Intel 80386 Reference Programmer's Manual
Table of Contents
Part I Applications Programming
Part II Systems Programming
Chapter 4 -- Systems Architecture
Chapter 6 -- Protection
Chapter 7 -- Multitasking
Chapter 8 -- Input/Output
Chapter 9 -- Exceptions and Interrupts
Chapter 10 -- Initialization
Chapter 14 -- 80386 Real-Address Mode
Chapter 16 -- Mixing 16-Bit and 32 Bit Code
Chapter 1 -- Introduction to the 80386
80386 - 32-битный процессор,оптимизированный для многозадачных операционных систем
и разработанный для приложений высокой производительности.
32-битные регистры поддерживают 32-битную адресацию и типы данных.
Процессор может адресовать память в диапазоне 4 гигабайт физической памяти
или 64 терабайт (2^(46) bytes) виртуальной памяти.
Встроенное управление памятью облегчает трансляцию адресов,многозадачность,защиту.
Специальные регистры создают возможность создавать точки прерывания при ROM-дебаге.
This chapter describes the 80386 application programming environment as
seen by assembly language programmers when the processor is executing in
protected mode. The chapter introduces programmers to those features of the
80386 architecture that directly affect the design and implementation of
80386 applications programs. Other chapters discuss 80386 features that
relate to systems programming or to compatibility with other processors of
the 8086 family.
The basic programming model consists of these aspects:
- Memory organization and segmentation
- Data types
- Registers
- Instruction format
- Operand selection
- Interrupts and exceptions
Note that input/output is not included as part of the basic programming
model. Systems designers may choose to make I/O instructions available to
applications or may choose to reserve these functions for the operating
system. For this reason, the I/O features of the 80386 are discussed in
Part II.
Chapter 2 -- Basic Programming Model
prev: Chapter 2 -- Basic Programming Model
next: 2.2 Data Types
2.1 Memory Organization and Segmentation
The physical memory of an 80386 system is organized as a sequence of 8-bit
bytes. Each byte is assigned a unique address that ranges from zero to a
maximum of 2^(32) -1 (4 gigabytes).
80386 programs, however, are independent of the physical address space.
This means that programs can be written without knowledge of how much
physical memory is available and without knowledge of exactly where in
physical memory the instructions and data are located.
The model of memory organization seen by applications programmers is
determined by systems-software designers. The architecture of the 80386
gives designers the freedom to choose a model for each task. The model of
memory organization can range between the following extremes:
- A "flat" address space consisting of a single array of up to 4
gigabytes.
- A segmented address space consisting of a collection of up to 16,383
linear address spaces of up to 4 gigabytes each.
Both models can provide memory protection. Different tasks may employ
different models of memory organization. The criteria that designers use to
determine a memory organization model and the means that systems programmers
use to implement that model are covered in Part -- Programming.
2.1.1 The "Flat" Model
In a "flat" model of memory organization, the applications programmer sees
a single array of up to 2^(32) bytes (4 gigabytes). While the physical
memory can contain up to 4 gigabytes, it is usually much smaller; the
processor maps the 4 gigabyte flat space onto the physical address space by
the address translation mechanisms described in
Chapter 5
. Applications
programmers do not need to know the details of the mapping.
A pointer into this flat address space is a 32-bit ordinal number that may
range from 0 to 2^(32) -1. Relocation of separately-compiled modules in this
space must be performed by systems software (e.g., linkers, locators,
binders, loaders).
2.1.2 The Segmented Model
In a segmented model of memory organization, the address space as viewed by
an applications program (called the logical address space) is a much larger
space of up to 2^(46) bytes (64 terabytes). The processor maps the 64
terabyte logical address space onto the physical address space (up to 4
gigabytes ) by the address translation mechanisms described in
Chapter 5
.
Applications programmers do not need to know the details of this mapping.
Applications programmers view the logical address space of the 80386 as a
collection of up to 16,383 one-dimensional subspaces, each with a specified
length. Each of these linear subspaces is called a segment. A segment is a
unit of contiguous address space. Segment sizes may range from one byte up
to a maximum of 2^(32) bytes (4 gigabytes).
A complete pointer in this address space consists of two parts (see
Figure 2-1
):
- A segment selector, which is a 16-bit field that identifies a
segment.
- An offset, which is a 32-bit ordinal that addresses to the byte level
within a segment.
During execution of a program, the processor associates with a segment
selector the physical address of the beginning of the segment. Separately
compiled modules can be relocated at run time by changing the base address
of their segments. The size of a segment is variable; therefore, a segment
can be exactly the size of the module it contains.
2.2 Data Types
Bytes, words, and doublewords are the fundamental data types (refer to
Figure 2-2
). A byte is eight contiguous bits starting at any logical
address. The bits are numbered 0 through 7; bit zero is the least
significant bit.
A word is two contiguous bytes starting at any byte address. A word thus
contains 16 bits. The bits of a word are numbered from 0 through 15; bit 0
is the least significant bit. The byte containing bit 0 of the word is
called the low byte; the byte containing bit 15 is called the high byte.
Each byte within a word has its own address, and the smaller of the
addresses is the address of the word. The byte at this lower address
contains the eight least significant bits of the word, while the byte at the
higher address contains the eight most significant bits.
A doubleword is two contiguous words starting at any byte address. A
doubleword thus contains 32 bits. The bits of a doubleword are numbered from
0 through 31; bit 0 is the least significant bit. The word containing bit 0
of the doubleword is called the low word; the word containing bit 31 is
called the high word.
Each byte within a doubleword has its own address, and the smallest of the
addresses is the address of the doubleword. The byte at this lowest address
contains the eight least significant bits of the doubleword, while the byte
at the highest address contains the eight most significant bits.
Figure 2-3
illustrates the arrangement of bytes within words anddoublewords.
Note that words need not be aligned at even-numbered addresses and
doublewords need not be aligned at addresses evenly divisible by four. This
allows maximum flexibility in data structures (e.g., records containing
mixed byte, word, and doubleword items) and efficiency in memory
utilization. When used in a configuration with a 32-bit bus, actual
transfers of data between processor and memory take place in units of
doublewords beginning at addresses evenly divisible by four; however, the
processor converts requests for misaligned words or doublewords into the
appropriate sequences of requests acceptable to the memory interface. Such
misaligned data transfers reduce performance by requiring extra memory
cycles. For maximum performance, data structures (including stacks) should
be designed in such a way that, whenever possible, word operands are aligned
at even addresses and doubleword operands are aligned at addresses evenly
divisible by four. Due to instruction prefetching and queuing within the
CPU, there is no requirement for instructions to be aligned on word or
doubleword boundaries. (However, a slight increase in speed results if the
target addresses of control transfers are evenly divisible by four.)
Although bytes, words, and doublewords are the fundamental types of
operands, the processor also supports additional interpretations of these
operands. Depending on the instruction referring to the operand, the
following additional data types are recognized:
-
Integer:
-
A signed binary numeric value contained in a 32-bit doubleword, 16-bit word,
or 8-bit byte. All operations assume a 2's complement representation. The
sign bit is located in bit 7 in a byte, bit 15 in a word, and bit 31 in a
doubleword. The sign bit has the value zero for positive integers and one
for negative. Since the high-order bit is used for a sign, the range of an
8-bit integer is -128 through +127; 16-bit integers may range from -32,768
through +32,767; 32-bit integers may range from -2^(31) through +2^(31) -1.
The value zero has a positive sign.
-
Ordinal:
-
An unsigned binary numeric value contained in a 32-bit doubleword,
16-bit word, or 8-bit byte. All bits are considered in determining
magnitude of the number. The value range of an 8-bit ordinal number
is 0-255; 16 bits can represent values from 0 through 65,535; 32 bits
can represent values from 0 through 2^(32) -1.
-
Near Pointer:
-
A 32-bit logical address. A near pointer is an offset within a segment.
Near pointers are used in either a flat or a segmented model of memory
organization.
-
Far Pointer:
-
A 48-bit logical address of two components: a 16-bit segment selector
component and a 32-bit offset component. Far pointers are used by
applications programmers only when systems designers choose a
segmented memory organization.
-
String:
-
A contiguous sequence of bytes, words, or doublewords. A string may
contain from zero bytes to 2^(32) -1 bytes (4 gigabytes).
-
Bit field:
-
A contiguous sequence of bits. A bit field may begin at any bit position
of any byte and may contain up to 32 bits.
-
Bit string:
-
A contiguous sequence of bits. A bit string may begin at any bit position
of any byte and may contain up to 2^(32) -1 bits.
-
BCD:
-
A byte (unpacked) representation of a decimal digit in the range 0 through 9.
Unpacked decimal numbers are stored as unsigned byte quantities. One
digit is stored in each byte. The magnitude of the number is determined from
the low-order half-byte; hexadecimal values 0-9 are valid and are
interpreted as decimal numbers. The high-order half-byte must be zero for
multiplication and division; it may contain any value for addition and
subtraction.
-
Packed BCD:
-
A byte (packed) representation of two decimal digits, each in the range
0 through 9. One digit is stored in each half-byte. The digit in the
high-order half-byte is the most significant. Values 0-9 are valid in each
half-byte. The range of a packed decimal byte is 0-99.
Figure 2-4
graphically summarizes the data types supported by the 80386.
The 80386 contains a total of sixteen registers that are of interest to the
applications programmer. As
Figure 2-5
shows, these registers may be
grouped into these basic categories:
- General registers. These eight 32-bit general-purpose registers are
used primarily to contain operands for arithmetic and logical
operations.
- Segment registers. These special-purpose registers permit systems
software designers to choose either a flat or segmented model of
memory organization. These six registers determine, at any given time,
which segments of memory are currently addressable.
- Status and instruction registers. These special-purpose registers are
used to record and alter certain aspects of the 80386 processor state.
2.3.1 General Registers
The general registers of the 80386 are the 32-bit registers EAX, EBX, ECX,
EDX, EBP, ESP, ESI, and EDI. These registers are used interchangeably to
contain the operands of logical and arithmetic operations. They may also be
used interchangeably for operands of address computations (except that ESP
cannot be used as an index operand).
As
Figure 2-5
shows, the low-order word of each of these eight registers
has a separate name and can be treated as a unit. This feature is useful for
handling 16-bit data items and for compatibility with the 8086 and 80286
processors. The word registers are named AX, BX, CX, DX, BP, SP, SI, and DI.
Figure 2-5
also illustrates that each byte of the 16-bit registers AX, BX,
CX, and DX has a separate name and can be treated as a unit. This feature is
useful for handling characters and other 8-bit data items. The byte
registers are named AH, BH, CH, and DH (high bytes); and AL, BL, CL, and DL
(low bytes).
All of the general-purpose registers are available for addressing
calculations and for the results of most arithmetic and logical
calculations; however, a few functions are dedicated to certain registers.
By implicitly choosing registers for these functions, the 80386 architecture
can encode instructions more compactly. The instructions that use specific
registers include: double-precision multiply and divide, I/O, string
instructions, translate, loop, variable shift and rotate, and stack
operations.
2.3.2 Segment Registers
The segment registers of the 80386 give systems software designers the
flexibility to choose among various models of memory organization.
Implementation of memory models is the subject of Part II -- Systems
Programming. Designers may choose a model in which applications programs do
not need to modify segment registers, in which case applications programmers
may skip this section.
Complete programs generally consist of many different modules, each
consisting of instructions and data. However, at any given time during
program execution, only a small subset of a program's modules are actually
in use. The 80386 architecture takes advantage of this by providing
mechanisms to support direct access to the instructions and data of the
current module's environment, with access to additional segments on demand.
At any given instant, six segments of memory may be immediately accessible
to an executing 80386 program. The segment registers CS, DS, SS, ES, FS, and
GS are used to identify these six current segments. Each of these registers
specifies a particular kind of segment, as characterized by the associated
mnemonics ("code," "data," or "stack") shown in
Figure 2-6
. Each register
uniquely determines one particular segment, from among the segments that
make up the program, that is to be immediately accessible at highest speed.
The segment containing the currently executing sequence of instructions is
known as the current code segment; it is specified by means of the CS
register. The 80386 fetches all instructions from this code segment, using
as an offset the contents of the instruction pointer. CS is changed
implicitly as the result of intersegment control-transfer instructions (for
example, CALL and
JMP), interrupts, and exceptions.
Subroutine calls, parameters, and procedure activation records usually
require that a region of memory be allocated for a stack. All stack
operations use the SS register to locate the stack. Unlike CS, the SS
register can be loaded explicitly, thereby permitting programmers to define
stacks dynamically.
The DS, ES, FS, and GS registers allow the specification of four data
segments, each addressable by the currently executing program. Accessibility
to four separate data areas helps programs efficiently access different
types of data structures; for example, one data segment register can point
to the data structures of the current module, another to the exported data
of a higher-level module, another to a dynamically created data structure,
and another to data shared with another task. An operand within a data
segment is addressed by specifying its offset either directly in an
instruction or indirectly via general registers.
Depending on the structure of data (e.g., the way data is parceled into one
or more segments), a program may require access to more than four data
segments. To access additional segments, the DS, ES, FS, and GS registers
can be changed under program control during the course of a program's
execution. This simply requires that the program execute an instruction to
load the appropriate segment register prior to executing instructions that
access the data.
The processor associates a base address with each segment selected by a
segment register. To address an element within a segment, a 32-bit offset is
added to the segment's base address. Once a segment is selected (by loading
the segment selector into a segment register), a data manipulation
instruction only needs to specify the offset. Simple rules define which
segment register is used to form an address when only an offset is
specified.
Stack operations are facilitated by three registers:
- The stack segment (SS) register. Stacks are implemented in memory. A
system may have a number of stacks that is limited only by the maximum
number of segments. A stack may be up to 4 gigabytes long, the maximum
length of a segment. One stack is directly addressable at a --
one located by SS. This is the current stack, often referred to simply
as "the" stack. SS is used automatically by the processor for all
stack operations.
- The stack pointer (ESP) register. ESP points to the top of the
push-down stack (TOS). It is referenced implicitly by PUSH and
POP
operations, subroutine calls and returns, and interrupt operations.
When an item is pushed onto the stack (see
Figure 2-7
), the processor
decrements ESP, then writes the item at the new TOS. When an item is
popped off the stack, the processor copies it from TOS, then
increments ESP. In other words, the stack grows down in memory toward
lesser addresses.
- The stack-frame base pointer (EBP) register. The EBP is the best
choice of register for accessing data structures, variables and
dynamically allocated work space within the stack. EBP is often used
to access elements on the stack relative to a fixed point on the stack
rather than relative to the current TOS. It typically identifies the
base address of the current stack frame established for the current
procedure. When EBP is used as the base register in an offset
calculation, the offset is calculated automatically in the current
stack segment (i.e., the segment currently selected by SS). Because
SS does not have to be explicitly specified, instruction encoding in
such cases is more efficient. EBP can also be used to index into
segments addressable via other segment registers.
The flags register is a 32-bit register named EFLAGS.
Figure 2-8
defines
the bits within this register. The flags control certain operations and
indicate the status of the 80386.
The low-order 16 bits of EFLAGS is named FLAGS and can be treated as a
unit. This feature is useful when executing 8086 and 80286 code, because
this part of EFLAGS is identical to the FLAGS register of the 8086 and the
80286.
The flags may be considered in three groups: the status flags, the control
flags, and the systems flags. Discussion of the systems flags is delayed
until Part II.
The status flags of the EFLAGS register allow the results of one
instruction to influence later instructions. The arithmetic instructions use
OF, SF, ZF, AF, PF, and CF. The SCAS (Scan String), CMPS (Compare String),
and LOOP instructions use ZF to signal that their operations are complete.
There are instructions to set, clear, and complement CF before execution of
an arithmetic instruction. Refer to Appendix C for definition of each
status flag.
2.3.4.2 Control Flag
The control flag DF of the EFLAGS register controls string instructions.
DF (Direction Flag, bit 10)
Setting DF causes string instructions to auto-decrement; that is, to
process strings from high addresses to low addresses. Clearing DF causes
string instructions to auto-increment, or to process strings from low
addresses to high addresses.
2.3.4.3 Instruction Pointer
The instruction pointer register (EIP) contains the offset address,
relative to the start of the current code segment, of the next sequential
instruction to be executed. The instruction pointer is not directly visible
to the programmer; it is controlled implicitly by control-transfer
instructions, interrupts, and exceptions.
As
Figure 2-9
shows, the low-order 16 bits of EIP is named IP and can be
used by the processor as a unit. This feature is useful when executing
instructions designed for the 8086 and 80286 processors.
The information encoded in an 80386 instruction includes a specification of
the operation to be performed, the type of the operands to be manipulated,
and the location of these operands. If an operand is located in memory, the
instruction must also select, explicitly or implicitly, which of the
currently addressable segments contains the operand.
80386 instructions are composed of various elements and have various
formats. The exact format of instructions is shown in Appendix B; the
elements of instructions are described below. Of these instruction elements,
only one, the opcode, is always present. The other elements may or may not
be present, depending on the particular operation involved and on the
location and type of the operands. The elements of an instruction, in order
of occurrence are as follows:
- Prefixes -- one or more bytes preceding an instruction that modify the
operation of the instruction. The following types of prefixes can be
used by applications programs:
- Segment override -- explicitly specifies which segment register an
instruction should use, thereby overriding the default
segment-register selection used by the 80386 for that instruction.
- Address size -- switches between 32-bit and 16-bit address
generation.
- Operand size -- switches between 32-bit and 16-bit operands.
- Repeat -- used with a string instruction to cause the instruction
to act on each element of the string.
- Opcode -- specifies the operation performed by the instruction. Some
operations have several different opcodes, each specifying a different
variant of the operation.
- Register specifier -- an instruction may specify one or two register
operands. Register specifiers may occur either in the same byte as the
opcode or in the same byte as the addressing-mode specifier.
- Addressing-mode specifier -- when present, specifies whether an operand
is a register or memory location; if in memory, specifies whether a
displacement, a base register, an index register, and scaling are to be
used.
- SIB (scale, index, base) byte -- when the addressing-mode specifier
indicates that an index register will be used to compute the address of
an operand, an SIB byte is included in the instruction to encode the
base register, the index register, and a scaling factor.
- Displacement -- when the addressing-mode specifier indicates that a
displacement will be used to compute the address of an operand, the
displacement is encoded in the instruction. A displacement is a signed
integer of 32, 16, or eight bits. The eight-bit form is used in the
common case when the displacement is sufficiently small. The processor
extends an eight-bit displacement to 16 or 32 bits, taking into
account the sign.
- Immediate operand -- when present, directly provides the value of an
operand of the instruction. Immediate operands may be 8, 16, or 32 bits
wide. In cases where an eight-bit immediate operand is combined in some
way with a 16- or 32-bit operand, the processor automatically extends
the size of the eight-bit operand, taking into account the sign.
2.5 Operand Selection
An instruction can act on zero or more operands, which are the data
manipulated by the instruction. An example of a zero-operand instruction is
NOP (no operation). An operand can be in any of these locations:
- In the instruction itself (an immediate operand)
- In a register (EAX, EBX, ECX, EDX, ESI, EDI, ESP, or EBP in the case
of 32-bit operands; AX, BX, CX, DX, SI, DI, SP, or BP in the case of
16-bit operands; AH, AL, BH, BL, CH, CL, DH, or DL in the case of 8-bit
operands; the segment registers; or the EFLAGS register for flag
operations)
- In memory
- At an I/O port
Immediate operands and operands in registers can be accessed more rapidly
than operands in memory since memory operands must be fetched from memory.
Register operands are available in the CPU. Immediate operands are also
available in the CPU, because they are prefetched as part of the
instruction.
Of the instructions that have operands, some specify operands implicitly;
others specify operands explicitly; still others use a combination of
implicit and explicit specification; for example:
-
Implicit operand: AAM
-
By definition, AAM (ASCII adjust for multiplication) operates on the
contents of the AX register.
-
Explicit operand: XCHG EAX, EBX
-
The operands to be exchanged are encoded in the instruction after the
opcode.
-
Implicit and explicit operands: PUSH COUNTER
-
The memory variable COUNTER (the explicit operand) is copied to the top of
the stack (the implicit operand).
Note that most instructions have implicit operands. All arithmetic
instructions, for example, update the EFLAGS register.
An 80386 instruction can explicitly reference one or two operands.
Two-operand instructions, such as MOV, ADD, XOR, etc., generally overwrite
one of the two participating operands with the result. A distinction can
thus be made between the source operand (the one unaffected by the
operation) and the destination operand (the one overwritten by the result).
For most instructions, one of the two explicitly specified --
the source or the -- be either in a register or in memory.
The other operand must be in a register or be an immediate source operand.
Thus, the explicit two-operand instructions of the 80386 permit operations
of the following kinds:
- Register-to-register
- Register-to-memory
- Memory-to-register
- Immediate-to-register
- Immediate-to-memory
Certain string instructions and stack manipulation instructions, however,
transfer data from memory to memory. Both operands of some string
instructions are in memory and are implicitly specified. Push and pop stack
operations allow transfer between memory operands and the memory-based
stack.
2.5.1 Immediate Operands
Certain instructions use data from the instruction itself as one (and
sometimes two) of the operands. Such an operand is called an immediate
operand. The operand may be 32-, 16-, or 8-bits long. For example:
SHR PATTERN, 2
One byte of the instruction holds the value 2, the number of bits by which
to shift the variable PATTERN.
TEST PATTERN, 0FFFF00FFH
A doubleword of the instruction holds the mask that is used to test the
variable PATTERN.
2.5.2 Register Operands
Operands may be located in one of the 32-bit general registers (EAX, EBX,
ECX, EDX, ESI, EDI, ESP, or EBP), in one of the 16-bit general registers
(AX, BX, CX, DX, SI, DI, SP, or BP), or in one of the 8-bit general
registers (AH, BH, CH, DH, AL, BL, CL,or DL).
The 80386 has instructions for referencing the segment registers (CS, DS,
ES, SS, FS, GS). These instructions are used by applications programs only
if systems designers have chosen a segmented memory model.
The 80386 also has instructions for referring to the flag register. The
flags may be stored on the stack and restored from the stack. Certain
instructions change the commonly modified flags directly in the EFLAGS
register. Other flags that are seldom modified can be modified indirectly
via the flags image in the stack.
2.5.3 Memory Operands
Data-manipulation instructions that address operands in memory must specify
(either directly or indirectly) the segment that contains the operand and
the offset of the operand within the segment. However, for speed and compact
instruction encoding, segment selectors are stored in the high speed segment
registers. Therefore, data-manipulation instructions need to specify only
the desired segment register and an offset in order to address a memory
operand.
An 80386 data-manipulation instruction that accesses memory uses one of the
following methods for specifying the offset of a memory operand within its
segment:
- Most data-manipulation instructions that access memory contain a byte
that explicitly specifies the addressing method for the operand. A
byte, known as the modR/M byte, follows the opcode and specifies
whether the operand is in a register or in memory. If the operand is
in memory, the address is computed from a segment register and any of
the following values: a base register, an index register, a scaling
factor, a displacement. When an index register is used, the modR/M
byte is also followed by another byte that identifies the index
register and scaling factor. This addressing method is the
most flexible.
- A few data-manipulation instructions implicitly use specialized
addressing methods:
- For a few short forms of
MOV that implicitly use the EAX register,
the offset of the operand is coded as a doubleword in the
instruction. No base register, index register, or scaling factor
are used.
- String operations implicitly address memory via DS:ESI,
(MOVS,
CMPS,
OUTS,
LODS,
SCAS) or via ES:EDI
(MOVS,
CMPS,
INS,
STOS).
- Stack operations implicitly address operands via SS:ESP
registers; e.g.,
PUSH,
POP,
PUSHA,
PUSHAD,
POPA,
POPAD,
PUSHF,
PUSHFD,
POPF,
POPFD,
CALL,
RET,
IRET,
IRETD,
exceptions, and interrupts.
2.5.3.1 Segment Selection
Data-manipulation instructions need not explicitly specify which segment
register is used. For all of these instructions, specification of a segment
register is optional. For all memory accesses, if a segment is not
explicitly specified by the instruction, the processor automatically chooses
a segment register according to the rules of Table 2-1. (If systems
designers have chosen a flat model of memory organization, the segment
registers and the rules that the processor uses in choosing them are not
apparent to applications programs.)
There is a close connection between the kind of memory reference and the
segment in which that operand resides. As a rule, a memory reference implies
the current data segment (i.e., the implicit segment selector is in DS).
However, ESP and EBP are used to access items on the stack; therefore, when
the ESP or EBP register is used as a base register, the current stack
segment is implied (i.e., SS contains the selector).
Special instruction prefix elements may be used to override the default
segment selection. Segment-override prefixes allow an explicit segment
selection. The 80386 has a segment-override prefix for each of the segment
registers. Only in the following special cases is there an implied segment
selection that a segment prefix cannot override:
- The use of ES for destination strings in string instructions.
- The use of SS in stack instructions.
- The use of CS for instruction fetches.
Table 2-1. Default Segment Register Selection Rules
Memory Reference Needed Segment Implicit Segment Selection Rule
Register
Used
Instructions Code (CS) Automatic with instruction prefetch
Stack Stack (SS) All stack pushes and pops. Any
memory reference that uses ESP or
EBP as a base register.
Local Data Data (DS) All data references except when
relative to stack or string
destination.
Destination Strings Extra (ES) Destination of string instructions.
2.5.3.2 Effective-Address Computation
The modR/M byte provides the most flexible of the addressing methods, and
instructions that require a modR/M byte as the second byte of the
instruction are the most common in the 80386 instruction set. For memory
operands defined by modR/M, the offset within the desired segment is
calculated by taking the sum of up to three components:
- A displacement element in the instruction.
- A base register.
- An index register. The index register may be automatically multiplied
by a scaling factor of 2, 4, or 8.
The offset that results from adding these components is called an effective
address. Each of these components of an effective address may have either a
positive or negative value. If the sum of all the components exceeds 2^(32),
the effective address is truncated to 32 bits.
Figure 2-10
illustrates the
full set of possibilities for modR/M addressing.
The displacement component, because it is encoded in the instruction, is
useful for fixed aspects of addressing; for example:
- Location of simple scalar operands.
- Beginning of a statically allocated array.
- Offset of an item within a record.
The base and index components have similar functions. Both utilize the same
set of general registers. Both can be used for aspects of addressing that
are determined dynamically; for example:
- Location of procedure parameters and local variables in stack.
- The beginning of one record among several occurrences of the same
record type or in an array of records.
- The beginning of one dimension of multiple dimension array.
- The beginning of a dynamically allocated array.
The uses of general registers as base or index components differ in the
following respects:
- ESP cannot be used as an index register.
- When ESP or EBP is used as the base register, the default segment is
the one selected by SS. In all other cases the default segment is DS.
The scaling factor permits efficient indexing into an array in the common
cases when array elements are 2, 4, or 8 bytes wide. The shifting of the
index register is done by the processor at the time the address is evaluated
with no performance loss. This eliminates the need for a separate shift or
multiply instruction.
The base, index, and displacement components may be used in any
combination; any of these components may be null. A scale factor can be used
only when an index is also used. Each possible combination is useful for
data structures commonly used by programmers in high-level languages and
assembly languages. Following are possible uses for some of the various
combinations of address components.
-
DISPLACEMENT
-
The displacement alone indicates the offset of the operand. This
combination is used to directly address a statically allocated scalar
operand. An 8-bit, 16-bit, or 32-bit displacement can be used.
-
BASE
-
The offset of the operand is specified indirectly in one of the general
registers, as for "based" variables.
-
BASE + DISPLACEMENT
-
A register and a displacement can be used together for two distinct
purposes:
- Index into static array when element size is not 2, 4, or 8 bytes.
The displacement component encodes the offset of the beginning of
the array. The register holds the results of a calculation to
determine the offset of a specific element within the array.
- Access item of a record. The displacement component locates an
within record. The base register selects one of several
occurrences of record, thereby providing a compact encoding for
this common function.
An important special case of this combination, is to access parameters
in the procedure activation record in the stack. In this case, EBP is
the best choice for the base register, because when EBP is used as a
base register, the processor automatically uses the stack segment
register (SS) to locate the operand, thereby providing a compact
encoding for this common function.
-
(INDEX * SCALE) + DISPLACEMENT
-
This combination provides efficient indexing into a static array when
the element size is 2, 4, or 8 bytes. The displacement addresses the
beginning of the array, the index register holds the subscript of the
desired array element, and the processor automatically converts the
subscript into an index by applying the scaling factor.
-
BASE + INDEX + DISPLACEMENT
-
Two registers used together support either a two-dimensional array (the
displacement determining the beginning of the array) or one of several
instances of an array of records (the displacement indicating an item
in the record).
-
BASE + (INDEX * SCALE) + DISPLACEMENT
-
This combination provides efficient indexing of a two-dimensional array
when the elements of the array are 2, 4, or 8 bytes wide.
The 80386 has two mechanisms for interrupting program execution:
- Exceptions are synchronous events that are the responses of the CPU
to certain conditions detected during the execution of an instruction.
- Interrupts are asynchronous events typically triggered by external
devices needing attention.
Interrupts and exceptions are alike in that both cause the processor to
temporarily suspend its present program execution in order to execute a
program of higher priority. The major distinction between these two kinds of
interrupts is their origin. An exception is always reproducible by
re-executing with the program and data that caused the exception, whereas an
interrupt is generally independent of the currently executing program.
Application programmers are not normally concerned with servicing
interrupts. More information on interrupts for systems programmers may be
found in
Chapter 9
. Certain exceptions , however, are of interest to
applications programmers, and many operating systems give applications
programs the opportunity to service these exceptions. However, the operating
system itself defines the interface between the applications programs and
the exception mechanism of the 80386.
Table 2-2 highlights the exceptions that may be of interest to applications
programmers.
- A divide error exception results when the instruction
DIV or
IDIV is
executed with a zero denominator or when the quotient is too large for
the destination operand . (Refer to
Chapter 3
for a discussion of DIV
and IDIV.)
- The debug exception may be reflected back to an applications program
if it results from the trap flag (TF).
- A breakpoint exception results when the instruction
INT 3 is executed.
This instruction is used by some debuggers to stop program execution at
specific points.
- An overflow exception results when the
INTO instruction is executed
and the OF (overflow) flag is set (after an arithmetic operation that
set the OF flag ) . (Refer to
Chapter 3
for a discussion of INTO) .
- A bounds check exception results when the
BOUND instruction is
executed and the array index it checks falls outside the bounds of the
array . (Refer to
Chapter 3
for a discussion of the BOUND instruction. )
- Invalid opcodes may be used by some applications to extend the
instruction set. In such a case, the invalid opcode exception presents
an opportunity to emulate the opcode.
- The "coprocessor not available" exception occurs if the program
contains instructions for a coprocessor, but no coprocessor is present
in the system.
- A coprocessor error is generated when a coprocessor detects an illegal
operation.
The instruction INT generates an interrupt whenever it is executed; the
processor treats this interrupt as an exception. The effects of this
interrupt (and the effects of all other exceptions) are determined by
exception handler routines provided by the application program or as part of
the systems software (provided by systems programmers). The
INT instruction
itself is discussed in
Chapter 3. Refer to
Chapter 9 for a more complete
description of exceptions.
Table 2-2. 80386 Reserved Exceptions and Interrupts
Vector Number Description
0 Divide Error
1 Debug Exceptions
2 NMI Interrupt
3 Breakpoint
4 INTO Detected Overflow
5 BOUND Range Exceeded
6 Invalid Opcode
7 Coprocessor Not Available
8 Double Exception
9 Coprocessor Segment Overrun
10 Invalid Task State Segment
11 Segment Not Present
12 Stack Fault
13 General Protection
14 Page Fault
15 (reserved)
16 Coprocessor Error
17-32 (reserved)
The registers designed for use by systems programmers fall into these
classes:
- EFLAGS
- Memory-Management Registers
- Control Registers
- Debug Registers
- Test Registers
4.1.1 Systems Flags
The systems flags of the EFLAGS register control I/O, maskable interrupts,
debugging, task switching, and enabling of virtual 8086 execution in a
protected, multitasking environment. These flags are highlighted in
Figure 4-1
.
-
IF
(Interrupt-Enable Flag, bit 9)
-
Setting IF allows the CPU to recognize external (maskable) interrupt
requests. Clearing IF disables these interrupts. IF has no effect on
either exceptions or nonmaskable external interrupts . Refer to
Chapter 9
for more details about interrupts .
-
NT
(Nested Task, bit 14)
-
The processor uses the nested task flag to control chaining of
interrupted and called tasks. NT influences the operation of the
IRET
instruction . Refer to
Chapter 7 and
Chapter 9
for more information on
nested tasks.
-
RF
(Resume Flag, bit 16)
-
The RF flag temporarily disables debug exceptions so that an instruction
can be restarted after a debug exception without immediately causing
another debug exception . Refer to
Chapter 12
for details .
-
TF
(Trap Flag, bit 8)
-
Setting TF puts the processor into single-step mode for debugging. In
this mode, the CPU automatically generates an exception after each
instruction, allowing a program to be inspected as it executes each
instruction. Single-stepping is just one of several debugging features of
the 80386 . Refer to
Chapter 12
for additional information .
-
VM
(Virtual 8086 Mode, bit 17)
-
When set, the VM flag indicates that the task is executing an 8086
program . Refer to
Chapter 14
for a detailed discussion of how the 80386
executes 8086 tasks in a protected, multitasking environment.
4.1.2 Memory-Management Registers
Four registers of the 80386 locate the data structures that control
segmented memory management:
-
GDTR
Global Descriptor Table Register
-
LDTR
Local Descriptor Table Register
-
These registers point to the segment descriptor tables GDT and LDT.
Refer to
Chapter 5
for an explanation of addressing via descriptor
tables.
-
IDTR
Interrupt Descriptor Table Register
-
This register points to a table of entry points for interrupt handlers
(the IDT ) . Refer to
Chapter 9
for details of the interrupt mechanism .
-
TR
Task Register
-
This register points to the information needed by the processor to define
the current task . Refer to
Chapter 7
for a description of the
multitasking features of the 80386.
4.1.3 Control Registers
Figure 4-2
shows the format of the 80386 control registers CR0, CR2, and
CR3. These registers are accessible to systems programmers only via variants
of the MOV instruction, which allow them to be loaded from or stored in
general registers; for example:
MOV EAX, CR0
MOV CR3, EBX
CR0 contains system control flags, which control or indicate conditions
that apply to the system as a whole, not to an individual task.
-
EM
(Emulation, bit 2)
-
EM indicates whether coprocessor functions are to be emulated. Refer to
Chapter 11
for details .
-
ET
(Extension Type, bit 4)
-
ET indicates the type of coprocessor present in the system (80287 or
80387 ) . Refer to
Chapter 11 and
Chapter 10 for details.
-
MP
(Math Present, bit 1)
-
MP controls the function of the WAIT instruction, which is used to
coordinate a coprocessor . Refer to
Chapter 11
for details .
-
PE
(Protection Enable, bit 0)
-
Setting PE causes the processor to begin executing in protected mode.
Resetting PE returns to real-address mode . Refer to
Chapter 14
and
Chapter 10
for more information on changing processor modes .
-
PG
(Paging, bit 31)
-
PG indicates whether the processor uses page tables to translate linear
addresses into physical addresses . Refer to
Chapter 5
for a description
of page translation; refer to
Chapter 10
for a discussion of how to set
PG.
-
TS
(Task Switched, bit 3)
-
The processor sets TS with every task switch and tests TS when
interpreting coprocessor instructions . Refer to
Chapter 11
for details .
CR2 is used for handling page faults when PG is set. The processor stores
in CR2 the linear address that triggers the fault . Refer to
Chapter 9
for a
description of page-fault handling.
CR3 is used when PG is set. CR3 enables the processor to locate the page
table directory for the current task . Refer to
Chapter 5
for a description
of page tables and page translation.
The debug registers bring advanced debugging abilities to the 80386,
including data breakpoints and the ability to set instruction breakpoints
without modifying code segments . Refer to
Chapter 12
for a complete
description of formats and usage.
4.1.5 Test Registers
The test registers are not a standard part of the 80386 architecture. They
are provided solely to enable confidence testing of the translation
lookaside buffer (TLB), the cache used for storing information from page
tables .
Chapter 12
explains how to use these registers .
Systems instructions deal with such functions as:
- Verification of pointer parameters (refer to
Chapter 6):
- Addressing descriptor tables (refer to
Chapter 5):
- Multitasking (refer to
Chapter 7):
- Coprocessing and Multiprocessing (refer to
Chapter 11):
- Input and Output (refer to
Chapter 8):
- Interrupt control (refer to
Chapter 9):
- Debugging (refer to
Chapter 12):
- TLB testing (refer to
Chapter 10):
- System Control:
The instructions SMSW and LMSW
are provided for compatibility with the
80286 processor. 80386 programs access the MSW in CR0 via variants of the
MOV instruction.
HLT stops the processor until receipt of an INTR or RESET
signal.
In addition to the chapters cited above, detailed information about each of
these instructions can be found in the instruction reference chapter,
Chapter 17
The 80386 transforms logical addresses (i.e., addresses as viewed by
programmers) into physical address (i.e., actual addresses in physical
memory) in two steps:
- Segment translation, in which a logical address (consisting of a
segment selector and segment offset) are converted to a linear address.
- Page translation, in which a linear address is converted to a physical
address. This step is optional, at the discretion of systems-software
designers.
These translations are performed in a way that is not visible to
applications programmers.
Figure 5-1
illustrates the two translations at a
high level of abstraction.
Figure 5-1
and the remainder of this chapter present a simplified view of
the 80386 addressing mechanism. In reality, the addressing mechanism also
includes memory protection features. For the sake of simplicity, however,
the subject of protection is taken up in another chapter,
Chapter 6.
Figure 5-2
shows in more detail how the processor converts a logical
address into a linear address.
To perform this translation, the processor uses the following data
structures:
- Descriptors
- Descriptor tables
- Selectors
- Segment Registers
5.1.1 Descriptors
The segment descriptor provides the processor with the data it needs to map
a logical address into a linear address. Descriptors are created by
compilers, linkers, loaders, or the operating system, not by applications
programmers.
Figure 5-3
illustrates the two general descriptor formats. All
types of segment descriptors take one of these formats. Segment-descriptor
fields are:
BASE: Defines the location of the segment within the 4 gigabyte linear
address space. The processor concatenates the three fragments of the base
address to form a single 32-bit value.
LIMIT: Defines the size of the segment. When the processor concatenates the
two parts of the limit field, a 20-bit value results. The processor
interprets the limit field in one of two ways, depending on the setting of
the granularity bit:
- In units of one byte, to define a limit of up to 1 megabyte.
- In units of 4 Kilobytes, to define a limit of up to 4 gigabytes. The
limit is shifted left by 12 bits when loaded, and low-order one-bits
are inserted.
Granularity bit: Specifies the units with which the LIMIT field is
interpreted. When thebit is clear, the limit is interpreted in units of one
byte; when set, the limit is interpreted in units of 4 Kilobytes.
TYPE: Distinguishes between various kinds of descriptors.
DPL (Descriptor Privilege Level): Used by the protection mechanism (refer
to
Chapter 6
) .
Segment-Present bit: If this bit is zero, the descriptor is not valid for
use in address transformation; the processor will signal an exception when a
selector for the descriptor is loaded into a segment register.
Figure 5-4
shows the format of a descriptor when the present-bit is zero. The operating
system is free to use the locations marked AVAILABLE. Operating systems that
implement segment-based virtual memory clear the present bit in either of
these cases:
- When the linear space spanned by the segment is not mapped by the
paging mechanism.
- When the segment is not present in memory.
Accessed bit: The processor sets this bit when the segment is accessed;
i.e., a selector for the descriptor is loaded into a segment register or
used by a selector test instruction. Operating systems that implement
virtual memory at the segment level may, by periodically testing and
clearing this bit, monitor frequency of segment usage.
Creation and maintenance of descriptors is the responsibility of systems
software, usually requiring the cooperation of compilers, program loaders or
system builders, and therating system.
Segment descriptors are stored in either of two kinds of descriptor table:
- The global descriptor table (GDT)
- A local descriptor table (LDT)
A descriptor table is simply a memory array of 8-byte entries that contain
descriptors, as
Figure 5-5
shows. A descriptor table is variable in length
and may contain up to 8192 (2^(13)) descriptors. The first entry of the GDT
(INDEX=0) is not used by the processor, however.
The processor locates the GDT and the current LDT in memory by means of the
GDTR and LDTR registers. These registers store the base addresses of the
tables in the linear address space and store the segment limits. The
instructions LGDT and
SGDT give access to the GDTR; the instructions
LLDT
and SLDT give access to the LDTR.
The selector portion of a logical address identifies a descriptor by
specifying a descriptor table and indexing a descriptor within that table.
Selectors may be visible to applications programs as a field within a
pointer variable, but the values of selectors are usually assigned (fixed
up) by linkers or linking loaders.
Figure 5-6
shows the format of a
selector.
Index: Selects one of 8192 descriptors in a descriptor table. The processor
simply multiplies this index value by 8 (the length of a descriptor), and
adds the result to the base address of the descriptor table in order to
access the appropriate segment descriptor in the table.
Table Indicator: Specifies to which descriptor table the selector refers. A
zero indicates the GDT; a one indicates the current LDT.
Requested Privilege Level: Used by the protection mechanism. (Refer to
Chapter 6)
Because the first entry of the GDT is not used by the processor, a selector
that has an index of zero and a table indicator of zero (i.e., a selector
that points to the first entry of the GDT), can be used as a null selector.
The processor does not cause an exception when a segment register (other
than CS or SS) is loaded with a null selector. It will, however, cause an
exception when the segment register is used to access memory. This feature
is useful for initializing unused segment registers so as to trap accidental
references.
The 80386 stores information from descriptors in segment registers, thereby
avoiding the need to consult a descriptor table every time it accesses
memory.
Every segment register has a "visible" portion and an "invisible" portion,
as
Figure 5-7
illustrates. The visible portions of these segment address
registers are manipulated by programs as if they were simply 16-bit
registers. The invisible portions are manipulated by the processor.
The operations that load these registers are normal program instructions
(previously described in
Chapter 3). These instructions are of two classes:
- Direct load instructions; for example, MOV,
POP,
LDS,
LSS,
LGS,
LFS.
These instructions explicitly reference the segment registers.
- Implied load instructions; for example, far
CALL and JMP. These
instructions implicitly reference the CS register, and load it with a
new value.
Using these instructions, a program loads the visible part of the segment
register with a 16-bit selector. The processor automatically fetches the
base address, limit, type, and other information from a descriptor table and
loads them into the invisible part of the segment register.
Because most instructions refer to data in segments whose selectors have
already been loaded into segment registers, the processor can add the
segment-relative offset supplied by the instruction to the segment base
address with no additional overhead.
In the second phase of address transformation, the 80386 transforms a
linear address into a physical address. This phase of address transformation
implements the basic features needed for page-oriented virtual-memory
systems and page-level protection.
The page-translation step is optional. Page translation is in effect only
when the PG bit of CR0 is set. This bit is typically set by the operating
system during software initialization. The PG bit must be set if the
operating system is to implement multiple virtual 8086 tasks, page-oriented
protection, or page-oriented virtual memory.
5.2.1 Page Frame
A page frame is a 4K-byte unit of contiguous addresses of physical memory.
Pages begin onbyte boundaries and are fixed in size.
5.2.2 Linear Address
A linear address refers indirectly to a physical address by specifying a
page table, a page within that table, and an offset within that page.
Figure 5-8
shows the format of a linear address.
Figure 5-9
shows how the processor converts the DIR, PAGE, and OFFSET
fields of a linear address into the physical address by consulting two
levels of page tables. The addressing mechanism uses the DIR field as an
index into a page directory, uses the PAGE field as an index into the page
table determined by the page directory, and uses the OFFSET field to address
a byte within the page determined by the page table.
A page table is simply an array of 32-bit page specifiers. A page table is
itself a page, and therefore contains 4 Kilobytes of memory or at most 1K
32-bit entries.
Two levels of tables are used to address a page of memory. At the higher
level is a page directory. The page directory addresses up to 1K page tables
of the second level. A page table of the second level addresses up to 1K
pages. All the tables addressed by one page directory, therefore, can
address 1M pages (2^(20)). Because each page contains 4K bytes 2^(12)
bytes), the tables of one page directory can span the entire physical
address space of the 80386 (2^(20) times 2^(12) = 2^(32)).
The physical address of the current page directory is stored in the CPU
register CR3, also called the page directory base register (PDBR). Memory
management software has the option of using one page directory for all
tasks, one page directory for each task, or some combination of the two.
Refer to
Chapter 10
for information on initialization of CR3 . Refer to
Chapter 7
to see how CR3 can change for each task .
5.2.4 Page-Table Entries
Entries in either level of page tables have the same format.
Figure 5-10
illustrates this format.
5.2.4.1 Page Frame Address
The page frame address specifies the physical starting address of a page.
Because pages are located on 4K boundaries, the low-order 12 bits are always
zero. In a page directory, the page frame address is the address of a page
table. In a second-level page table, the page frame address is the address
of the page frame that contains the desired memory operand.
5.2.4.2 Present Bit
The Present bit indicates whether a page table entry can be used in address
translation. P=1 indicates that the entry can be used.
When P=0 in either level of page tables, the entry is not valid for address
translation, and the rest of the entry is available for software use; none
of the other bits in the entry is tested by the hardware.
Figure 5-11
illustrates the format of a page-table entry when P=0.
If P=0 in either level of page tables when an attempt is made to use a
page-table entry for address translation, the processor signals a page
exception. In software systems that support paged virtual memory, the
page-not-present exception handler can bring the required page into physical
memory. The instruction that caused the exception can then be reexecuted.
Refer to
Chapter 9
for more information on exception handlers .
Note that there is no present bit for the page directory itself. The page
directory may be not-present while the associated task is suspended, but the
operating system must ensure that the page directory indicated by the CR3
image in the TSS is present in physical memory before the task is
dispatched . Refer to
Chapter 7
for an explanation of the TSS and task
dispatching.
These bits provide data about page usage in both levels of the page tables.
With the exception of the dirty bit in a page directory entry, these bits
are set by the hardware; however, the processor does not clear any of these
bits.
The processor sets the corresponding accessed bits in both levels of page
tables to one before a read or write operation to a page.
The processor sets the dirty bit in the second-level page table to one
before a write to an address covered by that page table entry. The dirty bit
in directory entries is undefined.
An operating system that supports paged virtual memory can use these bits
to determine what pages to eliminate from physical memory when the demand
for memory exceeds the physical memory available. The operating system is
responsible for testing and clearing these bits.
Refer to
Chapter 11
for how the 80386 coordinates updates to the accessed
and dirty bits in multiprocessor systems.
5.2.4.4 Read/Write and User/Supervisor Bits
These bits are not used for address translation, but are used for
page-level protection, which the processor performs at the same time as
address translation . Refer to
Chapter 6
where protection is discussed in
detail.
5.2.5 Page Translation Cache
For greatest efficiency in address translation, the processor stores the
most recently used page-table data in an on-chip cache. Only if the
necessary paging information is not in the cache must both levels of page
tables be referenced.
The existence of the page-translation cache is invisible to applications
programmers but not to systems programmers; operating-system programmers
must flush the cache whenever the page tables are changed. The
page-translation cache can be flushed by either of two methods:
- By reloading CR3 with a MOV instruction;
for example:
MOV CR3, EAX
- By performing a task switch to a TSS that has a different CR3 image
than the current TSS . (Refer to
Chapter 7
for more information on
task switching.)
Figure 5-12
combines
Figure 5-2
and
Figure 5-9
to summarize both phases
of the transformation from a logical address to a physical address when
paging is enabled. By appropriate choice of options and parameters to both
phases, memory-management software can implement several different styles of
memory management.
5.3.1 "Flat" Architecture
When the 80386 is used to execute software designed for architectures that
don't have segments, it may be expedient to effectively "turn off" the
segmentation features of the 80386. The 80386 does not have a mode that
disables segmentation, but the same effect can be achieved by initially
loading the segment registers with selectors for descriptors that encompass
the entire 32-bit linear address space. Once loaded, the segment registers
don't need to be changed. The 32-bit offsets used by 80386 instructions are
adequate to address the entire linear-address space.
5.3.2 Segments Spanning Several Pages
The architecture of the 80386 permits segments to be larger or smaller than
the size of a page (4 Kilobytes). For example, suppose a segment is used to
address and protect a large data structure that spans 132 Kilobytes. In a
software system that supports paged virtual memory, it is not necessary for
the entire structure to be in physical memory at once. The structure is
divided into 33 pages, any number of which may not be present. The
applications programmer does not need to be aware that the virtual memory
subsystem is paging the structure in this manner.
On the other hand, segments may be smaller than the size of a page. For
example, consider a small data structure such as a semaphore. Because of the
protection and sharing provided by segments (refer to
Chapter 6
) , it may be
useful to create a separate segment for each semaphore. But, because a
system may need many semaphores, it is not efficient to allocate a page for
each. Therefore, it may be useful to cluster many related segments within a
page.
5.3.4 Non-Aligned Page and Segment Boundaries
The architecture of the 80386 does not enforce any correspondence between
the boundaries of pages and segments. It is perfectly permissible for a page
to contain the end of one segment and the beginning of another. Likewise, a
segment may contain the end of one page and the beginning of another.
5.3.5 Aligned Page and Segment Boundaries
Memory-management software may be simpler, however, if it enforces some
correspondence between page and segment boundaries. For example, if segments
are allocated only in units of one page, the logic for segment and page
allocation can be combined. There is no need for logic to account for
partially used pages.
5.3.6 Page-Table per Segment
An approach to space management that provides even further simplification
of space-management software is to maintain a one-to-one correspondence
between segment descriptors and page-directory entries, as
Figure 5-13
illustrates. Each descriptor has a base address in which the low-order 22
bits are zero; in other words, the base address is mapped by the first entry
of a page table. A segment may have any limit from 1 to 4 megabytes.
Depending on the limit, the segment is contained in from 1 to 1K page
frames. A task is thus limited to 1K segments (a sufficient number for many
applications), each containing up to 4 Mbytes. The descriptor, the
corresponding page-directory entry, and the corresponding page table can be
allocated and deallocated simultaneously.
The purpose of the protection features of the 80386 is to help detect and
identify bugs. The 80386 supports sophisticated applications that may
consist of hundreds or thousands of program modules. In such applications,
the question is how bugs can be found and eliminated as quickly as possible
and how their damage can be tightly confined. To help debug applications
faster and make them more robust in production, the 80386 contains
mechanisms to verify memory accesses and instruction execution for
conformance to protection criteria. These mechanisms may be used or ignored,
according to system design objectives.
Protection in the 80386 has five aspects:
- Type checking
- Limit checking
- Restriction of addressable domain
- Restriction of procedure entry points
- Restriction of instruction set
The protection hardware of the 80386 is an integral part of the memory
management hardware. Protection applies both to segment translation and to
page translation.
Each reference to memory is checked by the hardware to verify that it
satisfies the protection criteria. All these checks are made before the
memory cycle is started; any violation prevents that cycle from starting and
results in an exception. Since the checks are performed concurrently with
address formation, there is no performance penalty.
Invalid attempts to access memory result in an exception. Refer to
Chapter 9
for an explanation of the exception mechanism . The present
chapter defines the protection violations that lead to exceptions.
The concept of "privilege" is central to several aspects of protection
(numbers 3, 4, and 5 in the preceeding list). Applied to procedures,
privilege is the degree to which the procedure can be trusted not to make a
mistake that might affect other procedures or data. Applied to data,
privilege is the degree of protection that a data structure should have
from less trusted procedures.
The concept of privilege applies both to segment protection and to page
protection.
All five aspects of protection apply to segment translation:
- Type checking
- Limit checking
- Restriction of addressable domain
- Restriction of procedure entry points
- Restriction of instruction set
The segment is the unit of protection, and segment descriptors store
protection parameters. Protection checks are performed automatically by the
CPU when the selector of a segment descriptor is loaded into a segment
register and with every segment access. Segment registers hold the
protection parameters of the currently addressable segments.
6.3.1 Descriptors Store Protection Parameters
Figure 6-1
highlights the protection-related fields of segment descriptors.
The protection parameters are placed in the descriptor by systems software
at the time a descriptor is created. In general, applications programmers do
not need to be concerned about protection parameters.
When a program loads a selector into a segment register, the processor
loads not only the base address of the segment but also protection
information. Each segment register has bits in the invisible portion for
storing base, limit, type, and privilege level; therefore, subsequent
protection checks on the same segment do not consume additional clock
cycles.
The TYPE field of a descriptor has two functions:
- It distinguishes among different descriptor formats.
- It specifies the intended usage of a segment.
Besides the descriptors for data and executable segments commonly used by
applications programs, the 80386 has descriptors for special segments used
by the operating system and for gates. Table 6-1 lists all the types defined
for system segments and gates. Note that not all descriptors define
segments; gate descriptors have a different purpose that is discussed later
in this chapter.
The type fields of data and executable segment descriptors include bits
which further define the purpose of the segment (refer to
Figure 6-1
):
- The writable bit in a data-segment descriptor specifies whether
instructions can write into the segment.
- The readable bit in an executable-segment descriptor specifies
whether instructions are allowed to read from the segment (for example,
to access constants that are stored with instructions). A readable,
executable segment may be read in two ways:
- Via the CS register, by using a CS override prefix.
- By loading a selector of the descriptor into a data-segment register
(DS, ES, FS,or GS).
Type checking can be used to detect programming errors that would attempt
to use segments in ways not intended by the programmer. The processor
examines type information on two kinds of occasions:
- When a selector of a descriptor is loaded into a segment register.
Certain segment registers can contain only certain descriptor types;
for example:
- The CS register can be loaded only with a selector of an executable
segment.
- Selectors of executable segments that are not readable cannot be
loaded into data-segment registers.
- Only selectors of writable data segments can be loaded into SS.
- When an instruction refers (implicitly or explicitly) to a segment
register. Certain segments can be used by instructions only in certain
predefined ways; for example:
- No instruction may write into an executable segment.
- No instruction may write into a data segment if the writable bit is
not set.
- No instruction may read an executable segment unless the readable bit
is set.
Table 6-1. System and Gate Descriptor Types
Code Type of Segment or Gate
0 -reserved
1 Available 286 TSS
2 LDT
3 Busy 286 TSS
4 Call Gate
5 Task Gate
6 286 Interrupt Gate
7 286 Trap Gate
8 -reserved
9 Available 386 TSS
A -reserved
B Busy 386 TSS
C 386 Call Gate
D -reserved
E 386 Interrupt Gate
F 386 Trap Gate
6.3.1.2 Limit Checking
The limit field of a segment descriptor is used by the processor to prevent
programs from addressing outside the segment. The processor's interpretation
of the limit depends on the setting of the G (granularity) bit. For data
segments, the processor's interpretation of the limit depends also on the
E-bit (expansion-direction bit) and the B-bit (big bit) (refer to Table
6-2).
When G=0, the actual limit is the value of the 20-bit limit field as it
appears in the descriptor. In this case, the limit may range from 0 to
0FFFFFH (2^(20) - 1 or 1 megabyte). When G=1, the processor appends 12
low-order one-bits to the value in the limit field. In this case the actual
limit may range from 0FFFH (2^(12) - 1 or 4 kilobytes) to 0FFFFFFFFH(2^(32)
- 1 or 4 gigabytes).
For all types of segments except expand-down data segments, the value of
the limit is one less than the size (expressed in bytes) of the segment. The
processor causes a general-protection exception in any of these cases:
- Attempt to access a memory byte at an address > limit.
- Attempt to access a memory word at an address >= limit.
- Attempt to access a memory doubleword at an address >= (limit-2).
For expand-down data segments, the limit has the same function but is
interpreted differently. In these cases the range of valid addresses is from
limit + 1 to either 64K or 2^(32) - 1 (4 Gbytes) depending on the B-bit. An
expand-down segment has maximum size when the limit is zero.
The expand-down feature makes it possible to expand the size of a stack by
copying it to a larger segment without needing also to update intrastack
pointers.
The limit field of descriptors for descriptor tables is used by the
processor to prevent programs from selecting a table entry outside the
descriptor table. The limit of a descriptor table identifies the last valid
byte of the last descriptor in the table. Since each descriptor is eight
bytes long, the limit value is N * 8 - 1 for a table that can contain up to
N descriptors.
Limit checking catches programming errors such as runaway subscripts and
invalid pointer calculations. Such errors are detected when they occur, so
that identification of the cause is easier. Without limit checking, such
errors could corrupt other modules; the existence of such errors would not
be discovered until later, when the corrupted module behaves incorrectly,
and when identification of the cause is difficult.
Table 6-2. Useful Combinations of E, G, and B Bits
Case: 1 2 3 4
Expansion Direction U U D D
G-bit 0 1 0 1
B-bit X X 0 1
Lower bound is:
0 X X
LIMIT+1 X
shl(LIMIT,12,1)+1 X
Upper bound is:
LIMIT X
shl(LIMIT,12,1) X
64K-1 X
4G-1 X
Max seg size is:
64K X
64K-1 X
4G-4K X
4G X
Min seg size is:
0 X X
4K X X
shl (X, 12, 1) = shift X left by 12 bits inserting one-bits on the right
6.3.1.3 Privilege Levels
The concept of privilege is implemented by assigning a value from zero to
three to key objects recognized by the processor. This value is called the
privilege level. The value zero represents the greatest privilege, the
value three represents the least privilege. The following
processor-recognized objects contain privilege levels:
- Descriptors contain a field called the descriptor privilege level
(DPL).
- Selectors contain a field called the requestor's privilege level
(RPL). The RPL is intended to represent the privilege level of
the procedure that originates a selector.
- An internal processor register records the current privilege level
(CPL). Normally the CPL is equal to the DPL of the segment that
the processor is currently executing. CPL changes as control is
transferred to segments with differing DPLs.
The processor automatically evaluates the right of a procedure to access
another segment by comparing the CPL to one or more other privilege levels.
The evaluation is performed at the time the selector of a descriptor is
loaded into a segment register. The criteria used for evaluating access to
data differs from that for evaluating transfers of control to executable
segments; therefore, the two types of access are considered separately in
the following sections.
Figure 6-2
shows how these levels of privilege can be interpreted as rings
of protection. The center is for the segments containing the most critical
software, usually the kernel of the operating system. Outer rings are for
the segments of less critical software.
It is not necessary to use all four privilege levels. Existing software
that was designed to use only one or two levels of privilege can simply
ignore the other levels offered by the 80386. A one-level system should use
privilege level zero; a two-level system should use privilege levels zero
and three.
To address operands in memory, an 80386 program must load the selector of a
data segment into a data-segment register (DS, ES, FS, GS, SS). The
processor automatically evaluates access to a data segment by comparing
privilege levels. The evaluation is performed at the time a selector for the
descriptor of the target segment is loaded into the data-segment register.
As
Figure 6-3
shows, three different privilege levels enter into this type
of privilege check:
- The CPL (current privilege level).
- The RPL (requestor's privilege level) of the selector used to specify
the target segment.
- The DPL of the descriptor of the target segment.
Instructions may load a data-segment register (and subsequently use the
target segment) only if the DPL of the target segment is numerically greater
than or equal to the maximum of the CPL and the selector's RPL. In other
words, a procedure can only access data that is at the same or less
privileged level.
The addressable domain of a task varies as CPL changes. When CPL is zero,
data segments at all privilege levels are accessible; when CPL is one, only
data segments at privilege levels one through three are accessible; when CPL
is three, only data segments at privilege level three are accessible. This
property of the 80386 can be used, for example, to prevent applications
procedures from reading or changing tables of the operating system.
Less common than the use of data segments is the use of code segments to
store data. Code segments may legitimately hold constants; it is not
possible to write to a segment described as a code segment. The following
methods of accessing data in code segments are possible:
- Load a data-segment register with a selector of a nonconforming,
readable, executable segment.
- Load a data-segment register with a selector of a conforming,
readable, executable segment.
- Use a CS override prefix to read a readable, executable segment whose
selector is already loaded in the CS register.
The same rules as for access to data segments apply to case 1. Case 2 is
always valid because the privilege level of a segment whose conforming bit
is set is effectively the same as CPL regardless of its DPL. Case 3 always
valid because the DPL of the code segment in CS is, by definition, equal to
CPL.
6.3.3 Restricting Control Transfers
With the 80386, control transfers are accomplished by the instructions
JMP,
CALL,
RET,
INT, and IRET,
as well as by the exception and interrupt
mechanisms . Exceptions and interrupts are special cases that
Chapter 9
covers. This chapter discusses only JMP,
CALL, and RET instructions.
The "near" forms of JMP, CALL, and
RET transfer within the current code
segment, and therefore are subject only to limit checking. The processor
ensures that the destination of the JMP,
CALL, or
RET instruction does not
exceed the limit of the current executable segment. This limit is cached in
the CS register; therefore, protection checks for near transfers require no
extra clock cycles.
The operands of the "far" forms of
JMP and
CALL refer to other segments;
therefore, the processor performs privilege checking. There are two ways a
JMP or
CALL can refer to another segment:
- The operand selects the descriptor of another executable segment.
- The operand selects a call gate descriptor. This gated form of
transfer is discussed in a later section on call gates.
As
Figure 6-4
shows, two different privilege levels enter into a privilege
check for a control transfer that does not use a call gate:
- The CPL (current privilege level).
- The DPL of the descriptor of the target segment.
Normally the CPL is equal to the DPL of the segment that the processor is
currently executing. CPL may, however, be greater than DPL if the conforming
bit is set in the descriptor of the current executable segment. The
processor keeps a record of the CPL cached in the CS register; this value
can be different from the DPL in the descriptor of the code segment.
The processor permits a
JMP or
CALL directly to another segment only if one
of the following privilege rules is satisfied:
- DPL of the target is equal to CPL.
- The conforming bit of the target code-segment descriptor is set, and
the DPL of the target is less than or equal to CPL.
An executable segment whose descriptor has the conforming bit set is called
a conforming segment. The conforming-segment mechanism permits sharing of
procedures that may be called from various privilege levels but should
execute at the privilege level of the calling procedure. Examples of such
procedures include math libraries and some exception handlers. When control
is transferred to a conforming segment, the CPL does not change. This is
the only case when CPL may be unequal to the DPL of the current executable
segment.
Most code segments are not conforming. The basic rules of privilege above
mean that, for nonconforming segments, control can be transferred without a
gate only to executable segments at the same level of privilege. There is a
need, however, to transfer control to (numerically) smaller privilege
levels; this need is met by the
CALL instruction when used with call-gate
descriptors, which are explained in the next section. The
JMP instruction
may never transfer control to a nonconforming segment whose DPL does not
equal CPL.
To provide protection for control transfers among executable segments
at different privilege levels, the 80386 uses gate descriptors. There are
four kinds of gate descriptors:
- Call gates
- Trap gates
- Interrupt gates
- Task gates
This chapter is concerned only with call gates. Task gates are used for
task switching , and therefore are discussed in
Chapter 7.
Chapter 9
explains how trap gates and interrupt gates are used by exceptions and
interrupts.
Figure 6-5
illustrates the format of a call gate. A call gate
descriptor may reside in the GDT or in an LDT, but not in the IDT.
A call gate has two primary functions:
- To define an entry point of a procedure.
- To specify the privilege level of the entry point.
Call gate descriptors are used by call and jump instructions in the same
manner as code segment descriptors. When the hardware recognizes that the
destination selector refers to a gate descriptor, the operation of the
instruction is expanded as determined by the contents of the call gate.
The selector and offset fields of a gate form a pointer to the entry point
of a procedure. A call gate guarantees that all transitions to another
segment go to a valid entry point, rather than possibly into the middle of a
procedure (or worse, into the middle of an instruction). The far pointer
operand of the control transfer instruction does not point to the segment
and offset of the target instruction; rather, the selector part of the
pointer selects a gate, and the offset is not used.
Figure 6-6
illustrates
this style of addressing.
As
Figure 6-7
shows, four different privilege levels are used to check the
validity of a control transfer via a call gate:
- The CPL (current privilege level).
- The RPL (requestor's privilege level) of the selector used to specify
the call gate.
- The DPL of the gate descriptor.
- The DPL of the descriptor of the target executable segment.
The DPL field of the gate descriptor determines what privilege levels can
use the gate. One code segment can have several procedures that are intended
for use by different privilege levels. For example, an operating system may
have some services that are intended to be used by applications, whereas
others may be intended only for use by other systems software.
Gates can be used for control transfers to numerically smaller privilege
levels or to the same privilege level (though they are not necessary for
transfers to the same level). Only
CALL instructions can use gates to
transfer to smaller privilege levels. A gate may be used by a
JMP
instruction only to transfer to an executable segment with the same
privilege level or to a conforming segment.
For a
JMP instruction to a nonconforming segment,
both of the following
privilege rules must be satisfied; otherwise, a general protection exception
results.
MAX (CPL,RPL) <= gate DPL
target segment DPL = CPL
For a CALL instruction (or for a
JMP instruction to a conforming segment),
both of the following privilege rules must be satisfied; otherwise, a
general protection exception results.
MAX (CPL,RPL) <= gate DPL
target segment DPL <= CPL
If the destination code segment of the call gate is at a different
privilege level than the CPL, an interlevel transfer is being requested.
To maintain system integrity, each privilege level has a separate stack.
These stacks assure sufficient stack space to process calls from less
privileged levels. Without them, a trusted procedure would not work
correctly if the calling procedure did not provide sufficient space on the
caller's stack.
The processor locates these stacks via the task state segment (see
Figure 6-8).
Each task has a separate TSS, thereby permitting tasks to have
separate stacks. Systems software is responsible for creating TSSs and
placing correct stack pointers in them. The initial stack pointers in the
TSS are strictly read-only values. The processor never changes them during
the course of execution.
When a call gate is used to change privilege levels, a new stack is
selected by loading a pointer value from the Task State Segment (TSS). The
processor uses the DPL of the target code segment (the new CPL) to index the
initial stack pointer for PL 0, PL 1, or PL 2.
The DPL of the new stack data segment must equal the new CPL; if it does
not, a stack exception occurs. It is the responsibility of systems software
to create stacks and stack-segment descriptors for all privilege levels that
are used. Each stack must contain enough space to hold the old SS:ESP, the
return address, and all parameters and local variables that may be required
to process a call.
As with intralevel calls, parameters for the subroutine are placed on the
stack. To make privilege transitions transparent to the called procedure,
the processor copies the parameters to the new stack. The count field of a
call gate tells the processor how many doublewords (up to 31) to copy from
the caller's stack to the new stack. If the count is zero, no parameters are
copied.
The processor performs the following stack-related steps in executing an
interlevel CALL.
- The new stack is checked to assure that it is large enough to hold
the parameters and linkages; if it is not, a stack fault occurs with
an error code of 0.
- The old value of the stack registers SS:ESP is pushed onto the new
stack as two doublewords.
- The parameters are copied.
- A pointer to the instruction after the
CALL instruction (the former
value of CS:EIP) is pushed onto the new stack. The final value of
SS:ESP points to this return pointer on the new stack.
Figure 6-9
illustrates the stack contents after a successful interlevel
call.
The TSS does not have a stack pointer for a privilege level 3 stack,
because privilege level 3 cannot be called by any procedure at any other
privilege level.
Procedures that may be called from another privilege level and that require
more than the 31 doublewords for parameters must use the saved SS:ESP link
to access all parameters beyond the last doubleword copied.
A call via a call gate does not check the values of the words copied onto
the new stack. The called procedure should check each parameter for
validity. A later section discusses how the ARPL,
VERR,
VERW,
LSL, and
LAR
instructions can be used to check pointer values.
The "near" forms of the
RET instruction transfer control within the current
code segment and therefore are subject only to limit checking. The offset of
the instruction following the corresponding
CALL, is popped from the stack.
The processor ensures that this offset does not exceed the limit of the
current executable segment.
The "far" form of the
RET instruction pops the return pointer that was
pushed onto the stack by a prior far
CALL instruction. Under normal
conditions, the return pointer is valid, because of its relation to the
prior CALL or
INT.
Nevertheless, the processor performs privilege checking
because of the possibility that the current procedure altered the pointer or
failed to properly maintain the stack. The RPL of the CS selector popped
off the stack by the return instruction identifies the privilege level of
the calling procedure.
An intersegment return instruction can change privilege levels, but only
toward procedures of lesser privilege. When the
RET instruction encounters a
saved CS value whose RPL is numerically greater than the CPL, an interlevel
return occurs. Such a return follows these steps:
- The checks shown in Table 6-3 are made, and CS:EIP and SS:ESP are
loaded with their former values that were saved on the stack.
- The old SS:ESP (from the top of the current stack) value is adjusted
by the number of bytes indicated in the
RET instruction. The resulting
ESP value is not compared to the limit of the stack segment. If ESP is
beyond the limit, that fact is not recognized until the next stack
operation. (The SS:ESP value of the returning procedure is not
preserved; normally, this value is the same as that contained in the
TSS.)
- The contents of the DS, ES, FS, and GS segment registers are checked.
If any of these registers refer to segments whose DPL is greater than
the new CPL (excluding conforming code segments), the segment register
is loaded with the null selector (INDEX = 0, TI = 0). The
RET
instruction itself does not signal exceptions in these cases;
however, any subsequent memory reference that attempts to use a
segment register that contains the null selector will cause a general
protection exception. This prevents less privileged code from
accessing more privileged segments using selectors left in the
segment registers by the more privileged procedure.
6.3.5 Some Instructions are Reserved for Operating System
Instructions that have the power to affect the protection mechanism or to
influence general system performance can only be executed by trusted
procedures. The 80386 has two classes of such instructions:
- Privileged instructions -- those used for system control.
- Sensitive instructions -- those used for I/O and I/O related
activities.
Table 6-3. Interlevel Return Checks
SF = Stack Fault
GP = General Protection Exception
NP = Segment-Not-Present Exception
Type of Check Exception Error Code
ESP is within current SS segment SF 0
ESP + 7 is within current SS segment SF 0
RPL of return CS is greater than CPL GP Return CS
Return CS selector is not null GP Return CS
Return CS segment is within descriptor
table limit GP Return CS
Return CS descriptor is a code segment GP Return CS
Return CS segment is present NP Return CS
DPL of return nonconforming code
segment = RPL of return CS, or DPL of
return conforming code segment <= RPL
of return CS GP Return CS
ESP + N + 15 is within SS segment
N Immediate Operand of RET N Instruction SF Return SS
SS selector at ESP + N + 12 is not null GP Return SS
SS selector at ESP + N + 12 is within
descriptor table limit GP Return SS
SS descriptor is writable data segment GP Return SS
SS segment is present SF Return SS
Saved SS segment DPL = RPL of saved
CS GP Return SS
Saved SS selector RPL = Saved SS
segment DPL GP Return SS
6.3.5.1 Privileged Instructions
The instructions that affect system data structures can only be executed
when CPL is zero. If the CPU encounters one of these instructions when CPL
is greater than zero, it signals a general protection exception. These
instructions include:
6.3.5.2 Sensitive Instructions
Instructions that deal with I/O need to be restricted but also need to be
executed by procedures executing at privilege levels other than zero. The
mechanisms for restriction of I/O operations are covered in detail in
Chapter 8, "Input/Output".
6.3.6 Instructions for Pointer Validation
Pointer validation is an important part of locating programming errors.
Pointer validation is necessary for maintaining isolation between the
privilege levels. Pointer validation consists of the following steps:
- Check if the supplier of the pointer is entitled to access the
segment.
- Check if the segment type is appropriate to its intended use.
- Check if the pointer violates the segment limit.
Although the 80386 processor automatically performs checks 2 and 3 during
instruction execution, software must assist in performing the first check.
The unprivileged instruction
ARPL is provided for this purpose. Software can
also explicitly perform steps 2 and 3 to check for potential violations
(rather than waiting for an exception). The unprivileged instructions
LAR,
LSL,
VERR, and
VERW are provided for this purpose.
LAR (Load Access Rights) is used to verify that a pointer refers to a
segment of the proper privilege level and type.
LAR has one operand
selector for a descriptor whose access rights are to be examined. The
descriptor must be visible at the privilege level which is the maximum of
the CPL and the selector's RPL. If the descriptor is visible,
LAR obtains a
masked form of the second doubleword of the descriptor, masks this value
with 00FxFF00H, stores the result into the specified 32-bit destination
register, and sets the zero flag. (The x indicates that the corresponding
four bits of the stored value are undefined.) Once loaded, the access-rights
bits can be tested. All valid descriptor types can be tested by the
LAR
instruction. If the RPL or CPL is greater than DPL, or if the selector is
outside the table limit, no access-rights value is returned, and the zero
flag is cleared. Conforming code segments may be accessed from any privilege
level.
LSL
(Load Segment Limit) allows software to test the limit of a descriptor.
If the descriptor denoted by the given selector (in memory or a register) is
visible at the CPL,
LSL loads the specified 32-bit register with a 32-bit,
byte granular, unscrambled limit that is calculated from fragmented limit
fields and the G-bit of that descriptor. This can only be done for segments
(data, code, task state, and local descriptor tables); gate descriptors are
inaccessible. (Table 6-4 lists in detail which types are valid and which
are not.) Interpreting the limit is a function of the segment type. For
example, downward expandable data segments treat the limit differently than
code segments do. For both
LAR and
LSL, the zero flag (ZF) is set if the
loading was performed; otherwise, the ZF is cleared.
Table 6-4. Valid Descriptor Types for LSL
Type Descriptor Type Valid?
Code
0 (invalid) NO
1 Available 286 TSS YES
2 LDT YES
3 Busy 286 TSS YES
4 286 Call Gate NO
5 Task Gate NO
6 286 Trap Gate NO
7 286 Interrupt Gate NO
8 (invalid) NO
9 Available 386 TSS YES
A (invalid) NO
B Busy 386 TSS YES
C 386 Call Gate NO
D (invalid) NO
E 386 Trap Gate NO
F 386 Interrupt Gate NO
6.3.6.1 Descriptor Validation
The 80386 has two instructions,
VERR and
VERW, which determine whether a
selector points to a segment that can be read or written at the current
privilege level. Neither instruction causes a protection fault if the result
is negative.
VERR (Verify for Reading)
verifies a segment for reading and loads ZF with
1 if that segment is readable from the current privilege level.
VERR checks
that:
- The selector points to a descriptor within the bounds of the GDT or
LDT.
- It denotes a code or data segment descriptor.
- The segment is readable and of appropriate privilege level.
The privilege check for data segments and nonconforming code segments is
that the DPL must be numerically greater than or equal to both the CPL and
the selector's RPL. Conforming segments are not checked for privilege level.
VERW (Verify for Writing)
provides the same capability as VERR for
verifying writability. Like the
VERR instruction,
VERW loads ZF if the
result of the writability check is positive. The instruction checks that the
descriptor is within bounds, is a segment descriptor, is writable, and that
its DPL is numerically greater or equal to both the CPL and the selector's
RPL. Code segments are never writable, conforming or not.
6.3.6.2 Pointer Integrity and RPL
The Requestor's Privilege Level (RPL) feature can prevent inappropriate use
of pointers that could corrupt the operation of more privileged code or data
from a less privileged level.
A common example is a file system procedure, FREAD (file_id, n_bytes,
buffer_ptr). This hypothetical procedure reads data from a file into a
buffer, overwriting whatever is there. Normally, FREAD would be available at
the user level, supplying only pointers to the file system procedures and
data located and operating at a privileged level. Normally, such a procedure
prevents user-level procedures from directly changing the file tables.
However, in the absence of a standard protocol for checking pointer
validity, a user-level procedure could supply a pointer into the file tables
in place of its buffer pointer, causing the FREAD procedure to corrupt them
unwittingly.
Use of RPL can avoid such problems. The RPL field allows a privilege
attribute to be assigned to a selector. This privilege attribute would
normally indicate the privilege level of the code which generated the
selector. The 80386 processor automatically checks the RPL of any selector
loaded into a segment register to determine whether the RPL allows access.
To take advantage of the processor's checking of RPL, the called procedure
need only ensure that all selectors passed to it have an RPL at least as
high (numerically) as the original caller's CPL. This action guarantees that
selectors are not more trusted than their supplier. If one of the selectors
is used to access a segment that the caller would not be able to access
directly, i.e., the RPL is numerically greater than the DPL, then a
protection fault will result when that selector is loaded into a segment
register.
ARPL
(Adjust Requestor's Privilege Level) adjusts the RPL field of a
selector to become the larger of its original value and the value of the RPL
field in a specified register. The latter is normally loaded from the image
of the caller's CS register which is on the stack. If the adjustment changes
the selector's RPL, ZF (the zero flag) is set; otherwise, ZF is cleared.
Two kinds of protection are related to pages:
- Restriction of addressable domain.
- Type checking.
6.4.1 Page-Table Entries Hold Protection Parameters
Figure 6-10
highlights the fields of PDEs and PTEs that control access to
pages.
The concept of privilege for pages is implemented by assigning each page to
one of two levels:
- Supervisor level (U/S=0) -- for the operating system and other systems
software and related data.
- User level (U/S=1) -- for applications procedures and data.
The current level (U or S) is related to CPL. If CPL is 0, 1, or 2, the
processor is executing at supervisor level. If CPL is 3, the processor is
executing at user level.
When the processor is executing at supervisor level, all pages are
addressable, but, when the processor is executing at user level, only pages
that belong to the user level are addressable.
6.4.1.2 Type Checking
At the level of page addressing, two types are defined:
- Read-only access (R/W=0)
- Read/write access (R/W=1)
When the processor is executing at supervisor level, all pages are both
readable and writable. When the processor is executing at user level, only
pages that belong to user level and are marked for read/write access are
writable; pages that belong to supervisor level are neither readable nor
writable from user level.
6.4.2 Combining Protection of Both Levels of Page Tables
For any one page, the protection attributes of its page directory entry may
differ from those of its page table entry. The 80386 computes the effective
protection attributes for a page by examining the protection attributes in
both the directory and the page table. Table 6-5 shows the effective
protection provided by the possible combinations of protection attributes.
6.4.3 Overrides to Page Protection
Certain accesses are checked as if they are privilege-level 0 references,
even if CPL = 3:
- LDT, GDT, TSS, IDT references.
- Access to inner stack during ring-crossing
CALL / INT.
When paging is enabled, the 80386 first evaluates segment protection, then
evaluates page protection. If the processor detects a protection violation
at either the segment or the page level, the requested operation cannot
proceed; a protection exception occurs instead.
For example, it is possible to define a large data segment which has some
subunits that are read-only and other subunits that are read-write. In this
case, the page directory (or page table) entries for the read-only subunits
would have the U/S and R/W bits set to x0, indicating no write rights for
all the pages described by that directory entry (or for individual pages).
This technique might be used, for example, in a UNIX-like system to define
a large data segment, part of which is read only (for shared data or ROMmed
constants). This enables UNIX-like systems to define a "flat" data space as
one large segment, use "flat" pointers to address within this "flat" space,
yet be able to protect shared data, shared files mapped into the virtual
space, and supervisor areas.
Table 6-5. Combining Directory and Page Protection
Page Directory Entry Page Table Entry Combined Protection
U/S R/W U/S R/W U/S R/W
S-0 R-0 S-0 R-0 S x
S-0 R-0 S-0 W-1 S x
S-0 R-0 U-1 R-0 S x
S-0 R-0 U-1 W-1 S x
S-0 W-1 S-0 R-0 S x
S-0 W-1 S-0 W-1 S x
S-0 W-1 U-1 R-0 S x
S-0 W-1 U-1 W-1 S x
U-1 R-0 S-0 R-0 S x
U-1 R-0 S-0 W-1 S x
U-1 R-0 U-1 R-0 U R
U-1 R-0 U-1 W-1 U R
U-1 W-1 S-0 R-0 S x
U-1 W-1 S-0 W-1 S x
U-1 W-1 U-1 R-0 U R
U-1 W-1 U-1 W-1 U W
Note
S -- Supervisor
R -- Read only
U -- User
W -- Read and Write
x indicates that when the combined U/S attribute is S, the R/W attribute
is not checked.
All the information the processor needs in order to manage a task is stored
in a special type of segment, a task state segment (TSS).
Figure 7-1
shows
the format of a TSS for executing 80386 tasks. (Another format is used for
executing 80286 tasks; refer to
Chapter 13.)
The fields of a TSS belong to two classes:
- A dynamic set that the processor updates with each switch from the
task. This set includes the fields that store:
- The general registers (EAX, ECX, EDX, EBX, ESP, EBP, ESI, EDI).
- The segment registers (ES, CS, SS, DS, FS, GS).
- The flags register (EFLAGS).
- The instruction pointer (EIP).
- The selector of the TSS of the previously executing task (updated
only when a return is expected).
- A static set that the processor reads but does not change. This set
includes the fields that store:
- The selector of the task's LDT.
- The register (PDBR) that contains the base address of the task's
page directory (read only when paging is enabled).
- Pointers to the stacks for privilege levels 0-2.
- The T-bit (debug trap bit) which causes the processor to raise a
debug exception when a task switch occurs . (Refer to
Chapter 12
for more information on debugging.)
- The I/O map base (refer to
Chapter 8 for more information on the
use of the I/O map).
Task state segments may reside anywhere in the linear space. The only case
that requires caution is when the TSS spans a page boundary and the
higher-addressed page is not present. In this case, the processor raises an
exception if it encounters the not-present page while reading the TSS during
a task switch. Such an exception can be avoided by either of two strategies:
- By allocating the TSS so that it does not cross a page boundary.
- By ensuring that both pages are either both present or both
not-present at the time of a task switch. If both pages are
not-present, then the page-fault handler must make both pages present
before restarting the instruction that caused the task switch.
The task state segment, like all other segments, is defined by a
descriptor.
Figure 7-2
shows the format of a TSS descriptor.
The B-bit in the type field indicates whether the task is busy. A type code
of 9 indicates a non-busy task; a type code of 11 indicates a busy task.
Tasks are not reentrant. The B-bit allows the processor to detect an attempt
to switch to a task that is already busy.
The BASE, LIMIT, and DPL fields and the G-bit and P-bit have functions
similar to their counterparts in data-segment descriptors. The LIMIT field,
however, must have a value equal to or greater than 103. An attempt to
switch to a task whose TSS descriptor has a limit less that 103 causes an
exception. A larger limit is permissible, and a larger limit is required if
an I/O permission map is present. A larger limit may also be convenient for
systems software if additional data is stored in the same segment as the
TSS.
A procedure that has access to a TSS descriptor can cause a task switch. In
most systems the DPL fields of TSS descriptors should be set to zero, so
that only trusted software has the right to perform task switching.
Having access to a TSS-descriptor does not give a procedure the right to
read or modify a TSS. Reading and modification can be accomplished only with
another descriptor that redefines the TSS as a data segment. An attempt to
load a TSS descriptor into any of the segment registers (CS, SS, DS, ES, FS,
GS) causes an exception.
TSS descriptors may reside only in the GDT. An attempt to identify a TSS
with a selector that has TI=1 (indicating the current LDT) results in an
exception.
The task register (TR) identifies the currently executing task by pointing
to the TSS.
Figure 7-3
shows the path by which the processor accesses the
current TSS.
The task register has both a "visible" portion (i.e., can be read and
changed by instructions) and an "invisible" portion (maintained by the
processor to correspond to the visible portion; cannot be read by any
instruction). The selector in the visible portion selects a TSS descriptor
in the GDT. The processor uses the invisible portion to cache the base and
limit values from the TSS descriptor. Holding the base and limit in a
register makes execution of the task more efficient, because the processor
does not need to repeatedly fetch these values from memory when it
references the TSS of the current task.
The instructions LTR and
STR are used to modify and read the visible
portion of the task register. Both instructions take one operand, a 16-bit
selector located in memory or in a general register.
LTR (Load task register)
loads the visible portion of the task register
with the selector operand, which must select a TSS descriptor in the GDT.
LTR
also loads the invisible portion with information from the TSS
descriptor selected by the operand.
LTR is a privileged instruction; it may
be executed only when CPL is zero.
LTR is generally used during system
initialization to give an initial value to the task register; thereafter,
the contents of TR are changed by task switch operations.
STR (Store task register)
stores the visible portion of the task register
in a general register or memory word.
STR is not privileged.
A task gate descriptor provides an indirect, protected reference to a TSS.
Figure 7-4
illustrates the format of a task gate.
The SELECTOR field of a task gate must refer to a TSS descriptor. The value
of the RPL in this selector is not used by the processor.
The DPL field of a task gate controls the right to use the descriptor to
cause a task switch. A procedure may not select a task gate descriptor
unless the maximum of the selector's RPL and the CPL of the procedure is
numerically less than or equal to the DPL of the descriptor. This constraint
prevents untrusted procedures from causing a task switch. (Note that when a
task gate is used, the DPL of the target TSS descriptor is not used for
privilege checking.)
A procedure that has access to a task gate has the power to cause a task
switch, just as a procedure that has access to a TSS descriptor. The 80386
has task gates in addition to TSS descriptors to satisfy three needs:
- The need for a task to have a single busy bit. Because the busy-bit
is stored in the TSS descriptor, each task should have only one such
descriptor. There may, however, be several task gates that select the
single TSS descriptor.
- The need to provide selective access to tasks. Task gates fulfill
this need, because they can reside in LDTs and can have a DPL that is
different from the TSS descriptor's DPL. A procedure that does not
have sufficient privilege to use the TSS descriptor in the GDT (which
usually has a DPL of 0) can still switch to another task if it has
access to a task gate for that task in its LDT. With task gates,
systems software can limit the right to cause task switches to
specific tasks.
- The need for an interrupt or exception to cause a task switch. Task
gates may also reside in the IDT, making it possible for interrupts
and exceptions to cause task switching. When interrupt or exception
vectors to an IDT entry that contains a task gate, the 80386 switches
to the indicated task. Thus, all tasks in the system can benefit from
the protection afforded by isolation from interrupt tasks.
Figure 7-5
illustrates how both a task gate in an LDT and a task gate in
the IDT can identify the same task.
The 80386 switches execution to another task in any of four cases:
- The current task executes a JMP or
CALL that refers to a TSS
descriptor.
- The current task executes a JMP or
CALL that refers to a task gate.
- An interrupt or exception vectors to a task gate in the IDT.
- The current task executes an
IRET when the NT flag is set.
JMP,
CALL,
IRET,
interrupts, and exceptions are all ordinary mechanisms of
the 80386 that can be used in circumstances that do not require a task
switch. Either the type of descriptor referenced or the NT (nested task) bit
in the flag word distinguishes between the standard mechanism and the
variant that causes a task switch.
To cause a task switch, a JMP or
CALL instruction can refer either to a TSS
descriptor or to a task gate. The effect is the same in either case: the
80386 switches to the indicated task.
An exception or interrupt causes a task switch when it vectors to a task
gate in the IDT. If it vectors to an interrupt or trap gate in the IDT, a
task switch does not occur . Refer to
Chapter 9
for more information on the
interrupt mechanism.
Whether invoked as a task or as a procedure of the interrupted task, an
interrupt handler always returns control to the interrupted procedure in the
interrupted task. If the NT flag is set, however, the handler is an
interrupt task, and the
IRET switches back to the interrupted task.
A task switching operation involves these steps:
- Checking that the current task is allowed to switch to the designated
task. Data-access privilege rules apply in the case of
JMP or CALL
instructions. The DPL of the TSS descriptor or task gate must be
numerically greater (e.g., lower privilege level)
than or equal to the maximum of CPL and the RPL of the gate selector.
Exceptions, interrupts, and
IRET are permitted to switch tasks
regardless of the DPL of the target task gate or TSS descriptor.
- Checking that the TSS descriptor of the new task is marked present
and has a valid limit. Any errors up to this point occur in the
context of the outgoing task. Errors are restartable and can be
handled in a way that is transparent to applications procedures.
- Saving the state of the current task. The processor finds the base
address of the current TSS cached in the task register. It copies the
registers into the current TSS (EAX, ECX, EDX, EBX, ESP, EBP, ESI,
EDI, ES, CS, SS, DS, FS, GS, and the flag register). The EIP field of
the TSS points to the instruction after the one that caused the task
switch.
- Loading the task register with the selector of the incoming task's
TSS descriptor, marking the incoming task's TSS descriptor as busy,
and setting the TS (task switched) bit of the MSW. The selector is
either the operand of a control transfer instruction or is taken from
a task gate.
- Loading the incoming task's state from its TSS and resuming
execution. The registers loaded are the LDT register; the flag
register; the general registers EIP, EAX, ECX, EDX, EBX, ESP, EBP,
ESI, EDI; the segment registers ES, CS, SS, DS, FS, and GS; and PDBR.
Any errors detected in this step occur in the context of the incoming
task. To an exception handler, it appears that the first instruction
of the new task has not yet executed.
Note that the state of the outgoing task is always saved when a task switch
occurs. If execution of that task is resumed, it starts after the
instruction that caused the task switch. The registers are restored to the
values they held when the task stopped executing.
Every task switch sets the TS (task switched) bit in the MSW (machine
status word). The TS flag is useful to systems software when a coprocessor
(such as a numerics coprocessor) is present. The TS bit signals that the
context of the coprocessor may not correspond to the current 80386 task.
Chapter 11
discusses the TS bit and coprocessors in more detail .
Exception handlers that field task-switch exceptions in the incoming task
(exceptions due to tests 4 thru 16 of Table 7-1) should be cautious about
taking any action that might load the selector that caused the exception.
Such an action will probably cause another exception, unless the exception
handler first examines the selector and fixes any potential problem.
The privilege level at which execution resumes in the incoming task is
neither restricted nor affected by the privilege level at which the outgoing
task was executing. Because the tasks are isolated by their separate address
spaces and TSSs and because privilege rules can be used to prevent improper
access to a TSS, no privilege rules are needed to constrain the relation
between the CPLs of the tasks. The new task begins executing at the
privilege level indicated by the RPL of the CS selector value that is loaded
from the TSS.
Table 7-1. Checks Made during a Task Switch
NP = Segment-not-present exception
GP = General protection fault
TS = Invalid TSS
SF = Stack fault
Validity tests of a selector check that the selector is in the proper
table (e.g., the LDT selector refers to the GDT), lies within the bounds of
the table, and refers to the proper type of descriptor (e.g., the LDT
selector refers to an LDT descriptor).
Test Test Description Exception Error Code Selects
1 Incoming TSS descriptor is
present NP Incoming TSS
2 Incoming TSS descriptor is
marked not-busy GP Incoming TSS
marked not-busy
3 Limit of incoming TSS is
greater than or equal to 103 TS Incoming TSS
-- All register and selector values are loaded --
4 LDT selector of incoming
task is valid TS Incoming TSS
5 LDT of incoming task is
present TS Incoming TSS
6 CS selector is valid TS Code segment
7 Code segment is present NP Code segment
8 Code segment DPL matches
CS RPL TS Code segment
9 Stack segment is valid GP Stack segment
10 Stack segment is present SF Stack segment
11 Stack segment DPL = CPL SF Stack segment
12 Stack-selector RPL = CPL GP Stack segment
13 DS, ES, FS, GS selectors are
valid GP Segment
14 DS, ES, FS, GS segments
are readable GP Segment
15 DS, ES, FS, GS segments
are present NP Segment
16 DS, ES, FS, GS segment DPL
>= CPL (unless these are
conforming segments) GP Segment
The back-link field of the TSS and the NT (nested task) bit of the flag
word together allow the 80386 to automatically return to a task that
CALLed
another task or was interrupted by another task. When a
CALL instruction, an
interrupt instruction, an external interrupt, or an exception causes a
switch to a new task, the 80386 automatically fills the back-link of the new
TSS with the selector of the outgoing task's TSS and, at the same time,
sets the NT bit in the new task's flag register. The NT flag indicates
whether the back-link field is valid. The new task releases control by
executing an IRET instruction. When interpreting an
IRET, the 80386 examines
the NT flag. If NT is set, the 80386 switches back to the task selected by
the back-link field. Table 7-2 summarizes the uses of these fields.
Table 7-2. Effect of Task Switch on BUSY, NT, and Back-Link
Affected Field Effect of JMP Effect of Effect of
Instruction CALL Instruction IRET Instruction
Busy bit of Set, must be Set, must be 0 Unchanged,
incoming task 0 before before must be set
Busy bit of Cleared Unchanged Cleared
outgoing task (already set)
NT bit of Cleared Set Unchanged
incoming task
NT bit of Unchanged Unchanged Cleared
outgoing task
Back-link of Unchanged Set to outgoing Unchanged
incoming task TSS selector
Back-link of Unchanged Unchanged Unchanged
outgoing task
7.6.1 Busy Bit Prevents Loops
The B-bit (busy bit) of the TSS descriptor ensures the integrity of the
back-link. A chain of back-links may grow to any length as interrupt tasks
interrupt other interrupt tasks or as called tasks call other tasks. The
busy bit ensures that the CPU can detect any attempt to create a loop. A
loop would indicate an attempt to reenter a task that is already busy;
however, the TSS is not a reentrable resource.
The processor uses the busy bit as follows:
- When switching to a task, the processor automatically sets the busy
bit of the new task.
- When switching from a task, the processor automatically clears the
busy bit of the old task if that task is not to be placed on the
back-link chain (i.e., the instruction causing the task switch is
JMP
or
IRET).
If the task is placed on the back-link chain, its busy bit
remains set.
- When switching to a task, the processor signals an exception if the
busy bit of the new task is already set.
By these actions, the processor prevents a task from switching to itself or
to any task that is on a back-link chain, thereby preventing invalid reentry
into a task.
The busy bit is effective even in multiprocessor configurations, because
the processor automatically asserts a bus lock when it sets or clears the
busy bit. This action ensures that two processors do not invoke the same
task at the same time . (Refer to
Chapter 11
for more on multiprocessing.)
7.6.2 Modifying Task Linkages
Any modification of the linkage order of tasks should be accomplished only
by software that can be trusted to correctly update the back-link and the
busy-bit. Such changes may be needed to resume an interrupted task before
the task that interrupted it. Trusted software that removes a task from the
back-link chain must follow one of the following policies:
- First change the back-link field in the TSS of the interrupting task,
then clear the busy-bit in the TSS descriptor of the task removed from
the list.
- Ensure that no interrupts occur between updating the back-link chain
and the busy bit.
The LDT selector and PDBR fields of the TSS give software systems designers
flexibility in utilization of segment and page mapping features of the
80386. By appropriate choice of the segment and page mappings for each task,
tasks may share address spaces, may have address spaces that are largely
distinct from one another, or may have any degree of sharing between these
two extremes.
The ability for tasks to have distinct address spaces is an important
aspect of 80386 protection. A module in one task cannot interfere with a
module in another task if the modules do not have access to the same address
spaces. The flexible memory management features of the 80386 allow systems
designers to assign areas of shared address space to those modules of
different tasks that are designed to cooperate with each other.
7.7.1 Task Linear-to-Physical Space Mapping
The choices for arranging the linear-to-physical mappings of tasks fall
into two general classes:
- One linear-to-physical mapping shared among all tasks.
When paging is not enabled, this is the only possibility. Without page
tables, all linear addresses map to the same physical addresses.
When paging is enabled, this style of linear-to-physical mapping
results from using one page directory for all tasks. The linear space
utilized may exceed the physical space available if the operating
system also implements page-level virtual memory.
- Several partially overlapping linear-to-physical mappings.
This style is implemented by using a different page directory for each
task. Because the PDBR (page directory base register) is loaded from
the TSS with each task switch, each task may have a different page
directory.
In theory, the linear address spaces of different tasks may map to
completely distinct physical addresses. If the entries of different page
directories point to different page tables and the page tables point to
different pages of physical memory, then the tasks do not share any physical
addresses.
In practice, some portion of the linear address spaces of all tasks must
map to the same physical addresses. The task state segments must lie in a
common space so that the mapping of TSS addresses does not change while the
processor is reading and updating the TSSs during a task switch. The linear
space mapped by the GDT should also be mapped to a common physical space;
otherwise, the purpose of the GDT is defeated.
Figure 7-6
shows how the
linear spaces of two tasks can overlap in the physical space by sharing
page tables.
7.7.2 Task Logical Address Space
By itself, a common linear-to-physical space mapping does not enable
sharing of data among tasks. To share data, tasks must also have a common
logical-to-linear space mapping; i.e., they must also have access to
descriptors that point into a shared linear address space. There are three
ways to create common logical-to-physical address-space mappings:
- Via the GDT. All tasks have access to the descriptors in the GDT. If
those descriptors point into a linear-address space that is mapped to
a common physical-address space for all tasks, then the tasks can
share data and instructions.
- By sharing LDTs. Two or more tasks can use the same LDT if the LDT
selectors in their TSSs select the same LDT segment. Those
LDT-resident descriptors that point into a linear space that is mapped
to a common physical space permit the tasks to share physical memory.
This method of sharing is more selective than sharing by the GDT; the
sharing can be limited to specific tasks. Other tasks in the system
may have different LDTs that do not give them access to the shared
areas.
- By descriptor aliases in LDTs. It is possible for certain descriptors
of different LDTs to point to the same linear address space. If that
linear address space is mapped to the same physical space by the page
mapping of the tasks involved, these descriptors permit the tasks to
share the common space. Such descriptors are commonly called
"aliases". This method of sharing is even more selective than the
prior two; other descriptors in the LDTs may point to distinct linear
addresses or to linear addresses that are not shared.
The 80386 allows input/output to be performed in either of two ways:
- By means of a separate I/O address space (using specific I/O
instructions)
- By means of memory-mapped I/O (using general-purpose operand
manipulation instructions).
8.1.1 I/O Address Space
The 80386 provides a separate I/O address space, distinct from physical
memory, that can be used to address the input/output ports that are used for
external 16 devices. The I/O address space consists of 2^(16) (64K)
individually addressable 8-bit ports; any two consecutive 8-bit ports can be
treated as a 16-bit port; and four consecutive 8-bit ports can be treated
as a 32-bit port. Thus, the I/O address space can accommodate up to 64K
8-bit ports, up to 32K 16-bit ports, or up to 16K 32-bit ports.
The program can specify the address of the port in two ways. Using an
immediate byte constant, the program can specify:
- 256 8-bit ports numbered 0 through 255.
- 128 16-bit ports numbered 0, 2, 4, . . . , 252, 254.
- 64 32-bit ports numbered 0, 4, 8, . . . , 248, 252.
Using a value in DX, the program can specify:
- 8-bit ports numbered 0 through 65535
- 16-bit ports numbered 0, 2, 4, . . . , 65532, 65534
- 32-bit ports numbered 0, 4, 8, . . . , 65528, 65532
The 80386 can transfer 32, 16, or 8 bits at a time to a device located in
the I/O space. Like doublewords in memory, 32-bit ports should be aligned at
addresses evenly divisible by four so that the 32 bits can be transferred in
a single bus access. Like words in memory, 16-bit ports should be aligned at
even-numbered addresses so that the 16 bits can be transferred in a single
bus access. An 8-bit port may be located at either an even or odd address.
The instructions IN and
OUT move data between a register and a port in the
I/O address space. The instructions INS and
OUTS move strings of data
between the memory address space and ports in the I/O address space.
8.1.2 Memory-Mapped I/O
I/O devices also may be placed in the 80386 memory address space. As long
as the devices respond like memory components, they are indistinguishable to
the processor.
Memory-mapped I/O provides additional programming flexibility. Any
instruction that references memory may be used to access an I/O port located
in the memory space. For example, the MOV instruction can transfer data
between any register and a port; and the AND, OR, and TEST instructions may
be used to manipulate bits in the internal registers of a device (see
Figure 8-1
). Memory-mapped I/O performed via the full instruction set maintains
the full complement of addressing modes for selecting the desired I/O
device (e.g., direct address, indirect address, base register, index
register, scaling).
Memory-mapped I/O, like any other memory reference, is subject to access
protection and control when executing in protected mode. Refer to
Chapter 6
for a discussion of memory protection.
The I/O instructions of the 80386 provide access to the processor's I/O
ports for the transfer of data to and from peripheral devices. These
instructions have as one operand the address of a port in the I/O address
space. There are two classes of I/O instruction:
- Those that transfer a single item (byte, word, or doubleword) located
in a register.
- Those that transfer strings of items (strings of bytes, words, or
doublewords) located in memory. These are known as "string I/O
instructions" or "block I/O instructions".
8.2.1 Register I/O Instructions
The I/O instructions IN and
OUTare provided to move data between I/O ports
and the EAX (32-bit I/O), the AX (16-bit I/O), or AL (8-bit I/O) general
registers. IN and
OUTinstructions address I/O ports either directly, with
the address of one of up to 256 port addresses coded in the instruction, or
indirectly via the DX register to one of up to 64K port addresses.
IN (Input from Port)
transfers a byte, word, or doubleword from an input
port to AL, AX, or EAX. If a program specifies AL with the
IN instruction,
the processor transfers 8 bits from the selected port to AL. If a program
specifies AX with the
IN instruction, the processor transfers 16 bits from
the port to AX. If a program specifies EAX with the
IN instruction, the
processor transfers 32 bits from the port to EAX.
OUT(Output to Port)
transfers a byte, word, or doubleword to an output
port from AL, AX, or EAX. The program can specify the number of the port
using the same methods as the IN instruction.
The block (or string) I/O instructions
INS and
OUTS move blocks of data
between I/O ports and memory space. Block I/O instructions use the DX
register to specify the address of a port in the I/O address space.
INS and
OUTSuse DX to specify:
- 8-bit ports numbered 0 through 65535
- 16-bit ports numbered 0, 2, 4, . . . , 65532, 65534
- 32-bit ports numbered 0, 4, 8, . . . , 65528, 65532
Block I/O instructions use either SI or DI to designate the source or
destination memory address. For each transfer, SI or DI are automatically
either incremented or decremented as specified by the direction bit in the
flags register.
INS and OUTS,
when used with repeat prefixes, cause block input or output
operations.
REP, the repeat prefix, modifies
INS and OUTS to provide a means
of transferring blocks of data between an I/O port and memory. These block
I/O instructions are string primitives (refer also to
Chapter 3
for more on
string primitives). They simplify programming and increase the speed of data
transfer by eliminating the need to use a separate
LOOP instruction or an
intermediate register to hold the data.
The string I/O primitives can operate on byte strings, word strings, or
doubleword strings. After each transfer, the memory address in ESI or EDI is
updated by 1 for byte operands, by 2 for word operands, or by 4 for
doubleword operands. The value in the direction flag (DF) determines whether
the processor automatically increments ESI or EDI (DF=0) or whether it
automatically decrements these registers (DF=1).
INS(Input String from Port)
transfers a byte or a word string element from
an input port to memory. The mnemonics
INSB,
INSW, and
INSD are variants
that explicitly specify the size of the operand. If a program specifies
INSB,
the processor transfers 8 bits from the selected port to the memory
location indicated by ES:EDI. If a program specifies
INSW, the processor
transfers 16 bits from the port to the memory location indicated by ES:EDI.
If a program specifies
INSD, the processor transfers 32 bits from the port
to the memory location indicated by ES:EDI. The destination segment register
choice (ES) cannot be changed for the
INS instruction. Combined with the
REP prefix,
INS moves a block of information from an input
port to a series of consecutive memory locations.
OUTS (Output String to Port)
transfers a byte, word, or doubleword string
element to an output port from memory. The mnemonics
OUTSB,
OUTSW, and
OUTSD
are variants that explicitly specify the size of the operand. If a program
specifies
OUTSB, the processor transfers 8 bits
from the memory location
indicated by ES:EDI to the the selected port. If a program specifies
OUTSW,
the processor transfers 16 bits from the memory location indicated by ES:EDI
to the the selected port. If a program specifies
OUTSD, the processor
transfers 32 bits from the memory location indicated by ES:EDI to the the
selected port. Combined with the
REP prefix,
OUTS moves a block of
information from a series of consecutive memory locations indicated by
DS:ESI to an output port.
Two mechanisms provide protection for I/O functions:
- The IOPL field in the EFLAGS register defines the right to use
I/O-related instructions.
- The I/O permission bit map of a 80386 TSS segment defines the right
to use ports in the I/O address space.
These mechanisms operate only in protected mode, including virtual 8086
mode; they do not operate in real mode. In real mode, there is no protection
of the I/O space; any procedure can execute I/O instructions, and any I/O
port can be addressed by the I/O instructions.
8.3.1 I/O Privilege Level
Instructions that deal with I/O need to be restricted but also need to be
executed by procedures executing at privilege levels other than zero. For
this reason, the processor uses two bits of the flags register to store the
I/O privilege level (IOPL). The IOPL defines the privilege level
needed to execute I/O-related instructions.
The following instructions can be executed only if CPL <= IOPL:
- IN -- Input
- INS -- Input String
- OUT -- Output
- OUTS -- Output String
- CLI -- Clear Interrupt-Enable Flag
- STI -- Set Interrupt-Enable
These instructions are called "sensitive" instructions, because they are
sensitive to IOPL.
To use sensitive instructions, a procedure must execute at a privilege
level at least as privileged as that specified by the IOPL (CPL <= IOPL). Any
attempt by a less privileged procedure to use a sensitive instruction
results in a general protection exception.
Because each task has its own unique copy of the flags register, each task
can have a different IOPL. A task whose primary function is to perform I/O
(a device driver) can benefit from having an IOPL of three, thereby
permitting all procedures of the task to perform I/O. Other tasks typically
have IOPL set to zero or one, reserving the right to perform I/O
instructions for the most privileged procedures.
A task can change IOPL only with the
POPF instruction; however, such
changes are privileged. No procedure may alter IOPL (the I/O privilege level
in the flag register) unless the procedure is executing at privilege level
0. An attempt by a less privileged procedure to alter IOPL does not result
in an exception; IOPL simply remains unaltered.
The POPF instruction may be used in addition to
CLI and STI to alter the
interrupt-enable flag (IF); however, changes to IF by
POPF are
IOPL-sensitive. A procedure may alter IF with a
POPF instruction only when
executing at a level that is at least as privileged as IOPL. An attempt by a
less privileged procedure to alter IF in this manner does not result in an
exception; IF simply remains unaltered.
8.3.2 I/O Permission Bit Map
The I/O instructions that directly refer to addresses in the processor's
I/O space are
IN,
INS,
OUT,
OUTS. The 80386 has the ability to selectively
trap references to specific I/O addresses. The structure that enables
selective trapping is the I/O Permission Bit Map in the TSS segment (see
Figure 8-2).
The I/O permission map is a bit vector. The size of the map
and its location in the TSS segment are variable. The processor locates the
I/O permission map by means of the I/O map base field in the fixed portion
of the TSS. The I/O map base field is 16 bits wide and contains the offset
of the beginning of the I/O permission map. The upper limit of the I/O
permission map is the same as the limit of the TSS segment.
In protected mode, when it encounters an I/O instruction
(IN,
INS,
OUT, or
OUTS), the processor first checks whether CPL <= IOPL.
If this condition is
true, the I/O operation may proceed. If not true, the processor checks the
I/O permission map. (In virtual 8086 mode, the processor consults the map
without regard for IOPL . Refer to
Chapter 15.)
Each bit in the map corresponds to an I/O port byte address; for example,
the bit for port 41 is found at I/O map base + 5, bit offset 1. The
processor tests all the bits that correspond to the I/O addresses spanned by
an I/O operation; for example, a doubleword operation tests four bits
corresponding to four adjacent byte addresses. If any tested bit is set,
the processor signals a general protection exception. If all the tested bits
are zero, the I/O operation may proceed.
It is not necessary for the I/O permission map to represent all the I/O
addresses. I/O addresses not spanned by the map are treated as if they had
one bits in the map. For example, if TSS limit is equal to I/O map base +
31, the first 256 I/O ports are mapped; I/O operations on any port greater
than 255 cause an exception.
If I/O map base is greater than or equal to TSS limit, the TSS segment has
no I/O permission map, and all I/O instructions in the 80386 program cause
exceptions when CPL > IOPL.
Because the I/O permission map is in the TSS segment, different tasks can
have different maps. Thus, the operating system can allocate ports to a task
by changing the I/O permission map in the task's TSS.
The processor associates an identifying number with each different type of
interrupt or exception.
The NMI and the exceptions recognized by the processor are assigned
predetermined identifiers in the range 0 through 31. Not all of these
numbers are currently used by the 80386; unassigned identifiers in this
range are reserved by Intel for possible future expansion.
The identifiers of the maskable interrupts are determined by external
interrupt controllers (such as Intel's 8259A Programmable Interrupt
Controller) and communicated to the processor during the processor's
interrupt-acknowledge sequence. The numbers assigned by an 8259A PIC can be
specified by software. Any numbers in the range 32 through 255 can be used.
Table 9-1 shows the assignment of interrupt and exception identifiers.
Exceptions are classified as faults, traps, or aborts depending on the way
they are reported and whether restart of the instruction that caused the
exception is supported.
-
Faults
-
Faults are exceptions that are reported "before" the
instruction causingthe exception. Faults are either detected before
the instruction begins to execute, or during execution of the
instruction. If detected during the instruction, the fault is
reported with the machine restored to a state that permits the
instruction to be restarted.
-
Traps
-
A trap is an exception that is reported at the instruction
boundary immediately after the instruction in which the
exception was detected.
-
Aborts
-
An abort is an exception that permits neither precise location
of the instruction causing the exception nor restart of the program
that caused the exception. Aborts are used to report severe errors,
such as hardware errors and inconsistent or illegal values in system
tables.
Table 9-1. Interrupt and Exception ID Assignments
Identifier Description
0 Divide error
1 Debug exceptions
2 Nonmaskable interrupt
3 Breakpoint (one-byte INT 3 instruction)
4 Overflow (INTO instruction)
5 Bounds check (BOUND instruction)
6 Invalid opcode
7 Coprocessor not available
8 Double fault
9 (reserved)
10 Invalid TSS
11 Segment not present
12 Stack exception
13 General protection
14 Page fault
15 (reserved)
16 Coprecessor error
17-31 (reserved)
32-255 Available for external interrupts via INTR pin
The processor services interrupts and exceptions only between the end of
one instruction and the beginning of the next. When the repeat prefix is
used to repeat a string instruction, interrupts and exceptions may occur
between repetitions. Thus, operations on long strings do not delay interrupt
response.
Certain conditions and flag settings cause the processor to inhibit certain
interrupts and exceptions at instruction boundaries.
9.2.1 NMI Masks Further NMIs
While an NMI handler is executing, the processor ignores further interrupt
signals at the NMI pin until the next IRET instruction is executed.
9.2.2 IF Masks INTR
The IF (interrupt-enable flag) controls the acceptance of external
interrupts signalled via the INTR pin. When IF=0, INTR interrupts are
inhibited; when IF=1, INTR interrupts are enabled. As with the other flag
bits, the processor clears IF in response to a RESET signal. The
instructions CLI and
STI alter the setting of IF.
CLI (Clear Interrupt-Enable Flag) and
STI (Set Interrupt-Enable Flag)
explicitly alter IF (bit 9 in the flag register). These instructions may be
executed only if CPL <= IOPL. A protection exception occurs if they are
executed when CPL > IOPL.
The IF is also affected implicitly by the following operations:
- The instruction
PUSHF stores all flags, including IF, in the stack
where they can be examined.
- Task switches and the instructions
POPF and
IRET load the flags
register; therefore, they can be used to modify IF.
- Interrupts through interrupt gates automatically reset IF, disabling
interrupts. (Interrupt gates are explained later in this chapter.)
9.2.3 RF Masks Debug Faults
The RF bit in EFLAGS controls the recognition of debug faults. This permits
debug faults to be raised for a given instruction at most once, no matter
how many times the instruction is restarted . (Refer to
Chapter 12
for more
information on debugging.)
9.2.4 MOV or POP to SS Masks Some Interrupts and Exceptions
Software that needs to change stack segments often uses a pair of
instructions; for example:
MOV SS, AX
MOV ESP, StackTop
If an interrupt or exception is processed after SS has been changed but
before ESP has received the corresponding change, the two parts of the stack
pointer SS:ESP are inconsistent for the duration of the interrupt handler or
exception handler.
To prevent this situation, the 80386, after both a
MOV to SS and a
POP to
SS instruction, inhibits NMI, INTR, debug exceptions, and single-step traps
at the instruction boundary following the instruction that changes SS. Some
exceptions may still occur; namely, page fault and general protection fault.
Always use the 80386
LSS instruction, and the problem will not occur.
If more than one interrupt or exception is pending at an instruction
boundary, the processor services one of them at a time. The priority among
classes of interrupt and exception sources is shown in
Table 9-2. The
processor first services a pending interrupt or exception from the class
that has the highest priority, transferring control to the first
instruction of the interrupt handler. Lower priority exceptions are
discarded; lower priority interrupts are held pending. Discarded exceptions
will be rediscovered when the interrupt handler returns control to the point
of interruption.
The interrupt descriptor table (IDT) associates each interrupt or exception
identifier with a descriptor for the instructions that service the
associated event. Like the GDT and LDTs, the IDT is an array of 8-byte
descriptors. Unlike the GDT and LDTs, the first entry of the IDT may contain
a descriptor. To form an index into the IDT, the processor multiplies the
interrupt or exception identifier by eight. Because there are only 256
identifiers, the IDT need not contain more than 256 descriptors. It can
contain fewer than 256 entries; entries are required only for interrupt
identifiers that are actually used.
The IDT may reside anywhere in physical memory. As
Figure 9-1
shows, the
processor locates the IDT by means of the IDT register (IDTR). The
instructions
LIDT and
SIDT operate on the IDTR. Both instructions have one
explicit operand: the address in memory of a 6-byte area.
Figure 9-2
shows
the format of this area.
LIDT (Load IDT register)
loads the IDT register with the linear base
address and limit values contained in the memory operand. This instruction
can be executed only when the CPL is zero. It is normally used by the
initialization logic of an operating system when creating an IDT. An
operating system may also use it to change from one IDT to another.
SIDT (Store IDT register)
copies the base and limit value stored in IDTR
to a memory location. This instruction can be executed at any privilege
level.
Table 9-2. Priority Among Simultaneous Interrupts and Exceptions
Priority Class of Interrupt or Exception
HIGHEST Faults except debug faults
Trap instructions INTO, INT n, INT 3
Debug traps for this instruction
Debug faults for next instruction
NMI interrupt
LOWEST INTR interrupt
The IDT may contain any of three kinds of descriptor:
- Task gates
- Interrupt gates
- Trap gates
Figure 9-3
illustrates the format of task gates and 80386 interrupt gates
and trap gates. (The task gate in an IDT is the same as the task gate
already discussed in
Chapter 7.)
Just as a CALL instruction can call either a procedure or a task, so an
interrupt or exception can "call" an interrupt handler that is either a
procedure or a task. When responding to an interrupt or exception, the
processor uses the interrupt or exception identifier to index a descriptor
in the IDT. If the processor indexes to an interrupt gate or trap gate, it
invokes the handler in a manner similar to a CALL to a call gate. If the
processor finds a task gate, it causes a task switch in a manner similar to
a CALL to a task gate.
9.6.1 Interrupt Procedures
An interrupt gate or trap gate points indirectly to a procedure which will
execute in the context of the currently executing task as illustrated by
Figure 9-4
. The selector of the gate points to an executable-segment
descriptor in either the GDT or the current LDT. The offset field of the
gate points to the beginning of the interrupt or exception handling
procedure.
The 80386 invokes an interrupt or exception handling procedure in much the
same manner as it
CALLs a procedure; the differences are explained in the
following sections.
Just as with a control transfer due to a CALL instruction, a control
transfer to an interrupt or exception handling procedure uses the stack to
store the information needed for returning to the original procedure. As
Figure 9-5
shows, an interrupt pushes the EFLAGS register onto the stack
before the pointer to the interrupted instruction.
Certain types of exceptions also cause an error code to be pushed on the
stack. An exception handler can use the error code to help diagnose the
exception.
9.6.1.2 Returning from an Interrupt Procedure
An interrupt procedure also differs from a normal procedure in the method
of leaving the procedure. The
IRET instruction is used to exit from an
interrupt procedure. IRET is similar to
RET except that
IRET increments EIP
by an extra four bytes (because of the flags on the stack) and moves the
saved flags into the EFLAGS register. The IOPL field of EFLAGS is changed
only if the CPL is zero. The IF flag is changed only if CPL <= IOPL.
Interrupts that vector through either interrupt gates or trap gates cause
TF (the trap flag) to be reset after the current value of TF is saved on the
stack as part of EFLAGS. By this action the processor prevents debugging
activity that uses single-stepping from affecting interrupt response. A
subsequent IRET instruction restores TF to the value in the EFLAGS image on
the stack.
The difference between an interrupt gate and a trap gate is in the effect
on IF (the interrupt-enable flag). An interrupt that vectors through an
interrupt gate resets IF, thereby preventing other interrupts from
interfering with the current interrupt handler. A subsequent
IRET
instruction restores IF to the value in the EFLAGS image on the stack. An
interrupt through a trap gate does not change IF.
9.6.1.4 Protection in Interrupt Procedures
The privilege rule that governs interrupt procedures is similar to that for
procedure calls: the CPU does not permit an interrupt to transfer control to
a procedure in a segment of lesser privilege (numerically greater privilege
level) than the current privilege level. An attempt to violate this rule
results in a general protection exception.
Because occurrence of interrupts is not generally predictable, this
privilege rule effectively imposes restrictions on the privilege levels at
which interrupt and exception handling procedures can execute. Either of the
following strategies can be employed to ensure that the privilege rule is
never violated.
- Place the handler in a conforming segment. This strategy suits the
handlers for certain exceptions (divide error, for example). Such a
handler must use only the data available to it from the stack. If it
needed data from a data segment, the data segment would have to have
privilege level three, thereby making it unprotected.
- Place the handler procedure in a privilege level zero segment.
9.6.2 Interrupt Tasks
A task gate in the IDT points indirectly to a task, as
Figure 9-6
illustrates. The selector of the gate points to a TSS descriptor in the GDT.
When an interrupt or exception vectors to a task gate in the IDT, a task
switch results. Handling an interrupt with a separate task offers two
advantages:
- The entire context is saved automatically.
- The interrupt handler can be isolated from other tasks by giving it a
separate address space, either via its LDT or via its page directory.
The actions that the processor takes to perform a task switch are discussed
in
Chapter 7.
The interrupt task returns to the interrupted task by
executing an IRET instruction.
If the task switch is caused by an exception that has an error code, the
processor automatically pushes the error code onto the stack that
corresponds to the privilege level of the first instruction to be executed
in the interrupt task.
When interrupt tasks are used in an operating system for the 80386, there
are actually two schedulers: the software scheduler (part of the operating
system) and the hardware scheduler (part of the processor's interrupt
mechanism). The design of the software scheduler should account for the fact
that the hardware scheduler may dispatch an interrupt task whenever
interrupts are enabled.
With exceptions that relate to a specific segment, the processor pushes an
error code onto the stack of the exception handler (whether procedure or
task). The error code has the format shown in
Figure 9-7
. The format of the
error code resembles that of a selector; however, instead of an RPL field,
the error code contains two one-bit items:
- The processor sets the EXT bit if an event external to the program
caused the exception.
- The processor sets the I-bit (IDT-bit) if the index portion of the
error code refers to a gate descriptor in the IDT.
If the I-bit is not set, the TI bit indicates whether the error code refers
to the GDT (value 0) or to the LDT (value 1). The remaining 14 bits are the
upper 14 bits of the segment selector involved. In some cases the error code
on the stack is null, i.e., all bits in the low-order word are zero.
The following sections describe each of the possible exception conditions
in detail. Each description classifies the exception as a fault, trap, or
abort. This classification provides information needed by systems
programmers for restarting the procedure in which the exception occurred:
-
Faults
-
The CS and EIP values saved when a fault is reported point to the
instruction causing the fault.
-
Traps
-
The CS and EIP values stored when the trap is reported point to the
instruction dynamically after the instruction causing the trap. If
a trap is detected during an instruction that alters program flow,
the reported values of CS and EIP reflect the alteration of program
flow. For example, if a trap is detected in a
JMP instruction, the
CS and EIP values pushed onto the stack point to the target of the
JMP, not to the instruction after the
JMP.
-
Aborts
-
An abort is an exception that permits neither precise location of
the instruction causing the exception nor restart of the program
that caused the exception. Aborts are used to report severe errors,
such as hardware errors and inconsistent or illegal values in
system tables.
9.8.1 Interrupt 0 -- Divide Error
The divide-error fault occurs during a DIV or an
IDIV instruction when the
divisor is zero.
9.8.2 Interrupt 1 -- Debug Exceptions
The processor triggers this interrupt for any of a number of conditions;
whether the exception is a fault or a trap depends on the condition:
- Instruction address breakpoint fault.
- Data address breakpoint trap.
- General detect fault.
- Single-step trap.
- Task-switch breakpoint trap.
The processor does not push an error code for this exception. An exception
handler can examine the debug registers to determine which condition caused
the exception . Refer to
Chapter 12
for more detailed information about
debugging and the debug registers.
9.8.3 Interrupt 3 -- Breakpoint
The INT 3 instruction causes this trap. The
INT 3 instruction is one byte
long, which makes it easy to replace an opcode in an executable segment with
the breakpoint opcode. The operating system or a debugging subsystem can use
a data-segment alias for an executable segment to place an
INT 3 anywhere it
is convenient to arrest normal execution so that some sort of special
processing can be performed. Debuggers typically use breakpoints as a way of
displaying registers, variables, etc., at crucial points in a task.
The saved CS:EIP value points to the byte following the breakpoint. If a
debugger replaces a planted breakpoint with a valid opcode, it must subtract
one from the saved EIP value before returning . Refer also to
Chapter 12
for
more information on debugging.
9.8.4 Interrupt 4 -- Overflow
This trap occurs when the processor encounters an
INTO instruction and the
OF (overflow) flag is set. Since signed arithmetic and unsigned arithmetic
both use the same arithmetic instructions, the processor cannot determine
which is intended and therefore does not cause overflow exceptions
automatically. Instead it merely sets OF when the results, if interpreted as
signed numbers, would be out of range. When doing arithmetic on signed
operands, careful programmers and compilers either test OF directly or use
the INTO instruction.
9.8.5 Interrupt 5 -- Bounds Check
This fault occurs when the processor, while executing a
BOUND instruction,
finds that the operand exceeds the specified limits. A program can use the
BOUND
instruction to check a signed array index against signed limits
defined in a block of memory.
9.8.6 Interrupt 6 -- Invalid Opcode
This fault occurs when an invalid opcode is detected by the execution unit.
(The exception is not detected until an attempt is made to execute the
invalid opcode; i.e., prefetching an invalid opcode does not cause this
exception.) No error code is pushed on the stack. The exception can be
handled within the same task.
This exception also occurs when the type of operand is invalid for the
given opcode. Examples include an intersegment
JMP referencing a register
operand, or an
LES instruction with a register source operand.
9.8.7 Interrupt 7 -- Coprocessor Not Available
This exception occurs in either of two conditions:
- The processor encounters an ESC (escape) instruction, and the EM
(emulate) bit ofCR0 (control register zero) is set.
- The processor encounters either the
WAIT instruction or an ESC
instruction, and both the MP (monitor coprocessor) and TS (task
switched) bits of CR0 are set.
Refer to
Chapter 11
for information about the coprocessor interface .
9.8.8 Interrupt 8 -- Double Fault
Normally, when the processor detects an exception while trying to invoke
the handler for a prior exception, the two exceptions can be handled
serially. If, however, the processor cannot handle them serially, it signals
the double-fault exception instead. To determine when two faults are to be
signalled as a double fault, the 80386 divides the exceptions into three
classes: benign exceptions, contributory exceptions, and page faults.
Table 9-3 shows this classification.
Table 9-4
shows which combinations of exceptions cause a double fault and
which do not.
The processor always pushes an error code onto the stack of the
double-fault handler; however, the error code is always zero. The faulting
instruction may not be restarted. If any other exception occurs while
attempting to invoke the double-fault handler, the processor shuts down.
Table 9-3. Double-Fault Detection Classes
Class ID Description
1 Debug exceptions
2 NMI
3 Breakpoint
Benign 4 Overflow
Exceptions 5 Bounds check
6 Invalid opcode
7 Coprocessor not available
16 Coprocessor error
0 Divide error
9 Coprocessor Segment Overrun
Contributory 10 Invalid TSS
Exceptions 11 Segment not present
12 Stack exception
13 General protection
Page Faults 14 Page fault
Table 9-4. Double-Fault Definition
SECOND EXCEPTION
Benign Contributory Page
Exception Exception Fault
Benign OK OK OK
Exception
FIRST Contributory OK DOUBLE OK
EXCEPTION Exception
Page
Fault OK DOUBLE DOUBLE
9.8.9 Interrupt 9 -- Coprocessor Segment Overrun
This exception is raised in protected mode if the 80386 detects a page or
segment violation while transferring the middle portion of a coprocessor
operand to the NPX . This exception is avoidable. Refer to
Chapter 11
for
more information about the coprocessor interface.
9.8.10 Interrupt 10 -- Invalid TSS
Interrupt 10 occurs if during a task switch the new TSS is invalid. A TSS
is considered invalid in the cases shown in
Table 9-5. An error code is
pushed onto the stack to help identify the cause of the fault. The EXT bit
indicates whether the exception was caused by a condition outside the
control of the program; e.g., an external interrupt via a task gate
triggered a switch to an invalid TSS.
This fault can occur either in the context of the original task or in the
context of the new task. Until the processor has completely verified the
presence of the new TSS, the exception occurs in the context of the original
task. Once the existence of the new TSS is verified, the task switch is
considered complete; i.e., TR is updated and, if the switch is due to a
CALL
or interrupt, the backlink of the new TSS is set to the old TSS. Any
errors discovered by the processor after this point are handled in the
context of the new task.
To insure a proper TSS to process it, the handler for exception 10 must be
a task invoked via a task gate.
Table 9-5. Conditions That Invalidate the TSS
Error Code Condition
TSS id + EXT The limit in the TSS descriptor is less than 103
LTD id + EXT Invalid LDT selector or LDT not present
SS id + EXT Stack segment selector is outside table limit
SS id + EXT Stack segment is not a writable segment
SS id + EXT Stack segment DPL does not match new CPL
SS id + EXT Stack segment selector RPL < > CPL
CS id + EXT Code segment selector is outside table limit
CS id + EXT Code segment selector does not refer to code
segment
CS id + EXT DPL of non-conforming code segment < > new CPL
CS id + EXT DPL of conforming code segment > new CPL
DS/ES/FS/GS id + EXT DS, ES, FS, or GS segment selector is outside
table limits
DS/ES/FS/GS id + EXT DS, ES, FS, or GS is not readable segment
9.8.11 Interrupt 11 -- Segment Not Present
Exception 11 occurs when the processor detects that the present bit of a
descriptor is zero. The processor can trigger this fault in any of these
cases:
- While attempting to load the CS, DS, ES, FS, or GS registers; loading
the SS register, however, causes a stack fault.
- While attempting loading the LDT register with an
LLDT instruction;
loading the LDT register during a task switch operation, however,
causes the "invalid TSS" exception.
- While attempting to use a gate descriptor that is marked not-present.
This fault is restartable. If the exception handler makes the segment
present and returns, the interrupted program will resume execution.
If a not-present exception occurs during a task switch, not all the steps
of the task switch are complete. During a task switch, the processor first
loads all the segment registers, then checks their contents for validity. If
a not-present exception is discovered, the remaining segment registers have
not been checked and therefore may not be usable for referencing memory. The
not-present handler should not rely on being able to use the values found
in CS, SS, DS, ES, FS, and GS without causing another exception. The
exception handler should check all segment registers before trying to resume
the new task; otherwise, general protection faults may result later under
conditions that make diagnosis more difficult. There are three ways to
handle this case:
- Handle the not-present fault with a task. The task switch back to the
interrupted task will cause the processor to check the registers as it
loads them from the TSS.
-
PUSH and
POP all segment registers. Each
POP causes the processor to
check the new contents of the segment register.
- Scrutinize the contents of each segment-register image in the TSS,
simulating the test that the processor makes when it loads a segment
register.
This exception pushes an error code onto the stack. The EXT bit of the
error code is set if an event external to the program caused an interrupt
that subsequently referenced a not-present segment. The I-bit is set if the
error code refers to an IDT entry, e.g., an
INT instruction referencing a
not-present gate.
An operating system typically uses the "segment not present" exception to
implement virtual memory at the segment level. A not-present indication in a
gate descriptor, however, usually does not indicate that a segment is not
present (because gates do not necessarily correspond to segments).
Not-present gates may be used by an operating system to trigger exceptions
of special significance to the operating system.
9.8.12 Interrupt 12 -- Stack Exception
A stack fault occurs in either of two general conditions:
- As a result of a limit violation in any operation that refers to the
SS register. This includes stack-oriented instructions such as
POP,
PUSH,
ENTER, and
LEAVE, as well as other memory references that
implicitly use SS (for example, MOV AX, [BP+6]).
ENTER causes this
exception when the stack is too small for the indicated local-variable
space.
- When attempting to load the SS register with a descriptor that is
marked not-present but is otherwise valid. This can occur in a task
switch, an interlevel
CALL, an interlevel return, an
LSS instruction,
or a MOV or POP instruction to SS.
When the processor detects a stack exception, it pushes an error code onto
the stack of the exception handler. If the exception is due to a not-present
stack segment or to overflow of the new stack during an interlevel
CALL, the
error code contains a selector to the segment in question (the exception
handler can test the present bit in the descriptor to determine which
exception occurred); otherwise the error code is zero.
An instruction that causes this fault is restartable in all cases. The
return pointer pushed onto the exception handler's stack points to the
instruction that needs to be restarted. This instruction is usually the one
that caused the exception; however, in the case of a stack exception due to
loading of a not-present stack-segment descriptor during a task switch, the
indicated instruction is the first instruction of the new task.
When a stack fault occurs during a task switch, the segment registers may
not be usable for referencing memory. During a task switch, the selector
values are loaded before the descriptors are checked. If a stack fault is
discovered, the remaining segment registers have not been checked and
therefore may not be usable for referencing memory. The stack fault handler
should not rely on being able to use the values found in CS, SS, DS, ES,
FS, and GS without causing another exception. The exception handler should
check all segment registers before trying to resume the new task; otherwise,
general protection faults may result later under conditions that make
diagnosis more difficult.
9.8.13 Interrupt 13 -- General Protection Exception
All protection violations that do not cause another exception cause a
general protection exception. This includes (but is not limited to):
- Exceeding segment limit when using CS, DS, ES, FS, or GS
- Exceeding segment limit when referencing a descriptor table
- Transferring control to a segment that is not executable
- Writing into a read-only data segment or into a code segment
- Reading from an execute-only segment
- Loading the SS register with a read-only descriptor (unless the
selector comes from the TSS during a task switch, in which case a TSS
exception occurs
- Loading SS, DS, ES, FS, or GS with the descriptor of a system segment
- Loading DS, ES, FS, or GS with the descriptor of an executable
segment that is not also readable
- Loading SS with the descriptor of an executable segment
- Accessing memory via DS, ES, FS, or GS when the segment register
contains a null selector
- Switching to a busy task
- Violating privilege rules
- Loading CR0 with PG=1 and PE=0.
- Interrupt or exception via trap or interrupt gate from V86 mode to
privilege level other than zero.
- Exceeding the instruction length limit of 15 bytes (this can occur
only if redundant prefixes are placed before an instruction)
The general protection exception is a fault. In response to a general
protection exception, the processor pushes an error code onto the exception
handler's stack. If loading a descriptor causes the exception, the error
code contains a selector to the descriptor; otherwise, the error code is
null. The source of the selector in an error code may be any of the
following:
- An operand of the instruction.
- A selector from a gate that is the operand of the instruction.
- A selector from a TSS involved in a task switch.
9.8.14 Interrupt 14 -- Page Fault
This exception occurs when paging is enabled (PG=1) and the processor
detects one of the following conditions while translating a linear address
to a physical address:
- The page-directory or page-table entry needed for the address
translation has zero in its present bit.
- The current procedure does not have sufficient privilege to access the
indicated page.
The processor makes available to the page fault handler two items of
information that aid in diagnosing the exception and recovering from it:
- An error code on the stack. The error code for a page fault has a
format different from that for other exceptions (see
Figure 9-8
). The
error code tells the exception handler three things:
- Whether the exception was due to a not present page or to an access
rights violation.
- Whether the processor was executing at user or supervisor level at
the time of the exception.
- Whether the memory access that caused the exception was a read or
write.
- CR2 (control register two). The processor stores in CR2 the linear
address used in the access that caused the exception (see
Figure 9-9).
The exception handler can use this address to locate the corresponding
page directory and page table entries. If another page fault can occur
during execution of the page fault handler, the handler should push CR2
onto the stack.
The processor may access any of four segments during a task switch:
- Writes the state of the original task in the TSS of that task.
- Reads the GDT to locate the TSS descriptor of the new task.
- Reads the TSS of the new task to check the types of segment
descriptors from the TSS.
- May read the LDT of the new task in order to verify the segment
registers stored in the new TSS.
A page fault can result from accessing any of these segments. In the latter
two cases the exception occurs in the context of the new task. The
instruction pointer refers to the next instruction of the new task, not to
the instruction that caused the task switch. If the design of the operating
system permits page faults to occur during task-switches, the page-fault
handler should be invoked via a task gate.
Special care should be taken to ensure that a page fault does not cause the
processor to use an invalid stack pointer (SS:ESP). Software written for
earlier processors in the 8086 family often uses a pair of instructions to
change to a new stack; for example:
MOV SS, AX
MOV SP, StackTop
With the 80386, because the second instruction accesses memory, it is
possible to get a page fault after SS has been changed but before SP has
received the corresponding change. At this point, the two parts of the stack
pointer SS:SP (or, for 32-bit programs, SS:ESP) are inconsistent.
The processor does not use the inconsistent stack pointer if the handling
of the page fault causes a stack switch to a well defined stack (i.e., the
handler is a task or a more privileged procedure). However, if the page
fault handler is invoked by a trap or interrupt gate and the page fault
occurs at the same privilege level as the page fault handler, the processor
will attempt to use the stack indicated by the current (invalid) stack
pointer.
In systems that implement paging and that handle page faults within the
faulting task (with trap or interrupt gates), software that executes at the
same privilege level as the page fault handler should initialize a new stack
by using the new
LSS instruction rather than an instruction pair shown
above. When the page fault handler executes at privilege level zero (the
normal case), the scope of the problem is limited to privilege-level zero
code, typically the kernel of the operating system.
9.8.15 Interrupt 16 -- Coprocessor Error
The 80386 reports this exception when it detects a signal from the 80287 or
80387 on the 80386's ERROR# input pin. The 80386 tests this pin only at the
beginning of certain ESC instructions and when it encounters a
WAIT
instruction while the EM bit of the MSW is zero (no emulation). Refer to
Chapter 11
for more information on the coprocessor interface.
Table 9-6 summarizes the exceptions recognized by the 386.
Table 9-6. Exception Summary
Description Interrupt Return Address Exception Function That Can Generate
Number Points to Type the Exception
Faulting
Instruction
Divide error 0 YES FAULT DIV, IDIV
Debug exceptions 1
Some debug exceptions are traps and some are faults. The exception
handler can determine which has occurred by examining DR6. (Refer to Chapter 12.)
Some debug exceptions are traps and some are faults. The exception
handler can determine which has occurred by examining DR6. (Refer to Chapter 12.) Any instruction
Breakpoint 3 NO TRAP One-byte INT 3
Overflow 4 NO TRAP INTO
Bounds check 5 YES FAULT BOUND
Invalid opcode 6 YES FAULT Any illegal instruction
Coprocessor not available 7 YES FAULT ESC, WAIT
Double fault 8 YES ABORT Any instruction that can
generate an exception
Coprocessor Segment
Overrun 9 NO ABORT Any operand of an ESC
instruction that wraps around
the end of a segment.
Invalid TSS 10 YES FAULT
An invalid-TSS fault is not restartable if it occurs during the
processing of an external interrupt. JMP, CALL, IRET, any interrupt
Segment not present 11 YES FAULT Any segment-register modifier
Stack exception 12 YES FAULT Any memory reference thru SS
General Protection 13 YES FAULT/ABORT
All GP faults are restartable. If the fault occurs while attempting to
vector to the handler for an external interrupt, the interrupted program is
restartable, but the interrupt may be lost. Any memory reference or code
fetch
Page fault 14 YES FAULT Any memory reference or code
fetch
Coprocessor error 16 YES FAULT
Coprocessor errors are reported as a fault on the first ESC or WAIT
instruction executed after the ESC instruction that caused the error. ESC, WAIT
Two-byte SW Interrupt 0-255 NO TRAP INT n
Table 9-7 summarizes the error information that is available with each
exception.
Table 9-7. Error-Code Summary
Description Interrupt Error Code
Number
Divide error 0 No
Debug exceptions 1 No
Breakpoint 3 No
Overflow 4 No
Bounds check 5 No
Invalid opcode 6 No
Coprocessor not available 7 No
System error 8 Yes (always 0)
Coprocessor Segment Overrun 9 No
Invalid TSS 10 Yes
Segment not present 11 Yes
Stack exception 12 Yes
General protection fault 13 Yes
Page fault 14 Yes
Coprocessor error 16 No
Two-byte SW interrupt 0-255 No
The contents of EAX depend upon the results of the power-up self test. The
self-test may be requested externally by assertion of BUSY# at the end of
RESET. The EAX register holds zero if the 80386 passed the test. A nonzero
value in EAX after self-test indicates that the particular 80386 unit is
faulty. If the self-test is not requested, the contents of EAX after RESET
is undefined.
DX holds a component identifier and revision number after RESET as
Figure 10-1
illustrates. DH contains 3, which indicates an 80386 component. DL
contains a unique identifier of the revision level.
Control register zero (CR0) contains the values shown in
Figure 10-2
. The
ET bit of CR0 is set if an 80387 is present in the configuration (according
to the state of the ERROR# pin after RESET). If ET is reset, the
configuration either contains an 80287 or does not contain a coprocessor. A
software test is required to distinguish between these latter two
possibilities.
The remaining registers and flags are set as follows:
EFLAGS =00000002H
IP =0000FFF0H
CS selector =000H
DS selector =0000H
ES selector =0000H
SS selector =0000H
FS selector =0000H
GS selector =0000H
IDTR:
base =0
limit =03FFH
All registers not mentioned above are undefined.
These settings imply that the processor begins in real-address mode with
interrupts disabled.
In real-address mode a few structures must be initialized before a program
can take advantage of all the features available in this mode.
10.2.1 Stack
No instructions that use the stack can be used until the stack-segment
register (SS) has been loaded. SS must point to an area in RAM.
10.2.2 Interrupt Table
The initial state of the 80386 leaves interrupts disabled; however, the
processor will still attempt to access the interrupt table if an exception
or nonmaskable interrupt (NMI) occurs. Initialization software should take
one of the following actions:
- Change the limit value in the IDTR to zero. This will cause a shutdown
if an exception or nonmaskable interrupt occurs. (Refer to the 80386
Hardware Reference Manual to see how shutdown is signalled externally.)
- Put pointers to valid interrupt handlers in all positions of the
interrupt table that might be used by exceptions or interrupts.
- Change the IDTR to point to a valid interrupt table.
10.2.3 First Instructions
After RESET, address lines A{31-20} are automatically asserted for
instruction fetches. This fact, together with the initial values of CS:IP,
causes instruction execution to begin at physical address FFFFFFF0H. Near
(intrasegment) forms of control transfer instructions may be used to pass
control to other addresses in the upper 64K bytes of the address space. The
first far (intersegment)
JMP or
CALL instruction causes A{31-20} to drop
low, and the 80386 continues executing instructions in the lower one
megabyte of physical memory. This automatic assertion of address lines
A{31-20} allows systems designers to use a ROM at the high end of
the address space to initialize the system.
Setting the PE bit of the MSW in CR0 causes the 80386 to begin executing in
protected mode. The current privilege level (CPL) starts at zero. The
segment registers continue to point to the same linear addresses as in real
address mode (in real address mode, linear addresses are the same physical
addresses).
Immediately after setting the PE flag, the initialization code must flush
the processor's instruction prefetch queue by executing a JMP instruction.
The 80386 fetches and decodes instructions and addresses before they are
used; however, after a change into protected mode, the prefetched
instruction information (which pertains to real-address mode) is no longer
valid. A JMP forces the processor to discard the invalid information.
Most of the initialization needed for protected mode can be done either
before or after switching to protected mode. If done in protected mode,
however, the initialization procedures must not use protected-mode features
that are not yet initialized.
10.4.1 Interrupt Descriptor Table
The IDTR may be loaded in either real-address or protected mode. However,
the format of the interrupt table for protected mode is different than that
for real-address mode. It is not possible to change to protected mode and
change interrupt table formats at the same time; therefore, it is inevitable
that, if IDTR selects an interrupt table, it will have the wrong format at
some time. An interrupt or exception that occurs at this time will have
unpredictable results. To avoid this unpredictability, interrupts should
remain disabled until interrupt handlers are in place and a valid IDT has
been created in protected mode.
10.4.2 Stack
The SS register may be loaded in either real-address mode or protected
mode. If loaded in real-address mode, SS continues to point to the same
linear base-address after the switch to protected mode.
10.4.3 Global Descriptor Table
Before any segment register is changed in protected mode, the GDT register
must point to a valid GDT. Initialization of the GDT and GDTR may be done in
real-address mode. The GDT (as well as LDTs) should reside in RAM, because
the processor modifies the accessed bit of descriptors.
10.4.4 Page Tables
Page tables and the PDBR in CR3 can be initialized in either real-address
mode or in protected mode; however, the paging enabled (PG) bit of CR0
cannot be set until the processor is in protected mode. PG may be set
simultaneously with PE, or later. When PG is set, the PDBR in CR3 should
already be initialized with a physical address that points to a valid page
directory. The initialization procedure should adopt one of the following
strategies to ensure consistent addressing before and after paging is
enabled:
10.4.5 First Task
The initialization procedure can run awhile in protected mode without
initializing the task register; however, before the first task switch, the
following conditions must prevail:
- There must be a valid task state segment (TSS) for the new task. The
stack pointers in the TSS for privilege levels numerically less than or
equal to the initial CPL must point to valid stack segments.
- The task register must point to an area in which to save the current
task state. After the first task switch, the information dumped in this
area is not needed, and the area can be used for other purposes.
$TITLE ('Initial Task')
NAME INIT
init_stack SEGMENT RW
DW 20 DUP(?)
tos LABEL WORD
init_stack ENDS
init_data SEGMENT RW PUBLIC
DW 20 DUP(?)
init_data ENDS
init_code SEGMENT ER PUBLIC
ASSUME DS:init_data
nop
nop
nop
init_start:
; set up stack
mov ax, init_stack
mov ss, ax
mov esp, offset tos
mov a1,1
blink:
xor a1,1
out 0e4h,a1
mov cx,3FFFh
here:
dec cx
jnz here
jmp SHORT blink
hlt
init_code ends
END init_start, SS:init_stack, DS:init_data
$TITLE('Protected Mode Transition -- 386 initialization')
NAME RESET
;*****************************************************************
; Upon reset the 386 starts executing at address 0FFFFFFF0H. The
; upper 12 address bits remain high until a FAR call or jump is
; executed.
;
; Assume the following:
;
;
; - a short jump at address 0FFFFFFF0H (placed there by the
; system builder) causes execution to begin at START in segment
; RESET_CODE.
;
;
; - segment RESET_CODE is based at physical address 0FFFF0000H,
; i.e. at the start of the last 64K in the 4G address space.
; Note that this is the base of the CS register at reset. If
; you locate ROMcode above this address, you will need to
; figure out an adjustment factor to address things within this
; segment.
;
;*****************************************************************
$EJECT ;
; Define addresses to locate GDT and IDT in RAM.
; These addresses are also used in the BLD386 file that defines
; the GDT and IDT. If you change these addresses, make sure you
; change the base addresses specified in the build file.
GDTbase EQU 00001000H ; physical address for GDT base
IDTbase EQU 00000400H ; physical address for IDT base
PUBLIC GDT_EPROM
PUBLIC IDT_EPROM
PUBLIC START
DUMMY segment rw ; ONLY for ASM386 main module stack init
DW 0
DUMMY ends
;*****************************************************************
;
; Note: RESET CODE must be USEl6 because the 386 initally executes
; in real mode.
;
RESET_CODE segment er PUBLIC USE16
ASSUME DS:nothing, ES:nothing
;
; 386 Descriptor template
DESC STRUC
lim_0_15 DW 0 ; limit bits (0..15)
bas_0_15 DW 0 ; base bits (0..15)
bas_16_23 DB 0 ; base bits (16..23)
access DB 0 ; access byte
gran DB 0 ; granularity byte
bas_24_31 DB 0 ; base bits (24..31)
DESC ENDS
; The following is the layout of the real GDT created by BLD386.
; It is located in EPROM and will be copied to RAM.
;
; GDT[O] ... NULL
; GDT[1] ... Alias for RAM GDT
; GDT[2] ... Alias for RAM IDT
; GDT[2] ... initial task TSS
; GDT[3] ... initial task TSS alias
; GDT[4] ... initial task LDT
; GDT[5] ... initial task LDT alias
;
; define entries in GDT and IDT.
GDT_ENTRIES EQU 8
IDT_ENTRIES EQU 32
; define some constants to index into the real GDT
GDT_ALIAS EQU 1*SIZE DESC
IDT_ALIAS EQU 2*SIZE DESC
INIT_TSS EQU 3*SIZE DESC
INIT_TSS_A EQU 4*SIZE DESC
INIT_LDT EQU 5*SIZE DESC
INIT_LDT_A EQU 6*SIZE DESC
;
; location of alias in INIT_LDT
INIT_LDT_ALIAS EQU 1*SIZE DESC
;
; access rights byte for DATA and TSS descriptors
DS_ACCESS EQU 010010010B
TSS_ACCESS EQU 010001001B
;
; This temporary GDT will be used to set up the real GDT in RAM.
Temp_GDT LABEL BYTE ; tag for begin of scratch GDT
NULL_DES DESC <> ; NULL descriptor
; 32-Gigabyte data segment based at 0
FLAT_DES DESC <0FFFFH,0,0,92h,0CFh,0>
GDT_eprom DP ? ; Builder places GDT address and limit
; in this 6 byte area.
IDT_eprom DP ? ; Builder places IDT address and limit
; in this 6 byte area.
;
; Prepare operand for loadings GDTR and LDTR.
TGDT_pword LABEL PWORD ; for temp GDT
DW end_Temp_GDT_Temp_GDT -1
DD 0
GDT_pword LABEL PWORD ; for GDT in RAM
DW GDT_ENTRIES * SIZE DESC -1
DD GDTbase
IDT_pword LABEL PWORD ; for IDT in RAM
DW IDT_ENTRIES * SIZE DESC -1
DD IDTbase
end_Temp_GDT LABEL BYTE
;
; Define equates for addressing convenience.
GDT_DES_FLAT EQU DS:GDT_ALIAS +GDTbase
IDT_DES_FLAT EQU DS:IDT_ALIAS +GDTbase
INIT_TSS_A_OFFSET EQU DS:INIT_TSS_A
INIT_TSS_OFFSET EQU DS:INIT_TSS
INIT_LDT_A_OFFSET EQU DS:INIT_LDT_A
INIT_LDT_OFFSET EQU DS:INIT_LDT
; define pointer for first task switch
ENTRY POINTER LABEL DWORD
DW 0, INIT_TSS
;******************************************************************
;
; Jump from reset vector to here.
START:
CLI ;disable interrupts
CLD ;clear direction flag
LIDT NULL_des ;force shutdown on errors
;
; move scratch GDT to RAM at physical 0
XOR DI,DI
MOV ES,DI ;point ES:DI to physical location 0
MOV SI,OFFSET Temp_GDT
MOV CX,end_Temp_GDT-Temp_GDT ;set byte count
INC CX
;
; move table
REP MOVS BYTE PTR ES:[DI],BYTE PTR CS:[SI]
LGDT tGDT_pword ;load GDTR for Temp. GDT
;(located at 0)
; switch to protected mode
MOV EAX,CR0 ;get current CRO
MOV EAX,1 ;set PE bit
MOV CRO,EAX ;begin protected mode
;
; clear prefetch queue
JMP SHORT flush
flush:
; set DS,ES,SS to address flat linear space (0 ... 4GB)
MOV BX,FLAT_DES-Temp_GDT
MOV US,BX
MOV ES,BX
MOV SS,BX
;
; initialize stack pointer to some (arbitrary) RAM location
MOV ESP, OFFSET end_Temp_GDT
;
; copy eprom GDT to RAM
MOV ESI,DWORD PTR GDT_eprom +2 ; get base of eprom GDT
; (put here by builder).
MOV EDI,GDTbase ; point ES:EDI to GDT base in RAM.
MOV CX,WORD PTR gdt_eprom +0 ; limit of eprom GDT
INC CX
SHR CX,1 ; easier to move words
CLD
REP MOVS WORD PTR ES:[EDI],WORD PTR DS:[ESI]
;
; copy eprom IDT to RAM
;
MOV ESI,DWORD PTR IDT_eprom +2 ; get base of eprom IDT
; (put here by builder)
MOV EDI,IDTbase ; point ES:EDI to IDT base in RAM.
MOV CX,WORD PTR idt_eprom +0 ; limit of eprom IDT
INC CX
SHR CX,1
CLD
REP MOVS WORD PTR ES:[EDI],WORD PTR DS:[ESI]
; switch to RAM GDT and IDT
;
LIDT IDT_pword
LGDT GDT_pword
;
MOV BX,GDT_ALIAS ; point DS to GDT alias
MOV DS,BX
;
; copy eprom TSS to RAM
;
MOV BX,INIT_TSS_A ; INIT TSS A descriptor base
; has RAM location of INIT TSS.
MOV ES,BX ; ES points to TSS in RAM
MOV BX,INIT_TSS ; get inital task selector
LAR DX,BX ; save access byte
MOV [BX].access,DS_ACCESS ; set access as data segment
MOV FS,BX ; FS points to eprom TSS
XOR si,si ; FS:si points to eprom TSS
XOR di,di ; ES:di points to RAM TSS
MOV CX,[BX].lim_0_15 ; get count to move
INC CX
;
; move INIT_TSS to RAM.
REP MOVS BYTE PTR ES:[di],BYTE PTR FS:[si]
MOV [BX].access,DH ; restore access byte
;
; change base of INIT TSS descriptor to point to RAM.
MOV AX,INIT_TSS_A_OFFSET.bas_0_15
MOV INIT_TSS_OFFSET.bas_0_15,AX
MOV AL,INIT_TSS_A_OFFSET.bas_16_23
MOV INIT_TSS_OFFSET.bas_16_23,AL
MOV AL,INIT_TSS_A_OFFSET.bas_24_31
MOV INIT_TSS_OFFSET.bas_24_31,AL
;
; change INIT TSS A to form a save area for TSS on first task
; switch. Use RAM at location 0.
MOV BX,INIT_TSS_A
MOV WORD PTR [BX].bas_0_15,0
MOV [BX].bas_16_23,0
MOV [BX].bas_24_31,0
MOV [BX].access,TSS_ACCESS
MOV [BX].gran,O
LTR BX ; defines save area for TSS
;
; copy eprom LDT to RAM
MOV BX,INIT_LDT_A ; INIT_LDT_A descriptor has
; base address in RAM for INIT_LDT.
MOV ES,BX ; ES points LDT location in RAM.
MOV AH,[BX].bas_24_31
MOV AL,[BX].bas_16_23
SHL EAX,16
MOV AX,[BX].bas_0_15 ; save INIT_LDT base (ram) in EAX
MOV BX,INIT_LDT ; get inital LDT selector
LAR DX,BX ; save access rights
MOV [BX].access,DS_ACCESS ; set access as data segment
MOV FS,BX ; FS points to eprom LDT
XOR si,si ; FS:SI points to eprom LDT
XOR di,di ; ES:DI points to RAM LDT
MOV CX,[BX].lim_0_15 ; get count to move
INC CX
;
; move initial LDT to RAM
REP MOVS BYTE PTR ES:[di],BYTE PTR FS:[si]
MOV [BX].access,DH ; restore access rights in
; INIT_LDT descriptor
;
; change base of alias (of INIT_LDT) to point to location in RAM.
MOV ES:[INIT_LDT_ALIAS].bas_0_15,AX
SHR EAX,16
MOV ES:[INIT_LDT_ALIAS].bas_16_23,AL
MOV ES:[INIT_LDT_ALIAS].bas_24_31,AH
;
; now set the base value in INIT_LDT descriptor
MOV AX,INIT_LDT_A_OFFSET.bas_0_15
MOV INIT_LDT_OFFSET.bas_0_15,AX
MOV AL,INIT_LDT_A_OFFSET.bas_16_23
MOV INIT_LDT_OFFSET.bas_16_23,AL
MOV AL,INIT_LDT_A_OFFSET.bas_24_31
MOV INIT_LDT_OFFSET.bas_24_31,AL
;
; Now GDT, IDT, initial TSS and initial LDT are all set up.
;
; Start the first task!
'
JMP ENTRY_POINTER
RESET_CODE ends
END START, SS:DUMMY,DS:DUMMY
The 80386 provides a mechanism for testing the Translation Lookaside Buffer
(TLB), the cache used for translating linear addresses to physical
addresses. Although failure of the TLB hardware is extremely unlikely, users
may wish to include TLB confidence tests among other power-up confidence
tests for the 80386.
Note
This TLB testing mechanism is unique to the 80386 and may not be
continued in the same way in future processors. Sortware that uses
this mechanism may be incompatible with future processors.
When testing the TLB it is recommended that paging be turned off (PG=0 in
CR0) to avoid interference with the test data being written to the TLB.
10.6.1 Structure of the TLB
The TLB is a four-way set-associative memory.
Figure 10-3
illustrates the
structure of the TLB. There are four sets of eight entries each. Each entry
consists of a tag and data. Tags are 24-bits wide. They contain the
high-order 20 bits of the linear address, the valid bit, and three attribute
bits. The data portion of each entry contains the high-order 20 bits of the
physical address.
10.6.2 Test Registers
Two test registers, shown in
Figure 10-4, are provided for the purpose of
testing. TR6 is the test command register, and TR7 is the test data
register. These registers are accessed by variants of the
MOV
instruction. A test register may be either the source operand or destination
operand. The
MOV instructions are defined in both
real-address mode and
protected mode. The test registers are privileged resources; in protected
mode, the
MOV instructions that access them can only
be executed at
privilege level 0. An attempt to read or write the test registers when
executing at any other privilege level causes a general
protection exception.
The test command register (TR6) contains a command and an address tag to
use in performing the command:
-
C
-
This is the command bit. There are two TLB testing commands:
write entries into the TLB, and perform TLB lookups. To cause an
immediate write into the TLB entry, move a doubleword into TR6
that contains a 0 in this bit. To cause an immediate TLB lookup,
move a doubleword into TR6 that contains a 1 in this bit.
-
Linear Address
-
On a TLB write, a TLB entry is allocated to this linear address;
the rest of that TLB entry is set per the value of TR7 and the
value just written into TR6. On a TLB lookup, the TLB is
interrogated per this value; if one and only one TLB entry
matches, the rest of the fields of TR6 and TR7 are set from the
matching TLB entry.
-
V
-
The valid bit for this TLB entry. The TLB uses the valid bit to
identify entries that contain valid data. Entries of the TLB
that have not been assigned values have zero in the valid bit.
All valid bits can be cleared by writing to CR3.
-
D, D#
-
The dirty bit (and its complement) for/from the TLB entry.
-
U, U#
-
The U/S bit (and its complement) for/from the TLB entry.
-
W, W#
-
The R/W bit (and its complement) for/from the TLB entry.
The meaning of these pairs of bits is given by Table 10-1,
where X represents D, U, or W.
The test data register (TR7) holds data read from or data to be written to
the TLB.
-
Physical Address
-
This is the data field of the TLB. On a write to the TLB, the
TLB entry allocated to the linear address in TR6 is set to this
value. On a TLB lookup, if HT is set, the data field (physical
address) from the TLB is read out to this field. If HT is not
set, this field is undefined.
-
HT
-
For a TLB lookup, the HT bit indicates whether the lookup was a
hit (HT := 1) or a miss (HT := 0). For a TLB write, HT must be set
to 1.
-
REP
-
For a TLB write, selects which of four associative blocks of the
TLB is to be written. For a TLB read, if HT is set,
REP reports
in which of the four associative blocks the tag was found; if HT
is not set, REP is undefined.
Table 10-1. Meaning of D, U, and W Bit Pairs
X X# Effect during Value of bit X
TLB Lookup after TLB Write
0 0 (undefined) (undefined)
0 1 Match if X=0 Bit X becomes 0
1 0 Match if X=1 Bit X becomes 1
1 1 (undefined) (undefined)
10.6.3 Test Operations
To write a TLB entry:
- Move a doubleword to TR7 that contains the desired physical address,
HT, and REP values. HT must contain 1.
REP must point to the
associative block in which to place the entry.
- Move a doubleword to TR6 that contains the appropriate linear
address, and values for V, D, U, and W. Be sure C=0 for "write"
command.
Be careful not to write duplicate tags; the results of doing so are
undefined.
To look up (read) a TLB entry:
- Move a doubleword to TR6 that contains the appropriate linear address
and attributes. Be sure C=1 for "lookup" command.
- Store TR7. If the HT bit in TR7 indicates a hit, then the other
values reveal the TLB contents. If HT indicates a miss, then the other
values in TR7 are indeterminate.
For the purposes of testing, the V bit functions as another bit of
addresss. The V bit for a lookup request should usually be set, so that
uninitialized tags do not match. Lookups with V=0 are unpredictable if any
tags are uninitialized.
The 80386 provides a one Mbyte + 64 Kbyte memory space for an 8086 program.
Segment relocation is performed as in the 8086: the 16-bit value in a
segment selector is shifted left by four bits to form the base address of a
segment. The effective address is extended with four high order zeros and
added to the base to form a linear address as
Figure 14-1
illustrates. (The
linear address is equivalent to the physical address, because paging is not
used in real-address mode.) Unlike the 8086, the resulting linear address
may have up to 21 significant bits. There is a possibility of a carry when
the base address is added to the effective address. On the 8086, the carried
bit is truncated, whereas on the 80386 the carried bit is stored in bit
position 20 of the linear address.
Unlike the 8086 and 80286, 32-bit effective addresses can be generated (via
the address-size prefix); however, the value of a 32-bit address may not
exceed 65535 without causing an exception. For full compatibility with 80286
real-address mode, pseudo-protection faults (interrupt 12 or 13 with no
error code) occur if an effective address is generated outside the range 0
through 65535.
The register set available in real-address mode includes all the registers
defined for the 8086 plus the new registers introduced by the 80386: FS, GS,
debug registers, control registers, and test registers. New instructions
that explicitly operate on the segment registers FS and GS are available,
and the new segment-override prefixes can be used to cause instructions to
utilize FS and GS for address calculations. Instructions can utilize 32-bit
operands through the use of the operand size prefix.
The instruction codes that cause undefined opcode traps (interrupt 6)
include instructions of the protected mode that manipulate or interrogate
80386 selectors and descriptors; namely,
VERR,
VERW,
LAR,
LSL,
LTR,
STR,
LLDT, and
SLDT.
Programs executing in real-address mode are able to take
advantage of the new applications-oriented instructions added to the
architecture by the introduction of the 80186/80188, 80286 and 80386:
- New instructions introduced by 80186/80188 and 80286.
- PUSH immediate data
- Push all and pop all (PUSHA and
POPA)
- Multiply immediate data
- Shift and rotate by immediate count
- String I/O
- ENTER and
LEAVE
- BOUND
- New instructions introduced by 80386.
- LSS,
LFS,
LGS instructions
- Long-displacement conditional jumps
- Single-bit instructions
- Bit scan
- Double-shift instructions
- Byte set on condition
- Move with sign/zero extension
- Generalized multiply
- MOV to and from control registers
- MOV to and from test registers
- MOV to and from debug registers
Interrupts and exceptions in 80386 real-address mode work as much as they
do on an 8086. Interrupts and exceptions vector to interrupt procedures via
an interrupt table. The processor multiplies the interrupt or exception
identifier by four to obtain an index into the interrupt table. The entries
of the interrupt table are far pointers to the entry points of interrupt or
exception handler procedures. When an interrupt occurs, the processor
pushes the current values of CS:IP onto the stack, disables interrupts,
clears TF (the single-step flag), then transfers control to the location
specified in the interrupt table. An IRET instruction at the end of the
handler procedure reverses these steps before returning control to the
interrupted procedure.
The primary difference in the interrupt handling of the 80386 compared to
the 8086 is that the location and size of the interrupt table depend on the
contents of the IDTR (IDT register). Ordinarily, this fact is not apparent
to programmers, because, after RESET, the IDTR contains a base address of 0
and a limit of 3FFH, which is compatible with the 8086. However, the
LIDT
instruction can be used in real-address mode to change the base and limit
values in the IDTR . Refer to
Chapter 9
for details on the IDTR , and the
LIDT and SIDT instructions.
If an interrupt occurs and the corresponding
entry of the interrupt table is beyond the limit stored in the IDTR, the
processor raises exception 8.
Real-address mode is in effect after a signal on the RESET pin. Even if the
system is going to be used in protected mode, the start-up program will
execute in real-address mode temporarily while initializing for protected
mode.
14.4.1 Switching to Protected Mode
The only way to leave real-address mode is to switch to protected mode. The
processor enters protected mode when a
MOV to CR0 instruction sets the PE
(protection enable) bit in CR0. (For compatibility with the 80286, the LMSW
instruction may also be used to set the PE bit.)
Refer to
Chapter 10
"Initialization" for other aspects of switching to
protected mode.
The processor reenters real-address mode if software clears the PE bit in
CR0 with a
MOV to CR0 instruction.
A procedure that attempts to do this,
however, should proceed as follows:
- If paging is enabled, perform the following sequence:
- Transfer control to linear addresses that have an identity mapping;
i.e., linear addresses equal physical addresses.
- Clear the PG bit in CR0.
- Move zeros to CR3 to clear out the paging cache.
- Transfer control to a segment that has a limit of 64K (FFFFH). This
loads the CS register with the limit it needs to have in real mode.
- Load segment registers SS, DS, ES, FS, and GS with a selector that
points to a descriptor containing the following values, which are
appropriate to real mode:
- Limit = 64K (FFFFH)
- Byte granular (G = 0)
- Expand up (E = 0)
- Writable (W = 1)
- Present (P = 1)
- Base = any value
- Disable interrupts. A
CLI instruction disables INTR interrupts. NMIs
can be disabled with external circuitry.
- Clear the PE bit.
- Jump to the real mode code to be executed using a far
JMP. This
action flushes the instruction queue and puts appropriate values in
the access rights of the CS register.
- Use the
LIDT
instruction to load the base and limit of the real-mode
interrupt vector table.
- Enable interrupts.
- Load the segment registers as needed by the real-mode code.
The 80386 reports some exceptions differently when executing in
real-address mode than when executing in protected mode. Table 14-1 details
the real-address-mode exceptions.
In general, the 80386 in real-address mode will correctly execute ROM-based
software designed for the 8086, 8088, 80186, and 80188. Following is a list
of the minor differences between 8086 execution on the 80386 and on an 8086.
- Instruction clock counts.
The 80386 takes fewer clocks for most instructions than the 8086/8088.
The areas most likely to be affected are:
- Delays required by I/O devices between I/O operations.
- Assumed delays with 8086/8088 operating in parallel with an 8087.
- Divide Exceptions Point to the
DIV instruction.
Divide exceptions on the 80386 always leave the saved CS:IP value
pointing to the instruction that failed. On the 8086/8088, the CS:IP
value points to the next instruction.
- Undefined 8086/8088 opcodes.
Opcodes that were not defined for the 8086/8088 will cause exception
6 or will execute one of the new instructions defined for the 80386.
- Value written by
PUSH SP.
The 80386 pushes a different value on the stack for
PUSH SP than the
8086/8088. The 80386 pushes the value of SP before SP is incremented
as part of the push operation; the 8086/8088 pushes the value of SP
after it is incremented. If the value pushed is important, replace
PUSH
SP instructions with the following three instructions:
PUSH BP
MOV BP, SP
XCHG BP, [BP]
This code functions as the 8086/8088
PUSH SP instruction on the 80386.
- Shift or rotate by more than 31 bits.
The 80386 masks all shift and rotate counts to the low-order five
bits. This MOD 32 operation limits the count to a maximum of 31 bits,
thereby limiting the time that interrupt response is delayed while
the instruction is executing.
- Redundant prefixes.
The 80386 sets a limit of 15 bytes on instruction length. The only
way to violate this limit is by putting redundant prefixes before an
instruction. Exception 13 occurs if the limit on instruction length
is violated. The 8086/8088 has no instruction length limit.
- Operand crossing offset 0 or 65,535.
On the 8086, an attempt to access a memory operand that crosses
offset 65,535 (e.g.,
MOV a word to offset 65,535) or offset 0 (e.g.,
PUSH
a word when SP = 1) causes the offset to wrap around modulo
65,536. The 80386 raises an exception in these -- 13 if
the segment is a data segment (i.e., if CS, DS, ES, FS, or GS is being
used to address the segment), exception 12 if the segment is a stack
segment (i.e., if SS is being used).
- Sequential execution across offset 65,535.
On the 8086, if sequential execution of instructions proceeds past
offset 65,535, the processor fetches the next instruction byte from
offset 0 of the same segment. On the 80386, the processor raises
exception 13 in such a case.
- LOCK is restricted to certain instructions.
The LOCK prefix and its corresponding output signal should only be
used to prevent other bus masters from interrupting a data movement
operation. The 80386 always asserts the LOCK signal during an XCHG
instruction with memory (even if the
LOCK prefix is not used).
LOCK
may only be used with the following 80386 instructions when they
update memory:
BTS,
BTR,
BTC,
XCHG,
ADD,
ADC,
SUB,
SBB,
INC,
DEC,
AND,
OR,
XOR,
NOT, and
NEG. An undefined-opcode exception
(interrupt 6) results from using
LOCK before any other instruction.
- Single-stepping external interrupt handlers.
The priority of the 80386 single-step exception is different from that
of the 8086/8088. The change prevents an external interrupt handler
from being single-stepped if the interrupt occurs while a program is
being single-stepped. The 80386 single-step exception has higher
priority that any external interrupt. The 80386 will still single-step
through an interrupt handler invoked by the
INT instructions or by an
exception.
- IDIV exceptions for quotients of 80H or 8000H.
The 80386 can generate the largest negative number as a quotient for
the IDIV instruction.
The 8086/8088 causes exception zero instead.
- Flags in stack.
The setting of the flags stored by
PUSHF, by interrupts, and by
exceptions is different from that stored by the 8086 in bit positions
12 through 15. On the 8086 these bits are stored as ones, but in
80386 real-address mode bit 15 is always zero, and bits 14 through 12
reflect the last value loaded into them.
- NMI interrupting NMI handlers.
After an NMI is recognized on the 80386, the NMI interrupt is masked
until an IRET instruction is executed.
- Coprocessor errors vector to interrupt 16.
Any 80386 system with a coprocessor must use interrupt vector 16 for
the coprocessor error exception. If an 8086/8088 system uses another
vector for the 8087 interrupt, both vectors should point to the
coprocessor-error exception handler.
- Numeric exception handlers should allow prefixes.
On the 80386, the value of CS:IP saved for coprocessor exceptions
points at any prefixes before an ESC instruction. On 8086/8088
systems, the saved CS:IP points to the ESC instruction.
- Coprocessor does not use interrupt controller.
The coprocessor error signal to the 80386 does not pass through an
interrupt controller (an 8087
INT signal does). Some instructions in
a coprocessor error handler may need to be deleted if they deal with
the interrupt controller.
- Six new interrupt vectors.
The 80386 adds six exceptions that arise only if the 8086 program has
a hidden bug. It is recommended that exception handlers be added that
treat these exceptions as invalid operations. This additional
software does not significantly affect the existing 8086 software
because the interrupts do not normally occur. These interrupt
identifiers should not already have been used by the 8086 software,
because they are in the range reserved by Intel. Table 14-2 describes
the new 80386 exceptions.
- One megabyte wraparound.
The 80386 does not wrap addresses at 1 megabyte in real-address mode.
On members of the 8086 family, it possible to specify addresses
greater than one megabyte. For example, with a selector value 0FFFFH
and an offset of 0FFFFH, the effective address would be 10FFEFH (1
Mbyte + 65519). The 8086, which can form adresses only up to 20 bits
long, truncates the high-order bit, thereby "wrapping" this address
to 0FFEFH. However, the 80386, which can form addresses up to 32
bits long does not truncate such an address.
Table 14-1. 80386 Real-Address Mode Exceptions
Description Interrupt Function that Can Return Address
Number Generate the Exception Points to Faulting
Instruction
Divide error 0 DIV, IDIV YES
Debug exceptions 1 All
Some debug exceptions point to the faulting instruction, others to the
next instruction. The exception handler can determine which has occurred by
examining DR6.
Breakpoint 3 INT NO
Overflow 4 INTO NO
Bounds check 5 BOUND YES
Invalid opcode 6 Any undefined opcode or LOCK YES
used with wrong instruction
Coprocessor not available 7 ESC or WAIT YES
Interrupt table limit too small 8 INT vector is not within IDTR YES
limit
Reserved 9-12
Stack fault 12 Memory operand crosses offset YES
0 or 0FFFFH
Pseudo-protection exception 13 Memory operand crosses offset YES
0FFFFH or attempt to execute
past offset 0FFFFH or
instruction longer than 15
bytes
Reserved 14,15
Coprocessor error 16 ESC or WAIT YES
Coprocessor errors are reported on the first ESC or WAIT instruction
after the ESC instruction that caused the error.
Two-byte SW interrupt 0-255 INT n NO
Table 14-2. New 80386 Exceptions
Interrupt Function
Identifier
5 A BOUND instruction was executed with a register value outside
the limit values.
6 An undefined opcode was encountered or LOCK was used improperly
before an instruction to which it does not apply.
7 The EM bit in the MSW is set when an ESC instruction was
encountered. This exception also occurs on a WAIT instruction
if TS is set.
8 An exception or interrupt has vectored to an interrupt table
entry beyond the interrupt table limit in IDTR. This can occur
only if the LIDT instruction has changed the limit from the
default value of 3FFH, which is enough for all 256 interrupt
IDs.
12 Operand crosses extremes of stack segment, e.g., MOV operation
at offset 0FFFFH or push with SP=1 during PUSH, CALL, or INT.
13 Operand crosses extremes of a segment other than a stack
segment; or sequential instruction execution attempts to
proceed beyond offset 0FFFFH; or an instruction is longer than
15 bytes (including prefixes).
The few differences that exist between 80386 real-address mode and 80286
real-address mode are not likely to affect any existing 80286 programs
except possibly the system initialization procedures.
14.8.1 Bus Lock
The 80286 processor implements the bus lock function differently than the
80386. Programs that use forms of memory locking specific to the 80286 may
not execute properly if transported to a specific application of the 80386.
The
LOCK
prefix and its corresponding output signal should only be used to
prevent other bus masters from interrupting a data movement operation.
LOCK
may only be used with the following 80386 instructions when they modify
memory. An undefined-opcode exception results from using
LOCK before any
other instruction.
- Bit test and change:
BTS,
BTR,
BTC.
- Exchange:
XCHG.
- One-operand arithmetic and logical:
INC,
DEC,
NOT, and
NEG.
- Two-operand arithmetic and logical:
ADD,
ADC,
SUB,
SBB,
AND,
OR,
XOR.
A locked instruction is guaranteed to lock only the area of memory defined
by the destination operand, but may lock a larger memory area. For example,
typical 8086 and 80286 configurations lock the entire physical memory space.
With the 80386, the defined area of memory is guranteed to be locked against
access by a processor executing a locked instruction on exactly the same
memory area, i.e., an operand with identical starting address and identical
length.
14.8.2 Location of First Instruction
The starting location is 0FFFFFFF0H (sixteen bytes from end of 32-bit
address space) on the 80386 rather than 0FFFFF0H (sixteen bytes from end of
24-bit address space) as on the 80286. Many 80286 ROM initialization
programs will work correctly in this new environment. Others can be made to
work correctly with external hardware that redefines the signals on
A{31-20}.
14.8.3 Initial Values of General Registers
On the 80386, certain general registers may contain different values after
RESET than on the 80286. This should not cause compatibility problems,
because the content of 8086 registers after RESET is undefined. If
self-test is requested during the reset sequence and errors are detected in
the 80386 unit, EAX will contain a nonzero value. EDX contains the component
and revision identifier . Refer to
Chapter 10
for more information .
14.8.4 MSW Initialization
The 80286 initializes the MSW register to FFF0H, but the 80386 initializes
this register to 0000H. This difference should have no effect, because the
bits that are different are undefined on the 80286. Programs that read the
value of the MSW will behave differently on the 80386 only if they depend on
the setting of the undefined, high-order bits.
The features of the architecture that permit the 80386 to work equally well
with 32-bit and 16-bit address and operand sizes include:
- The D-bit (default bit) of code-segment descriptors, which determines
the default choice of operand-size and address-size for the
instructions of a code segment. (In real-address mode and V86 mode,
which do not use descriptors, the default is 16 bits.) A code segment
whose D-bit is set is known as a USE32 segment; a code segment whose
D-bit is zero is a USE16 segment. The D-bit eliminates the need to
encode the operand size and address size in instructions when all
instructions use operands and effective addresses of the same size.
- Instruction prefixes that explicitly override the default choice of
operand size and address size (available in protected mode as well as
in real-address mode and V86 mode).
- Separate 32-bit and 16-bit gates for intersegment control transfers
(including call gates, interrupt gates, and trap gates). The operand
size for the control transfer is determined by the type of gate, not by
the D-bit or prefix of the transfer instruction.
- Registers that can be used both for 32-bit and 16-bit operands and
effective-address calculations.
- The B-bit (big bit) of data-segment descriptors, which determines the
size of stack pointer (32-bit ESP or 16-bit SP) used by the CPU for
implicit stack references.
The 80386 has two instruction prefixes that allow mixing of 32-bit and
16-bit operations within one segment:
- The operand-size prefix (66H)
- The address-size prefix (67H)
These prefixes reverse the default size selected by the D-bit. For example,
the processor can interpret the word-move instruction MOV mem, reg in any of
four ways:
- In a USE32 segment:
- Normally moves 32 bits from a 32-bit register to a 32-bit
effective address in memory.
- If preceded by an operand-size prefix, moves 16 bits from a 16-bit
register to 32-bit effective address in memory.
- If preceded by an address-size prefix, moves 32 bits from a 32-bit
register to a16-bit effective address in memory.
- If preceded by both an address-size prefix and an operand-size
prefix, moves 16 bits from a 16-bit register to a 16-bit effective
address in memory.
- In a USE16 segment:
- Normally moves 16 bits from a 16-bit register to a 16-bit
effective address in memory.
- If preceded by an operand-size prefix, moves 32 bits from a 32-bit
register to 16-bit effective address in memory.
- If preceded by an address-size prefix, moves 16 bits from a 16-bit
register to a32-bit effective address in memory.
- If preceded by both an address-size prefix and an operand-size
prefix, moves 32 bits from a 32-bit register to a 32-bit effective
address in memory.
These examples illustrate that any instruction can generate any combination
of operand size and address size regardless of whether the instruction is in
a USE16 or USE32 segment. The choice of the USE16 or USE32 attribute for a
code segment is based upon these criteria:
- The need to address instructions or data in segments that are larger
than 64 Kilobytes.
- The predominant size of operands.
- The addressing modes desired. (Refer to
Chapter 17 for an explanation
of the additional addressing modes that are available when 32-bit
addressing is used.)
Choosing a setting of the D-bit that is contrary to the predominant size of
operands requires the generation of an excessive number of operand-size
prefixes.
Because the choice of operand size and address size is defined in code
segments and their descriptors, data segments can be shared freely among
both USE16 and USE32 code segments. The only limitation is the one imposed
by pointers with 16-bit offsets, which can only point to the first 64
Kilobytes of a segment. When a data segment that contains more than 64
Kilobytes is to be shared among USE32 and USE16 segments, the data that is
to be accessed by the USE16 segments must be located within the first 64
Kilobytes.
A stack that spans addresses less than 64K can be shared by both USE16 and
USE32 code segments. This class of stacks includes:
- Stacks in expand-up segments with G=0 and B=0.
- Stacks in expand-down segments with G=0 and B=0.
- Stacks in expand-up segments with G=1 and B=0, in which the stack is
contained completely within the lower 64 Kilobytes. (Offsets greater
than 64K can be used for data, other than the stack, that is not
shared.)
The B-bit of a stack segment cannot, in general, be used to change the size
of stack used by a USE16 code segment. The size of stack pointer used by the
processor for implicit stack references is controlled by the B-bit of the
data-segment descriptor for the stack. Implicit references are those caused
by interrupts, exceptions, and instructions such as
PUSH,
POP,
CALL, and
RET.
One might be tempted, therefore, to try to increase beyond 64K the
size of the stack used by 16-bit code simply by supplying a larger stack
segment with the B-bit set. However, the B-bit does not control explicit
stack references, such as accesses to parameters or local variables. A USE16
code segment can utilize a "big" stack only if the code is modified so that
all explicit references to the stack are preceded by the address-size
prefix, causing those references to use 32-bit addressing.
In big, expand-down segments (B=1, G=1, and E=1), all offsets are greater
than 64K, therefore USE16 code cannot utilize such a stack segment unless
the code segment is modified to employ 32-bit addressing . (Refer to
Chapter 6
for a review of the B , G, and E bits .)
When transferring control among procedures in USE16 and USE32 code
segments, programmers must be aware of three points:
Clearly, 16-bit effective addresses cannot be used to address data or code
located beyond 64K in a 32-bit segment, nor can large 32-bit parameters be
squeezed into a 16-bit word; however, except for these obvious limits, most
interfacing problems between 16-bit and 32-bit modules can be solved. Some
solutions involve inserting interface procedures between the procedures in
question.
16.4.1 Size of Code-Segment Pointer
For control-transfer instructions that use a pointer to identify the next
instruction (i.e., those that do not use gates), the size of the offset
portion of the pointer is determined by the operand-size attribute. The
implications of the use of two different sizes of code-segment pointer are:
- JMP, CALL, or RET from 32-bit segment to 16-bit segment is always
possible using a 32-bit operand size.
- JMP, CALL, or RET from 16-bit segment using a 16-bit operand size
cannot address the target in a 32-bit segment if the address of the
target is greater than 64K.
An interface procedure can enable transfers from USE16 segments to 32-bit
addresses beyond 64K without requiring modifications any more extensive than
relinking or rebinding the old programs. The requirements for such an
interface procedure are discussed later in this chapter.
16.4.2 Stack Management for Control Transfers
Because stack management is different for 16-bit CALL/RET than for 32-bit
CALL/RET, the operand size of RET must match that of
CALL. (Refer to
Figure 16-1
.) A 16-bit CALL pushes the 16-bit IP and
(for calls between privilege levels)
the 16-bit SP register. The corresponding RET must also use a 16-bit
operand size to POP these 16-bit values from the stack into the 16-bit
registers. A 32-bit CALL pushes the 32-bit EIP and (for interlevel calls)
the 32-bit ESP register. The corresponding RET must also use a 32-bit
operand size to POP these 32-bit values from the stack into the 32-bit
registers. If the two halves of a CALL/RET pair do not have matching operand
sizes, the stack will not be managed correctly and the values of the
instruction pointer and stack pointer will not be restored to correct
values.
When the CALL and its corresponding RET are in segments that have D-bits
with the same values (i.e., both have 32-bit defaults or both have 16-bit
defaults), there is no problem. When the CALL and its corresponding RET are
in segments that have different D-bit values, however, programmers (or
program development software) must ensure that the CALL and RET match.
There are three ways to cause a 16-bit procedure to execute a 32-bit call:
- Use a 16-bit call to a 32-bit interface procedure that then uses a
32-bit call to invoke the intended target.
- Bind the 16-bit call to a 32-bit call gate.
- Modify the 16-bit procedure, inserting an operand-size prefix before
the call, thereby changing it to a 32-bit call.
Likewise, there are three ways to cause a 32-bit procedure to execute a
16-bit call:
- Use a 32-bit call to a 32-bit interface procedure that then uses a
16-bit call to invoke the intended target.
- Bind the 32-bit call to a 16-bit call gate.
- Modify the 32-bit procedure, inserting an operand-size prefix before
the call, thereby changing it to a 16-bit call. (Be certain that the
return offset does not exceed 64K.)
Programmers can utilize any of the preceding methods to make a CALL in a
USE16 segment match the corresponding RET in a USE32 segment, or to make a
CALL in a USE32 segment match the corresponding RET in a USE16 segment.
When the selector of the pointer referenced by a CALL instruction selects a
segment descriptor, the operand-size attribute in effect for the CALL
instruction is determined by the D-bit in the segment descriptor and by any
operand-size instruction prefix.
When the selector of the pointer referenced by a CALL instruction selects a
gate descriptor, the type of call is determined by the type of call gate. A
call via an 80286 call gate (descriptor type 4) always has a 16-bit
operand-size attribute; a call via an 80386 call gate (descriptor type 12)
always has a 32-bit operand-size attribute. The offset of the target
procedure is taken from the gate descriptor; therefore, even a 16-bit
procedure can call a procedure that is located more than 64 kilobytes from
the base of a 32-bit segment, because a 32-bit call gate contains a 32-bit
target offset.
An unmodified 16-bit code segment that has run successfully on an 8086 or
real-mode 80286 will always have a D-bit of zero and will not use
operand-size override prefixes; therefore, it will always execute 16-bit
versions of CALL. The only modification needed to make a16-bit procedure
effect a 32-bit call is to relink the call to an 80386 call gate.
16.4.2.2 Changing Size of Call
When adding 32-bit gates to 16-bit procedures, it is important to consider
the number of parameters. The count field of the gate descriptor specifies
the size of the parameter string to copy from the current stack to the stack
of the more privileged procedure. The count field of a 16-bit gate specifies
the number of words to be copied, whereas the count field of a 32-bit gate
specifies the number of doublewords to be copied; therefore, the 16-bit
procedure must use an even number of words as parameters.
16.4.3 Interrupt Control Transfers
With a control transfer due to an interrupt or exception, a gate is always
involved. The operand-size attribute for the interrupt is determined by the
type of IDT gate.
A 386 interrupt or trap gate (descriptor type 14 or 15) to a 32-bit
interrupt procedure can be used to interrupt either 32-bit or 16-bit
procedures. However, it is not generally feasible to permit an interrupt or
exception to invoke a 16-bit handler procedure when 32-bit code is
executing, because a 16-bit interrupt procedure has a return offset of only
16-bits on its stack. If the 32-bit procedure is executing at an address
greater than 64K, the 16-bit interrupt procedure cannot return correctly.
16.4.4 Parameter Translation
When segment offsets or pointers (which contain segment offsets) are passed
as parameters between 16-bit and 32-bit procedures, some translation is
required. Clearly, if a 32-bit procedure passes a pointer to data located
beyond 64K to a 16-bit procedure, the 16-bit procedure cannot utilize it.
Beyond this natural limitation, an interface procedure can perform any
format conversion between 32-bit and 16-bit pointers that may be needed.
Parameters passed by value between 32-bit and 16-bit code may also require
translation between 32-bit and 16-bit formats. Such translation requirements
are application dependent. Systems designers should take care to limit the
range of values passed so that such translations are possible.
16.4.5 The Interface Procedure
Interposing an interface procedure between 32-bit and 16-bit procedures can
be the solution to any of several interface requirements:
- Allowing procedures in 16-bit segments to transfer control to
instructions located beyond 64K in 32-bit segments.
- Matching of operand size for CALL/RET.
- Parameter translation.
Interface procedures between USE32 and USE16 segments can be constructed
with these properties:
- The procedures reside in a code segment whose D-bit is set, indicating
a default operand size of 32-bits.
- All entry points that may be called by 16-bit procedures have offsets
that are actually less than 64K.
- All points to which called 16-bit procedures may return also lie
within 64K.
The interface procedures do little more than call corresponding procedures
in other segments. There may be two kinds of procedures:
- Those that are called by 16-bit procedures and call 32-bit procedures.
These interface procedures are called by 16-bit CALLs and use the
operand-size prefix before RET instructions to cause a 16-bit RET.
CALLs to 32-bit segments are 32-bit calls (by default, because the
D-bit is set), and the 32-bit code returns with 32-bit RET
instructions.
- Those that are called by 32-bit procedures and call 16-bit procedures.
These interface procedures are called by 32-bit CALL instructions, and
return with 32-bit RET instructions (by default, because the D-bit is
set). CALLs to 16-bit procedures use the operand-size prefix;
procedures in the 16-bit code return with 16-bit RET instructions.
|