Z P M 3 by Simeon Cran ======================== A Z80 coded CP/M 3.0 compatible BDOS replacement. The first public release: 27/3/92 This document dated: 24/4/92 Distributed at: Z-Node 62 (Perth, Western Australia) V21,V22,V22bis 09 450 0200 WELCOME TO ZPM3 ~~~~~~~~~~~~~~~ Welcome to the best CP/M compatible operating system for Z80 based computers with banked memory. The best? Yes, we believe so. CP/M 3.0 has had bad press, but the fact is that it is faster than CP/M 2.2 ever was, and it offered more integrated facilities. Perhaps it was all the Z80 replacement BDOSes for CP/M 2.2 which stole the limelight from CP/M 3.0, or was it just that few computers had the required banked memory? Whatever the reason for CP/M 3.0's lack of success in the marketplace, there are still plenty of users who will stand by its wonderful facilities and speed. For those users ZPM3 provides the long awaited Z80 coded update. ZPM3 offers all the good things that CP/M 3.0 does, and then it offers more. Because ZPM3 is written in Z80 code rather than the 8080 code of CP/M 3.0, it can do everything that CP/M 3.0 does, but in much less space. With the extra space recovered, ZPM3 packs in a number of new facilities. Yet the whole package fits in exactly the same space as CP/M 3.0 so you can directly replace your old CP/M 3.0 BDOS with ZPM3 without a worry. ZPM3 is also fast. Faster, in fact, than CP/M 3.0. This is possible because the rich Z80 instruction set allows many algorithms to be implemented more efficiently. In addition, the extra space available in ZPM3 has been put to use to further optimise the code. Lots of small optimisations smooth the execution flow, so ZPM3 becomes the fastest operating system on most banked CP/M computers. THE FEATURES ~~~~~~~~~~~~ ZPM3, in addition to complete CP/M 3.0 compatibility, offers the following features: Random Read Bug fixed. ++++++++++++++++++++++ Maybe you didn't know, but CP/M 3.0 has a bug. It affects random reads under very specific circumstances, and can result in a program thinking that you don't have some pieces of data in a file when in fact you do. The bug would occur very, very rarely, but it is real. ZPM3 finally squashes it. Protected SCB User code +++++++++++++++++++++++ The System Control Block of CP/M 3.0 was a revolution at the time. ZCPR has a system environment and most other operating systems have other similar structures, but the SCB of CP/M 3.0 was one of the very first. Unfortunately, Digital Research never properly documented it, and some programmers found things out about it that weren't quite true and started programming accordingly. As well, because it is available in the TPA bank, runaway programs can overwrite it causing problems. Mostly though, the SCB will survive, or at least any problems will be so obvious that the user will realise that a crash has occurred and will reboot. A real problem exists with the CP/M 3.0 code however when the user value is written over with a value above 15. Many programs now directly write to this byte, and if they put a value in that is above 15, all sorts of havoc can happen with the disk system. Actually, CP/M 3.0 will handle user areas above 15 with this method, and all seems ok until the operating system mistakes one of these directory entries as an XFCB. Simply put, user areas above 15 must not be used with CP/M 3.0. ZPM3 has code which prevents these problems, making the system even more stable. Obsoleted Trap system. ++++++++++++++++++++++ One of the problems of the banked operating system was that it was possible to redirect the BIOS to code below common memory, in which case the banked BDOS could not access it. One solution is to call all BIOS code from common memory, but this involves a bank switch for every BIOS call, and this slows things down considerably. CP/M 3.0 got around the problem by providing special code just below the SCB. If you redirected the BIOS, you also had to change this code which caused a bank switch when your new BIOS routine was called. When you removed the redirection, you also had to restore the special code. This system has major drawbacks. For a start, if you redirect the BIOS, then another program redirects your redirection, then you remove your first redirection (along with the special code), the bank switch won't happen for the second redirection and the system will crash. If a CP/M 2.2 program tried to do the redirection, it would know nothing about CP/M 3.0 and would not adjust the special code, so a crash would result in that case too. The special code was called the "Trap System" as it was meant to trap redirection (as long as you set the trap). ZPM3 has eliminated the need for the traps. They are still there, and programs can still fiddle with them, but it doesn't matter how they are set, they are ignored. There is simply no need for them anymore. And this has been achieved without a performance penalty. In fact, in the case of a program which sets the traps but forgets to restore them, performance is now much better. Semi-Permanent Read Only status for drives. +++++++++++++++++++++++++++++++++++++++++++ In recent years, a trend in CP/M 2.2 is to make drives which have been set read only to remain that way until explicitly changed by function 37. ZPM3 now adopts this logic. Previously a control-C would return a read only drive to read write. The advantage is that a program can now make a drive read only for a session and know that it will stay that way. ZCPR compatible function 152 ++++++++++++++++++++++++++++ Function 152 is the CP/M 3.0 parser. It was a great innovation at the time as parsing is one of the more tedious aspects of programming for CP/M. Unfortunately, almost as soon as it appeared, it was made obsolete by the fact that it didn't handle references to user number (DU references). A line such as A:FILE.TYP would be correctly parsed, but A3:FILE.TYP would not. CP/M 3.0 programs would often parse the drive and user separately, then give function 152 the line without the DU: reference. All this extra work should not have been necessary if CP/M 3.0 had included user number parsing. ZPM3 parses the user number, and goes even further by handling named directories for ZCPR. This is possible as long as you set a special word in the SCB which tells ZPM3 where to find the ZCPR system environment descriptor. ZCCP, a companion CCP for ZPM3, handles this automatically, but for Z3PLUS users, a special utility is available which automatically sets this word. The result is that CP/M 3.0 programs will not balk at DU: references and ZPM3 aware programs can use the full DU: and DIR: facilities of function 152. It has also made the brilliant ZCCP code possible. New Functions 54 and 55 +++++++++++++++++++++++ Datestamps in CP/M 3.0 are wonderful, but difficult to manipulate. Two new functions make them easier to handle and at the same time give compatibility to Z80DOS aware programs. Function 54 (Get Stamp) returns a Z80DOS compatible datestamp. Any program (such as many directory programs) which recognise the Z80DOS standard can make use of function 54. There is only one slight difference between Z80DOS datestamps and ZPM3's which you should be aware of. Z80DOS will return a correct datestamp after any successful open or search of any extent. ZPM3 can only return a correct datestamp after a successful open or search of the first extent of the file. This is because CP/M 3.0 datestamps are only saved for the first extents of each file, in order to provide the highest performance. Even more interesting is Function 55 (Use Stamp) which provides a mechanism for changing datestamps on files. Trying to do this with CP/M 3.0 was virtually impossible because it involved direct sector writes. With Function 55 you can simply set the stamp and then write. Wheel protected files +++++++++++++++++++++ If you are using a ZCPR system (ZCCP or Z3PLUS), ZPM3 has access to the wheel byte and supports wheel protected files. Such files act normally if the wheel is set (signifying a priveleged user), but if the wheel is not set, the files can not be changed. This is of most benefit to BBS systems. The implementation is virtually the same as most current Z80 CP/M 2.2 compatible BDOSes. Better error messages +++++++++++++++++++++ CP/M 3.0 introduced the best error messages that CP/M had ever had. ZPM3 goes further. The main difference you will notice is that the user number as well as the drive is shown in the error message. This is invaluable in helping you identify which file might have caused a problem. Function 10 history buffer and improved editing. ++++++++++++++++++++++++++++++++++++++++++++++++ Function 10 is used by the CCP to input command lines. Many other programs use function 10 for input. CP/M 3.0 introduced a history buffer for function 10. You press control-W and you were returned the last command. It is a great facility, but because it only remembers one command it is rather limited. There have been RSXes written which give a much larger history buffer, but RSXes take up extra program memory so are undesirable. ZPM3 gives a large (approximately 250 bytes) history buffer which can store multiple commands. It also makes very intelligent use of the buffer so that identical commands are not stored twice, and commands of less than three characters are not stored. The history buffer takes up no additional memory, and is always available. For security, it is possible to clear the history buffer so that other users can not see what commands you have used. The ZPM3 history buffer feature is so good, that for many users, the ZPM3 upgrade is completely justified by it. As part of the history buffer system, ZPM3 also offers a facility called Automatic Command Prompting. This can be disabled, or can be made switchable from the keyboard. When it is on, ZPM3 tries to fill in the rest of your command based on what commands you used most recently. It is like magic, and can save you typing out complicated commands many times. In effect, it looks through the history buffer for you and finds the command it thinks you want. As you keep typing, if it turns out that the command doesn't match anymore, it will try to match another command, and if it can't, it lets you make the command by yourself. This facility is quite amazing to watch. And to integrate the history buffer and the automatic command prompting, function 10 has the best command line editing you'll find anywhere. Most of the control keys do something when you are editing a function 10 line, and for the most part they mimic the standard WordStar/NewWord/ZDE functions. You can jump to different words in the command, delete individual words, delete individual letters, insert letters, and a whole lot more. Here is a list of what the various control keys do for function 10: A Move left one word B Go to the beginning or end of the line C Warm boot if at start of line, otherwise nothing D Go right one character E Go backwards one command in the history buffer F Go right one word G Delete current character H Destructive backspace I J Enter line K Delete all to the right L M Enter line N O P Toggle printing Q Toggle automatic command prompting (if enabled) R S Go left one character T Delete current word U Add current line to history buffer V Clear line and delete from history buffer W Go forwards one command in the history buffer X Delete all to the left Y Clear the whole line Z CPMLDR.REL bug fixed. +++++++++++++++++++++ If you have ever tried to use the CPMLDR.REL code supplied with CP/M 3.0 to load a CPM3.SYS file larger than 16k, you have probably come across the CPMLDR.REL bug. The computer probably crashed, and you were left wondering what you did wrong in your bios. Well CPMLDR.REL has a bug. To solve this for you ZPM3 comes with ZPM3LDR.REL which directly replaces CPMLDR.REL. It is also somewhat better in that all the messages, and the fcb for loading CPM3.SYS, are at the start of the file along with plenty of spare room. As a result you can easily patch the signon and error messages to say whatever you like and even change the FCB to load a file called something other than CPM3.SYS. All About the Random Read Bug. ============================== Never heard of it? Well it's there in CP/M 3.0. I spent a lot of time trying to work out what it was and just why it was happening, and if you are interested, here are the details. CP/M 3.0 uses the Record Count byte of an active FCB a little differently from the way CP/M 2.2 does. It is mentioned in the CP/M 3.0 manuals that the record count may contain numbers greater than 128, but in such a case it implies that the record count is really 128. CP/M 2.2 would not return record counts greater than 128. The reason for the use of the record count in this way is to help speed up some of the logic used to find records in a file. It works very well for sequential access. When it comes to random access, the system has some failings. The idea behind CP/M 3.0's unusual use of the record count is to keep the record count of the last logical extent of the current physical extent always in the Record Count byte. When accessing extents before the last one, bit 7 of the byte is set. That way it will always be at least 128 for logical extents before the last (which CP/M 3.0 translates to mean equal to 128), and the lower 7 bits are used as convenient storage for the record count of the last logical extent. This is particularly convenient because it means there is no need to go and read the directory entry again when it comes time to read the last logical extent. I hope you have followed that! In sequential access, this scheme is great. The problem occurs with random access. In this case it is possible to access a logical extent which has no records in it. This could be any logical extent past the last one. In such a case the record count must be returned as 0 (which is correct). If we then go back to a previous logical extent in the same physical extent, CP/M 3.0 gets confused and assumes that there must be 128 records in that extent because the one we just came from had no records and we are now accessing an earlier extent. You're probably well and truly lost by now! Anyhow, the assumption that CP/M 3.0 makes is quite wrong. The record count ends up being set to 128, a read is allowed to go ahead as if nothing was wrong, no error is returned, and the record count remains incorrectly set until a different physical extent is opened. The result could be chaos, but mostly it just means that a program returns the wrong information. Remember, a logical extent is always 16k. A physical extent can be a multiple of 16k and is all the data described by one directory entry. If your system has physical extents which are 16k, you would never have the problem because a new physical extent would be properly opened for every new logical extent that was accessed. Typically though, a physical extent is 32k, so it holds 2 logical extents. The problem won't arise until the file grows past the 32k mark in such a case. And when the file gets over 48k the problem can't occur again until it gets over 64k... and so on. Even then, it can only happen if reads are attempted to particular extents in a particular order. So you shouldn't be too surprised if the bug hasn't been too noticeable to you. ZPM3 squashes the bug once and for all by using the correct logic. In the situation where the bug would normally occur, ZPM3 makes sure it gets the correct record count information, and the reads return the correct record count every time. If you are interested in seeing a demonstration of the bug in action (on CP/M 3.0) and comparing it with ZPM3, there is a file floating around various bulletin boards which contains demonstrations for the bug and an RSX to fix it. The RSX is a less than perfect way of overcoming the bug, although it seems to work. However, now that you have ZPM3, you don't need to worry. LEGALS and SUCH =============== The ZPM3 package is supplied free of charge, on the condition that you don't use it to make money. If you want to use it commercially you must contact me to get the OK (and negotiate our fee). If you find anyone (except myself) charging money for ZPM3, please inform me! Nobody is making any guarantees about this software. None at all. If it causes your house to burn down, or a divorce, or just a bad day, this is unfortunate, regrettable, but there is nothing that I can or will do about it. You have been warned. The ZPM3 package must only be distributed in the form that you found it. Do not change or add anything. Don't even change it into a different type of archive. Just leave it alone. However you are free to distribute it to as many places and people that you can. Just don't charge for it. If in using ZPM3 you find that it doesn't act as described, please forward the details to me so that either the ZPM3 code or the documentation can be changed. If you would like further details, please forward your specific questions to me. SJC. As a service to all our ZPM3 fans, the latest version of the ZPM3 package can now be ordered. At this stage we can only supply IBM formatted 3.5 inch 720k disks, however if you are keen enough that shouldn't matter. ZPM3 remains free, however this service will cost you $15 Australian (for the disk, copying, postage and packing) to most places in the Western World (others by arrangement). This is a good way to guarantee you have the latest version, and to guarantee that your package has not been corrupted by some unscrupulous person. When we fill your order, we will make sure to include the latest demonstration copy of MYZ80 - the fastest and best Z80 emulator for IBM AT (and better) compatibles. MYZ80 can run ZPM3 with ease. It also handles ZCPR and CP/M 2.2. And yes, we do mean FASTEST. Send your international money order to: Software by Simeon ZPM3 Package 2 Maytone Ave Killara NSW Australia 2071 Your order will be promptly filled.