Night Owl 6

home *** CD-ROM | disk | FTP | other *** search

/ Night Owl 6 / Night Owl's Shareware - PDSI-006 - Night Owl Corp (1990).iso / 010a / hdts535a.zip / HDTEST.DOC < prev next >

Wrap

Text File | 1991-10-20 | 157KB | 2,662 lines

Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ HDTEST - Hard Disk Test/Repair Utility ______________________________________ Copyright (1984-1991), P. R. Fletcher _____________________________________ All Rights Reserved ___________________ Distributed by: Computer Counselors, Inc. 1515 West Montgomery Avenue Rosemont PA 19010 _______ ____|__ | (r) --| | |------------------- | ____|__ | Association of | | |_| Shareware |__| o | Professionals -----| | |--------------------- |___|___| MEMBER Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ TABLE OF CONTENTS _________________ User-Supported Software 1 Introduction 1 Getting Started - Installation 5 - Testing a Disk 5 Hardware and Software Requirements 7 Limitations and Restrictions 9 Command Line Options - HDCHEK 10 - HDTEST 13 What does HDTEST do? 16 Incremental Disk Testing 18 Using the On-Disk Pre-Boot 18 Using the HD programs with SpeedStor and Disk Manager 21 Special considerations with SpeedStor 22 Dealing with Bad Clusters 22 Running HDTEST from a .BAT file 24 Bugs and Other Unpleasant Subjects 25 Changes from Previous Versions 25 Appendix A - Summary of Flag Settings 31 Appendix B - A Hardware Tutorial 33 - Disk Error Reports 36 Appendix C - Example Program Runs 41 Registration Form 44 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ USER-SUPPORTED SOFTWARE _______________________ Much of the software which is available for your computer has, in the past, been produced and sold on the assumption that most computer users are crooks who will take any available opportunity to cheat software writers and vendors out of their rightful rewards. Such software was copy-protected within an inch of its (and your computer's) life and was priced at a level which still guaranteed its producers a profit even if four or five illegal copies were made for each one sold. Thankfully, prices of commercial software have (for the most part) come down somewhat, and copy-protection is becoming much less prevalent, but buying commercial software remains something of a lottery - you often cannot be sure that the software does what you want in the way you want it done until you have paid for it. User- supported software (Shareware) is different. You may have got a copy of this package free from a friend or a bulletin board or (for a nominal fee) from a users' group or software library. In any case, you have a perfectly legal working copy of the HDTEST package. You are encouraged to make as many copies of the package as you like and to distribute them to all your friends. If, after trying the program out, you find that it is useful and want to continue to keep and use it, you must register your copy by sending the registration fee ($35.00, unless you are already a registered user of DIREDIT, in which case it is $25.00) together with the completed registration form, to the address on the form. Only your conscience can make you do this ____ (although you are probably breaking the law if you don't), but bear in mind that the distributors of "user-supported" software are making the assumption that you are honest and willing to pay a reasonable price for a useful program, many commercial software vendors, on the other hand, would still apparently rather believe you are a crook. You can decide who you would rather prove right. Registered users are entitled to support, upgrades, and help with HDTEST-related problems (including telephone support) for a year from the date of registration. This program is produced by a member of the Association of Shareware Professionals (ASP). ASP wants to make sure that the shareware principle works for you. If you are unable to resolve a shareware- related problem with an ASP member by contacting the member directly, ASP may be able to help. The ASP Ombudsman can help you resolve a dispute or problem with an ASP member, but does not provide technical support for members' products. Please write to the ASP Ombudsman at 545 Grover Road, Muskegon, MI 49442-9427 or send a Compuserve message via CompuServe Mail to ASP Ombudsman 70007,3536" INTRODUCTION ____________ HDTEST was originally written to allow comprehensive Read/Write testing of hard disks on IBM PCs, XTs, ATs, and "true" compatibles. The V4.10 release removed many of the hardware restrictions imposed by earlier (V2.xx and V3.xx) versions of the program and made it useful for testing most MS-DOS-compatible storage devices currently in use on Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ IBM-compatible computers, including removable (floppy and hard) disks and logically partitioned hard disks. The V4.34 release included BIOS- level support for DOS 3.3 (and above) "extended devices" and more options for the handling of totally unreadable sectors. The V4.53 release added even more extensive support for IBM-incompatible hardware and removed the requirement that ANSI.SYS be loaded to support a proper screen display. Subsequent releases added full support for the large (> 32Mb) devices allowed by DOS 4.0, the ability to locate bad spots physically (by track, head, and sector) on the disk, and some support for the use of .BAT files to run HDTEST on multiple disks at one "sitting". The V5.20 release added INT 13 level support for hard disks with more than 1024 heads which have been partitioned by the SpeedStor or Disk Manager utilities, as well as full support for proprietary Disk Manager and SpeedStor partitions under DOS versions prior to 4.00 (or Compaq DOS 3.31). The V5.35 release included a number of changes and improvements, mainly aimed at allowing the programs to be used on even bigger partitions, but also including support for MS DOS 5.0 and (some) support for DR DOS 5.0. Although many programs, including IBM's diagnostics, allow "surface scanning" and seek testing of hard disks, few that are available to the average user test whether the whole disk can retain the data that is or will be written to it. HDTEST's principal "claim to fame" is the ability thoroughly to test the magnetic media in a device which contains data while preserving the data (although data loss is _________________________ unavoidable if an area which is in use has become unreadable). HDTEST ____ tests the disk much more thoroughly than programs such as DFixer, Mace, and Norton's DT. Mace and DT, like many public domain "disk test" programs, simply read every cluster on the disk and report on any errors encountered. DFixer does better, writing 4 test patterns to every free cluster and checking that they can be read back correctly, but it makes no attempt to test clusters that are already in use - the very ones which contain your precious data! Disk Technician (TM), a product which has come to market more recently, is more closely comparable to HDTEST than any other disk testing program that I know about. It is, however, much more expensive, is (functionally, at least) copy-protected, and is not available as ___ Shareware! In addition to its testing function, HDTEST has been reported to be useful as an automatic disk repair utility for both hard and floppy disks which have developed (especially) intermittent or persistent read errors, preventing or impeding their normal use. Although HDTEST cannot read the totally unreadable, it tries much harder that DOS does and will almost always, in the course of its operation, allow you to recover the entire readable portion(s) of a file with one or more ________ unreadable clusters in it, optionally either remapping all of what it reads from the disk or replacing bad sectors (the default) or clusters with NULLs. HDTEST is not an alternative to backing up your floppies ___ and/or hard disk - it may sometimes be able to save your neck when you _________ forgot or your backup went bad too! On hard disks which have developed intermittent read errors, running HDTEST can sometimes totally cure ___ Page 2 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ the problem, presumably because the "flaky" sectors get rewritten and the minor misalignments and/or weakening of the magnetic "image" which were causing the trouble are corrected when the data is rewritten. In its default ("slow") mode, HDTEST writes 20 different test patterns to every cluster on the disk and checks that each pattern can be read _____ back correctly, while preserving (and eventually restoring) the contents of the clusters that contain data. HDTEST can even (on many devices) detect hard disk errors which are minor enough to be correctable (and corrected) by the disk controller's own "error- correcting code" (ECC). These errors are normally totally invisible to programs running under DOS and will not usually be picked up by other disk test utilities. They may, however, be early signs of problems that will become significant later, and HDTEST consequently marks these clusters, too, as bad. As long as HDTEST can read the data from a flaky cluster correctly once (and it tries quite hard) it will be able to save the file of which that cluster is a part intact, while ______ marking the cluster so that DOS does not try to use it again. The program was designed to minimize the likelihood of user error causing data loss on a good (or slightly bad) disk, and to be as easy as possible to use - it could not be made very fast, since multiple bit patterns have to be written to and read from each cluster, and this takes a significant time even on a fast hard disk. The program does, however, have a less sensitive "fast" test mode, which writes only two test patterns to each block and is recommended for use when testing/repairing floppy disks and/or to correct intermittent problems or rule out gross problems on a hard disk. The program can be used to test/repair single and double-sided 8- and 9-sector minifloppies (if you have plenty of patience!), as well as ______ AT-format 1.2Mb minifloppies, 8" standard floppies, and 3.5" microfloppies (with the same proviso regarding a need for patience). HDTEST's companion program, HDCHEK, simply examines the logical structure of the system and target device and gives a short report on what it finds. It does not write to the disk, and is provided ___ principally to allow a quick and easy, but fairly reliable, check on the compatibility between HDTEST and a given software/hardware configuration. As part of its report, HDCHEK notes at what level (DOS interrupts only, INT 13, or direct entry to the BIOS) it was able to access the target device, which has a bearing on HDTEST's ability to detect certain types of errors. In general, if HDCHEK runs __________ successfully on a system and target device, then HDTEST can be expected to do so, although the latter program requires much more available memory. The HDTEST package is distributed as "User-supported Software". If you intend to continue using the program, you are required to become a Registered user. Send a completed copy of the Registration Form (which is included in this documentation), together with a check or money order for $35.00 (+ $5.00 for shipping outside the continent of North America) to me at the address which appears on the form. Please also add $2.50 (or equivalent) if you want the software on a 1.2 Mb HD Page 3 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ floppy or a 3.5" minifloppy, rather than a 360 Kb DSDD floppy, and specify which you want on the Reg. form. Residents of Pennsylvania should also add 6% Sales Tax to their order total. Checks and money orders in denominations other than U.S. dollars are acceptable, but please add a further $15.00 for bank charges before using the current exchange rate to calculate the fee in the currency in which you are paying. On receiving your Registration, I will send you a personalized copy of the latest, enhanced, version of the HDTEST package (currently V5.39), and register you for a year's support. Registration entitles you to be notified of upgrades, bug fixes, newly discovered incompati- bilities, etc., and to ask for (and get) necessary help in adapting the program to work on your hardware. I cannot guarantee to make HDTEST work with every possible combination of hardware and software, but I will promise to try to solve problems as they come up, and I will return your Registration fee if I cannot support HDTEST on your system. Support is available TO REGISTERED USERS ONLY by mail, ________________________ BIXmail, Compuserve EMail and telephone. Please also bear in mind that my willingness and ability to continue to support and enhance these programs will be directly dependent on users' willingness to pay the Registration fee and (to some extent) inversely dependent on the number of questions I get that could be answered by reading the docu- mentation(!). The HD programs were written principally in C, compiled by the Microsoft C compiler (Version 6.00A) using its "small" memory model, and linked with code from the distributed Microsoft C object libraries using the Microsoft Segmented Executable Linker (Version 5.13). Portions of the distributed program are consequently Copyright (C) by Microsoft Corp., 1985 - 1990. All rights reserved. These portions are used under the terms of a license from Microsoft Corp. Some low-level functions were written in assembly language and assembled by the Microsoft MASM Macro Assembler (Version 6.00). These programs were written with care and have been tested on a number of different systems. I cannot, however, give any guarantee as to their performance on a specific system, nor can I accept liability for any actual or consequential loss resulting from their use. The programs may only be used and/or copied in their original form. Unauthorized modifications and/or the use of all or part of the programs in other software, commercial or otherwise, are strictly prohibited. The HD programs owe a lot of their value to suggestions and bug reports (more of the former than the latter, fortunately) from Registered users. They were originally written in response to a "challenge" issued by Roedy Green on Bix, and their further development has also been significantly influenced by his input. I would also like to acknowledge the invaluable help of ltl, wlmoore, and mike123 on Bix in beta-testing V5.xx of the program. Page 4 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ GETTING STARTED - INSTALLATION ______________________________ 1) If you received the HDTEST package on a disk, use the DOS DISKCOPY program to make a duplicate of the distribution disk, put the original away in a safe place, and use the duplicate as your working disk. Otherwise, use COPY to save a copy of the HDTEST distribution files on a floppy disk, and put it away. 2) HDTEST does not require any special installation procedure. You may run HDCHEK and HDTEST from a working floppy disk or from any subdirectory on your hard disk. If you wish to keep and run the programs on your hard disk, you will probably find it most convenient to COPY them to a subdirectory which is one of those included in your DOS PATH (see your DOS Reference Manual). Only the two main .EXE files (HDTEST.EXE and HDCHEK.EXE) need normally be kept on-line. BOOTLOAD.COM is only used or needed to create a special boot disk, if you find you need to use one. 3) If you wish to be able to refer to this documentation on-line, you should follow the instructions in the README.1ST file in the distribution package to produce a .TXT file (without underlining) and keep that file in your documentation subdirectory. GETTING STARTED - TESTING A DISK ________________________________ This section gives step by step instructions for running HDCHEK and HDTEST, under IBM or "generic" MS-DOS, on the C: drive of a standard IBM PC, XT, AT, or 100% compatible. Subsequent sections of this documentation go into much more detail about some of the program's other features and options, and should help to explain and let you deal with any problems which you may encounter. Note that you should never attempt to run HDTEST under a multi-tasking or multi-user _____ operating system, nor activate any "pop-up" program (e.g. SideKick) while the program is running. HDTEST is also incompatible with programs (such as TIMEPARK) which are designed to move the heads of a hard disk to a safe (or new!) area of the disk after a period of inactivity. Such programs should be disabled or unloaded before running HDTEST. 1) Make sure that you have current backups of all the files on drive C:. This is particularly important when running the program for the first time on a system. 2) Run DOS's CHKDSK program on drive C: and correct any errors reported by it before proceeding further. 3) Make the drive and/or directory which contains HDCHEK.EXE and HDTEST.EXE your current drive and/or directory. Page 5 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ 4) If you are not running HDTEST from your A: drive, insert a formatted, non-write-protected floppy disk in drive A. HDTEST will write an ASCII log file called HDTESTC0.LOG to the disk in drive A - any other files on it will be undisturbed. 5) Give the command: "HDCHEK C:". After a certain amount of disk activity has taken place, and certainly within a minute or two, your screen should look like one of the examples given in Appendix C. If any errors are reported, DO NOT PROCEED until you have read the rest ______________ of the documentation and/or are sure you know what is going on. 6) If HDCHEK reported no errors, give the command "HDTEST C: A:". ____________________________ After some disk activity, the program will prompt for replies to three questions - press <Return> in response to each prompt. The program will report its progress to the screen and will record any problems encountered in the log file on the floppy disk. Depending on the size of your DOS partition and the access time of the disk, a complete run of HDTEST (in its normal, "slow", mode) will take between 2 and 5 hours. Your intervention will only be required if a recovered bad _____ cluster is found in a "hidden", "system", or write-protected file, when you will be asked to approve remapping; otherwise, bad clusters will be automatically remapped. HDTEST can be aborted at any time by pressing <Esc> or <Ctrl-Break> and responding to the "Are you sure...." prompt by pressing <Return>. 7) If HDTEST finishes testing without finding any bad clusters, it will simply record that fact and the finishing time in the log file and return to the DOS prompt. Otherwise, when read/write testing has been completed (or terminated, by entering <Esc> or <Ctrl-Break>), you will be prompted for permission to complete the remapping process by updating the disk FATs and Directories. Pressing <Return> will allow this to take place, and you will then be returned to the DOS prompt. 8) N.B. If HDTEST reported and remapped any bad clusters and you are ____ ___ using the DOS (3.3 and above) FASTOPEN utility, or any other directory- or file-cacheing software, you should reboot your system after running HDTEST and before proceeding with any other work on it. ______ ___ 9) In any event, if HDTEST reported and remapped some bad clusters, you should read the section on "Dealing with bad clusters" later in this documentation. Page 6 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ HARDWARE & SOFTWARE REQUIREMENTS ________________________________ HDTEST and HDCHEK make a number of assumptions about the environment in which they are operating. They will, generally speaking, abort with an informative error message if they find one of these assumptions to be untrue! They require MS/PC-DOS 2.x or above and (for HDTEST) up to about 320 Kb of free memory (the exact amount required depends on the size and other characteristics of the device being tested) to run - the presence of more memory may (slightly) increase HDTEST's speed when testing devices with large cluster sizes. The device to be tested must be usable as a normal DOS device - commands such as DIR, COPY, and CHKDSK should be valid for it and for files and subdirectories on it. It need not, however, be bootable. Only the DOS-accessible partition(s) of a multi-operating-system disk can be tested, but all the partitions of a disk that is divided into multiple DOS-accessible ______________ partitions can be tested (one by one). This version of HDTEST fully supports DOS 3.3 (and above) extended partitions, and will test them as thoroughly as it does normal DOS partitions. It also allows BIOS- level testing of the DOS-compatible "proprietary" partitions created by SpeedStor and Disk Manager. HDTEST's support for other disk partitioning software is provided only to the extent that DOS support is provided by the producers of that software. If you wish to use HDTEST to test a new hard disk which is to be ____ configured to support multiple operating systems, you should first configure it as one or more DOS-accessible partitions, test it in that configuration, and then reconfigure it as you intend to use it. If you __ have low-level disk formatting software which allows you to specify bad sectors and/or tracks to be "locked out", you can use HDTEST's /G option on the DOS partition(s) to produce a list of defects, which can then be used in reformatting the disk. Note that HDTEST can only be ____ used as a test for areas of the disk that are to be devoted to other ____ operating systems - when you repartition the disk, any remapping and/or marking of bad clusters which HDTEST may have done in the DOS partition is lost, unless you use information given by the /G option to redo the low-level format. _________ Ideally, for normal hard disks, the BIOS handler for interrupt 13 should be capable of handling (at least) normal disk reads and writes (sub-functions 2 and 3) in a way that is compatible with that documented for the XT hard disk adapter (and AT BIOS) in IBM's technical documentation. This should be no problem for most disk systems on IBM PCs and most compatibles. HDTEST may, however, be forced to use the (somewhat more limiting) DOS device interface when used with non-bootable hard disks and/or with hard disks which require special device drivers (other than those provided with the SpeedStor and Disk Manager partitioning software) to be installed, and may also be forced to use the DOS device interface when dealing with disks partitioned by means other that the DOS 3.3 (and above) FDISK utilities. Page 7 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ Early V4.xx releases of HDCHEK and HDTEST would abort if run on small (<= 10 Mb) hard disks which had been patched to use a cluster size smaller than the DOS standard (and 16-bit FATs). This deficiency was corrected in later V4.xx releases, but very small (less than 1024 clusters) partitions may still cause problems if 16-bit FATs are used (there is no good reason why such partitions should use 16-bit FATs). By default, if run under DOS 3.0 or above on an IBM INT 13 compatible hard disk, The HD programs attempt to locate and use the ROM BIOS INT 13 entry point, rather than the normal INT 13 vector (which DOS 3.x and above intercepts and interferes with). By default, for machines with 8088, 8086, 80188, 80186, or equivalent NEC CPUs the programs assume that, if a ROM is present at absolute address C8000H, it ___________________ contains the hard disk BIOS and has initialization code starting at ___ offset 3. This assumption is true for all hard disks which "plug in" to the IBM PC and XT. It may not be true for all clones which come with hard disks installed, although it is likely to be true for most of them. Also by default, the programs assume that the hard disk BIOS in systems with 80286 or 80386 processors is in the "system" ROM at absolute address F0000H, but make no more specific assumptions about __ its exact location. Problems (up to and including system halts) may occur if none of the special flags is set and your system is configured in a way that is not consistent with these assumptions. Such problems are almost guaranteed for PC or XT systems if the initialization code does not start at offset 3 in the ROM. If you know that your hard disk ROM is located somewhere other than at the default segment address and/or does not obey the "rules", you can use the /H flag to tell the programs where to look for it. Unless you know that your system is particularly non-IBM-compatible, however, it is usually worth trying to run the HD programs in their default mode first. Problem hardware should respond to one of the approaches described in the "Using the On-Disk Preboot" section of this documentation. The use of all the command line flags is discussed in more detail below. Note that, since HDTEST remaps bad clusters to free space, there is not much point in running it on a disk which has none (as shown by DIR or CHKDSK), since the program will abort immediately if it finds a bad cluster. Unless you have a really flaky disk, however, you should not need more than a couple of hundred Kb of free space to remap any reasonably probable number of new bad clusters. This version of the HD programs does not require an ANSI console driver (such as ANSI.SYS) to be loaded. The programs will, however, continue to function correctly if such a driver is present. Page 8 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ LIMITATIONS AND RESTRICTIONS ____________________________ HDTEST still has a few (mainly theoretical) limitations, most of which result from the way data is stored internally by the program. None of them are "etched in stone", although some could only be avoided at the cost of slowing down what is already (and necessarily) far from a fast program. The known limitations are as follows: 1) Maximum cluster size = 65,528 bytes. I do not believe that a DOS disk can ever have clusters larger than 64 Kb, but my limit is a few less than that and may, since sector and cluster sizes tend to run in powers of 2, effectively limit cluster size to 32 Kb - this should be enough for most people!. 2) Because of certain limitations of the chips used to support direct memory access (DMA, the method used to transfer data from disk to memory and vice versa) in the IBM PC and most of its early clones and compatibles, HDTEST may be unable to use available memory efficiently when cluster sizes (and the buffers needed to hold them) are large. This inefficiency can result in "Insufficient Memory" or "Memory Allocation" errors, even on systems with more than the minimum 320 Kb of free memory prescribed above. The only cure for this is to free up (or buy!) more memory (remove RAMdisks, remove resident software, etc.). Work is in progress on a version of HDTEST which can make more extensive use of LIM 4.0 expanded memory for its buffers. Contact me if you would like a preliminary copy of this version. 3) The current version of HDTEST will run, in its default mode, on ATs and most AT clones, but may not be able to detect corrected ECC errors when run in default mode on some of the clones - a warning message will be displayed when the program initializes itself if corrected ECC errors are not being detected. Appropriate use of the extended form of the /H flag or the /U flag, as described in the "Using the On-Disk Preboot" section of this documentation, will allow detection of corrected ECC errors on almost all hardware. The program will not ___ detect corrected ECC errors when invoked with the /I flag (under DOS 3.0 or above) or with the /D flag. 4) For reasons which are poorly understood (at least by the author!), HDTEST runs very slowly on Epson Equity II systems, particularly when ____ used to test "plug-in" hard drives of the "Hardcard" or "Flashcard" variety. For this hardware, the estimated run times given elsewhere in the documentation need to be multiplied by about 2.5. There is some __________ ___ reason to think that this problem may be related to an excessively "optimistic" interleave setting on (at least) some of these systems. 5) HDTEST or HDCHEK will, by default, abort with an informative error message if the program detects that the (80286 or better) CPU that it is running on is not in "real" mode, since this normally means that a (possibly multitasking) OS other than MS/PC-DOS is running and that HDTEST cannot be guaranteed unchallenged access to the hardware. The presence of LIM 4.0 expanded memory emulators such as QEMM-386 or Page 9 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ 386^Max, which use the 80386's "virtual 8086" mode to allow remapping of extended memory into unoccupied space in the area between 640 Kb and 1 Mb, will cause a similar abort, unless the driver is inactive (turned "OFF"). This safety feature can be bypassed by the use of the /V command line option (see below). 6) HDCHEK and HDTEST will operate normally (except than the /U or /H flags will sometimes be needed to use the low-level interface) on (very large) hard disks with more than 1024 cylinders, provided that the partition being tested does not begin or end above cylinder 1023. ______ HDTEST now supports the SpeedStor and Disk Manager extensions to INT 13, which allow access to cylinders beyond cylinder 1023. To use the HD programs on partitions which span or start above cylinder 1024, follow the instructions in the section of this document entitled: "Using the HD programs with SpeedStor and Disk Manager". Otherwise, by default, if the partition starts below and ends above track 1023, HDCHEK will issue a warning and HDTEST will abort unless the /D flag is used. If the partition starts above cylinder 1023, both HDCHEK and HDTEST will, by default, abort. COMMAND LINE OPTIONS ____________________ Always run HDCHEK on any new hardware/software configuration before ______ you attempt to run HDTEST (at least the first time). The latter program has been written as conservatively as possible, but it is not possible to predict the behavior of every conceivable combination of hardware and operating system when dealing with a program which bypasses most of the controls which normally protect the user from destroying his world. HDCHEK is invoked with the command line: HDCHEK device [optional parameters] "optional parameters" may include one or more of the following: [/A] [/Cn | /CQ] [/D | /Ex | /I | /U] [/Hnnnn[:nnnn]] [/M] [/V] "device" is the device to be tested (e.g. A:, B:, etc. - the colons are permitted but not required). For versions of HDCHEK from V4.35, ___ the spaces between the device identifier and the flags (and between different flags) are optional - the command lines "HDCHEK C:/C2/D" and "HDCHEK C: /C2 /D" are equivalent. The /A(T) flag is used to tell the HD program that it is being run on an IBM AT. Setting the /A flag allows the programs to deal correctly with a known bug in the IBM AT floppy disk controller BIOS when HDTEST is used on 1.2 Mb "AT format" 5.25" floppy disks. If the /A flag is set and the CPU is not an 80286 or better, you will be asked to confirm that you wish to use AT defaults (HD BIOS at F0000H). N.B. the definition of the /A flag changed (slightly) at V4.71 - it ___________________________________________________________________ need now only be used if you are testing a 1.2 Mb floppy disk on a __________________________________________________________________ Page 10 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ "vanilla" IBM AT. It may still be used to change the default BIOS _________________________________________________________________ segment to F0000H on a PC or XT class machine, but the use of the /H ____________________________________________________________________ flag for this purpose is preferred. ___________________________________ The /C flag has two variants. It is used, immediately followed by a single decimal digit ("n") in systems which do not support IBM's INT 11 "equipment determination" BIOS function (e.g. the Tandy 2000). The parameter ("n") is the number of internal floppy drives attached to the system, which would otherwise be determined by the program, using INT 11. The second variant should be used on Compaq computers - here, the flag is followed by a "Q", instead of a number. This variant enables large device support (as for IBM DOS 4.00) under Compaq DOS 3.31. The /D(OS interrupts) flag forces HDCHEK to use the hardware- independent DOS interrupt 25 for access to the target device. It is usually not required, since the program is usually smart enough to _______ _______ recognize devices which have to be accessed this way (e.g. logical disk partitions other than those created as DOS 3.3 (and above) "Extended Devices", non-bootable hard disks, some removable-media hard disks). If the /D flag is not used, the program will issue a warning ___ message if it finds it has to use the DOS interrupts. N.B. the definition of the /D flag changed at V4.00 - the new /I flag _____________________________________________________________________ has the same effect as the old (pre-V4.00) /D flag. Only one of these _____________________________________________________________________ flags may be present on the command line. _________________________________________ The /E(xtended INT 13) flag permits HDCHEK to recognize "proprietary" SpeedStor or Disk Manager partitions and allows the use of the "extended" INT 13 definitions which permit SpeedStor and Disk Manager to support disks with more than 1024 cylinders. Use /ES to use SpeedStor extensions and /ED for Disk Manager extensions. See the "Using the HD programs with SpeedStor and Disk Manager" section of this documentation for further details. The /I(nterrupt 13) flag has no effect under DOS 2.x. Under DOS 3.0 or above, it forces HDCHEK to use the existing INT 13 vector (which it always does under DOS 2.x), and prevents the program from trying to determine the ROM BIOS entry point; the flag should be used if you know that your system's fixed disk BIOS uses INT 13 but does not follow IBM's "rules" or if the program crashes the system when invoked without it. The /H(ex BIOS address) flag is normally used (with a single numeric argument) to tell the HD programs that the ROM BIOS of the device being tested is located at nnnn, which is interpreted as a Hexadecimal segment address, rather than at C800, which is the default for PC and XT class machines, or F000, which is the default for systems using a 80286 or better CPU. There must be no space between the H and the number. The extended form of the /H flag (e.g. /HF000:80D5) forces the HD programs to use the given address as the ROM BIOS INT 13 entry ____________________________ Page 11 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ point. Its use is documented in more detail in the "Using the On-Disk _____ Preboot" section of this documentation. The /M(ulti-track read problem) flag forces HDCHEK to split up disk accesses that would otherwise require the BIOS to read data from more than one track in a single "bite". Its use is documented in more detail in the "Using the On-Disk Preboot" section of this documentation. The /U(se INT 68) flag forces the HD programs to use the existing INT 68 vector (which must have been set up to point to an appropriate location!) instead of locating and using the INT 13 BIOS entry point. The use of this flag is documented in more detail in the "Using the On-Disk Preboot" section of this documentation. The /V(irtual mode) flag bypasses the code which prevents the HD programs from working when the (80286 or better) CPU is not in "real mode". It should only be used when the programs are being run under a ____ single-tasking operating system which is using "virtual mode" only for ____ memory-mapping. Certain LIM 4.0 expanded memory emulators for 386 machines use the processor's virtual mode for this purpose. The two most well-known examples of this type of software are QEMM-386 and 386^Max. The HD programs are "DESQview aware", and will abort if run under DESQview, even if the /V flag is used. They will not be able to protect you from the (probably disastrous) consequences of using the /V flag to run them under other multi-tasking OSes (such as VM-386). After a short delay, during most of which the specified device will be active, a description of the device's logical structure will be disp- layed. After a further delay, and more disk activity, a message will be printed giving the number of files and subdirectories found on the disk, and the program will exit back to DOS. Note that the total number of files and subdirectories reported by HDCHEK will be greater than the sum of the counts reported by CHKDSK. This is because HDCHEK counts the "." and ".." entries in subdirectories and CHKDSK doesn't. Errors and/or problems encountered, if any, will be reported on in some detail. A number of example screen displays and log files produced by running HDCHEK (and HDTEST) on different systems will be found in Appendix C at the end of this documentation. If HDCHEK tells you that you have a bad 8088 chip, believe it - some early 8088s had a bug which allowed them to be interrupted while changing their stack frame* - this is not benign, and can cause all sorts of rare, otherwise almost inexplicable, problems. The only real cure is to _____________________________________________________________________ * See "Chips in Transition", by Bob Smith. PC Tech J. 4:4 (Apr. 1986): __________ 56-69. The code which the HD programs use to identify the CPU and diagnose this problem is adapted from the CPUID program described in this article and from code from subsequent, related, articles in that (now defunct) magazine. Page 12 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ replace the CPU (which is, fortunately, not a particularly expensive proposition these days). If HDCHEK reports a discrepancy between the two FATs, HDTEST will abort. The FIXFAT program (on the distribution disk) may be able to correct this problem - see FIXFAT.DOC, also on the distribution disk. Always run the DOS CHKDSK program immediately before running HDTEST on a disk. If CHKDSK reports any problems, its /F option (or other means) should be used to correct them before HDTEST is run. ____________________ HDTEST is invoked with the command line: HDTEST device1 [device2[:pathname]] [optional parameters] "optional parameters" may include one or more of the following: [/A] [/B] [/Cn | /CQ] [/D | /Ex | /I | /U] [/F] [/G[V]] [/Hnnnn[:nnnn]] [/M] [/N] [/P] [/R[n]] [/Snnnnn] [/V] [/X] [/Yn] [/Z[n]] "device1" is the name of the device to be tested, as for HDCHEK. "device2" (which, if present, must be different from "device1") is the ____ name of a device on which a log file is to be written. The device name may be immediately followed by a file- or path-name for the log file. If only a device name (with or without a colon) is specified for the log file, a file called HDTESTdn.LOG will be written to the default directory on the specified device. "d" in this file specification is replaced by the single-character specifier of the device being tested, while "n" is replaced by the lowest decimal digit which results in the creation of a new file (previous log files will not normally be ___ ___ overwritten). If all ten possible log file names for this test device have already been used, the program will prompt for a new name. If a directory path (ending in a backslash (\)) is specified as device2, HDTESTdn.LOG will be created in the specified directory. If a full pathname is specified (not ending in a backslash), a file with that ___ pathname will be opened as the log file. In this case (only), any pre- existing file of the same name will be overwritten. If "device2" is not specified, errors are reported to the console terminal. The DOS console redirection facility should not be used to redirect them ___ elsewhere. For versions of HDTEST from V4.35, the spaces between the device identifier and the flags (and between different flags) are optional - the command lines "HDTEST C:/A/F" and "HDTEST C: /A /F" are equivalent. The /A, /C, /D, /E, /H, /I, /M, /U, and /V flags have the same meaning and effect that they do for HDCHEK. When used for HDTEST, the /D flag forces the use of INT 25 to read from and INT 26 to write to the target disk. Page 13 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ The /B(oot/Partition sector test) flag forces the program to include the boot/partition sector (track 0, head 0, sector 1) in the DOS area for the purposes of read/write testing. The default action is only to perform read and seek tests on this sector, even if read/write testing of the DOS area is specified. Since the boot/partition sector is effectively read-only under normal circumstances (it is only written _______ to during low-level formatting, and by FDISK and other disk __ partitioning programs), read/write testing is less vital than for the rest of the disk. In addition, at least one hard disk controller (a Western Digital board which allows a single disk to be addressed as two "physical" devices) will not allow test patterns to be written to this sector. Certain devices (chiefly RAMdisks) appear to DOS to be hard disks but do not, in fact, possess boot/partition sectors. On these devices, HDTEST will report check/test results on the boot/partition sector while actually reading/writing the device's boot sector - the latter sector will simply end up being checked/tested twice under these circumstances. The /F(ast mode) flag causes the program to use only two test patterns (all 1s and all 0s) for read/write testing. It is intended for use on floppy disks and for "quick and dirty" testing of other devices. The /G(ive error details) flag causes the program to determine the location of errors occurring during testing of the disk's data area at the sector level (they are normally handled at the cluster level), and to report on the physical location of all bad sectors found. The report is written to the log device or file after all testing is ____________________ finished. By default, only an error code is given for each bad sector ________ - if a V (for "verbose"!) is appended to the flag, the error code is expanded to a report similar to that given for each bad cluster. This flag is ineffective (and ignored) if HDTEST is using DOS interrupts, either because the /D flag was set or because the program was not able to access the disk at the BIOS level. The /N(o Query) flag disables the individual prompts for confirmation which the program normally issues before remapping clusters which are marked as "Reserved" or which are part of a file which is marked as "System", "Hidden", or "Readonly". The /N flag does not prevent the ___ program from prompting for confirmation before updating the disk. The /P(rint) flag causes the log file to be directed to the PRN device (normally LPT1), and overrides any "device2" specification. The /R(epeat)n flag causes HDTEST to go through n complete testing cycles. Omitting n, or setting it to zero, causes HDTEST to run continuously until terminated by pressing <Esc>). Instead of stopping after completing the testing of allocated data clusters, the program immediately starts again with the DOS area. Any bad clusters found and flagged in the first pass should be reported as "already marked as bad" in subsequent passes. The /R flag implicitly sets the /N flag and disables the program's incremental testing functionality. Page 14 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ The /S(tart)nnnnn flag became largely obsolete with version 5.32 of HDTEST. It causes HDTEST to start testing occupied clusters at cluster ________ number nnnnn (expressed as a decimal number). There must be no space _______ between the "S" and the number. This option was mainly useful when a previous run of HDTEST had been interrupted (by pressing <Esc> or <Ctrl-C>), since it allowed you to start testing occupied clusters ________ where you had left off. In order to be sure that data was not remapped to a bad area, HDTEST always tested all empty clusters before starting ______ ___ to test occupied clusters. For this reason, using the /S flag did not result in a large time saving if the target disk was nearly empty, and it made the most difference on a full, or nearly full, disk. From version 5.32, HDTEST defaults to "picking up where it left off" after a run is interrupted. See the "Incremental Disk Testing" section for further details. The /Y(es, please remap)n flag causes HDTEST to remap up to n (a decimal number) bad clusters without prompting for confirmation. It is effective only if HDTEST is allowed to run to completion, and is intended to allow the program to be run unattended on multiple disks from a .BAT file. If more than n bad clusters are found or if the program is aborted (by pressing <Esc>), the normal prompt for confirmation is issued before any bad clusters are remapped. See "Running HDTEST from a .BAT file". The /X(clude bad clusters) flag causes HDTEST not to test clusters ___ that are already marked as bad by FORMAT or previous runs of HDTEST (the default is to test all clusters every time the program is run). ___ This saves some time when testing a disk with a substantial number of known defects. The marked clusters will be reported as bad but will not be retested. The /X flag overrides the /G flag for previously marked clusters - no bad sector report will appear for these clusters if both flags are used. The /Z(ero bad sectors)n flag controls how HDTEST deals with sectors that could not be read correctly (due to unrecoverable read errors) when it remaps bad clusters. Versions of HDTEST up to and including V4.10 (somewhat cavalierly) disregarded the remaining contents of the entire cluster if one or more sectors in it were unreadable, and remapped it to a cluster of NULLs. The default action is now to replace only the unreadable sector(s) with NULLs, transferring the _________ remaining sectors intact to the remapped cluster. There are, however, two other options, which can be chosen by setting n to 0 or 2 (if n is omitted, 1, or any other number, the default action is taken). Setting n to 0 uses the old default (remaps the bad cluster to a cluster of ___ _______ NULLs), while setting n to 2 goes to the other extreme and copies all data read (some of which is almost certainly invalid) to the remapped cluster. It is always safe and often advisable to use the /Z2 option provided that you do not try to use any non-ASCII files that have been ______________________________________________________________________ remapped with unrecovered read errors and the /Z2 option. Remapped _________________________________________________________ ASCII files can be inspected with an editor and fixed up where necessary - non-ASCII files (e.g. .EXE, .COM, .WKS files) may appear to be undamaged (the program may run, the worksheet be readable, etc.) Page 15 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ but they can still contain small changes which will cause trouble later, and they cannot easily be checked. The large "holes" which /Z options 0 and 1 leave in such files usually allows it to be much more obvious that they are defective! WHAT DOES HDTEST DO? ____________________ The following discussion assumes that both any immediately previous run of HDTEST on the target device and the current run was and is allowed to run to completion. HDTEST's new (V5.32) incremental testing functionality is discussed in the "Incremental Disk Testing" section of this documentation. After displaying its copyright message, HDTEST reads the FATs and scans the root directory and all subdirectory trees of the specified device before initiating a short dialogue. All questions are of the "Yes/No" type and should be answered by typing "Y" or "N" (case does not matter) and pressing the <Return> key. Pressing the <Return> key alone is equivalent to giving the default response (which appears in square brackets at the end of each prompt). HDTEST then proceeds to test the specified device. By default, the DOS area (boot sector, FATs, and root directory) is not subjected to Read/Write testing, since failures in this area usually make themselves apparent fairly rapidly. The user can, however, choose to run these tests on the DOS area too, or to limit testing to read and seek tests only over the whole partition. In its reports to the screen, HDTEST uses the convention that read- only testing is "Checking" while read/write testing is "Testing" - it always "checks" the boot/partition sector and the rest of the DOS area, but only "tests" them if requested to do so. In "checking" or "testing" a block (sector or cluster), HDTEST first reads it (retrying, if necessary, up to five times) and then does the longest possible seek away from it before attempting to read it again. "Checking" is complete at this point. "Testing" continues with the program sequentially writing 20 (or 2, for "fast mode") different bit patterns to the block and checking that each one can be read correctly, before rewriting the original contents of the block (and reading it again to ensure that it has been restored correctly). Errors that occur at any point in this process are reported and (if the block is in the data area - errors in the DOS area are always treated as fatal) the block is remapped. If an error is detected in a "Reserved" cluster or in one that is part of a file which is marked as "System", "Hidden", or "Readonly", the user is normally asked for confirmation before the cluster is remapped, otherwise it is assumed that remapping is desired. HDTEST keeps a copy of the FAT and a map of the device's directory structure in memory. A bad cluster is remapped by transferring its contents to a free cluster and (if the bad cluster was in use) Page 16 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ changing all necessary FAT and directory entries in memory so that the pointer chain which maintains the structure of the affected file or subdirectory remains valid. The bad cluster's FAT entry is then altered to identify it as bad to DOS. The device's FATs and ________ directories are not altered until the whole device has been scanned, and then only after the user has confirmed that it is OK to proceed. This means that a power failure and/or an inadvertent power-off which occurs while the program is running will cause the loss of, at most, one cluster of data and of no FAT or root directory information (unless you are very unlucky and it happens after you have confirmed ____ that you want a damaged data area to be remapped and while this is being done). If a power failure and/or power-off occurs while a cluster which contains the directory information of a subdirectory is being tested, part or all of that subdirectory will probably become inaccessible to DOS, although the data in it will remain on the disk and be potentially recoverable with the aid of (e.g.) the Norton Utilities (TM). You can safely interrupt the testing process part way through (by pressing the <Esc> or <Ctrl-Break> key) with no risk of leaving a partially updated FAT or directory. If you do interrupt the program in this way, you will be given the option of updating the disk, so that bad clusters which were detected before it was interrupted can be remapped. The program will also create a file (HDTESTED.MP0) in the root directory of the target disk which a subsequent run of HDTEST can read to identify clusters that have already been tested. See the "Incremental Disk Testing" section of this documentation for more discussion of this area. Errors which occur in the program's initialization phase cause it to abort with an error message and (sometimes) a hexadecimal error code. The meaning of the error code is context-dependent and it is useful only to technical support personnel (i.e. the author!). Errors detected during read-write testing only cause the program to abort if they occur during testing of the DOS area, or if remapping is impossible (only likely to happen if the disk is full). In any event, a description of the error is written to the log file (if any) and/or to the screen. This specifies the cluster or (for the DOS area) sector number and the name of the affected file (if any), and defines the type(s) of error which occurred. Appendix B contains a discussion of some of the more technical aspects of data storage on disks (floppy and hard), and includes more information about some of the possible errors that may be encountered. HDTEST takes a very long time to run to completion because of the ____ large number of reads and writes it does for each cluster. A typical time for full testing ("slow" mode) of an XT disk (10 Mb) would be 2 hrs; full disk testing on a 30 Mb Enhanced AT takes at least 4 hrs. Note also that using the /D or /G flags will substantially slow down the testing of disks with significant numbers of bad spots. Using the /X flag as well will allow the physical location of new defects to be Page 17 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ reported without requiring that known bad areas be exhaustively retested. INCREMENTAL DISK TESTING ________________________ In response to a considerable number of requests, mainly from users with very large disks and/or partitions, on which HDTEST took a very long time to run to completion, the program has been revised to allow true "incremental" testing of the data area. When a run of HDTEST is interrupted by the user pressing <Esc> or <Ctrl-Break> during testing of the data area, the program writes a special file (HDTESTED.MP0) in the root directory of the target disk before exiting. This file contains a bitmap of the data area of the disk, with set bits indicating clusters that have been tested, and a six-byte header which a subsequent run of HDTEST can use to ensure that the file has not been corrupted. After HDTEST writes the file, the program gives it the "READ ONLY" attribute, so that it will not be accidentally deleted or modified. If a subsequent invocation of HDTEST finds HDTESTED.MP0 in the root directory of the target disk, the program will ask if you want to use the information in the file to allow it effectively to start where the previous run left off. If the file is less than 14 days old, the default answer to this query is "Y"; if the file is older, the default answer is "N". If the new invocation of HDTEST runs to completion, HDTESTED.MP0 will be deleted; if the new invocation is again interrupted, the file will be updated with bits set for any additional tested clusters. Note that every run of HDTEST will check or test the entire DOS area of the disk. Only the data area is potentially subject to incremental testing. Note also that it does not matter if files are written or deleted between "incremental" runs of HDTEST. Newly freed clusters will be tested before any occupied _____ clusters, even if the previous run had included all (then) free clusters, and previously free clusters now included in files will be skipped, but the program will not become confused or miss any untested clusters because the contents of the disk may have changed. USING THE ON-DISK PRE-BOOT __________________________ All "normal" PCs and almost all clones capable of running IBM PC-DOS (rather than "generic" MS-DOS) should also run the HD programs with no problems, and should allow them to locate and use the ROM BIOS INT 13 entry point with no help from the user. Less compatible systems have presented more problems in the past, but current Registered versions of HDTEST are distributed on a disk whose boot sector has been modified to allow identification and use of the "real" entry point on almost any MS-DOS-compatible hardware. This modified boot sector will ____ be correctly duplicated if you back your distribution disk up by using the DOS "DISKCOPY" command - it will not be copied by a "COPY *.*" ___ command. This initially presented a significant potential problem for users of the Shareware version of the program, but the distribution packages of all current releases of HDTEST also contain a program Page 18 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ (BOOTLOAD.COM) which will modify the boot sector of any removable ___ disk, allowing you to create a special pre-boot on the media of your choice. Further documentation for this program is to be found (as BOOTLOAD.DOC) in the distribution package. To use the special pre-boot, first insert your copy of the HDTEST distribution disk in the boot drive (normally drive A), then (re)boot your system as you normally do (either by switching it on or by pressing the <Ctrl>-<Alt>-<Del> combination or its equivalent for your hardware). After performing its self-test, the system's BIOS will load and pass control to the code on the boot sector of the floppy. On a normally FORMATTed floppy disk, this code attempts to locate the MS- DOS system files at the beginning of the disk's root directory and to load and execute them. The HDTEST boot disk will not load DOS (even if you SYS the disk) - it simply reads and reports on the contents of the INT 13 vector (which DOS has not yet had a chance to tamper with, since it has not been loaded) and conditionally transfers the vector to the INT 68 location. The INT 68 location is normally unused in MS- DOS systems - it is defined that way by IBM - and is usually loaded either with zeroes or with a pointer to some "do-nothing" code by the BIOS bootstrap. The HD programs use it for the "real" INT 13 vector, normally saving the original contents of the location on entry and restoring them on exit. This behavior is modified by specifying the /U switch on the command line. The HDTEST boot sector code will only transfer the "real" INT 13 BIOS entry point to the INT 68 location if the latter contains either zeroes or the same vector as INT 69. In the ______ __ latter case (since there is no point in having two "real" interrupt vectors pointing at the same place), it is assumed that both are pointing at "do-nothing" code. If the INT 68 vector is neither 0:0 nor _______ ___ the same as the INT 69 vector, the preboot code will leave it alone and report that the INT 13 vector was not transferred, otherwise it will confirm that the transfer has taken place - in any event, you will be given an opportunity to write down the vector and can then remove the disk and press a key to allow the boot process to complete. If you need to boot from a floppy disk, for some reason, a new, bootable, disk can be inserted before you press the key to continue. If the preboot reported that the INT 13 vector was successfully transferred, you should simply specify the /U flag on the HDTEST or HDCHEK command line, in addition to the "short" form of the /H flag ______________ (if the BIOS ROM is at an unusual location for the class of machine). The requirement for the /H flag with unusual hardware allows the program to do some additional "reasonableness checking". Unless some other software interferes with it, the INT 68 vector will then Page 19 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ continue to be set up correctly for use by the HD programs until the system is switched off or rebooted. If the preboot was not able to transfer the vector, or if you know that a device driver or resident program loaded by your DOS uses INT 68, you should invoke the HD programs with the "long" form of the /H flag, specifying the full address displayed by the preboot. No error checking is attempted in this case - if you mistype the address, the first time the program attempts to access the disk, it will jump to a (effectively) random location in ROM, with potentially disastrous results. If it is feasible, use of the pre-boot disk and the /U flag is preferable (in terms of safety, if for no other reason) to that of the "long" form of /H. The following example may help to clarify matters. You have a clone system with a BIOS ROM location that is non-standard. The HDTEST boot disk reports that the INT 13 entry point is at E000:13AC, and that the vector has been successfully transferred. You may now run HDTEST with the command line: HDTEST C: A: /U /HE000 ____ You may also, now or later, with or without prior use of the pre-boot, _______________ run HDTEST with the command line: HDTEST C: A: /HE000:13AC _________ Note that the /H flag is not needed with the /U flag if the segment ___ address of the hard disk BIOS is the usual value for the system type ___________________ (as determined by the processor). For a system based on an 8088, 8086, 80188, or 80186 processor, this "usual" value is C800. For all other systems, the "usual" value is F000. It is important to note that the /Ex and /U (or /H) command line options are incompatible, since the former forces the HD programs to use the third-party INT 13 driver entry point, rather than the BIOS entry point. If you are unfortunate enough to have a system which has both a non-standard BIOS location and one or more partitions ____ ___ controlled by SpeedStor or Disk Manager, you will need to use the /U or /H flags when testing the "vanilla" DOS partition(s) and the appropriate form of the /Ex flag when testing the "third-party" partitions. One further problem may be encountered - certain hard disk BIOSes do not exactly comply with IBM standards in dealing with disk accesses that cross track boundaries, and will produce frequent errors when such accesses are attempted. It was this "multi-track read" problem which resulted in a number of previously apparently 99.99% IBM- compatible systems failing to run IBM PC-DOS 3.3 (which was the first version of DOS actually to attempt such accesses). IBM issued a "fix" for this problem, in the form of a program which, when run on the "bad" system, creates a device driver which protects the BIOS from Page 20 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ being asked to do something that it can't. This device driver does its job by intercepting INT 13, and will, of course, do nothing to help the HD programs if they are bypassing that interrupt. Use the /M flag with the HD programs if you have a system on which PC-DOS 3.3 or above will not run without being "fixed" as described. It forces the programs to split up the multi-track accesses into manageable chunks - this can be expected to cause a substantial slowing of the programs' operation. USING THE HD PROGRAMS WITH SPEEDSTOR AND DISK MANAGER _____________________________________________________ The HD programs will operate without problems under DOS 2.x through 4.01 on almost all hard disks with up to 1024 cylinders, as well (usually) as on larger ESDI disks which use "sector translation" to appear to have less than this magic number of cylinders. An increasing number of disks now coming to market have more than this number of cylinders but do not have the ability to perform "sector translation". If their full extent is to be usable under DOS, these disks must be partitioned with the use of special non-DOS utilities (the two major contenders are SpeedStor and OnTrack's Disk Manager), and access to (at least) the "high" partitions requires a special driver to be loaded in CONFIG.SYS. In addition, many of those who used such third- party drivers to support partitions greater than 32 MB in extent under DOS versions prior to 4.0 on intermediate sized disks continue to use them for this purpose. Versions of the HD programs prior to V5.00 were only able to handle "third-party" partitions at the DOS level. The current version understands the partitioning schemes used by the SpeedStor (SS) and Disk Manager (DM) software, and can, if necessary, use the special drivers' INT 13 extensions to test the proprietary partitions as thoroughly as it does "normal" DOS partitions. In addition to using the /Ex command line flag (q.v.), it is necessary to load a special device driver, HDTESTDD.BIN, which is supplied with the HDTEST package, in CONFIG.SYS. The line: DEVICE=[drive:][path\]HDTESTDD.BIN should be added to CONFIG.SYS immediately after the line which loads ______________________________________ the partitioning software's driver. The HDTESTDD driver takes up only __________________________________ 128 bytes of memory, and the command to load it may be made a permanent part of your CONFIG.SYS - the driver does not intercept or chain into any interrupts and will have no effect on the normal functioning of your system. It does, however, allow the HD programs to locate and use the partitioning driver's INT 13 entry point, which they may not be able to obtain in any other way. Either SS or DM can be used to set up and format disks with one or (under DOS 3.3 and above) more normal DOS partitions, as well as, or sometimes instead of, one or more of their own proprietary ones. The /Ex command line switch should only be used for the proprietary ____ partitions - its use with normal DOS ones will slow the programs down somewhat and may make HDTEST less sensitive to minor problems. Page 21 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ N.B. The SS or DM driver (SSTOR.SYS or DMDRVR.BIN) should always be ____ the first block device driver in CONFIG.SYS. Drivers for other block _____ devices such as RAMdisks, external floppy drives, tape drives, or CD- ROM drives should always follow both it and the HDTESTDD.BIN driver. The HD programs assume that the device names of any SS or DM partitions follow in alphabetical order immediately after those of any _________________ "normal" DOS floppy and hard disk partitions present. If any other devices are interposed, the HD programs will not be able to operate at anything but the DOS interrupt level. Reports that earlier versions of the HD programs could cause potential or actual data loss if run on disks with more than 1024 tracks appear to have resulted from attempts to use them on improperly partitioned ______________________ drives. Problems with ESDI disks using sector translation were, however, eventually tracked down to a bug in the HD programs' initialization code which was causing the incorrect interpretation of partition table data in extended partitions, while problems with large disks using third-party partitioning drivers seem to have resulted from the incorrect designation of a partition ending above cylinder 1023 as a DOS partition - a mistake which would have resulted in major problems from any attempt to access the area above track 1023. There ___ now appears to be no evidence that earlier versions of HDTEST could damage data on properly partitioned disks of any size. ____________________ SPECIAL CONSIDERATIONS WITH SPEEDSTOR _____________________________________ SpeedStor V6.0 and above uses a rather complex partition tree structure to support more than the 8 partitions per disk allowed by earlier versions of the software. SpeedStor also always names the partitions which it controls in alphabetical order as they appear on the physical disk, whether or not this is the order in which pointers to them are found while traversing the directory tree. There are, consequently, some circumstances under which the HD programs and SpeedStor may "disagree" about the order in which its partitions are named. This will generally cause HDTEST to "fall back" to DOS interrupt level when testing the disputed partitions. This problem will generally only be encountered when the "owner" of a partition is changed after a disk has been partitioned, and/or one or more partitions have been deleted and a larger number have been added in the middle of a disk. There is no "quick fix" for this problem. The best solution is to back up all the partitions on the "offending" device, and delete and recreate them "from scratch". This will ensure that SpeedStor and HDTEST agree on the order in which the partitions are named. DEALING WITH BAD CLUSTERS _________________________ It is not unusual for a new hard disk to have a small, constant, _____ ________ number of bad clusters. IBM's quality control standards allow, for example, 5% of the surface area of a disk to be bad (!). These pre- Page 22 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ existing bad clusters, which correspond to imperfections in the magnetic coating on the disk, will normally be detected by the manufacturer's routine tests. Their locations will appear on the "badspot list", which comes with the drive, and the whole track which contains the defect will be marked as bad during low-level formatting. The marked bad sectors will be detected by the DOS FORMAT program, which will mark the relevant clusters as bad, and HDTEST will report them as "already marked as bad". This should not be regarded as evidence of any real problem with the disk, although some suppliers (not including IBM) will replace such disks under warranty if pressed ___ to do so. It is also not unusual for a disk to develop a few more ___ defects as it ages. When these do develop, they should also subsequently be constant in size and position, and will be reported as "already marked as bad" by HDTEST on runs subsequent to that during which they are first found. The occasional appearance (and subsequent __________ persistence) of a new bad cluster is, therefore, not necessarily evidence of serious trouble developing. However, if more than one or two new bad clusters are reported by HDTEST, and especially if two consecutive runs (or passes, if the /R flag is set) of HDTEST report different bad clusters, there is a good chance that the hard disk or _________ controller is developing a serious problem - the disk should be backed up immediately and the disk and/or controller sent for repair at the earliest opportunity. New bad clusters in the data area are automatically mapped out by HDTEST, and no further action is necessary. Bad sectors in the DOS area can be handled in two ways: 1) If you are able to do a low-level format on your hard disk (most _________ makers of "add-on" hard disks supply a method of doing this, either in the form of a short program, or by giving instructions for running a program in the ROM BIOS from DEBUG), back up the entire hard disk ______ (including any non-DOS partitions), run the low-level format, use FDISK to repartition the disk, and then run the DOS FORMAT program. If HDTEST now reports no errors in the DOS area you can restore all your files and continue to use the disk. You should, however, use HDTEST to test (at least) the DOS area fairly frequently for a month or so to be sure that the problems are not going to recur. 2) If you do not have the means to do a low-level format, or if you follow the above instructions and HDTEST continues to report errors in the DOS area, your only other choice is to back up the DOS partition, reconfigure the disk with FDISK so that the new DOS partition starts above the bad area, reFORMAT it, and restore all of the old files to the new DOS partition (which will unavoidably be a little smaller than the old). If the hard disk's boot/partition sector (track 0, Head 0, sector 1) becomes bad, the disk is useless and must be replaced. This sector is only subjected to read/write testing if the /B flag is specified on __________ the command line, so any reported errors (in sector -1) during normal __ testing represent a problem reading the sector and are likely to be _______ Page 23 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ significant. Problems encountered during read/write testing of this sector do not always indicate a significant problem - certain ___ controller cards may not allow test patterns to be written to it, resulting in a report of multiple errors on read/write testing. Consequently, if multiple errors are reported on read/write testing of the boot/partition sector, but read-only testing proceeds without problems and the rest of the disk seems to be OK, there is probably no cause for concern. RUNNING HDTEST FROM A .BAT FILE _______________________________ Versions of HDTEST from V4.75 can be run from a .BAT file (e.g. for unattended testing of multiple disks) without the user losing the ability to "escape" cleanly from the program. The /Y flag also allows the user to specify a maximum number of new bad clusters for each disk to be remapped "automatically", so that the .BAT file will normally just keep going but will pause and wait for input if something terrible seems to have happened to a disk. In order to run HDTEST from a .BAT file, it is necessary to prepare both the .BAT file with your ____ usual HDTEST command lines, generally including the /Y and /N flags, and a second ASCII file containing the answers to HDTEST's initial ___ questions, one per line. DOS's input redirection ("<") command is used to "feed" these answers to HDTEST. If the "answer" file is called (e.g.) ANSWERS.TXT, each line in the .BAT file will look like: HDTEST dev1 dev2 /opt1 /opt2 <ANSWERS.TXT ____________ If you do not wish to read/write test the DOS area of the disk, your answer file can consist of three blank lines. If you do wish to __ read/write test the DOS area, the first line of the answer file can be blank, the next two must start with a "y" and the fourth can be blank. ___ In order to allow HDTEST to be interrupted from the keyboard, if necessary, the program terminates input redirection before starting testing the disk, so anything in the answer file after the answers to _____ HDTEST's initial questions will be ignored. On exit, HDTEST now returns a meaningful ERRORLEVEL, which can be tested from a .BAT file by means of DOS's "IF ERRORLEVEL" syntax. Versions of the program prior to V4.81 returned an ERRORLEVEL of 1 on aborts and 0 otherwise - HDCHEK still does this. The current version of HDTEST normally returns an ERRORLEVEL equal to the number of new bad clusters found. On aborts it returns 254, and, if deliberately terminated by the user (either by answering "N" to the last question in the introductory dialogue or by pressing and confirming <Esc> during testing), it returns 255. You should not normally use HDTEST's incremental testing functionality (see "Incremental Disk Testing" section) when running the program from a .BAT file - if HDTEST finds HDTESTED.MP0 in the root directory while being run from a .BAT file, it will not ask whether or not to use the information in it. It will simply use it if the file is less than 14 Page 24 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ days old, and not otherwise (these would be the default responses if the program prompted for them). Note that this application of HDTEST is really intended for advanced users, and I am aware that this description is somewhat cursory - more help in using it can, if necessary, be obtained directly from me. BUGS AND OTHER UNPLEASANT SUBJECTS __________________________________ Although, like all programmers, I like to think that my product is perfect, it is possible that there are a few bugs lurking in the depths of the program, and likely that some combination of software and hardware exists which HDTEST cannot figure out and/or cope with. If you think you have found a bug and/or if you cannot get either of the programs to work on your system, do the following: 1) Reread this documentation! Make sure that you are doing exactly what the documentation says to do. Make sure (if you are testing a drive with removable media) that the media is not write-protected. 2) Boot your system from a "virgin" DOS disk - preferably a copy of your DOS distribution disk with only any additional drivers necessary to support the partitioning of your hard disk (and HDTESTDD.BIN) added in CONFIG.SYS - and try to run the HD programs again. If they run in this environment but not in your normal DOS environment, try reconstructing the system under which they failed by adding things one by one until the problem recurs. If you find a specific piece of software which causes the trouble, let me know about the incompatibility. 3) If the above approaches do not help, or if HDCHEK runs but HDTEST bombs (the reverse should only happen if HDTEST aborts with an "insufficient memory" error), send me as detailed a description of the problem as possible and I will do my best to diagnose and fix it. A bug report should always include a full description of the system ______ being used (hardware and software), as well as a copy of the output ___ from a run of HDCHEK. I can be contacted by U.S. mail at the address given earlier in the documentation, by BIXMail to pfletcher, or by Compuserve EMail to 72255,305. CHANGES FROM PREVIOUS VERSIONS ______________________________ The first version of the HD programs actually released (not including beta-test versions) was V2.60. The programs have grown and developed gradually, for the most part, but the releases listed below have constituted the most significant "watersheds" - features added and changes made between the listed releases are treated here as if first _______ included in the next listed release, so some individuals for whom this ____ Page 25 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ is an upgrade may have packages with some, but not all, of the features listed for a given release. Note that the V4.45 Shareware version of the program is not listed here - it was identical to V4.41 except that it included the "fix" for a deficiency which was corrected at V4.70 in the Registered version. Similarly, the V4.98 Shareware version contains all the logic error "fixes" documented for V5.20, together with fixes for one or two other bugs and deficiencies discovered during work on that version. Unless otherwise noted, both programs were affected by the listed changes. Version 5.35a (this version) _____________ 1) Bugs in code supporting "incremental" disk testing identified and fixed (HDTEST only). Version 5.35 ____________ 1) Support for "incremental" disk testing added. 2) Support for MS/PC DOS 5.0 added. 3) Limited LIM 4.0 support added (up to 64 KB of LIM memory will be used, if present, sparing the equivalent amount of normal DOS memory). 4) Directory tree buffering scheme redesigned to allow more files per partition. 5) Miscellaneous minor inaccuracies in the text of error messages corrected. Version 5.23 ____________ 1) Option to proceed after bad 8088 chip identified added. 2) Minor changes in error messages. 3) Rearranged code and link order to reduce program .EXE size. Version 5.20 ____________ 1) Support for "extended" INT 13 used by Disk Manager and SpeedStor added. 2) Support for SpeedStor and Disk Manager large partitions under DOS 3.xx added. 3) Ability to abort HDTEST cleanly by entering <Ctrl-C> added. Page 26 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ 4) Bug causing erroneous report of "more than 1024 tracks" on failure to access disk via BIOS fixed. 5) Bug causing system to hang at unexpected "Press any key to continue...." prompt if run from .BAT file fixed. 6) Logic errors causing improper "parsing" of long extended partition chains and chains containing DM and SpeedStor partitions corrected. Version 4.95 ____________ 1) Added code to prevent undetected "wrap-around" when using existing INT 13 on a disk with more than 1024 cylinders. 2) Deficiency allowing "queueing" of keystrokes in typeahead buffer fixed. <Esc> now produces (almost) immediate response even if other key(s) have been pressed first (HDTEST). 3) Copyright screen displayed for minimum of 2 seconds (may be longer if program is being run on a large device). 4) Previous restriction on testable FAT size (was 64 Kb) removed. 5) /V option added to bypass "virtual mode" prohibition. Version 4.84 ____________ 1) Bug causing the programs to fail if a subdirectory were physically located beyond the 32 Mb boundary on a large (> 32 Mb) device located and corrected. 2) Deficiency causing "fallback" to DOS interrupt level under some circumstances when attempting to test "extended devices" on a second physical hard drive fixed. The Shareware V4.45 also included this "fix". 3) Log file specification capabilities enhanced (HDTEST). 4) DOS 4.0 support added (including BIOS-level access to large partitions). 5) Meaningful ERRORLEVEL returned on exit (see "Running HDTEST from a .BAT file) (HDTEST). 6) Minor improvements made to memory allocation algorithms. 7) Minor changes made to final screen & file output format (HDTEST). 8) Minor change to reporting of initialization errors (HDCHEK). Page 27 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ 9) Fully automatic setting of default ROM BIOS location by CPU type. Meaning of and requirement for /A flag modified. 10) Support for large devices under Compaq DOS 3.31 added (see description of /C flag). 11) /Y flag and .BAT file support added (HDTEST). 12) /G option added (HDTEST). 13) /X option added (HDTEST). Version 4.54 ____________ 1) Deficiency in CPU recognition code (which caused fast AT clones sometimes not to appear to be ATs) fixed. ___ 2) ANSI.SYS requirement removed. 3) Semi-automatic CPU recognition added. 4) Pre-boot functionality, /M option, /U option, and extended (segment:offset) form of /H option added. Version 4.41 ____________ 1) Deficiency (introduced at V4.00) causing HDCHEK to issue an incorrect error report on an abort resulting from a FAT comparison error corrected. 2) Minor deficiency, causing HDTEST to create an empty log file on the specified device when both a log device and the /P option were specified, corrected. 3) Work-around for hardware multi-sector read ECC problem (see Appendix B) added (HDTEST). 4) Support for small (<= 10 Mb) hard disks patched to have small clusters and 16-bit FATs added. 5) Requirement that command line option flags be separated by one or more spaces or tabs removed. 6) /R option added (to HDTEST) for multi-pass disk testing. Page 28 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ Version 4.30 ____________ 1) Support for BIOS-level access to DOS 3.3 extended partitions added. 2) /Z option added (to HDTEST) to control remapping of unreadable sectors in bad clusters. Version 4.10 ____________ 1) Assorted minor changes made to screen displays. 2) /C option added to allow the programs to be used on some non-IBM compatible MS-DOS systems. 3) /D option redefined - old /D option became new /I option. 4) /F option added (to HDTEST). 5) "Short" form of /H option added. Version 3.13 ____________ 1) Assorted minor bugs and deficiencies in error reporting and termination routines fixed (HDTEST). 2) Minor bug affecting reporting of multiple corrected ECC errors fixed (HDTEST). 3) Disk reset added before retry on all errors (HDTEST). 4) Absolute limit on number of files and directories on device removed. 5) Display of total sector/cluster count during testing added (HDTEST). 6) /A option added. 7) /B option added (HDTEST). Version 2.73 ____________ 1) Bug (in V2.60) which could cause bad termination and possible subdirectory inconsistencies if a sector which was not in use was found to be bad, fixed (HDTEST). 2) Bug (in V2.70) which caused HDTEST to enter a terminal input loop if ANSI.SYS was not loaded and <Esc> was entered to abort it, fixed. Page 29 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ 3) Minor bug in V2.60 (which resulted in a failed directory scan if the program was run on a write-protected device) fixed. 4) Internal changes made to avoid interference with expanded memory device handlers. 5) Ability to update disk after HDTEST terminated by user (typing <Esc>) added. 6) /S option to start checking occupied clusters at a given cluster number added (HDTEST). Page 30 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ APPENDIX A - SUMMARY OF FLAG SETTINGS _____________________________________ /A(T) - used to tell the HD program that it is being run on an IBM AT (HDCHEK and HDTEST). /A is not equivalent to /HF000. /B(oot/Partition sector test) - used to include the hard disk boot/partition sector in the DOS area for the purposes of read/write testing (HDTEST only). /C(lone BIOS)n - used in systems which do not support the IBM BIOS's INT 11 "equipment determination" BIOS function (e.g. the Tandy 2000). The parameter ("n") is the number of floppy drives attached to the system, which is otherwise determined automatically (HDCHEK and HDTEST). /CQ (Compaq) - used to indicate that the program is running on a Compaq system. If DOS 3.31 (or above) is in use, large (> 32 Mb) devices are supported (HDCHEK and HDTEST). /E(xtended INT 13)x - allows testing of "proprietary" partitions created by SpeedStor and Disk Manager partitioning software, and the use of the "extended" INT 13 definitions which permit SpeedStor and Disk Manager to support disks with more than 1024 cylinders. Use /ES to use SpeedStor extensions and /ED for Disk Manager extensions. /D(OS interrupts) forces the program to use only DOS interrupts (INT 25 and INT 26) for access to the target device. May not be combined with the /I, /H, or /U flag (HDCHEK and HDTEST). /F(ast mode) limits read/write testing to the use of two patterns only (HDTEST only). /G(ive error details) allows the physical disk location of bad sectors to be identified (HDTEST only). /H(ex BIOS address)nnnn[:nnnn] - used to tell the HD programs that the ROM BIOS of the device being tested is located at nnnn, which is interpreted as a Hexadecimal segment address, rather than at the (CPU- dependent) default. The extended (segment:offset) form of the command is used to force the programs to use the given address as the INT 13 entry point (HDCHEK and HDTEST). /I(nterrupt 13) - forces the program to use the existing INT 13 vector. This is the default under DOS 2.x. May not be combined with the /D, /H, or /U flag (HDCHEK and HDTEST). /M(ulti-track read problem) - bypasses the multi-track read problem which also prevents certain otherwise IBM-compatible clones from running PC-DOS 3.3 (HDCHEK and HDTEST). /N(o Query) - disables the individual prompts for confirmation which the program normally issues before remapping "special" clusters - does Page 31 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ not prevent the program from prompting for confirmation before ___ updating the disk (HDTEST only). /P(rint) - causes the log file to be directed to the PRN device (HDTEST only). /R(epeat)n - causes HDTEST to perform n complete testing cycles (or to continue until interrupted for n=0) (HDTEST only). /S(tart from)nnnnn - causes HDTEST to start testing occupied clusters ________ at cluster number nnnnn (expressed as a decimal number) (HDTEST only). /U(se existing INT 68 vector) - causes the HD programs to use the existing INT 68 vector as the INT 13 entry point (HDCHEK and HDTEST). ______ /V(irtual mode) - bypasses normal prohibition on running program if 80286 or better CPU is operating in "virtual mode". Allows use with QEMM-386 and 386^Max. /X(clude marked bad sectors) - causes HDTEST to report on but not retest clusters that are already marked as bad (HDTEST only). /Y(es, please remap)n - allows HDTEST to update the disk by remapping up to n new bad sectors without prompting for confirmation (HDTEST _______ only). /Z(ap bad sectors)n - controls remapping of unreadable sectors in bad clusters (HDTEST only) - n may have one of the following values: 0 - Remap clusters with any unreadable sectors to all NULLs 1 (default) - Remap only actual unreadable sectors to NULLs 2 - Transfer all data read to the new cluster Page 32 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ APPENDIX B - A HARDWARE TUTORIAL ________________________________ Users of earlier versions of this software repeatedly requested that I provide a more detailed explanation of the significance of some of the error messages displayed by HDTEST when it encounters a bad sector or cluster on the disk. It is not possible to do this without assuming (or imparting) a more detailed understanding of the "nitty-gritty" details of data storage on hard and floppy disks than most people want or need, let alone possess. This Appendix is not, therefore, "required reading" for the safe or effective use of the HD programs. It is intended as a tutorial for those who would like to understand some- thing more of what is really going on on their hard or floppy disk ______ drive when the little red light comes on. For the purposes of what follows, it is (fortunately) not necessary to know or understand exactly how changes in magnetic flux on the surface of a hard (or floppy) disk are translated into binary data, or vice versa. That could be the subject of another long dissertation. I will simply take for granted that binary data, conceptually organized into 8-bit bytes, can be read and written by the disk drive hardware under the command of its controller (which usually resides on a separate card plugged into one of the computer's expansion slots). I will be describing how blocks of data are physically arranged on a disk and how the disk controller goes about ensuring that the data it reads has not been corrupted since it was written. I propose to start off by referring to some of the error-detection and error-correction techniques and procedures that are applied to data stored on magnetic media. These techniques and procedures are necessary because, although magnetic media are very reliable and the technology of recording and playback is well understood, there always remains the possibility that a defect in the medium or some other intercurrent problem may cause the sequence of bytes read back from a disk or tape to be something other than a perfect duplicate of the sequence that was stored. If a single copy of the raw data were all that were recorded, there would be no way of immediately detecting (let alone correcting) such an error. It might be brought to light when the file that was changed caused problems when it is used by or executed as a program, but it might not become apparent until it had propagated itself through and caused problems or errors in multiple other programs or files. This would clearly be an unacceptable state of affairs. In order to prevent this from occurring, then, there should be some means of checking data which is read in from storage for errors, and/or (preferably) of correcting any errors which may have occurred before any use is made of the data. ______ Although error checking and/or correction can (and sometimes does) take place at any level from that of the application program down to that of the device controller, I am only going to describe the means that are used by hard and floppy disk controllers. When a block of data is passed to the controller to be written to the disk, a series of arithmetical operations is performed in sequence on all the bytes Page 33 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ in the block, to produce a 2-byte (for diskettes) or multi-byte (for hard drives) code. The diskette controller implements what is known as a cyclic redundancy check (CRC) algorithm, which is optimized to allow detection of all single bit errors and many longer error bursts (it is _________ reasonable to assume that errors will most probably either affect a single bit at random or, if they result from media defects, a number of adjacent bits). The longer (at least 32-bit) error correcting code (ECC) calculated by a hard disk controller gives enough information to allow detection and correction of, at least, all single bit errors and ______________ detection of most likely longer error bursts. The CRC or ECC is saved on disk with the data and is read back in by the controller when the data is read. The controller then compares the saved CRC or ECC with one newly calculated from the data. If the two are identical, the data is assumed to be valid and a successful read is signalled. If there is a discrepancy, a diskette controller always returns the data with a "CRC error", since the CRC code does not contain enough information to allow the incorrect bit or bits to be identified. On a hard disk, if the ECC comparison suggests a correctable error, it will be corrected, and the read call will return with a "Corrected ECC Error" code but with (presumably) valid data in the transfer buffer. If a more extensive error is detected, on the other hand, the call returns with a "Uncorrectable ECC Error" code (which is the same as the CRC error code). The 32-bit ECC used by IBM and many compatible disk manufacturers allows correction of single-bit errors only. Manufacturers of other, ____ usually more expensive, controllers use longer and more complex ECCs, to allow the correction of some multi-bit errors, but the necessity to minimize the time taken to calculate the ECC (which is part of "access time") sets tight limits on how complex it can be. ECC error correction is unfortunately not infallible. It is possible for a ___ complex, uncorrectable error to result in the same ECC disparity as a __ different, correctable, error, so that invalid "corrected" data can be ______ returned. The longer and more complex the ECC, the less likely this becomes, but there is no way of completely avoiding this ambiguity, short of keeping multiple copies of the data on the disk. In addition, some hard disk controllers* apparently abort with the "corrected ECC error" code immediately on encountering a sector containing such an ___________ error, even if it is not the last sector in the block specified by a ___ (multi-sector) read request. This non-intuitive (to say the least of it) behavior can combine with DOS >= 3.x's habit of ignoring corrected ________ ECC errors to result in apparently inexplicable disk-related problems, up to and including system crashes (if the file in which the problem occurred contained executable code). This hardware deficiency may be related to the problems which have plagued users of DOS 3.3 on some (previously) apparently IBM-compatible non-IBM hardware, since these also result from (similar but distinct) deficiencies in the clones' _____________________________________________________________________ * This behavior is known to occur with some early Western Digital controllers (1002A-WX1, 1002S-WX2) but it may occur with others, too. Page 34 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ handling of multi-sector disk reads. Prior to V4.36, HDTEST did retry ___ corrected ECC errors which were reported on the first read of a cluster, but would eventually (if the corrected error persisted) accept the data read as valid and copy and remap it to a "good" cluster. This could cause trouble if the controller had the "immediate abort on corrected ECC error" problem described above and the ECC ___ error were not in the last sector of the cluster, since the sectors after the "bad" one would not have been read, and would be lost (actually replaced by the corresponding sectors from the previous ________ cluster read) after remapping. On a machine with this problem, DOS would not, of course, have done any better than the earlier versions of HDTEST at reading the bad cluster, and HDTEST at least warned you that there was a problem, but the behavior described was certainly not ideal. From V4.36, HDTEST rereads clusters in which a persistent corrected ECC error is reported on the initial read sector by sector, ________________ to ensure that all the data in the cluster is transferred when it is ___ remapped. In most cases (as long as the ECC error was really ______ correctable), this will save the cluster (and the file of which it is a part) intact. DOS allows the programmer and/or computer user to think of and deal with disk data in terms of files, lines, and/or records, but data is always moved between a disk and memory a sector at a time. Any disk read or write request passed to DOS eventually gets translated into a command to read or write one or more sectors on the disk, commencing at a specified head, track, and sector. A floppy disk may have either one or two usable surfaces, corresponding to the top and bottom of the ________ disk. A hard disk usually has at least two surfaces, and may have more, since multiple physical "platters" are frequently incorporated in a single device. Each surface is normally accessed by and corresponds to a single read/write head, which moves radially across ____ it, although one surface on newer disks is often reserved for "servo" tracks, which are written with a special pattern that allows the controller to ensure that the read/write heads are kept exactly aligned with the data tracks. For a single device, the (data) heads are, by convention, numbered sequentially from zero up to one less than the total number present. Each surface is further divided into a number of circumferential tracks (or cylinders) and each track is ______ _________ divided into a number of sectors. Again by convention, the tracks on _______ each surface are also numbered sequentially from zero, with track zero being the outermost track. The sectors on each track are normally numbered from one to a number corresponding to the total number of ___ sectors on the track, but adjacent sectors do not normally have ___ consecutive sector numbers. A typical arrangement of the seventeen sectors on a single track of an XT hard disk, for example, would be as follows: 1,8,15,5,12,2,9,16,6,13,3,10,17,7,14,4,11. This "inter- ______ leaving" of sectors helps to make multi-sector disk reads and writes _______ more efficient. The controller requires a finite time to process the data from one sector before it can start reading (or writing) the next. During this time, of course, the disk continues to rotate and one or more of the sectors which physically "follow" the one that was last accessed pass under the read/write head unread. Ideally, the Page 35 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ interleave (number of physical sectors separating consecutively numbered sectors) should be such that the next sector is just about to come under the read-write head when the controller is free to deal with it, but, because this time is not exactly predictable, it is better to be too conservative (as IBM usually is) and make it one or two sectors longer than necessary than to be too optimistic. Too small an interleave value is very undesirable, because it results in the controller having to wait almost a complete disk rotation time for each sector it misses to come round again. At the beginning of each disk sector is an address mark, a short, ____________ recognizable (to the controller) bit sequence containing the track and sector number. The process of writing the address marks for each sector on a disk is known as "low-level" or "sector" formatting. The DOS FORMAT program does this when it is run on floppy disks, but on hard disks the DOS program merely makes sure that all required address marks are present - they are created by the disk manufacturer or supplier's low-level format program, which may or may not be available to the end user. After each address mark there follows a short gap (to allow the controller time to recognize and respond to the address mark), then the CRC or ECC data, then the raw data. A sector normally contains 512 bytes of raw data, and DOS 4.xx requires that this be the ________ case, but DOS versions prior to 4.00 (and HDTEST) can recognize and deal appropriately with devices with non-standard sector sizes. When a sector read/write command is passed to the disk controller, it (the controller) causes the disk's read/write head to move (seek) to ____ the specified track, waits a few milliseconds for it to settle down, and then starts to "pay attention to" the data passing underneath it. The controller locates and reads the address marks to confirm that the read/write head is positioned at the correct track and to identify each sector. If data is being read from the disk, the raw data and recorded CRC or ECC from the requested sector(s) are transferred into the disk controller's internal buffers and (assuming that no errors are detected), thence to main memory. If data is being written, the controller derives the error-checking data from the raw data and writes both to the disk. Some of this process is under the direct control of the controller hardware, but it is supervised by the device's ROM BIOS, which is responsible for error-reporting. Errors can occur at almost any stage of the read/write process, but (if the disk is working at all) usually result from media defects or problems developing in the mechanical components of the hard disk drive. Media defects, once present, are generally constant in size and position (although jarring a disk while it is being written to may sometimes generate a temporary bad sector). _________ HDTEST DISK ERROR REPORTS _________________________ When HDTEST detects an error during read/write testing of a sector or cluster which has not previously been marked as bad, a (normally) four-line error message is printed, either to the screen (by default), Page 36 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ to the printer (/P option), or to the logging file (if an output device was specified). The first line of the message simply indicates the sector or cluster number at which the error occurred. The second line indicates what sort of error was detected and during what phase of read/write testing it occurred. The third line indicates whether HDTEST was able to recover and/or remap the data from the bad sector or cluster, and the fourth indicates either the file which a bad cluster formed part of or that the cluster was not in use. If it was not possible to read one or more sectors from a cluster which contained data (unrecoverable read error), a further two-line message is printed which notes how many sectors were unreadable and how the _______ data from the bad cluster has been remapped (which depends on whether a /Zn option was specified on the command line). If the /M flag is not specified when HDTEST is run on a system that suffers from the "multi- track read" problems discussed in the main part of the documentation, multiple bad clusters may be reported with no errors persisting on __ sector-by sector reading. These reports should disappear if the /M flag is used. If the "verbose" form of the /G option is used, the bad sector report will contain the equivalent of the second and third lines of the bad cluster report for each bad sector. Note that, if the /X option is not ___________________ used to prevent testing of previously flagged bad clusters, these may give rise to spurious warnings of possible data loss in the third line ________ of the bad sector report. If no unrecoverable bad clusters were ______ ________ reported, such warnings can safely be ignored. If the default form of the /G option is used, the bad sector report gives only a Hex error code for each bad sector. The LSB of this code is the value returned (in AH) by INT 13. The MSB of the code depends on the phase of testing during which the error occurred. In general, if the first digit (of a four digit code) is C or above, an unrecoverable error occurred, but error codes FFFD through FFFF have special meanings and usually indicate recovered errors. When the program detects an error during read/write testing of a sector in the DOS area, only the first three lines described above are printed and the program aborts. The remainder of this Appendix contains more details of and further information on the possible contents of lines two and three of HDTEST's disk error message. The second line of the disk read/write error message will usually take the form: "[Error Type] occurred during [phase of testing]". Possible values for [Error Type] are as follows (in alphabetical order): Address Mark Not Found - this is a fairly common error reported for bad blocks which are present from "scratch" on a new disk, since a significant media defect may well not retain sector formatting data and/or the manufacturer may have made sure that a known bad track on the disk will not be used by deliberately not writing address marks there. If new bad clusters are reported with this error message, it is likely that the hard disk surface has been physically damaged (perhaps Page 37 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ by a shock to the drive) but possible that the controller has a problem. Bad Controller - this means what it says, but check that all boards and cables are properly seated before giving up. Bad Sector - the BIOS was instructed to read a sector number which should not (according to its configuration data) exist on the device. This indicates either an internal program error or that the program is having a problem determining the physical layout of the disk. Check the values reported by HDCHEK against those in the disk documentation. Bad Track - analogous to "Bad Sector". Buffer Comparison - the data read from a sector differs from that written to it. This should not happen, since it suggests either that the device's error-detection functionality has failed or that it has been overwhelmed. If it does occur, the most likely causes are controller and memory problems (in that order of likelihood). Corrected ECC - as discussed at some length above, the data retrieved from a hard disk was incorrect but the device's error-correction hardware was (probably) able to correct it. DMA Boundary Violation - should not occur (HDTEST positions its buffers to avoid crossing segment boundaries). Suggests either an internal program error or a controller/BIOS problem. Hardware Timeout - a seek or other operation was not completed within the maximum time allowed by the BIOS. Check connections, reseat boards, otherwise suspect the mechanics of the disk. Init Fail - should not occur (on a working disk). Internal - internal program error. Make sure that your working copy of HDTEST has not been corrupted. The disk is probably not defective. ________ Invalid Command - should not occur. Treat as Internal Error. Multiple Errors - one or more additional errors occurred while HDTEST was attempting to recover from an initial one. Multiple phase report lines will be printed but the error types cannot be further specified. Recovered - HDTEST recovered from the error and did not report on the error type (should not occur). Reset Fail - The device could not be reset (after an error). Usually indicates major problems with the disk or controller, but check connections, etc. Sector Not Found - the requested sector's address mark was not found. Likely to result from a media defect. Page 38 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ Seek Fail - after a seek to a track, the controller found that the read/write head was not positioned correctly. Usually indicates a significant problem with the (drive) hardware. Sense Fail - the controller cannot find the drive! Check connections, etc. Uncorrected ECC - data error too major to be corrected. Usually a media defect. This message will also be produced if a CRC error is detected on a floppy disk. Undefined - the BIOS cannot figure out what went wrong, but something did! Unknown - likewise! Write Protection - HDTEST is attempting to write to a write-protected device. Possible forms for the rest of the second line are as follows: ... Error (unrecovered) occurred during first read - the cluster could not be read without error in five tries before a seek to the other "end" of the disk. ... Error occurred during first read - the error occurred during the first read but was recovered after one or more retries. ... Error occurred during read after seek - the error occurred when the cluster was read after a seek away. ... Error occurred during pattern read/writes - the error occurred during the pattern read/write phase of testing. ... Error occurred during final write - the error occurred when the saved copy of the cluster was being restored to the disk. ... Error occurred during read after final write - the error occurred when the restored cluster was being read back. ... (Buffer comparison) error occurred after final read - the copy of the cluster read back after restoration to the disk differed from its original memory image. There are only three basic forms of the third line: This error is (or "These errors are") recoverable (no data loss has occurred). - the bad cluster was read successfully and has been successfully restored to its original form, but it remains suspect and will (with your permission) be remapped. Page 39 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ This error is unrecoverable, but remapping is possible. - the bad cluster was read successfully but has not been restored to its ___ original form. Remapping should be successful, but the data will be lost if this is not permitted. This error is unrecoverable, and data may be lost! - the bad cluster was not read successfully and will be remapped as provided by the value of the /Z switch parameter if remapping is permitted. In any event, some data loss has almost certainly occurred. Page 40 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ APPENDIX C - EXAMPLE PROGRAM RUNS _________________________________ Screen display from successful run of HDCHEK on partition occupying the whole of a 20 Mb hard disk: Device C: appears to be the DOS partition of a hard disk It contains 41531 sectors It starts at absolute sector 17 on the disk {THIS IS CORRECT} _______________ The total partition size is approximately 20 Mb There are 4 heads and 17 sectors/track Sector size is 512 bytes Each of the 2 (16 bit) FATs occupies 41 sectors The Root Directory occupies 32 sectors The data area starts at DOS sector 115, and contains 10354 clusters, each consisting of 4 sectors The directory scan was successful; a total of 1260 files and subdirectories was found. Screen display from successful run of HDCHEK on a freshly FORMATted DSDD 9-sector floppy disk: Device A: is not a hard disk - using DOS interrupts Device A: is a floppy disk The disk contains 720 sectors There are 2 heads and 9 sectors/track Sector size is 512 bytes Each of the 2 (12 bit) FATs occupies 2 sectors The Root Directory occupies 7 sectors The data area starts at DOS sector 12, and contains 354 clusters, each consisting of 2 sectors The directory scan was successful; a total of 0 files and subdirectories was found. Screen display from run of HDCHEK on 20 Mb DOS partition - directory structure bad or unrecognizable: Device C: appears to be the DOS partition of a hard disk It contains 41531 sectors It starts at absolute sector 17 on the disk The total partition size is approximately 20 Mb There are 4 heads and 17 sectors/track Sector size is 512 bytes Each of the 2 (16 bit) FATs occupies 41 sectors The Root Directory occupies 32 sectors The data area starts at DOS sector 115, and contains 10354 clusters, each consisting of 4 sectors The preliminary directory scan failed. Page 41 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ Screen display from run of HDCHEK on 20 Mb DOS partition - FATs not identical: FATS are different FAT discrepancy at byte 20F - FAT 1 has FF; FAT 2 has F8 Continue comparison (Y/N)? [Y]: n {user input} _ Device C: appears to be the DOS partition of a hard disk It contains 41531 sectors It starts at absolute sector 17 on the disk The total partition size is approximately 20 Mb There are 4 heads and 17 sectors/track Sector size is 512 bytes Each of the 2 (16 bit) FATs occupies 41 sectors The Root Directory occupies 32 sectors The data area starts at DOS sector 115, and contains 10354 clusters, each consisting of 4 sectors Screen display from run of HDCHEK - program unable to deduce disk structure - disk not recognizable as a DOS device: Unrecognized media descriptor byte! - see error code Error Code = F1 {F1 is not a media descriptor byte used by DOS} Page 42 Documentation for HDTEST V5.35a Copyright 1991, P. R. Fletcher ________________________________ _______________________________ Log file from successful run of HDTEST on 20 Mb DOS partition: HDTEST Log File - Wed Jan 07 11:44:52 1986 Testing Device C: No new bad clusters identified - HDTEST finished. Wed Jan 07 16:10:41 1987 Log file from run of HDTEST on 20 Mb DOS partition - one bad cluster detected and remapped: HDTEST Log File - Mon Jan 05 21:09:32 1987 Testing Device C: Read/Write error at cluster number 7623!! Corrected ECC error occurred during pattern read/writes This error is recoverable (no data loss has occurred). Bad cluster is in file \MYFILES\DATA5\FFT1234.DAT 1 new bad clusters identified, marked, and remapped - HDTEST finished. Tue Jan 06 05:08:04 1987 Page 43 REGISTRATION FORM - HDTEST V5.35 ________________________________ Please complete this form and return it with a check or money order for $35.00 (+ $5.00 for shipping outside North America) or equivalent to: Computer Counselors, Inc. 1515 West Montgomery Avenue Rosemont PA 19010 U.S.A. None of the information on this form will be passed on to any other individual or commercial organization. Pennsylvania residents must pay Sales Tax (currently 6%) on all orders. -------------------------------------------------------------------------- This information is required to process your Registration: Last Name:____________________________ Other Names:______________________ Address:______________________________ Town/City:________________________ State:_____________ Post Code:____________ Telephone:_________________ (evenings/weekends) I understand that this software is supplied without warranty, express or implied, and agree not to hold its author liable for any direct or consequential loss arising out of my use of it. Signature:______________________________ ------------------------------------------------------------------------- You are requested, but not required, also to answer the following questions: On What type(s) of computer are you using HDTEST? _________________________ CPU (8088, 8086, etc.)? __________ Amount of RAM (kB)? ____________ Type & Size of Hard Disk(s)? ____________________________________ Make and Type of Hard Disk Controller(s)? ______________________________ Do you own a Modem? Y N Are you a member of BIX? Y N Compuserve? Y N How did you obtain/hear about this program? Did you think that the $35.00 registration fee for this program was: Very Good Value? About Right? Excessively High? Are there any other utility programs that you would like to see made available as "User Supported Software"? What sort of programs?