blob: 94848734ac66399379ea5da7991de2e7f1465253 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001$Id: README.aha152x,v 1.2 1999/12/25 15:32:30 fischer Exp fischer $
2Adaptec AHA-1520/1522 SCSI driver for Linux (aha152x)
3
John Anthony Kazos Jrbe2a6082007-05-09 08:50:42 +02004Copyright 1993-1999 Jürgen Fischer <fischer@norbit.de>
Linus Torvalds1da177e2005-04-16 15:20:36 -07005TC1550 patches by Luuk van Dijk (ldz@xs4all.nl)
6
7
8In Revision 2 the driver was modified a lot (especially the
9bottom-half handler complete()).
10
11The driver is much cleaner now, has support for the new
12error handling code in 2.3, produced less cpu load (much
13less polling loops), has slightly higher throughput (at
14least on my ancient test box; a i486/33Mhz/20MB).
15
16
17CONFIGURATION ARGUMENTS:
18
19IOPORT base io address (0x340/0x140)
20IRQ interrupt level (9-12; default 11)
21SCSI_ID scsi id of controller (0-7; default 7)
22RECONNECT allow targets to disconnect from the bus (0/1; default 1 [on])
23PARITY enable parity checking (0/1; default 1 [on])
24SYNCHRONOUS enable synchronous transfers (0/1; default 1 [on])
25DELAY: bus reset delay (default 100)
26EXT_TRANS: enable extended translation (0/1: default 0 [off])
27 (see NOTES)
28
29COMPILE TIME CONFIGURATION (go into AHA152X in drivers/scsi/Makefile):
30
31-DAUTOCONF
32 use configuration the controller reports (AHA-152x only)
33
34-DSKIP_BIOSTEST
35 Don't test for BIOS signature (AHA-1510 or disabled BIOS)
36
37-DSETUP0="{ IOPORT, IRQ, SCSI_ID, RECONNECT, PARITY, SYNCHRONOUS, DELAY, EXT_TRANS }"
38 override for the first controller
39
40-DSETUP1="{ IOPORT, IRQ, SCSI_ID, RECONNECT, PARITY, SYNCHRONOUS, DELAY, EXT_TRANS }"
41 override for the second controller
42
43-DAHA152X_DEBUG
44 enable debugging output
45
46-DAHA152X_STAT
47 enable some statistics
48
49
50LILO COMMAND LINE OPTIONS:
51
52aha152x=<IOPORT>[,<IRQ>[,<SCSI-ID>[,<RECONNECT>[,<PARITY>[,<SYNCHRONOUS>[,<DELAY> [,<EXT_TRANS]]]]]]]
53
54 The normal configuration can be overridden by specifying a command line.
55 When you do this, the BIOS test is skipped. Entered values have to be
56 valid (known). Don't use values that aren't supported under normal
57 operation. If you think that you need other values: contact me.
58 For two controllers use the aha152x statement twice.
59
60
61SYMBOLS FOR MODULE CONFIGURATION:
62
63Choose from 2 alternatives:
64
651. specify everything (old)
66
67aha152x=IOPORT,IRQ,SCSI_ID,RECONNECT,PARITY,SYNCHRONOUS,DELAY,EXT_TRANS
68 configuration override for first controller
69
70
71aha152x1=IOPORT,IRQ,SCSI_ID,RECONNECT,PARITY,SYNCHRONOUS,DELAY,EXT_TRANS
72 configuration override for second controller
73
742. specify only what you need to (irq or io is required; new)
75
76io=IOPORT0[,IOPORT1]
77 IOPORT for first and second controller
78
79irq=IRQ0[,IRQ1]
80 IRQ for first and second controller
81
82scsiid=SCSIID0[,SCSIID1]
83 SCSIID for first and second controller
84
85reconnect=RECONNECT0[,RECONNECT1]
86 allow targets to disconnect for first and second controller
87
88parity=PAR0[PAR1]
89 use parity for first and second controller
90
91sync=SYNCHRONOUS0[,SYNCHRONOUS1]
92 enable synchronous transfers for first and second controller
93
94delay=DELAY0[,DELAY1]
95 reset DELAY for first and second controller
96
97exttrans=EXTTRANS0[,EXTTRANS1]
98 enable extended translation for first and second controller
99
100
101If you use both alternatives the first will be taken.
102
103
104NOTES ON EXT_TRANS:
105
106SCSI uses block numbers to address blocks/sectors on a device.
107The BIOS uses a cylinder/head/sector addressing scheme (C/H/S)
108scheme instead. DOS expects a BIOS or driver that understands this
109C/H/S addressing.
110
111The number of cylinders/heads/sectors is called geometry and is required
112as base for requests in C/H/S addressing. SCSI only knows about the
113total capacity of disks in blocks (sectors).
114
115Therefore the SCSI BIOS/DOS driver has to calculate a logical/virtual
116geometry just to be able to support that addressing scheme. The geometry
117returned by the SCSI BIOS is a pure calculation and has nothing to
118do with the real/physical geometry of the disk (which is usually
119irrelevant anyway).
120
121Basically this has no impact at all on Linux, because it also uses block
122instead of C/H/S addressing. Unfortunately C/H/S addressing is also used
123in the partition table and therefore every operating system has to know
124the right geometry to be able to interpret it.
125
126Moreover there are certain limitations to the C/H/S addressing scheme,
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300127namely the address space is limited to up to 255 heads, up to 63 sectors
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128and a maximum of 1023 cylinders.
129
130The AHA-1522 BIOS calculates the geometry by fixing the number of heads
131to 64, the number of sectors to 32 and by calculating the number of
132cylinders by dividing the capacity reported by the disk by 64*32 (1 MB).
133This is considered to be the default translation.
134
135With respect to the limit of 1023 cylinders using C/H/S you can only
136address the first GB of your disk in the partition table. Therefore
137BIOSes of some newer controllers based on the AIC-6260/6360 support
138extended translation. This means that the BIOS uses 255 for heads,
13963 for sectors and then divides the capacity of the disk by 255*63
140(about 8 MB), as soon it sees a disk greater than 1 GB. That results
141in a maximum of about 8 GB addressable diskspace in the partition table
142(but there are already bigger disks out there today).
143
144To make it even more complicated the translation mode might/might
145not be configurable in certain BIOS setups.
146
147This driver does some more or less failsafe guessing to get the
148geometry right in most cases:
149
150- for disks<1GB: use default translation (C/32/64)
151
152- for disks>1GB:
153 - take current geometry from the partition table
154 (using scsicam_bios_param and accept only `valid' geometries,
155 ie. either (C/32/64) or (C/63/255)). This can be extended translation
156 even if it's not enabled in the driver.
157
158 - if that fails, take extended translation if enabled by override,
159 kernel or module parameter, otherwise take default translation and
160 ask the user for verification. This might on not yet partitioned
161 disks.
162
163
164REFERENCES USED:
165
166 "AIC-6260 SCSI Chip Specification", Adaptec Corporation.
167
168 "SCSI COMPUTER SYSTEM INTERFACE - 2 (SCSI-2)", X3T9.2/86-109 rev. 10h
169
170 "Writing a SCSI device driver for Linux", Rik Faith (faith@cs.unc.edu)
171
172 "Kernel Hacker's Guide", Michael K. Johnson (johnsonm@sunsite.unc.edu)
173
174 "Adaptec 1520/1522 User's Guide", Adaptec Corporation.
175
176 Michael K. Johnson (johnsonm@sunsite.unc.edu)
177
178 Drew Eckhardt (drew@cs.colorado.edu)
179
180 Eric Youngdale (eric@andante.org)
181
182 special thanks to Eric Youngdale for the free(!) supplying the
183 documentation on the chip.