A Overview

 Floppy discs are flat discs made from a similar material to cassette
tape, and are magnetised to store information. Floppy discs and disc
drives come in many different sizes and formats. The most popular sizes
now are 5.25 inch and 3.5 inch, but 8 inch discs are still available.

 There are three things which determine how much information a disc can
store; the number of tracks supported by the disc drive, whether the disc
drive is single or double sided, and the recording density.

 Information is written on to floppy discs in a series of concentric
bands, or tracks. Most disc drives will be able to read and write 80 of
these tracks, but some older drives may only support 40 tracks. An 80
track disc will store twice as much information as a 40 track disc.

 Double sided disc drives have two read/write heads, one on each side of
the disc, and so are able to record twice as much information as a single
sided drive, which has only one read/write head. Note: single sided discs
cannot be inverted to use the reverse side of the disc, except for 3.5
inch discs.

 The recording density determines how much information can be stored on
each track of a disc. Single density recording allows 10 sectors of 256
bytes each to be recorded on a typical disc, whereas double density
recording allows up to 18 sectors per track.

 The Intel 8271 Floppy Disc Controller chip which HDFS is designed to work
with can only read and write single density discs. HDFS is designed to
work with single or double sided, 40 or 80 track disc drives. The amount
of information which can be stored on a disc is given in the table below.

		Sides:	Single	Double
	Tracks:	 40	 100k	 200k
		 80	 200k	 400k

B Installation

 When you have inserted the HDFS ROM and turned the machine on, you should
see a message like:

	BBC Computer 32K
	Hierarchical DFS
	BASIC
	>

 This message indicates that the HDFS is selected as the current filing
system. It is possible to have the DNFS installed as well as the HDFS . In
this case, the ROM in the higher priority socket will take control by
default. If the HDFS is in the lower priority socket, it may be selected
by the key combination H-BREAK. The DNFS may be selected by pressing
D-BREAK . The HDFS also allows commands to be prefixed by 'H', so it can
also be selected unambiguously by the command *HDISC.

 There are several link switches at the bottom right hand corner of the
keyboard PCB; some of these can be used to set options to HDFS. Links 3
and 4 are used to set the disc drive characteristics, and links 1 and 2
are used to set the default stepping configuration. The disc drive timing
link positions are:

	 Link Step Settle Head
	 3 4  time  time  load
	 1 1   4     16     0
	 1 0   6     16     0
	 0 1   6     50    32
	 0 0  24     20    64

 The drive stepping configuration links are:
	 Link
	 1 2 Action
	 1 0 Double stepping enabled
	 0 1 Single stepping enabled
	 0 0 Automatic double stepping enabled

C Command Summary
 This section contains a list of the HDFS commands and their syntax.

 ACCESS afsp (LXWR)
 ATTRIB fsp (load exec) (len) (LXWR)
 BUILD fsp (A)
 CLOSE
 COMPACT (dsp) (D)
 COPY <fsp fsp/afsp dsp> (C)
 CREATE fsp (load exec) (len) (LXWR) (D)
 DEDIT (drv) start (+) ((+) (R)
 DELETE fsp
 DFIND (drv) start (+) <string/{hex}>
 DIR (dsp)
 DISC
 DISK
 DRIVE drv (S/D/A) (C)
 DUMP fsp
 ENABLE
 EXTEND fsp
 FEDIT fsp (offs) (R)
 FFIND fsp (offs) <string/{hex}>
 FORMAT drv <40/80> <X/A> <S/D> (skew)
 FREE (dsp)
 GOIO addr
 INFO afsp
 KEYS
 LIB (dsp)
 MEDIT start (+) ((+) (R)
 MFIND start (+) <string/{hex}>
 MODE dec
 MOVE start (+) addr
 READ (drv) addr start (+) (D)
 RENAME <fsp fsp/afsp dsp> (C)
 ROMS
 TITLE string (dsp)
 TYPE fsp (C)
 VDU <string/{((+/-) (,/;)
 VERIFY drv <40/80> <X/A> <S/D>
 WILD :$^.*?
 WIPE afsp (A)
 WRITE (drv) addr start (+) (D)


D 8271 commands

 This section describes the operations supported by the 8271 floppy disc
controller chip. These commands can be sent to the 8271 using the OSWORD
&7F call, where the operation code must have &40 added to it. Refer to the
8271 FDC data sheet for more information. Some of the commands in the
table below operate on multiple sectors; for these commands, the sectors
must all be on the same track. Other data transfer commands operate on
single sectors only. The commands supported by the 8271 FDC are:

 Command		Op   P1     P2      P3        P4          P5

Scan multi		00 Track  Sector  Len/Sec  Scan step  Field length

Scan deleted multi	04 Track  Sector  Len/Sec  Scan step  Field length

Write			0A Track  Sector

Write multi		0B Track  Sector  Len/Sec

Write deleted		0E Track  Sector

Write deleted multi	0F Track  Sector  Len/Sec

Read			12 Track  Sector

Read multi		13 Track  Sector  Len/Sec

Read deleted		16 Track  Sector

Read deleted multi	17 Track  Sector  Len/Sec

Read track ID		1B Track     0     NumID

Verify deleted		1E Track  Sector

Verify deleted multi	1F Track  Sector  Len/Sec

Format multi		23 Track  Gap 3   Len/Sec      Gap 5      Gap 1

Seek			29 Track

Read status		2C

Specify		35  What  Step/    Settle/     Load/
				  Track     Track      Track

Write register		3A  Reg.   Data

Read register		3D  Reg.

The parameters for these commands, and their meanings are:

Track		This is the track number (normal range 0-79)

Sector		This is the start sector on the track (normal range 0-9)

Len/Sec	Bits 0-4 indicate the number of sectors accessed (0-9) Bits
		5-7 are sector size (2 for 256 byte sectors)

NumID		This is the number of ID fields to transfer.

Scan step	Bits 6-7 indicate the scan condition (0 EQ, 1 GEQ, 2 LEQ)
		Bits 0-5 are the offset to the next sector in a multi scan.

Field length	This is the number of bytes to compare when scanning.

Gap 1		This is the formatting gap between the last sector and the
		index mark.

Gap 3		This is the syncronisation field formatting gap.

Gap 5		This is the formatting gap between the ID field and sector.

What		This value indicates what the other parameters in the
		command mean. The values &10 and &18 load surface zero and
		one bad tracks, and &0D sets the drive parameters.

Step		This is the disc step rate.

Settle		This is the disc head settle time. Load Bits 0-3 are the
		head load time, Bits 4-7 are the index unload count.
Reg.		This is the register number to read or write.

Data		This is the data to read or write to a register.

 The result byte returned by the 8271 contains the result value of the
command. Bit 5 of the result is a flag which indicates that deleted data
was found on the disc. Bits 1-4 contain a completion code which indicates
what the final status of the command was:

 00 Good completion or scan not met
 02 Scan met equal
 04 Scan met not equal
 08 Clock error
 0A Late DMA
 0C ID CRC error
 0E Data CRC error
 10 Drive not ready
 12 Write protect
 14 Track 0 not found
 16 Write fault
 18 Sector not found

 Note that the scan commands do not behave as specified in the 8271 data
sheet. The memory address of the scan key is not automatically reset each
time the scan repeats, and so the scan key should be duplicated throughout
a memory buffer the same size as the number of sectors being scanned. This
behaviour is the same under DNFS and HDFS.


E Catalogue Format

The format of HDFS and Acorn (DNFS) catalogues differ slightly. The
catalogue formats are similar enough that HDFS discs can be catalogued
using DNFS, but no changes should be made to them. There are two sectors
allocated to each catalogue:

Sector 0: Bytes   Data DNFS		HDFS

	     0	   Title char 1		b7: Sectors bit 10

	    1-7    Title chars 2-8

	     8	    Filename char 1	b7: Start sec bit 10
	     9	    Filename char 2	b7: Length bit 18
	     A	    Filename char 3
	     B	    Filename char 4	b7: Directory
	     C	    Filename char 5	b7: Not readable
	     D	    Filename char 6	b7: Not writable
	     E	    Filename char 7	b7: Not executable
	     F	    Directory char
		b7: Locked		b7: Not deletable

Sector 1: Bytes    Data DNFS		HDFS

	    0-3	    Title chars 9-12
	     4	    Cycle number	Key number
	     5	    Entries*8
	     6	    b0-1: Sectors bits 8-9
		    b2:   0		b2: Disc sides 1/2
		    b3:   0		b3: 1
		    b4-5: Auto-boot option
		    b6-7: 0
	     7	    Sectors bits 0-7

	    8-9	    Load address bits 0-15
	    A-B	    Exec address bits 0-15
	    C-D	    Length bits 0-15
	    E b0-1: Start sector bits 8-9
	      b2-3: Load address bits 16-17
	      b4-5: File length  bits 16-17
	      b6-7: Exec address bits 16-17
	    F	    Start sector bits 0-7


F Memory Usage

F.1 Zero page
 Several zero page areas are used by HDFS. The NMI workspace from &A0 to
&A7 is used when the NMI area is claimed for disc transfers. The operating
system scratch workspace from &AC to &AF is used for parsing commands.
The filing system scratch workspace from &B0 to &BF is used as the main
zero page workspace. The contents are used during HDFS command execution,
but are not held over between commands. The zero page filing system
workspace from &C0 to &CF is fully used while HDFS is active.

 This area should not be changed by user programs:

	C0-C3	Drive information table
	C4-C5	Directory start sector
	C6   	Copy of current drive
	C7	Last drive accessed
	C8-C9	Command line parameters
	CA	Last catalogue type
	CB	Root key of last drive
	CC	Tube flag
	CD	Message flag (*OPT 1 value)
	CE	Number of retries (*OPT 2 value)
	CF	Sideways ROM number (*OPT 3 value)

F.2 NMI workspace
 The NMI workspace in &D00-&D9E is claimed for disc and memory transfers.
The NMI allocation regime is non-greedy; the workspace is returned to its
previous owner as soon as possible. The zero page NMI workspace is also
used when the NMI workspace is claimed.

F.3 Absolute workspace
 The absolute workspace in &E00-&11FF is claimed for use by most
operations. HDFS is greedy about absolute workspace; it will not release
control of the absolute workspace until requested by another claimant.
Absolute workspace addresses &E00-&FFF are used to hold a copy of the
current catalogue. This copy is used instead of reading the catalogue from
the disc if the disc is still spinning when an operation is started. The
absolute workspace in &1000-&11FF is used for various purposes:

	1000-101F Scratch filename 1
	101F-103F Scratch filename 2
	1040-105F Current directory
	1060-107F Current library
	1080-1083 Copy of drive information table
	1084-1089 Character table
	108A-108F File handle table
	1090-109F Command workspace
	10A0-10DF String space
	10E0-10F5 File info workspace

	1100-1115 Open file info 1
	1116-112B Open file info 2
	112C-1141 Open file info 3
	1142-1157 Open file info 4
	1158-116D Open file info 5
	116E-1183 Open file info 6
	11A0-11AF OSGBPB workspace
	11B3-11FF Open file workspace

F.4 Private workspace
 Seven pages of private workspace are allocated; these will usually be
from &1200-&17FF . The first six pages are used to keep memory buffers for
open files. The last page of private workspace, which is usually in
addresses &1800-&18FF, is used for a copy of vital information which is
normally kept in the absolute workspace &1000-&11FF. This is used when the
filing system is shut down and re-activated. This memory page contains:

 00-1F Current directory
 20-3F Current library
 40-43 Drive table
 44-49 Character table
 4A-4F File handle table
 50-65 Open file info 1
 66-7B Open file info 2
 7C-91 Open file info 3
 92-A7 Open file info 4
 A8-BD Open file info 5
 BE-D3 Open file info 6
 D4    Private workspace validity flag


G Benchmarks

 The benchmarks in the tables below indicate how the performance of HDFS
compares to DNFS. There are four columns in each table; the first three
columns are results for HDFS operating on a native HDFS root directory, a
single level sub-directory, and an Acorn format directory . The final
column contains results for Acorn's DNFS 1.20. All of the times shown are
in seconds, most of them timed over 10 repetitions. All of the tests were
performed on discs formatted with a skew of 3.

G.1 SAVE 32K

In this test, &8000 bytes were saved to a file. Several variants of this
test were performed. In the tables below, 'IO' indicates that the data was
taken from the I/O processor, and 'Tube' indicates that the data was taken
from the second processor. The keyword 'Spun' indicates that the disc was
spinning when the test started, so the drive start-up time is not counted
in the timing. The word 'Stop' indicates that the disc was stopped when
the test was started, and the word 'New' indicates that a new file entry
was used, rather than writing over an old file.

		HDFS	HDFS	HDFS	DNFS
SAVE 32K	root	sub-dir	acorn	1.20

IO, Spun	4.52	4.66	3.83	3.84

IO, Stop	4.82	5.06	4.43	4.40

Tube, Spun	4.54	4.62	3.83	3.87

Tube, Stop	4.73	5.17	4.42	4.49

IO, Spun, New	4.39	4.81	3.84	3.72

IO, Stop, New	4.73	5.04	4.43	4.33

Tube, Spun, New 4.41	4.83	3.84	3.75

Tube, Stop, New 4.85	5.11	4.44	4.37


G.2 LOAD 32K
 This test measured the time taken to load &8000 bytes from a file.

		HDFS	HDFS	HDFS	DNFS
LOAD 32K	root	sub-dir	acorn	1.20

IO, Spun	3.44	4.03	3.44	3.43

IO, Stop	4.39	4.70	4.30	4.31

Tube, Spun	3.80	4.04	3.80	3.43

Tube, Stop	4.39	4.70	4.30	4.31

G.3 OPEN
 The time taken to open a file was measured by this test. There are three
variants on this test, "In" (open for input only) "Out" (open for output
only) and "Update" (open for input and output).

		HDFS	HDFS	HDFS	DNFS
OPEN		root	sub-dir	acorn	1.20

In, Spun	0.01	0.40	0.00	0.01

In, Stop	0.50	0.75	0.59	0.54

Out, Spun	0.39	0.79	0.39	0.19

Out, Stop	0.91	1.15	0.99	0.74

Update, Spun	0.01	0.39	0.01	0.00

Update, Stop	0.50	0.74	0.58	0.55

G.4 BPUT 32K
 This test measured the time taken to write &8000 bytes into a file, one
at a time. There is one variant on this test, where the second processor's
fast BPUT call was used.

		HDFS	HDFS	HDFS	DNFS
BPUT 32K	root	sub-dir	acorn	1.20

Out		54.32	54.36	54.33	55.02

Up		76.04	75.75	76.28	76.03

Fast, Out	28.80	29.23	28.42	29.41

Fast, Update	49.92	49.75	50.46	50.00

G.5 BGET 32K
 The time taken to read &8000 bytes from a file one at a time was measured
by this test.

		HDFS	HDFS	HDFS	DNFS
BGET 32K	root	sub-dir	acorn	1.20

Get		52.70	52.50	52.71	52.86

G.6 MOVE
 This test measured how fast the sequential file pointer can be moved.

		HDFS	HDFS	HDFS	DNFS
MOVE		root	sub-dir	acorn	1.20

Out		29.15	29.87	29.81	29.60

Up		29.80	29.87	29.80	29.26

G.7 GBPB
 There were two variants of this test, where &8000 bytes were read or
written to a file using the OSGBPB call.

		HDFS	HDFS	HDFS	DNFS
GBPB		root	sub-dir	acorn	1.20

Read, IO	29.11	29.17	29.12	29.24

Read, Tube	29.14	29.17	29.11	29.24

Write, IO	29.28	29.40	29.28	29.27

Write, Tube	29.27	29.39	29.27	29.28

G.8 SHIFT
 This test measured how fast the sequential file pointer could be moved
while performing file operations.

		HDFS	HDFS	HDFS	DNFS
MOVE2 		root	sub-dir	acorn	1.20

Read		0.40	0.48	0.40	0.39

Write		0.62	0.68	0.62	0.62