangled versions of it. Since 2.2.2 this is done correctly (by taking the BIOS idea of H and S, and computing C = capacity/(H*S)). Usually this problem is solved by setting the disk to Normal in the BIOS setup (or, even better, to None, not mentioning it at all to the BIOS). If that is impossible because you have to boot from it or use it also with DOS/Windows, and upgrading to 2.2.2 or later is not an option, use kernel boot parameters.
If a BIOS reports 16320/16/63, then this is usually done in order to get 1024/255/63 after translation.
There is an additional problem here. If the disk was partitioned
using a geometry translation, then the kernel may at boot time
see this geometry used in the partition table, and report
hda: [PTBL] [1027/255/63]
. This is bad, because now the
disk is only 8.4 GB. This was fixed in 2.3.21. Again, kernel
boot parameters will help.
Many disks have jumpers that allow you to choose between a 15-head an a 16-head geometry. The default settings will give you a 16-head disk. Sometimes both geometries address the same number of sectors, sometimes the 15-head version is smaller. There may be a good reason for this setup: Petri Kaukasoina writes: `A 10.1 Gig IBM Deskstar 16 GP (model IBM-DTTA-351010) was jumpered for 16 heads as default but this old PC (with AMI BIOS) didn't boot and I had to jumper it for 15 heads. hdparm -i tells RawCHS=16383/15/63 and LBAsects=19807200. I use 20960/15/63 to get the full capacity.' For the jumper settings, see http://www.storage.ibm.com/techsup/hddtech/hddtech.htm.
Many disks have jumpers that allow you to make the disk appear smaller than it is. A silly thing to do, and probably no Linux user ever wants to use this, but some BIOSes crash on big disks. The usual solution is to keep the disk entirely out of the BIOS setup. But this may be feasible only if the disk is not your boot disk.
The first serious limit was the 4096 cylinder limit (that is, with 16 heads and 63 sectors/track, 2.11 GB). For example, a Fujitsu MPB3032ATU 3.24 GB disk has default geometry 6704/15/63, but can be jumpered to appear as 4092/16/63, and then reports LBAcapacity 4124736 sectors, so that the operating system cannot guess that it is larger in reality. In such a case (with a BIOS that crashes if it hears how big the disk is in reality, so that the jumper is required) one needs boot parameters to tell Linux about the size of the disk.
That is unfortunate. Most disks can be jumpered so as to appear as a 2 GB disk and then report a clipped geometry like 4092/16/63 or 4096/16/63, but still report full LBAcapacity. Such disks will work well, and use full capacity under Linux, regardless of jumper settings.
A more recent limit is the 33.8 GB limit. Linux kernels older than 2.3.21 need a patch to be able to cope with IDE disks larger than this. Some disks larger than this limit can be jumpered to appear as a 33.8 GB disk. For example, the IBM Deskstar 37.5 GB (DPTA-353750) with 73261440 sectors (corresponding to 72680/16/63, or 4560/255/63) can be jumpered to appear as a 33.8 GB disk, and then reports geometry 16383/16/63 like any big disk, but LBAcapacity 66055248 (corresponding to 65531/16/63, or 4111/255/63). Unfortunately the jumper seems to be too effective - it not only influences what the drive reports to the system, but it also influences actual I/O: Petr Soucek reports that for this particular disk boot parameters do not help - with jumper present every access to sector 66055248 or more gives an I/O error. Thus, on a motherboard with Award 4.51PG BIOS one cannot use this disk as boot disk and also use the disk to full capacity. See also the BIOS 33.8 GB limit.
The HDIO_GETGEO
ioctl returns the number of cylinders in a short.
This means that if you have more than 65535 cylinders, the number is
truncated, and (for a typical SCSI setup with 1 MiB cylinders)
a 80 GiB disk may appear as a 16 GiB one.
Once one recognizes what the problem is, it is easily avoided.
Drives larger than 33.8 GB will not work with kernels older than 2.3.21. The details are as follows. Suppose you bought a new IBM-DPTA-373420 disk with a capacity of 66835440 sectors (34.2 GB). Pre-2.3.21 kernels will tell you that the size is 769*16*63 = 775152 sectors (0.4 GB), which is a bit disappointing. And giving command line parameters hdc=4160,255,63 doesn't help at all - these are just ignored. What happens? The routine idedisk_setup() retrieves the geometry reported by the disk (which is 16383/16/63) and overwrites what the user specified on the command line, so that the user data is used only for the BIOS geometry. The routine current_capacity() or idedisk_capacity() recomputes the cylinder number as 66835440/(16*63)=66305, but since this is stored in a short, it becomes 769. Since lba_capacity_is_ok() destroyed id->cyls, every following call to it will return false, so that the disk capacity becomes 769*16*63. For several kernels a patch is available. A patch for 2.0.38 can be found at ftp.kernel.org. A patch for 2.2.12 can be found at www.uwsg.indiana.edu (some editing may be required to get rid of the html markup). The 2.2.14 kernels do support these disks. In the 2.3.* kernel series, there is support for these disks since 2.3.21. One can also `solve' the problem in hardware by using a jumper to clip the size to 33.8 GB. In many cases a BIOS upgrade will be required if one wants to boot from the disk.
Above, we saw the structure of the MBR (sector 0): boot loader code followed by 4 partition table entries of 16 bytes each, followed by an AA55 signature. Partition table entries of type 5 or F or 85 (hex) have a special significance: they describe extended partitions: blobs of space that are further partitioned into logical partitions. (So, an extended partition is only a box, it cannot be used itself, one uses the logical partitions inside.) Only the location of the first sector of an extended partition is important. This first sector contains a partition table with four entries: one a logical partition, one an extended partition, and two unused. In this way one gets a chain of partition table sectors, scattered over the disk, where the first one describes three primary partitions and the extended partition, and each following partition table sector describes one logical partition and the location of the next partition table sector.
It is important to understand this: When people do something stupid while partitioning a disk, they want to know: Is my data still there? And the answer is usually: Yes. But if logical partitions were created then the partition table sectors describing them are written at the beginning of these logical partitions, and data that was there before is lost.
The program sfdisk will show the full chain. E.g.,
# sfdisk -l -x /dev/hda
Disk /dev/hda: 16 heads, 63 sectors, 33483 cylinders
Units = cylinders of 516096 bytes, blocks of 1024 bytes, counting from 0
Device Boot Start End #cyls #blocks Id System
/dev/hda1 0+ 101 102- 51376+ 83 Linux
/dev/hda2 102 2133 2032 1024128 83 Linux
/dev/hda3 2134 33482 31349 15799896 5 Extended
/dev/hda4 0 - 0 0 0 Empty
/dev/hda5 2134+ 6197 4064- 2048224+ 83 Linux
- 6198 10261 4064 2048256 5 Extended
- 2134 2133 0 0 0 Empty
- 2134 2133 0 0 0 Empty
/dev/hda6 6198+ 10261 4064- 2048224+ 83 Linux
- 10262 16357 6096 3072384 5 Extended
- 6198 6197 0 0 0 Empty
- 6198 6197 0 0 0 Empty
...
/dev/hda10 30581+ 33482 2902- 1462576+ 83 Linux
- 30581 30580 0 0 0 Empty
- 30581 30580 0 0 0 Empty
- 30581 30580 0 0 0 Empty
#
It is possible to construct bad partition tables.
Many kernels get into a loop if some extended partition points back
to itself or to an earlier partition in the chain.
It is possible to have two extended partitions in one of these
partition table sectors so that the partition table chain forks.
(This can happen for example with an fdisk that does not recognize
each of 5, F, 85 as an extended partition, and creates a 5 next to an F.)
No standard fdisk type program can handle such situations, and some
handwork is required to repair them.
The Linux kernel will accept a fork at the outermost level.
That is, you can have two chains of logical partitions.
Sometimes this is useful - for example, one can use type 5 and be
seen by DOS, and the other type 85, invisible for DOS, so that
DOS FDISK will not crash because of logical partitions past cylinder 1024.
Usually one needs sfdisk
to create such a setup.
Many people think they have problems, while in fact nothing is wrong. Or, they think that the problems they have are due to disk geometry, while in fact disk geometry has nothing to do with the matter. All of the above may have sounded complicated, but disk geometry handling is extremely easy: do nothing at all, and all is fine; or perhaps give LILO the keyword `linear' if it doesn't get past `LI' when booting. Watch the kernel boot messages, and remember: the more you fiddle with geometries (specifying heads and cylinders to LILO and fdisk and on the kernel command line) the less likely it is that things will work. Roughly speaking, all is fine by default.
And remember: nowhere in Linux is disk geometry used, so no problem you have while running Linux can be caused by disk geometry. Indeed, disk geometry is used only by LILO and by fdisk. So, if LILO fails to boot the kernel, that may be a geometry problem. If different operating systems do not understand the partition table, that may be a geometry problem. Nothing else. In particular, if mount doesnt seem to work, never worry about disk geometry - the problem is elsewhere.
It is quite possible that a disk gets the wrong geometry. The Linux kernel asks the BIOS about hd0 and hd1 (the BIOS drives numbered 80H and 81H) and assumes that this data is for hda and hdb. But on a system that boots from SCSI, the first two disks may well be SCSI disks, and thus it may happen that the fifth disk, which is the first IDE disk hda, gets assigned a geometry belonging to sda. Such things are easily solved by giving boot parameters `hda=C,H,S' for the appropriate numbers C, H and S, either at boot time or in /etc/lilo.conf.
`I have two identical 10 GB IBM disks. However, fdisk gives different sizes for them. Look:
# fdisk -l /dev/hdb
Disk /dev/hdb: 255 heads, 63 sectors, 1232 cylinders
Units = cylinders of 16065 * 512 bytes
Device Boot Start End Blocks Id System
/dev/hdb1 1 1232 9896008+ 83 Linux native
# fdisk -l /dev/hdd
Disk /dev/hdd: 16 heads, 63 sectors, 19650 cylinders
Units = cylinders of 1008 * 512 bytes
Device Boot Start End Blocks Id System
/dev/hdd1 1 19650 9903568+ 83 Linux native
How come?'
What is happening here? Well, first of all these drives
really are 10gig: hdb has size 255*
63*
1232*
512 = 10133544960,
and hdd has size 16*
63*
19650*
512 = 10141286400, so, nothing
is wrong and the kernel sees both as 10.1 GB.
Why the difference in size? That is because the kernel gets
data for the first two IDE disks from the BIOS, and the BIOS
has remapped hdb to have 255 heads (and 16*
19650/255=1232 cylinders).
The rounding down here costs almost 8 MB.
If you would like to remap hdd in the same way, give the kernel boot parameters `hdd=1232,255,63'.
fdisk will tell you how many blocks there are on the disk. If you make a filesystem on the disk, say with mke2fs, then this filesystem needs some space for bookkeeping - typically something like 4% of the filesystem size, more if you ask for a lot of inodes during mke2fs. For example:
# sfdisk -s /dev/hda9
4095976
# mke2fs -i 1024 /dev/hda9
mke2fs 1.12, 9-Jul-98 for EXT2 FS 0.5b, 95/08/09
...
204798 blocks (5.00%) reserved for the super user
...
# mount /dev/hda9 /somewhere
# df /somewhere
Filesystem 1024-blocks Used Available Capacity Mounted on
/dev/hda9 3574475 13 3369664 0% /mnt
# df -i /somewhere
Filesystem Inodes IUsed IFree %IUsed Mounted on
/dev/hda9 4096000 11 4095989 0% /mnt
#
We have a partition with 4095976 blocks, make an ext2 filesystem
on it, mount it somewhere and find that it only has 3574475 blocks -
521501 blocks (12%) was lost to inodes and other bookkeeping.
Note that the difference between the total 3574475 and the 3369664
available to the user are the 13 blocks in use plus the 204798
blocks reserved for root. This latter number can be changed by tune2fs.
This `-i 1024' is only reasonable for news spools and the like,
with lots and lots of small files. The default would be:
# mke2fs /dev/hda9
# mount /dev/hda9 /somewhere
# df /somewhere
Filesystem 1024-blocks Used Available Capacity Mounted on
/dev/hda9 3958475 13 3753664 0% /mnt
# df -i /somewhere
Filesystem Inodes IUsed IFree %IUsed Mounted on
/dev/hda9 1024000 11 1023989 0% /mnt
#
Now only 137501 blocks (3.3%) are used for inodes, so that we have
384 MB more than before. (Apparently, each inode takes 128 bytes.)
On the other hand, this filesystem can have at most 1024000 files
(more than enough), against 4096000 (too much) earlier.