Here is a copy of an article that recently appeared in a periodical magizine. The informatin it contains and the opinions expressed in it are not mine. I'm sure you will find it useful and informitive.
In last month's article on the P9, we described the method used to access extended memory (memory beyond the lower 1 megabyte) from real-mode programs running on a 286. This method requires switching to protected mode to perform access, and then resetting the processor to return to real mode. We have since learned that there is another way. An undocumented 286 instruction, LOADALL, allows all of the processor's registers (including protected mode registers and hidden internal registers) to be loaded, even when operating in real mode. By changing the value of the descriptor cache base register, a program can select a segment beyond the lower 1Mbyte. LOADALL also has other uses, as described later in this article.
---------------------------------------------------------------------- Physical Memory Address CPU register (hex) ====================================================================== 800-805 none 806-807 MSW (Machine Status Word) 808-815 None 816-817 TR (Task Register) 818-819 Flag Word 81A-81B IP (Instruction Pointer) 81C-81D LDT (Local Descriptor Table) 81E-81F DS (Data Segment Selector) 820-821 SS (Stack Segment Selector) 822-823 CS (Code Segment Selector) 824-825 ES (Extra Segment Selector) 826-827 DI (Destination Index) 818-829 SI (Source Index) 82A-82B BP (Base Pointer) 82C-82D SP (Stack Pointer) 82E-82F BX (Data Register B) 830-831 DX (Data Register D) 832-833 CX (Data Register C) 834-835 AX (Accumulator) 836-83B ES Descriptor Cache 83C-841 CS Descriptor Cache 842-847 SS Descriptor Cache 848-84D DS Descriptor Cache 84E-853 GDTR (Global Descriptor Table Register) 854-859 LDT Descriptor Cache 85A-85F IDTR (Interrupt Descriptor Table Register) 860-865 TSS (Task State Segment) Descriptor Cache ======================================================================= Table 1. LOADALL data area format -----------------------------------------------------------------------
Originally included by Intel for chip testing, Microsoft is now using this instruction in their RAM Drive program and in OS/2's compatibility box. While this instruction is probably appropriate only for use in operating systems and system-level utilities, it is important because it provides a set of capabilities that are not otherwise available in a 286-based system.
We have received a copy of a document that describes LOADALL. No company name is shown, but is almost surely written by Intel. The 15-page document describes in detail how to use the instruction, so calling it "undocumented" is not quite correct -- it is documented, but not in the data sheet. By restricting access to this documentation, Intel gives unfair advantage to Microsoft and other large customers.
---------------------------------------------------------------------- Descriptor Cache Format Bytes 0-2 24-bit segment base address Byte 3 Access rights byte. Format is the same as the access rights byte in a descriptor, except that the "present" bit becomes a "valid" bit. If a "valid" bit is not set, any memory reference using the descriptor will cause exception 13 with error code of 0. Bytes 4-5 16-bit segment size GDTR and IDTR Format Bytes 0-2 24-bit base address Byte 3 0 Bytes 4-5 16-bit segment limit ---------------------------------------------------------------------- Table 2. Descriptor Cache Formats
Intel's position is that this instruction is not useful to most users, and can easily lead to machine crashes if not used properly. Intel says that they will provide documentation for the instruction on a "need-to-know" basis, presumably only after a non-disclosure agreement has been signed. Most developers, of course, would not know to ask. The information we are presenting here should be enough for you to evaluate the usefulness of the instruction; if you intend to use it in a product, you should contact Intel for the full documentation.
The opcode for LOADALL is 0F05 hex. No operands follow the instruction; LOADALL gets its data from a 102-byte block of memory starting at the fixed address 800 hex, as shown in Table 1. Table 2 shows the format of the descriptor cache entries. These are hidden registers that cannot otherwise be modified by the programmer; they are set automatically when a descriptor is read from the descriptor table.
The following sequence is required to access high memory from real mode using LOADALL:
Thus while it is appealing not to have to go to protected mode and back to access high memory from real mode, this isn't much better. LOADALL requires 195 clock cycles on a machine with no wait states, or 19.5 us at 10Mhz.
LOADALL can also be used to allow programs to be executed in high memory, even though the processor is in real mode. A special paragraph ID (such as FFFF) is used to indicate when a program is running in extended memory. Interrupt service routines must check the segment register to see if any contain the special ID, and if so, they must reload the registers using LOADALL to restore the segment base addresses before returning from the interrupt.
LOADALL has other uses as well. It can be used to switch back to a protected mode task from real mode, serving as a fast intermode context switch. OS/2 presumably uses it to return to protected mode from a real-mode task running in the compatibility box.
LOADALL can also be used to emulate real mode from protected mode. By setting the privilege level of all local and global descriptor table entries lower than that of the current program, a protection violation occurs whenever the program attempts to load a segment register. (Actually, loading the segment register with a value of 0000 to 0003 does not cause an exception, and this case must be handled specially.) The exception handler then uses LOADALL to set the base address for the segment.
There are many other complexities to this emulation, as described in the LOADALL document. The emulation is imperfect, and Microsoft apparently decided to use the "reset to real mode" technique (as described in our article last month on the P9), rather than trying to emulate real mode from within protected mode.
LOADALL performs no checking on the values loaded into the registers, so no exception will occur even if an illegal value is loaded. Thus, the processor can potential be put into a strange state. If an illegal descriptor value is set, no exception occurs from the execution of LOADALL. An exception will occur, however, when an access using that descriptor is attempted.
LOADALL can be executed in protected mode, but only at the most privileged level (level 0). Thus it does not violate the protection. Unfortunately, LOADALL cannot be used to switch back to real mode from protected mode.
Early versions of 286 (A1 and B1 steppings) have bugs which affect the use of LOADALL. Thus, the techniques described in this article may not work properly on older systems.
LOADALL is not implemented on the 80386, so Microsoft has included code in OS/2 and in the current version of RAM Drive that tests for the processor type. If it's a 286, LOADALL is used, and if it's a 386, the native 386 mechanisms are used. If you use LOADALL, you should also perform this check, or your software will not run on 386-based machines.
Using LOADALL is certainly fraught with peril, and requires careful program design and knowledge of all the implications. We have spoken with a number of software vendors that were aware of the instruction, and all had decided not to use it due to the risks and complexity involved. We don't recommend its use to most people, but we do feel that everyone should have access to the information so they can make their own choice about whether or not to use the instruction.