ghidra/GhidraDocs/languages/html/sleigh_constructors.html

2279 lines
111 KiB
HTML
Raw Permalink Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1">
<title>7. Constructors</title>
<link rel="stylesheet" type="text/css" href="Frontpage.css">
<link rel="stylesheet" type="text/css" href="languages.css">
<meta name="generator" content="DocBook XSL Stylesheets V1.78.1">
<link rel="home" href="sleigh.html" title="SLEIGH">
<link rel="up" href="sleigh.html" title="SLEIGH">
<link rel="prev" href="sleigh_tokens.html" title="6. Tokens and Fields">
<link rel="next" href="sleigh_context.html" title="8. Using Context">
</head>
<body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF">
<div class="navheader">
<table width="100%" summary="Navigation header">
<tr><th colspan="3" align="center">7. Constructors</th></tr>
<tr>
<td width="20%" align="left">
<a accesskey="p" href="sleigh_tokens.html">Prev</a> </td>
<th width="60%" align="center"> </th>
<td width="20%" align="right"> <a accesskey="n" href="sleigh_context.html">Next</a>
</td>
</tr>
</table>
<hr>
</div>
<div class="sect1">
<div class="titlepage"><div><div><h2 class="title" style="clear: both">
<a name="sleigh_constructors"></a>7. Constructors</h2></div></div></div>
<p>
Fields are the basic building block for family symbols. The mechanisms
for building up from fields to the
root <span class="emphasis"><em>instruction</em></span> symbol are
the <span class="emphasis"><em>constructor</em></span> and <span class="emphasis"><em>table</em></span>.
</p>
<p>
A <span class="emphasis"><em>constructor</em></span> is the unit of syntax for building
new symbols. In essence a constructor describes how to build a new
family symbol, by describing, in turn, how to build a new display
meaning, how to build a new semantic meaning, and how encodings map to
these new meanings. A <span class="emphasis"><em>table</em></span> is a set of one or
more constructors and is the final step in creating a new family
symbol identifier associated with the pieces defined by
constructors. The name of the table is this new identifier, and it is
this identifier which can be used in the syntax for subsequent
constructors.
</p>
<p>
The difference between a constructor and table is slightly confusing
at first. In short, the syntactical elements described in this
chapter, for combining existing symbols into new symbols, are all used
to describe a single constructor. Specifications for multiple
constructors are combined to describe a single table. Since many
tables are built with only one constructor, it is natural and correct
to think of a constructor as a kind of table in and of itself. But it
is only the table that has an actual family symbol identifier
associated with it. Most of this chapter is devoted to describing how
to define a single constructor. The issues involved in combining
multiple constructors into a single table are addressed in <a class="xref" href="sleigh_constructors.html#sleigh_tables" title="7.8. Tables">Section 7.8, &#8220;Tables&#8221;</a>.
</p>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="idm140526920750848"></a>7.1. The Five Sections of a Constructor</h3></div></div></div>
<p>
A single complex statement in the specification file describes a
constructor. This statement is always made up of five distinct
sections that are listed below in the order in which the must occur.
</p>
<div class="informalexample"><div class="orderedlist"><ol class="orderedlist compact" type="1">
<li class="listitem">
Table Header
</li>
<li class="listitem">
Display Section
</li>
<li class="listitem">
Bit Pattern Sections
</li>
<li class="listitem">
Disassembly Actions Section
</li>
<li class="listitem">
Semantics Actions Section
</li>
</ol></div></div>
<p>
The full set of rules for correctly writing each section is long and
involved, but for any given constructor in a real specification file,
the syntax typically fits on a single line. We describe each section
in turn.
</p>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="idm140526920746272"></a>7.2. The Table Header</h3></div></div></div>
<p>
Every constructor must be part of a table, which is the element with
an actual family symbol identifier associated with it. So each
constructor starts with the identifier of the table it belongs to
followed by a colon &#8216;:&#8217;.
</p>
<div class="informalexample"><pre class="programlisting">
mode1: <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
The above line starts the definition of a constructor that is part of
the table identified as <span class="emphasis"><em>mode1</em></span>. If the identifier
has not appeared before, a new table is created. If other constructors
have used the identifier, the new constructor becomes an additional
part of that same table. A constructor in the
root <span class="emphasis"><em>instruction</em></span> table is defined by omitting the
identifier.
</p>
<div class="informalexample"><pre class="programlisting">
: <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
The identifier <span class="emphasis"><em>instruction</em></span> is actually reserved
for the root table, but should not be used in the table header as the
SLEIGH parser uses the blank identifier to help distinguish assembly
mnemonics from operands (see <a class="xref" href="sleigh_constructors.html#sleigh_mnemonic" title="7.3.1. Mnemonic">Section 7.3.1, &#8220;Mnemonic&#8221;</a>).
</p>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_display_section"></a>7.3. The Display Section</h3></div></div></div>
<p>
The <span class="emphasis"><em>display section</em></span> consists of all characters
after the table header &#8216;:&#8217; up to the SLEIGH
keyword <span class="bold"><strong>is</strong></span>. The section&#8217;s primary
purpose is to assign disassembly display meaning to the
constructor. The section&#8217;s secondary purpose is to define local
identifiers for the pieces out of which the constructor is being
built. Characters in the display section are treated as literals with
the following exceptions.
</p>
<div class="informalexample"><div class="itemizedlist"><ul class="itemizedlist compact" style="list-style-type: bullet; ">
<li class="listitem" style="list-style-type: disc">
Legal identifiers are not treated literally unless
<div class="orderedlist"><ol class="orderedlist compact" type="a">
<li class="listitem">
The identifier is surrounded by double quotes.
</li>
<li class="listitem">
The identifier is considered a mnemonic (see below).
</li>
</ol></div>
</li>
<li class="listitem" style="list-style-type: disc">
The character &#8216;^&#8217; has special meaning.
</li>
<li class="listitem" style="list-style-type: disc">
White space is trimmed from the beginning and end of the section.
</li>
<li class="listitem" style="list-style-type: disc">
Other sequences of white space characters are condensed into a single space.
</li>
</ul></div></div>
<p>
</p>
<p>
In particular, all punctuation except &#8216;^&#8217; loses its special
meaning. Those identifiers that are not treated as literals are
considered to be new, initially undefined, family symbols. We refer to
these new symbols as the <span class="emphasis"><em>operands</em></span> of the constructor. And for root
constructors, these operands frequently correspond to the natural
assembly operands. Thinking of it as a family symbol, the
constructor&#8217;s display meaning becomes the string of literals itself,
with each identifier replaced with the display meaning of the symbol
corresponding to that identifier.
</p>
<div class="informalexample"><pre class="programlisting">
mode1: ( op1 ),op2 is <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
In the above example, a constructor for
table <span class="emphasis"><em>mode1</em></span> is being built out of two pieces,
symbol <span class="emphasis"><em>op1</em></span> and
symbol <span class="emphasis"><em>op2</em></span>. The characters &#8216;(&#8216;, &#8217;)&#8217;, and &#8216;,&#8217;
become literal parts of the disassembly display for symbol
mode1. After the display strings for <span class="emphasis"><em>op1</em></span>
and <span class="emphasis"><em>op2</em></span> are found, they are inserted into the
string of literals, forming the constructor&#8217;s display string. The
white space characters surrounding the <span class="emphasis"><em>op1</em></span>
identifier are preserved as part of this string.
</p>
<p>
The identifiers <span class="emphasis"><em>op1</em></span> and <span class="emphasis"><em>op2</em></span>
are local to the constructor and can mask global symbols with the same
names. The symbols will (must) be defined in the following sections,
but only their identifiers are established in the display section.
</p>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_mnemonic"></a>7.3.1. Mnemonic</h4></div></div></div>
<p>
If the constructor is part of the root instruction table, the first
string of characters in the display section that does not contain
white space is treated as the <span class="emphasis"><em>literal mnemonic</em></span> of
the instruction and is not considered a local symbol identifier even
if it is legal.
</p>
<div class="informalexample"><pre class="programlisting">
:and (var1) is <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
In the above example, the string &#8220;var1&#8221; is treated as a symbol
identifier, but the string &#8220;and&#8221; is considered to be the mnemonic of
the instruction.
</p>
<p>
There is nothing that special about the mnemonic. As far as the
display meaning of the constructor is concerned, it is just a sequence
of literal characters. Although the current parser does not concern
itself with this, the mnemonic of any assembly language instruction in
general is used to guarantee the uniqueness of the assembly
representation. It is conceivable that a forward engineering engine
built on SLEIGH would place additional requirements on the mnemonic to
assure uniqueness, but for reverse engineering applications there is
no such requirement.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920716688"></a>7.3.2. The '^' character</h4></div></div></div>
<p>
The &#8216;^&#8217; character in the display section is used to separate
identifiers from other characters where there shouldn&#8217;t be white space
in the disassembly display. This can be used in any manner but is
usually used to attach display characters from a local symbol to the
literal characters of the mnemonic.
</p>
<div class="informalexample"><pre class="programlisting">
:bra^cc op1,op2 is <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
In the above example, &#8220;bra&#8221; is treated as literal characters in the
resulting display string followed immediately, with no intervening
spaces, by the display string of the local
symbol <span class="emphasis"><em>cc</em></span>. Thus the whole constructor actually
has three operands, denoted by the three
identifiers <span class="emphasis"><em>cc</em></span>, <span class="emphasis"><em>op1</em></span>,
and <span class="emphasis"><em>op2</em></span>.
</p>
<p>
If the &#8216;^&#8217; is used as the first (non-whitespace) character in the
display section of a base constructor, this inhibits the first
identifier in the display from being considered the mnemonic, as
described in <a class="xref" href="sleigh_constructors.html#sleigh_mnemonic" title="7.3.1. Mnemonic">Section 7.3.1, &#8220;Mnemonic&#8221;</a>. This allows
specification of less common situations, where the first part of the
mnemonic, rather than perhaps a later part, needs to be considered as
an operand. An initial &#8216;^&#8217; character can also facilitate certain
recursive constructions.
</p>
</div>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_bit_pattern"></a>7.4. The Bit Pattern Section</h3></div></div></div>
<p>
Syntactically, this section comes between the
keyword <span class="bold"><strong>is</strong></span> and the delimiter for the
following section, either an &#8216;{&#8216; or an &#8216;[&#8216;. The <span class="emphasis"><em>bit pattern
section</em></span> describes a
constructor&#8217;s <span class="emphasis"><em>pattern</em></span>, the subset of possible
instruction encodings that the designer wants
to <span class="emphasis"><em>match</em></span> the constructor being defined.
</p>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920705248"></a>7.4.1. Constraints</h4></div></div></div>
<p>
The patterns required for processor specifications can almost always
be described as a mask and value pair. Given a specific instruction
encoding, we can decide if the encoding matches our pattern by looking
at just the bits specified by the <span class="emphasis"><em>mask</em></span> and seeing
if they match a specific <span class="emphasis"><em>value</em></span>. The fields, as
defined in <a class="xref" href="sleigh_tokens.html#sleigh_defining_tokens" title="6.1. Defining Tokens and Fields">Section 6.1, &#8220;Defining Tokens and Fields&#8221;</a>, typically give us
our masks. So to construct a pattern, we can simply require that the
field take on a specific value, as in the example below.
</p>
<div class="informalexample"><pre class="programlisting">
:halt is opcode=0x15 { <span class="weak">...</span>
</pre></div>
<p>
Assuming the symbol <span class="emphasis"><em>opcode</em></span> was defined as a field, this says that a
root constructor with mnemonic &#8220;halt&#8221; matches any instruction where
the bits defining this field have the value 0x15. The equation
&#8220;opcode=0x15&#8221; is called a <span class="emphasis"><em>constraint</em></span>.
</p>
<p>
The standard bit encoding of the integer is used when restricting the
value of a field. This encoding is used even if
an <span class="bold"><strong>attach</strong></span> statement has assigned a
different meaning to the field. The alternate meaning does not apply
within the pattern. This can be slightly confusing, particularly in
the case of an <span class="bold"><strong>attach values</strong></span>
statement, which provides an alternate integer interpretation of the
field.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_ampandor"></a>7.4.2. The '&amp;' and '|' Operators</h4></div></div></div>
<p>
More complicated patterns are built out of logical operators. The
meaning of these are fairly straightforward. We can force two or more
constraints to be true at the same time, a <span class="emphasis"><em>logical
and</em></span> &#8216;&amp;&#8217;, or we can require that either one constraint or
another must be true, a <span class="emphasis"><em>logical or</em></span> &#8216;|&#8217;. By using these with
constraints and parentheses for grouping, arbitrarily complicated
patterns can be constructed.
</p>
<div class="informalexample"><pre class="programlisting">
:nop is (opcode=0 &amp; mode=0) | (opcode=15) { <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
Of the two operators, the <span class="emphasis"><em>logical and</em></span> is much
more common. The SLEIGH compiler typically can group together several
constraints that are combined with this operator into a single
efficient mask/value check, so this operator is to be preferred if at
all possible. The <span class="emphasis"><em>logical or</em></span> operator usually
requires two or more mask/value style checks to correctly implement.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920691312"></a>7.4.3. Defining Operands and Invoking Subtables</h4></div></div></div>
<p>
The principle way of defining a constructor operand, left undefined
from the display section, is done in the bit pattern section. If an
operand&#8217;s identifier is used by itself, not as part of a constraint,
then the operand takes on both the display and semantic definition of
the global symbol with the same identifier. The syntax is slightly
confusing at first. The identifier must appear in the pattern as if it
were a term in a sequence of constraints but without the operator and
right-hand side of the constraint.
</p>
<div class="informalexample"><pre class="programlisting">
define token instr(32)
opcode = (0,5)
r1 = (6,10)
r2 = (11,15);
attach variables [ r1 r2 ] [ reg0 reg1 reg2 reg3 ];
:add r1,r2 is opcode=7 &amp; r1 &amp; r2 { <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
This is a typical example. The <span class="emphasis"><em>add</em></span> instruction
must have the bits in the <span class="emphasis"><em>opcode</em></span> field set
specifically. But it also uses two fields in the instruction which
specify registers. The <span class="emphasis"><em>r1</em></span>
and <span class="emphasis"><em>r2</em></span> identifiers are defined to be local
because they appear in the display section, but their use in the
pattern section of the definition links the local symbols with the
global register symbols defined as fields with attached registers. The
constructor is essentially saying that it is building the
full <span class="emphasis"><em>add</em></span> instruction encoding out of the register
fields <span class="emphasis"><em>r1</em></span> and <span class="emphasis"><em>r2</em></span> but is not
specifying their value.
</p>
<p>
The syntax makes a little more sense keeping in mind this principle:
</p>
<div class="informalexample"><div class="itemizedlist"><ul class="itemizedlist compact" style="list-style-type: bullet; "><li class="listitem" style="list-style-type: disc">
The pattern must somehow specify all the bits and symbols
being used by the constructor, even if the bits are not restricted
to specific values.
</li></ul></div></div>
<p>
The linkage from local symbol to global symbol will happen for any
global identifier which represents a family symbol, including table
symbols. This is in fact the principle mechanism for recursively
building new symbols from old symbols. For those familiar with grammar
parsers, a SLEIGH specification is in part a grammar
specification. The terminal symbols, or tokens, are the bits of an
instruction, and the constructors and tables are the non-terminating
symbols. These all build up to the root instruction table, the
grammar&#8217;s start symbol. So this link from local to global is simply a
statement of the grouping of old symbols into the new constructor.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920679904"></a>7.4.4. Variable Length Instructions</h4></div></div></div>
<p>
There are some additional complexities to designing a specification
for a processor with variable length instructions. Some initial
portion of an instruction must always be parsed. But depending on the
fields in this first portion, additional portions of varying lengths
may need to be read. The key to incorporating this behavior into a
SLEIGH specification is the token. Recall that all fields are built on
top of a token which is defined to be a specific number of bytes. If a
processor has fixed length instructions, the specification needs to
define only a single token representing the entire instruction, and
all fields are built on top of this one token. For processors with
variable length instructions however, more than one token needs to be
defined. Each token has different fields defined upon it, and the
SLEIGH compiler can distinguish which tokens are involved in a
particular constructor by examining the fields it uses. The tokens
that are actually used by any matching constructors determine the
final length of the instruction. SLEIGH has two operators that are
specific to variable length instruction sets and that give the
designer control over how tokens fit together.
</p>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920676432"></a>7.4.4.1. The ';' Operator</h5></div></div></div>
<p>
The most important operator for patterns defining variable length
instructions is the concatenation operator &#8216;;&#8217;. When building a
constructor with fields from two or more tokens, the pattern must
explicitly define the order of the tokens. In terms of the logic of
the pattern expressions themselves, the &#8216;;&#8217; operator has the same
meaning as the &#8216;&amp;&#8217; operator. The combined expression matches only if
both subexpressions are true. However, it also requires that the
subexpressions involve multiple tokens and explicitly indicates an
order for them.
</p>
<div class="informalexample"><pre class="programlisting">
define token base(8)
op=(0,3)
mode=(4,4)
reg=(5,7);
define token immtoken(16)
imm16 = (0,15);
:inc reg is op=2 &amp; reg { <span class="weak">...</span>
:add reg,imm16 is op=3 &amp; reg; imm16 { <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
In the above example, we see the definitions of two different
tokens, <span class="emphasis"><em>base</em></span>
and <span class="emphasis"><em>immtoken</em></span>. For the first
instruction, <span class="emphasis"><em>inc</em></span>, the constructor uses
fields <span class="emphasis"><em>op</em></span> and <span class="emphasis"><em>reg</em></span>, both
defined on <span class="emphasis"><em>base</em></span>. Thus, the pattern applies
constraints to just a single byte, the size of base, in the
corresponding encoding. The second
instruction, <span class="emphasis"><em>add</em></span>, uses
fields <span class="emphasis"><em>op</em></span> and <span class="emphasis"><em>reg</em></span>, but it
also uses field <span class="emphasis"><em>imm16</em></span> contained
in <span class="emphasis"><em>immtoken</em></span>. The &#8216;;&#8217; operator indicates that
token <span class="emphasis"><em>base</em></span> (via its fields) comes first in the
encoding, followed by <span class="emphasis"><em>immtoken</em></span>. The constraints
on <span class="emphasis"><em>base</em></span> will therefore correspond to constraints
on the first byte of the encoding, and the constraints
on <span class="emphasis"><em>immtoken</em></span> will apply to the second and third
bytes. The length of the final encoding for <span class="emphasis"><em>add</em></span>
will be 3 bytes, the sum of the lengths of the two tokens.
</p>
<p>
If two pattern expressions are combined with the &#8216;&amp;&#8217; or &#8216;|&#8217; operator,
where the concatenation operator &#8216;;&#8217; is also being used, the designer
must make sure that the tokens underlying each expression are the same
and come in the same order. In the example <span class="emphasis"><em>add</em></span>
instruction for instance, the &#8216;&amp;&#8217; operator combines the &#8220;op=3&#8221; and
&#8220;reg&#8221; expressions. Both of these expressions involve only the
token <span class="emphasis"><em>base</em></span>, so the matching requirement is
satisfied. The &#8216;&amp;&#8217; and &#8216;|&#8217; operators can combine expressions built out
of more than one token, but the tokens must come in the same
order. Also these operators have higher precedence than the &#8216;;&#8217;
operator, so parentheses may be necessary to get the intended meaning.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920661120"></a>7.4.4.2. The '...' Operator</h5></div></div></div>
<p>
The ellipsis operator &#8216;...&#8217; is used to satisfy the token matching
requirements of the &#8216;&amp;&#8217; and &#8216;|&#8217; operators (described in the previous
section), when the operands are of different lengths. The ellipsis is
a unary operator applied to a pattern expression that extends its
token length before it is combined with another expression. Depending
on what side of the expression the ellipsis is applied, the
expression's tokens are either right or left justified within the
extension.
</p>
<div class="informalexample"><pre class="programlisting">
addrmode: reg is reg &amp; mode=0 { <span class="weak">...</span>
addrmode: #imm16 is mode=1; imm16 { <span class="weak">...</span>
:xor &#8220;A&#8221;,addrmode is op=4 ... &amp; addrmode { <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
Extending the example from the previous section, we add a
subtable <span class="emphasis"><em>addrmode</em></span>, representing an operand that
can be encoded either as a register, if <span class="emphasis"><em>mode</em></span> is
set to zero, or as an immediate value, if
the <span class="emphasis"><em>mode</em></span> bit is one. If the immediate value mode
is selected, the operand is built by reading an additional two bytes
directly from the instruction encoding. So
the <span class="emphasis"><em>addrmode</em></span> table can represent a 1 byte or a 3
byte encoding depending on the mode. In the
following <span class="emphasis"><em>xor</em></span>
instruction, <span class="emphasis"><em>addrmode</em></span> is used as an operand. The
particular instruction is selected by encoding a 4 in
the <span class="emphasis"><em>op</em></span> field, so it requires a constraint on that
field in the pattern expression. Since the instruction uses
the <span class="emphasis"><em>addrmode</em></span> operand, it must combine the
constraint on <span class="emphasis"><em>op</em></span> with the pattern
for <span class="emphasis"><em>addrmode</em></span>. But <span class="emphasis"><em>op</em></span>
involves only the token <span class="emphasis"><em>base</em></span>,
while <span class="emphasis"><em>addrmode</em></span> may also
involve <span class="emphasis"><em>immtoken</em></span>. The ellipsis operator resolves
the conflict by extending the <span class="emphasis"><em>op</em></span> constraint to be
whatever the length of <span class="emphasis"><em>addrmode</em></span> turns out to be.
</p>
<p>
Since the <span class="emphasis"><em>op</em></span> constraint occurs to the left of the
ellipsis, it is considered left justified, and the matching
requirement for &#8216;&amp;&#8217; will insist that <span class="emphasis"><em>base</em></span> is the
first token in all forms of <span class="emphasis"><em>addrmode</em></span>. This allows
the <span class="emphasis"><em>xor</em></span> instruction's constraint
on <span class="emphasis"><em>op</em></span> and the <span class="emphasis"><em>addrmode</em></span>
constraint on <span class="emphasis"><em>mode</em></span> to be combined into
constraints on a single byte in the final encoding.
</p>
</div>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_invisible_operands"></a>7.4.5. Invisible Operands</h4></div></div></div>
<p>
It is not necessary for a global symbol, which is needed by a
constructor, to appear in the display section of the definition. If
the global identifier is used in the pattern section as it would be
for a normal operand definition but the identifier was not used in the
display section, then the constructor defines an <span class="emphasis"><em>invisible
operand</em></span>. Such an operand behaves and is parsed exactly like
any other operand but there is absolutely no visible indication of the
operand in the final display of the assembly instruction. The one
common type of instruction that uses this is the relative branch (see
<a class="xref" href="sleigh_constructors.html#sleigh_relative_branches" title="7.5.1. Relative Branches">Section 7.5.1, &#8220;Relative Branches&#8221;</a>) but it is otherwise needed
only in more esoteric instructions. It is useful in situations where
you need to break up the parsing of an instruction along lines that
don&#8217;t quite match the assembly.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920640560"></a>7.4.6. Empty Patterns</h4></div></div></div>
<p>
Occasionally there is a need for an empty pattern when building
tables. An empty pattern matches everything. There is a predefined
symbol <span class="emphasis"><em>epsilon</em></span> which has been traditionally used
to indicate an empty pattern.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920638720"></a>7.4.7. Advanced Constraints</h4></div></div></div>
<p>
A constraint does not have to be of the form &#8220;field = constant&#8221;,
although this is almost always what is needed. In certain situations,
it may be more convenient to use a different kind of
constraint. Special care should be taken when designing these
constraints because they can substantially deviate from the mask/value
model used to implement most constraints. These more general
constraints are implemented by splitting it up into smaller states
which can be modeled as a mask/value pair. This is all done
automatically, and the designer may inadvertently create huge numbers
of parsing states for a single constraint.
</p>
<p>
A constraint can actually be built out of arbitrary
expressions. These <span class="emphasis"><em>pattern expressions</em></span> are more
commonly used in disassembly actions and are defined in
<a class="xref" href="sleigh_constructors.html#sleigh_general_actions" title="7.5.2. General Actions and Pattern Expressions">Section 7.5.2, &#8220;General Actions and Pattern Expressions&#8221;</a>, but they can also be used in
constraints. So in general, a constraint is any equation where the
left-hand side is a single family symbol, the right-hand side is an
arbitrary pattern expression, and the constraint operator is one of
the following:
</p>
<div class="informalexample">
<div class="table">
<a name="constraints.htmltable"></a><p class="title"><b>Table 3. Constraint Operators</b></p>
<div class="table-contents"><table width="50%" frame="box" rules="all">
<col width="50%">
<col width="50%">
<thead><tr>
<td><span class="bold"><strong>Operator Name</strong></span></td>
<td><span class="bold"><strong>Syntax</strong></span></td>
</tr></thead>
<tbody>
<tr>
<td>Integer equality</td>
<td>=</td>
</tr>
<tr>
<td>Integer inequality</td>
<td>!=</td>
</tr>
<tr>
<td>Integer less-than</td>
<td>&lt;</td>
</tr>
<tr>
<td>Integer greater-than</td>
<td>&gt;</td>
</tr>
</tbody>
</table></div>
</div>
<br class="table-break">
</div>
<p>
For a particular instruction encoding, each variable evaluates to a
specific integer depending on the encoding. A constraint is <span class="emphasis"><em>satisfied</em></span>
if, when all the variables are evaluated, the equation is true.
</p>
<div class="informalexample"><pre class="programlisting">
:xor r1,r2 is opcode=0xcd &amp; r1 &amp; r2 { r1 = r1 ^ r2; }
:clr r1 is opcode=0xcd &amp; r1 &amp; r2=r1 { r1 = 0; }
</pre></div>
<p>
</p>
<p>
The above example illustrates a situation that does come up
occasionally. A processor uses an exclusive-or instruction to clear a
register by setting both operands of the instruction to the same
register. The first line in the example illustrates such an
instruction. However, processor documentation stipulates, and analysts
prefer, that, in this case, the disassembler should print a
pseudo-instruction <span class="emphasis"><em>clr</em></span>. The distinguishing
feature of <span class="emphasis"><em>clr</em></span> from <span class="emphasis"><em>xor</em></span> is
that the two fields, specifying the two register inputs
to <span class="emphasis"><em>xor</em></span>, are equal. The easiest way to specify
this special case is with the general constraint,
&#8220;<span class="emphasis"><em>r2</em></span> = <span class="emphasis"><em>r1</em></span>&#8221;, as in the second
line of the example. The SLEIGH compiler will implement this by
enumerating all the cases where <span class="emphasis"><em>r2</em></span>
equals <span class="emphasis"><em>r1</em></span>, creating as many states as there are
registers. But the specification itself, at least, remains compact.
</p>
</div>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_disassembly_actions"></a>7.5. Disassembly Actions Section</h3></div></div></div>
<p>
After the bit pattern section, there can optionally be a section for
doing dynamic calculations, which must be between square brackets. For
certain kinds of instructions, there is a need to calculate values
that depend on the specific bits of the instruction, but which cannot
be obtained as an integer interpretation of a field or by building
with an <span class="bold"><strong>attach values</strong></span> statement. So
SLEIGH provides a mechanism to build values of arbitrary
complexity. This section is not intended to emulate the execution of
the processor (this is the job of the semantic section) but is
intended to produce only those values that are needed at disassembly
time, usually for part of the disassembly display.
</p>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_relative_branches"></a>7.5.1. Relative Branches</h4></div></div></div>
<p>
The canonical example of an action at disassembly time is a branch
relocation. A jump instruction encodes the address of where it jumps
to as a relative offset to the instruction&#8217;s address, for
instance. But when we display the assembly, we want to show the
absolute address of the jump destination. The correct way to specify
this is to reserve an identifier in the display section which
represents the absolute address, but then, instead of defining it in
the pattern section, we define it in the disassembly action section as
a function of the current address and the relative offset.
</p>
<div class="informalexample"><pre class="programlisting">
jmpdest: reloc is simm8 [ reloc=inst_next + simm8*4; ] { <span class="weak">...</span>
</pre></div>
<p>
</p>
<p>
The identifier <span class="emphasis"><em>reloc</em></span> is reserved in the display
section for this constructor, but the identifier is not defined in the
pattern section. Instead, an invisible
operand <span class="emphasis"><em>simm8</em></span> is defined which is attached to a
global field definition. The <span class="emphasis"><em>reloc</em></span> identifier is
defined in the action section as the integer obtained by adding a
multiple of <span class="emphasis"><em>simm8</em></span>
to <span class="emphasis"><em>inst_next</em></span>, a symbol predefined to be equal to
the address of the following instruction (see
<a class="xref" href="sleigh_symbols.html#sleigh_predefined_symbols" title="5.2. Predefined Symbols">Section 5.2, &#8220;Predefined Symbols&#8221;</a>). Now <span class="emphasis"><em>reloc</em></span>
is a specific symbol with both semantic and display meaning equal to
the desired absolute address. This address is calculated separately,
at disassembly time, for every instruction that this constructor
matches.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_general_actions"></a>7.5.2. General Actions and Pattern Expressions</h4></div></div></div>
<p>
In general, the disassembly actions are encoded as a sequence of
assignments separated by semicolons. The left-hand side of each
statement must be a single operand identifier, and the right-hand side
must be a <span class="emphasis"><em>pattern expression</em></span>. A <span class="emphasis"><em>pattern
expression</em></span> is made up of both integer constants and family
symbols that have retained their semantic meaning as integers, and it
is built up out of the following typical operators:
</p>
<div class="informalexample">
<div class="table">
<a name="patexp.htmltable"></a><p class="title"><b>Table 4. Pattern Expression Operators</b></p>
<div class="table-contents"><table width="50%" frame="box" rules="all">
<col width="50%">
<col width="50%">
<thead><tr>
<td><span class="bold"><strong>Operator Name</strong></span></td>
<td><span class="bold"><strong>Syntax</strong></span></td>
</tr></thead>
<tbody>
<tr>
<td>Integer addition</td>
<td>+</td>
</tr>
<tr>
<td>Integer subtraction</td>
<td>-</td>
</tr>
<tr>
<td>Integer multiplication</td>
<td>*</td>
</tr>
<tr>
<td>Integer division</td>
<td>/</td>
</tr>
<tr>
<td>Left-shift</td>
<td>&lt;&lt;</td>
</tr>
<tr>
<td>Arithmetic right-shift</td>
<td>&gt;&gt;</td>
</tr>
<tr>
<td>Bitwise and</td>
<td>
<div class="informaltable">
<a name="bitwiseand.htmltable"></a><table frame="none"><tbody>
<tr>
<td>$and</td>
</tr>
<tr>
<td>&amp; (within square brackets)</td>
</tr>
</tbody></table>
</div>
</td>
</tr>
<tr>
<td>Bitwise or</td>
<td>
<div class="informaltable">
<a name="bitwiseor.htmltable"></a><table frame="none"><tbody>
<tr>
<td>$or</td>
</tr>
<tr>
<td>| (within square brackets)</td>
</tr>
</tbody></table>
</div>
</td>
</tr>
<tr>
<td>Bitwise xor</td>
<td>
<div class="informaltable">
<a name="bitwisexor.htmltable"></a><table frame="none"><tbody>
<tr>
<td>$xor</td>
</tr>
<tr>
<td>^</td>
</tr>
</tbody></table>
</div>
</td>
</tr>
<tr>
<td>Bitwise negation</td>
<td>~</td>
</tr>
</tbody>
</table></div>
</div>
<br class="table-break">
</div>
<p>
For the sake of these expressions, integers are considered signed
values of arbitrary precision. Expressions can also make use of
parentheses. A family symbol can be used in an expression, only if it
can be resolved to a particular specific symbol. This generally means
that a global family symbol, such as a field, must be attached to a
local identifier before it can be used.
</p>
<p>
The left-hand side of an assignment statement can be a context
variable (see <a class="xref" href="sleigh_tokens.html#sleigh_context_variables" title="6.4. Context Variables">Section 6.4, &#8220;Context Variables&#8221;</a>). An
assignment to such a variable changes the context in which the current
instruction is being disassembled and can potentially have a drastic
effect on how the rest of the instruction is disassembled. An
assignment of this form is considered local to the instruction and
will not affect how other instructions are parsed. The context
variable is reset to its original value before parsing other
instructions. The disassembly action may also contain one or
more <span class="bold"><strong>globalset</strong></span> directives, which
cause changes to context variables to become more permanent. This
directive is distinct from the operators in a pattern expression and
must be invoked as a separate statement. See
<a class="xref" href="sleigh_context.html" title="8. Using Context">Section 8, &#8220;Using Context&#8221;</a>, for a discussion of how to
effectively use context variables and
<a class="xref" href="sleigh_context.html#sleigh_global_change" title="8.3. Global Context Change">Section 8.3, &#8220;Global Context Change&#8221;</a>, for details of
the <span class="bold"><strong>globalset</strong></span> directive.
</p>
<p>
Note that there are two syntax forms for the logical operators in a
pattern expression. When an expression is used as part of a
constraint, the &#8220;$and&#8221; and &#8220;$or&#8221; forms of the operators must be used
in order to distinguish the bitwise operators from the special pattern
combining operators, &#8216;&amp;&#8217; and &#8216;|&#8217; (as described in
<a class="xref" href="sleigh_constructors.html#sleigh_ampandor" title="7.4.2. The '&amp;' and '|' Operators">Section 7.4.2, &#8220;The '&amp;' and '|' Operators&#8221;</a>). However inside the square braces
of the disassembly action section, &#8216;&amp;&#8217; and &#8216;|&#8217; are interpreted as
the usual logical operators.
</p>
</div>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_with_block"></a>7.6. The With Block</h3></div></div></div>
<p>
To avoid tedious repetition and to ease the maintenance of specifications
already having many, many constructors and tables, the <span class="emphasis"><em>with
block</em></span> is provided. It is a syntactic construct that allows a
designer to apply a table header, bit pattern constraints, and/or disassembly
actions to a group of constructors. The block starts at the
<span class="bold"><strong>with</strong></span> directive and ends with a closing brace.
All constructors within the block are affected:
</p>
<div class="informalexample"><pre class="programlisting">
with op1 : mode=1 [ mode=2; ] {
:reg is reg &amp; ind=0 [ mode=1; ] { <span class="weak">...</span> }
:[reg] is reg &amp; ind=1 { <span class="weak">...</span> }
}
</pre></div>
<p>
In the example, both constructors are added to the table identified by
<span class="emphasis"><em>op1</em></span>. Both require the context field
<span class="emphasis"><em>mode</em></span> to be equal to 1. The listed constraints take the
form described in <a class="xref" href="sleigh_constructors.html#sleigh_bit_pattern" title="7.4. The Bit Pattern Section">Section 7.4, &#8220;The Bit Pattern Section&#8221;</a>, and they are joined to
those given in the constructor statement as if prepended using &#8216;&amp;&#8217;. Similarly,
the actions take the form described in <a class="xref" href="sleigh_constructors.html#sleigh_disassembly_actions" title="7.5. Disassembly Actions Section">Section 7.5, &#8220;Disassembly Actions Section&#8221;</a>
and are prepended to the actions given in the constructor statement. Prepending
the actions allows the statement to override actions in the with block. Both
technically occur, but only the last one has a noticeable effect. The above
example could have been equivalently specified:
</p>
<div class="informalexample"><pre class="programlisting">
op1:reg is mode=1 &amp; reg &amp; ind=0 [ mode=2; mode=1; ] { <span class="weak">...</span> }
op1:[ref] is mode=1 &amp; reg &amp; ind=1 [ mode=2; ] { <span class="weak">...</span> }
</pre></div>
<p>
</p>
<p>
The three parts (table header, bit pattern section, and disassembly actions
section) of the with block are all optional. Any of them may be omitted,
though omitting all of them is rather pointless. With blocks may also be nested.
The innermost with block having a table header specifies the default header of
the constructors it contains. The constraints and actions are combined outermost
to innermost, left to right.
Note that when a with block has a table header specifying a table that does not
yet exist, the table is created immediately. Inside a with block that has a
table header, a nested with block may specify the <span class="emphasis"><em>instruction</em></span>
table by name, as in "with instruction : {<span class="weak">...</span>}".
Inside such a block, the rule regarding mnemonic literals is restored (see
<a class="xref" href="sleigh_constructors.html#sleigh_mnemonic" title="7.3.1. Mnemonic">Section 7.3.1, &#8220;Mnemonic&#8221;</a>).
</p>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_semantic_section"></a>7.7. The Semantic Section</h3></div></div></div>
<p>
The final section of a constructor definition is the <span class="emphasis"><em>semantic
section</em></span>. This is a description of how the processor would manipulate
data if it actually executed an instruction that matched the
constructor. From the perspective of a single constructor, the basic
idea is that all the operands for the constructor have been defined in
the bit pattern or disassembly action sections as either specific or
family symbols. In context, all the family symbols map to specific
symbols, and the semantic section uses these and possibly other global
specific symbols in statements that describe the action of the
constructor. All specific symbols have a varnode associated with them,
so within the semantic section, symbols are manipulated as if they
were varnodes.
</p>
<p>
The semantic section for one constructor is surrounded by curly braces
&#8216;{&#8216; and &#8216;}&#8217; and consists of zero or more statements separated by
semicolons &#8216;;&#8217;. Most statements are built up out of C-like syntax,
where the variables are the symbols visible to the constructor. There
is a direct correspondence between each type of operator used in the
statements and a p-code operation. The SLEIGH compiler generates
p-code operations and varnodes corresponding to the SLEIGH operators
and symbols by collapsing the syntax trees represented by the
statements and creating temporary storage within
the <span class="emphasis"><em>unique</em></span> space when it needs to.
</p>
<div class="informalexample"><pre class="programlisting">
:add r1,r2 is opcode=0x26 &amp; r1 &amp; r2 { r1 = r1 + r2; }
</pre></div>
<p>
</p>
<p>
The above example generates exactly one integer addition
operation, <span class="emphasis"><em>INT_ADD</em></span>, where the input varnodes
are <span class="emphasis"><em>r1</em></span> and <span class="emphasis"><em>r2</em></span> and the output
varnode is <span class="emphasis"><em>r1</em></span>.
</p>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920530304"></a>7.7.1. Expressions</h4></div></div></div>
<p>
Expressions are built out of symbols and the binary and unary
operators listed in <a class="xref" href="sleigh_ref.html#syntaxref.htmltable" title="Table 5. Semantic Expression Operators and Syntax">Table 5, &#8220;Semantic Expression Operators and Syntax&#8221;</a> in the
Appendix. All expressions evaluate to an integer, floating point, or
boolean value, depending on the final operation of the expression. The
value is then used depending on the kind of statement. Most of the
operators require that their input and output varnodes all be the same
size (see <a class="xref" href="sleigh_constructors.html#sleigh_varnode_sizes" title="7.7.3. Varnode Sizes">Section 7.7.3, &#8220;Varnode Sizes&#8221;</a>). The operators all
have a precedence, which is used by the SLEIGH compiler to determine
the ordering of the final p-code operations. Parentheses can be used
within expressions to affect this order.
</p>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920527872"></a>7.7.1.1. Arithmetic, Logical and Boolean Operators</h5></div></div></div>
<p>
For the most part these operators should be familiar to software
developers. The only real differences arise from the fact that
varnodes are typeless. So for instance, there has to be separate
operators to distinguish between dividing unsigned numbers &#8216;/&#8217;,
dividing signed numbers &#8216;s/&#8217;, and dividing floating point numbers
&#8216;f/&#8217;.
</p>
<p>
Carry, borrow, and overflow calculations are implemented with separate
operations, rather than having indirect effects with the arithmetic
operations. Thus
the <span class="emphasis"><em>INT_CARRY</em></span>, <span class="emphasis"><em>INT_SCARRY</em></span>,
and <span class="emphasis"><em>INT_SBORROW</em></span> operations may be unfamiliar to
some people in this form (see the descriptions in the Appendix).
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_star_operator"></a>7.7.1.2. The '*' Operator</h5></div></div></div>
<p>
The dereference operator, which generates <span class="emphasis"><em>LOAD</em></span>
operations (and <span class="emphasis"><em>STORE</em></span> operations), has slightly
unfamiliar syntax. The &#8216;*&#8217; operator, as is usual in many programming
languages, indicates that the affected variable is a pointer and that
the expression is <span class="emphasis"><em>dereferencing</em></span> the data being
pointed to. Unlike most languages, in SLEIGH, it is not immediately
clear what address space the variable is pointing into because there
may be multiple address spaces defined. In the absence of any other
information, SLEIGH assumes that the variable points into
the <span class="emphasis"><em>default</em></span> space, as labeled in the definition
of one of the address spaces with
the <span class="bold"><strong>default</strong></span> attribute. If that is not
the space desired, the default can be overridden by putting the
identifier for the space in square brackets immediately after the &#8216;*&#8217;.
</p>
<p>
It is also frequently not clear what the size of the dereferenced data
is because the pointer variable is typeless. The SLEIGH compiler can
frequently deduce what the size must be by looking at the operation in
the context of the entire statement (see
<a class="xref" href="sleigh_constructors.html#sleigh_varnode_sizes" title="7.7.3. Varnode Sizes">Section 7.7.3, &#8220;Varnode Sizes&#8221;</a>). But in some situations, this
may not be possible, so there is a way to specify the size
explicitly. The operator can be followed by a colon &#8216;:&#8217; and an integer
indicating the number of bytes being dereferenced. This can be used
with or without the address space override. We give an example of each
kind of override in the example below.
</p>
<div class="informalexample"><pre class="programlisting">
:load r1,[r2] is opcode=0x99 &amp; r1 &amp; r2 { r1 = * r2; }
:load2 r1,[r2] is opcode=0x9a &amp; r1 &amp; r2 { r1 = *[other] r2; }
:load3 r1,[r2] is opcode=0x9b &amp; r1 &amp; r2 { r1 = *:2 r2; }
:load4 r1,[r2] is opcode=0x9c &amp; r1 &amp; r2 { r1 = *[other]:2 r2; }
</pre></div>
<p>
Keep in mind that the address represented by the pointer is not a byte
address if the <span class="bold"><strong>wordsize</strong></span> attribute is
set to something other than one.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920515552"></a>7.7.1.3. Extension</h5></div></div></div>
<p>
Most processors have instructions that extend small values into big
values, and many instructions do these minor data manipulations
implicitly. In keeping with the p-code philosophy, these operations
must be specified explicitly with the <span class="emphasis"><em>INT_ZEXT</em></span>
and <span class="emphasis"><em>INT_SEXT</em></span> operators in the semantic
section. The <span class="emphasis"><em>INT_ZEXT</em></span>, does a
so-called <span class="emphasis"><em>zero extension</em></span>. The low-order bits are
copied from the input, and any remaining high-order bits in the result
are set to zero. The <span class="emphasis"><em>INT_SEXT</em></span>, does
a <span class="emphasis"><em>signed extension</em></span>. The low-order bits are copied
from the input, but any remaining high-order bits in the result are
set to the value of the high-order bit of the
input. The <span class="emphasis"><em>INT_ZEXT</em></span> operation is invoked with
the <span class="bold"><strong>zext</strong></span> operator, and
the <span class="emphasis"><em>INT_SEXT</em></span> operation is invoked with
the <span class="bold"><strong>sext</strong></span> operator.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920508832"></a>7.7.1.4. Truncation</h5></div></div></div>
<p>
There are two forms of syntax indicating a truncation of the input
varnode. In one the varnode is followed by a colon &#8216;:&#8217; and an integer
indicating the number of bytes to copy into the output, starting with
the least significant byte. In the second form, the varnode is
followed by an integer, surrounded by parentheses, indicating the
number of least significant bytes to truncate from the input. This
second form doesn&#8217;t directly specify the size of the output, which
must be inferred from context.
</p>
<div class="informalexample"><pre class="programlisting">
:split r1,lo,hi is opcode=0x81 &amp; r1 &amp; lo &amp; hi {
lo = r1:4;
hi = r1(4);
}
</pre></div>
<p>
This is an example using both forms of truncation to split a large
value <span class="emphasis"><em>r1</em></span> into two smaller
pieces, <span class="emphasis"><em>lo</em></span>
and <span class="emphasis"><em>hi</em></span>. Assuming <span class="emphasis"><em>r1</em></span> is an 8
byte value, <span class="emphasis"><em>lo</em></span> receives the least significant
half and <span class="emphasis"><em>hi</em></span> receives the most significant half.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_bitrange_operator"></a>7.7.1.5. Bit Range Operator</h5></div></div></div>
<p>
A specific subrange of bits within a varnode can be explicitly
referenced. Depending on the range, this may amount to just a
variation on the truncation syntax described earlier. But for this
operator, the size and boundaries of the range do not have to be
restricted to byte alignment.
</p>
<div class="informalexample"><pre class="programlisting">
:bit3 r1,r2 is op=0x7e &amp; r1 &amp; r2 { r1 = zext(r2[3,1]); }
</pre></div>
<p>
</p>
<p>
A varnode, <span class="emphasis"><em>r2</em></span> in this example, is immediately
followed by square brackets &#8216;[&#8217; and &#8216;]&#8217; indicating a bit range, and
within the brackets, there are two parameters separated by a
comma. The first parameter is an integer indicating the least
significant bit of the resulting bit range. The bits of the varnode
are labeled in order of significance, with the least significant bit
of the varnode being 0. The second parameter is an integer indicating
the number of bits in the range. In the example, a single bit is
extracted from <span class="emphasis"><em>r2</em></span>, and its value is extended to
fill <span class="emphasis"><em>r1</em></span>. Thus <span class="emphasis"><em>r1</em></span> takes
either the value 0 or 1, depending on bit 3
of <span class="emphasis"><em>r2</em></span>.
</p>
<p>
There are some caveats associated with using this operator. Bit range
extraction is really a pseudo operator, as real p-code can only work
with memory down to byte resolution. The bit range operator will
generate some combination
of <span class="emphasis"><em>INT_RIGHT</em></span>, <span class="emphasis"><em>INT_AND</em></span>,
and <span class="emphasis"><em>SUBPIECE</em></span> to simulate the extraction of
smaller or unaligned pieces. The &#8220;r2[3,1]&#8221; from the example generates
the following p-code, for instance.
</p>
<div class="informalexample"><pre class="programlisting">
u1 = INT_RIGHT r2,#3
u2 = SUBPIECE u1,0
u3 = INT_AND u2,#0x1
</pre></div>
<p>
</p>
<p>
The result of any bit range operator still has a size in bytes. This
size is always the minimum number of bytes needed to contain the
resulting bit range, and if there are any extra bits in the result
these are automatically set to zero.
</p>
<p>
This operator can also be used on the left-hand side of assignments
with similar behavior and caveats (see <a class="xref" href="sleigh_constructors.html#sleigh_bitrange_assign" title="7.7.2.7. Bit Range Assignments">Section 7.7.2.7, &#8220;Bit Range Assignments&#8221;</a>).
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_addressof"></a>7.7.1.6. Address-of Operator</h5></div></div></div>
<p>
There is an <span class="emphasis"><em>address-of</em></span> operator for generating
the address offset of a selected varnode as an integer value for use
in expressions. Use of this operator is a little subtle because it
does <span class="emphasis"><em>not</em></span> generate a p-code operation that
calculates the desired value. The address is only calculated at
disassembly time and not during execution. The operator can only be
used if the symbol referenced has a static address.
</p>
<div class="warning" style="margin-left: 0.5in; margin-right: 0.5in;">
<h3 class="title">Warning</h3>
<p> The current SLEIGH compiler cannot distinguish when
the symbol has an address that can always be resolved during
disassembly. So improper use may not be flagged as an error, and the
specification may produce unexpected results.
</p>
</div>
<p>
There &#8216;&amp;&#8217; operator in front of a symbol invokes this function. The
ampersand can also be followed by a colon &#8216;:&#8217; and an integer
explicitly indicating the size of the resulting constant as a varnode.
</p>
<div class="informalexample"><pre class="programlisting">
:copyr r1 is op=0x3b &amp; r1 { tmp:4 = &amp;r1 + 4; r1 = *[register]tmp;}
</pre></div>
<p>
</p>
<p>
The above is a contrived example of using the address-of operator to
copy from a register that is not explicitly indicated by the
instruction. This example constructs the address of the register
following <span class="emphasis"><em>r1</em></span> within
the <span class="emphasis"><em>register</em></span> space, and then
loads <span class="emphasis"><em>r1</em></span> with data from that address. The net
effect of all this is that the register
following <span class="emphasis"><em>r1</em></span> is copied
into <span class="emphasis"><em>r1</em></span>, even though it is not mentioned directly
in the instruction. Notice that the address-of operator only produces
the offset portion of the address, and to copy the desired value, the
&#8216;*&#8217; operator must have a <span class="emphasis"><em>register</em></span> space override.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920484032"></a>7.7.1.7. Managed Code Operations</h5></div></div></div>
<p>
SLEIGH provides basic support for instructions where encoding and context
don't provide a complete description of the semantics. This is the case
typically for <span class="emphasis"><em>managed code</em></span> instruction sets where generation
of the semantic details of an instruction may be deferred until run-time. Support for
these operators is architecture dependent, otherwise they just act as black-box
functions.
</p>
<p>
The constant pool operator, <span class="bold"><strong>cpool</strong></span>,
returns sizes, offsets, addresses, and other structural constants. It behaves like a
<span class="emphasis"><em>query</em></span> to the architecture about these constants. The first
parameter is generally an <span class="emphasis"><em>object reference</em></span>, and additional parameters
are constants describing the particular query. The operator returns the requested value.
In the following example, an object reference
<span class="emphasis"><em>regParamC</em></span> and the encoded constant <span class="emphasis"><em>METHOD_INDEX</em></span>
are sent as part of a query to obtain the final destination address of an object method.
</p>
<div class="informalexample"><pre class="programlisting">
:invoke_direct METHOD_INDEX,regParamC
is inst0=0x70 ; N_PARAMS=1 &amp; METHOD_INDEX &amp; regParamC
{
iv0 = regParamC;
destination:4 = cpool( regParamC, METHOD_INDEX, $(CPOOL_METHOD));
call [ destination ];
}
</pre></div>
<p>
</p>
<p>
If object memory allocation is an atomic feature of the instruction set, the specification
designer can use the <span class="bold"><strong>newobject</strong></span> functional operator to
implement it in SLEIGH. It takes one
or two parameters. The first parameter is a <span class="emphasis"><em>class reference</em></span> or other value
describing the object to be allocated, and the second parameter is an optional count of the number
of objects to allocate. It returns a pointer to the allocated object.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_userdef_op"></a>7.7.1.8. User-Defined Operations</h5></div></div></div>
<p>
Any identifier that has been defined as a new p-code operation, using
the <span class="bold"><strong>define pcodeop</strong></span> statement, can be
invoked as an operator using functional syntax. The SLEIGH compiler
assumes that the operator can take an arbitrary number of inputs, and
if used in an expression, the compiler assumes the operation returns
an output. Using this syntax of course generates the particular p-code
operation reserved for the identifier.
</p>
<div class="informalexample"><pre class="programlisting">
define pcodeop arctan;
<span class="weak">...</span>
:atan r1,r2 is opcode=0xa3 &amp; r1 &amp; r2 { r1 = arctan(r2); }
</pre></div>
<p>
</p>
</div>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920471120"></a>7.7.2. Statements</h4></div></div></div>
<p>
We describe the types of semantic statements that are allowed in SLEIGH.
</p>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_assign_statements"></a>7.7.2.1. Assignment Statements and Temporary Variables</h5></div></div></div>
<p>
Of course SLEIGH allows assignment statements with the &#8216;=&#8217; operator,
where the right-hand side is an arbitrary expression and the left-hand
side is the varnode being assigned. The assigned varnode can be any
specific symbol in the scope of the constructor, either a global
symbol or a local operand.
</p>
<p>
In SLEIGH, the keyword <span class="bold"><strong>local</strong></span>
is used to allocate temporary variables. If an assignment
statement is prepended with <span class="bold"><strong>local</strong></span>,
and the identifier on the left-hand side of an assignment does not match
any symbol in the scope of the constructor, a named temporary varnode is
created in the <span class="emphasis"><em>unique</em></span> address space to hold the
result of the expression. The new symbol becomes part of the local
scope of the constructor, and can be referred to in the following
semantic statements. The size of the new varnode is calculated by
examining the statement in context (see
<a class="xref" href="sleigh_constructors.html#sleigh_varnode_sizes" title="7.7.3. Varnode Sizes">Section 7.7.3, &#8220;Varnode Sizes&#8221;</a>). It is also possible to
explicitly indicate the size by using the colon &#8216;:&#8217; operator followed
by an integer size in bytes. The following examples demonstrate the
temporary variable <span class="emphasis"><em>tmp</em></span> being defined using both
forms.
</p>
<div class="informalexample"><pre class="programlisting">
:swap r1,r2 is opcode=0x41 &amp; r1 &amp; r2 {
local tmp = r1;
r1 = r2;
r2 = tmp;
}
:store r1,imm is opcode=0x42 &amp; r1 &amp; imm {
local tmp:4 = imm+0x20;
*r1 = tmp;
}
</pre></div>
<p>
</p>
<p>
The <span class="bold"><strong>local</strong></span> keyword can also be used
to declare a named temporary varnode, without an assignment statement.
This is useful for temporaries that are immediately passed into a macro.
</p>
<div class="informalexample"><pre class="programlisting">
:pushflags r1 is opcode=0x43 &amp; r1 {
local tmp:4;
packflags(tmp);
* r1 = tmp;
r1 = r1 - 4;
}
</pre></div>
<p>
</p>
<div class="warning" style="margin-left: 0.5in; margin-right: 0.5in;">
<h3 class="title">Warning</h3>
<p>Currently, the SLEIGH compiler does not need the
<span class="bold"><strong>local</strong></span> keyword to create a temporary
variable. For any assignment statement, if the left-hand side has a new
identifier, a new temporary symbol will be created using this identifier.
Unfortunately, this can cause SLEIGH to blindly accept assignment statements
where the left-hand side identifier is a misspelling of an existing symbol.
Use of the <span class="bold"><strong>local</strong></span> keyword is preferred
and may be enforced in future compiler versions.
</p>
</div>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920458176"></a>7.7.2.2. Storage Statements</h5></div></div></div>
<p>
SLEIGH supports fairly standard <span class="emphasis"><em>storage statement</em></span>
syntax to complement the load operator. The left-hand side of an
assignment statement uses the &#8216;*&#8217; operator to indicate a dynamic
storage location, followed by an arbitrary expression to calculate the
location. This syntax of course generates the
p-code <span class="emphasis"><em>STORE</em></span> operator as the final step of the
statement.
</p>
<div class="informalexample"><pre class="programlisting">
:sta [r1],r2 is opcode=0x20 &amp; r1 &amp; r2 { *r1 = r2; }
:stx [r1],r2 is opcode=0x21 &amp; r1 &amp; r2 { *[other] r1 = r2; }
:sti [r1],imm is opcode=0x22 &amp; r1 &amp; imm { *:4 r1 = imm; }
</pre></div>
<p>
</p>
<p>
The same size and address space considerations that apply to the &#8216;*&#8217;
operator when it is used as a load operator also apply when it is used
as a store operator, see
<a class="xref" href="sleigh_constructors.html#sleigh_star_operator" title="7.7.1.2. The '*' Operator">Section 7.7.1.2, &#8220;The '*' Operator&#8221;</a>. Unless explicit modifiers are
given, the default address space is assumed as the storage
destination, and the size of the data being stored is calculated from
context. Keep in mind that the address represented by the pointer is
not a byte address if the <span class="bold"><strong>wordsize</strong></span>
attribute is set to something other than one.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920452240"></a>7.7.2.3. Exports</h5></div></div></div>
<p>
The semantic section doesn&#8217;t just specify how to generate p-code for a
constructor. Except for those constructors in the root table, this
section also associates a semantic meaning to the table symbol the
constructor is part of, allowing the table to be used as an operand in
other tables. The mechanism for making this association is
the <span class="emphasis"><em>export</em></span> statement. This must be the last
statement in the section and consists of
the <span class="bold"><strong>export</strong></span> keyword followed by the
specific symbol to be associated with the constructor. In general, the
constructor will have a sequence of assignment statements building a
final value, and then the varnode containing the value will be
exported. However, anything can be exported.
</p>
<div class="informalexample"><pre class="programlisting">
mode: reg++ is addrmode=0x2 &amp; reg { tmp=reg; reg=reg+1; export tmp; }
</pre></div>
<p>
</p>
<p>
This is an example of a post-increment addressing mode that would be
used to build more complicated instructions. The constructor
increments a register <span class="emphasis"><em>reg</em></span> but stores a copy of its
original value in <span class="emphasis"><em>tmp</em></span>. The
varnode <span class="emphasis"><em>tmp</em></span> is then exported, associating it with
the table symbol <span class="emphasis"><em>mode</em></span>. When this constructor is
matched, as part of a more complicated instruction, the
symbol <span class="emphasis"><em>mode</em></span> will represent the original semantic
value of <span class="emphasis"><em>reg</em></span> but with the standard post-increment
side-effect.
</p>
<p>
The table symbol associated with the constructor becomes
a <span class="emphasis"><em>reference</em></span> to the varnode being exported, not a
copy of the value. If the table symbol is written to, as the left-hand
side of an assignment statement, in some other constructor, the
exported varnode is affected. A constant can be exported if its size
as a varnode is given explicitly with the &#8216;:&#8217; operator.
</p>
<p>
It is not legal to put a full expression in
an <span class="bold"><strong>export</strong></span> statement, any expression
must appear in an earlier statement. However, a single &#8216;&amp;&#8217;
operator is allowed as part of the statement and it behaves as it
would in a normal expression (see
<a class="xref" href="sleigh_constructors.html#sleigh_addressof" title="7.7.1.6. Address-of Operator">Section 7.7.1.6, &#8220;Address-of Operator&#8221;</a>). It causes the address of the
varnode being modified to be exported as an integer constant.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920441008"></a>7.7.2.4. Dynamic References</h5></div></div></div>
<p>
The only other operator allowed as part of
an <span class="bold"><strong>export</strong></span> statement, is the &#8216;*&#8217;
operator. The semantic meaning of this operator is the same as if it
were used in an expression (see
<a class="xref" href="sleigh_constructors.html#sleigh_star_operator" title="7.7.1.2. The '*' Operator">Section 7.7.1.2, &#8220;The '*' Operator&#8221;</a>), but it is worth examining the
effects of this form of export in detail. Bearing in mind that
an <span class="bold"><strong>export</strong></span> statement exports
a <span class="emphasis"><em>reference</em></span>, using the &#8216;*&#8217; operator in the
statement exports a <span class="emphasis"><em>dynamic reference</em></span>. The
varnode being modified by the &#8216;*&#8217; is interpreted as a pointer to
another varnode. It is this varnode being pointed to which is
exported, even though the address may be dynamic and cannot be
determined at disassembly time. This is not the same as dereferencing
the pointer into a temporary variable that is then exported. The
dynamic reference can be both read
and <span class="emphasis"><em>written</em></span>. Internally, the SLEIGH compiler
keeps track of the pointer and inserts a <span class="emphasis"><em>LOAD</em></span>
or <span class="emphasis"><em>STORE</em></span> operation when the symbol associated
with the dynamic reference is referred to in other constructors.
</p>
<div class="informalexample"><pre class="programlisting">
mode: reg[off] is addr=1 &amp; reg &amp; off {
ea = reg + off;
export *:4 ea;
}
dest: reloc is abs [ reloc = abs * 4; ] {
export *[ram]:4 reloc;
}
</pre></div>
<p>
</p>
<p>
In the first example, the effective address of an operand is
calculated from a register <span class="emphasis"><em>reg</em></span> and a field of the
instruction <span class="emphasis"><em>off</em></span>. The constructor does not export
the resulting pointer <span class="emphasis"><em>ea</em></span>, it exports the location
being pointed to by <span class="emphasis"><em>ea</em></span>. Notice the size of this
location (4) is given explicitly with the &#8216;:&#8217; modifier. The &#8216;*&#8217;
operator can also be used on constant pointers. In the second example,
the constant operand <span class="emphasis"><em>reloc</em></span> is used as the offset
portion of an address into the <span class="emphasis"><em>ram</em></span> address
space. The constant <span class="emphasis"><em>reloc</em></span> is calculated at
disassembly time from the instruction
field <span class="emphasis"><em>abs</em></span>. This is a very common construction for
jump destinations (see <a class="xref" href="sleigh_constructors.html#sleigh_relative_branches" title="7.5.1. Relative Branches">Section 7.5.1, &#8220;Relative Branches&#8221;</a>) but
can be used in general. This particular combination of a disassembly
time action and a dynamic export is a very general way to construct a
family of varnodes.
</p>
<p>
Dynamic references are a key construction for effectively separating
addressing mode implementations from instruction semantics at higher
levels.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920427360"></a>7.7.2.5. Branching Statements</h5></div></div></div>
<p>
This section discusses statements that generate p-code branching
operations. These are listed in <a class="xref" href="sleigh_ref.html#branchref.htmltable" title="Table 7. Branching Statements">Table 7, &#8220;Branching Statements&#8221;</a>, in the Appendix.
</p>
<p>
There are six forms covering the gamut of typical assembly language
branches, but in terms of actual semantics there are really only
three. With p-code,
</p>
<div class="informalexample"><div class="itemizedlist"><ul class="itemizedlist compact" style="list-style-type: bullet; ">
<li class="listitem" style="list-style-type: disc">
<span class="emphasis"><em>CALL</em></span> is semantically equivalent to <span class="emphasis"><em>BRANCH</em></span>,
</li>
<li class="listitem" style="list-style-type: disc">
<span class="emphasis"><em>CALLIND</em></span> is semantically equivalent to <span class="emphasis"><em>BRANCHIND</em></span>, and
</li>
<li class="listitem" style="list-style-type: disc">
<span class="emphasis"><em>RETURN</em></span> is semantically equivalent to <span class="emphasis"><em>BRANCHIND</em></span>.
</li>
</ul></div></div>
<p>
The reason for this is that calls and returns imply the presence of
some sort of a stack. Typically an assembly language call instruction
does several separate actions, manipulating a stack pointer, storing a
return value, and so on. When translating the call instruction into
p-code, these actions must be implemented with explicit
operations. The final step of the instruction, the actual jump to the
destination of the call is now just a branch, stripped of its implied
meaning. The <span class="emphasis"><em>CALL</em></span>, <span class="emphasis"><em>CALLIND</em></span>,
and <span class="emphasis"><em>RETURN</em></span> operations, are kept as distinct from
their <span class="emphasis"><em>BRANCH</em></span> counterparts in order to provide
analysis software a hint as to the higher level meaning of the branch.
</p>
<p>
There are actually two fundamentally different ways of indicating a
destination for these branch operations. By far the most common way to
specify a destination is to give the <span class="emphasis"><em>address</em></span> of a
machine instruction. It bears repeating here that there is typically
more than one p-code operation per machine instruction. So specifying
a <span class="emphasis"><em>destination address</em></span> really means that the
destination is the first p-code operation for the (translated) machine
instruction at that address. For most cases, this is the only kind of
branching needed. The rarer case of <span class="emphasis"><em>p-code
relative</em></span> branching is discussed in the following section
(<a class="xref" href="sleigh_constructors.html#sleigh_pcode_relative" title="7.7.2.6. P-code Relative Branching">Section 7.7.2.6, &#8220;P-code Relative Branching&#8221;</a>), but for the remainder of
this section, we assume the destination is ultimately given as an
address.
</p>
<p>
There are two ways to specify a branching operation&#8217;s destination
address; directly and indirectly. Where a direct address is needed, as
for the <span class="emphasis"><em>BRANCH</em></span>, <span class="emphasis"><em>CBRANCH</em></span>,
and <span class="emphasis"><em>CALL</em></span> instructions, The specification can give
the integer offset of the jump destination within the address space of
the current instruction. Optionally, the offset can be followed by the
name of another address space in square brackets, if the destination
is in another address space.
</p>
<div class="informalexample"><pre class="programlisting">
:reset is opcode=0x0 { goto 0x1000; }
:modeshift is opcode=0x1 { goto 0x0[codespace]; }
</pre></div>
<p>
</p>
<p>
Of course, most branching instructions encode the destination of the
jump within the instruction somehow. So the jump destination is almost
always represented by an operand symbol and its associated
varnode. For a direct branch, the destination is given by the address
space and the offset defining the varnode. In this case, the varnode
itself is really just an annotation of the jump destination and not
used as a variable. The best way to define varnodes which annotate
jump destinations in this way is with a dynamic export.
</p>
<div class="informalexample"><pre class="programlisting">
dest: rel is simm8 [ rel = inst_next + simm8*4; ] {
export *[ram]:4 rel;
}
</pre></div>
<p>
</p>
<p>
In this example, the operand <span class="emphasis"><em>rel</em></span> is defined with
a disassembly action in terms of the address of the following
instruction, <span class="emphasis"><em>inst_next</em></span>, and a field specifying a
relative relocation, <span class="emphasis"><em>simm8</em></span>. The resulting
exported varnode has <span class="emphasis"><em>rel</em></span> as its offset
and <span class="emphasis"><em>ram</em></span> as its address space, by virtue of the
dynamic form of the export. The symbol associated with this
varnode, <span class="emphasis"><em>dest</em></span>, can now be used in branch
operations.
</p>
<div class="informalexample"><pre class="programlisting">
:jmp dest is opcode=3 &amp; dest {
goto dest;
}
:call dest is opcode=4 &amp; dest {
*:4 sp = inst_next;
sp=sp-4;
call dest;
}
</pre></div>
<p>
</p>
<p>
The above examples illustrate the direct forms of
the <span class="bold"><strong>goto</strong></span>
and <span class="bold"><strong>call</strong></span> operators, which generate
the p- code <span class="emphasis"><em>BRANCH</em></span> and <span class="emphasis"><em>CALL</em></span>
operations respectively. Both these operations take a single
annotation varnode as input, indicating the destination address of the
jump. Notice the explicit manipulation of a stack
pointer <span class="emphasis"><em>sp</em></span>, for the call
instruction. The <span class="emphasis"><em>CBRANCH</em></span> operation takes two
inputs, a boolean value indicating whether or not the branch should be
taken, and a destination annotation.
</p>
<div class="informalexample"><pre class="programlisting">
:bcc dest is opcode=5 &amp; dest { if (carryflag==0) goto dest; }
</pre></div>
<p>
</p>
<p>
As in the above example, the <span class="emphasis"><em>CBRANCH</em></span> operation
takes two inputs, a boolean value indicating whether or operation is
invoked with the <span class="bold"><strong>if goto</strong></span> operation
takes two inputs, a boolean value indicating whether or syntax. The
condition of the <span class="bold"><strong>if</strong></span> operation takes
two inputs, a boolean value indicating whether or can be any semantic
expression that results in a boolean value. The destination must be an
annotation varnode.
</p>
<p>
The
operators <span class="emphasis"><em>BRANCHIND</em></span>, <span class="emphasis"><em>CALLIND</em></span>,
and <span class="emphasis"><em>RETURN</em></span> all have the same semantic meaning and
all use the same syntax to specify an indirect address.
</p>
<div class="informalexample"><pre class="programlisting">
:b [reg] is opcode=6 &amp; reg {
goto [reg];
}
:call (reg) is opcode=7 &amp; reg {
*:4 sp = inst_next;
sp=sp-4;
call [reg];
}
:ret is opcode=8 {
sp=sp+4;
tmp:4 = * sp;
return [tmp];
}
</pre></div>
<p>
</p>
<p>
Square brackets surround the varnode containing the
address. Currently, any indirect address must be in the address space
containing the branch instruction. The offset of the destination
address is taken dynamically from the varnode. The size of the varnode
must match the size of the destination space.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_pcode_relative"></a>7.7.2.6. P-code Relative Branching</h5></div></div></div>
<p>
In some cases, the semantics of an instruction may require
branching <span class="emphasis"><em>within</em></span> the semantics of a single
instruction, so specifying a destination address is too course. In
this case, SLEIGH is capable of <span class="emphasis"><em>p-code relative</em></span>
branching. Individual p-code operations can be identified by
a <span class="emphasis"><em>label</em></span>, and this label can be used as the
destination specifier, after the <span class="bold"><strong>goto</strong></span>
keyword. A <span class="emphasis"><em>label</em></span>, within the semantic section, is
any identifier surrounded by the &#8216;&lt;&#8217; and &#8216;&gt;&#8217; characters. If this
construction occurs at the beginning of a statement, we say the label
is <span class="emphasis"><em>defined</em></span>, and that identifier is now associated
with the first p-code operation corresponding to the following
statement. Any label must be defined exactly once in this way. When
the construction is used as a destination, immediately after
a <span class="bold"><strong>goto</strong></span>
or <span class="bold"><strong>call</strong></span>, this is referred to as a
label reference. Of course the p-code destination meant by a label
reference is the operation at the point where the label was
defined. Multiple references to the same label are allowed.
</p>
<div class="informalexample"><pre class="programlisting">
:sum r1,r2,r3 is opcode=7 &amp; r1 &amp; r2 &amp; r3 {
tmp:4 = 0;
r1 = 0;
&lt;loopstart&gt;
r1 = r1 + *r2;
r2 = r2 + 4;
tmp = tmp + 1;
if (tmp &lt; r3) goto &lt;loopstart&gt;;
}
</pre></div>
<p>
</p>
<p>
In the example above, the string &#8220;loopstart&#8221; is the label identifier
which appears twice; once at the point where the label is defined at
the top of the loop, after the initialization, and once as a reference
where the conditional branch is made for the loop.
</p>
<p>
References to labels can refer to p-code that occurs either before or
after the branching statement. But label references can only be used
in a branching statement, they cannot be used as a varnode in other
expressions. The label identifiers are local symbols and can only be
referred to within the semantic section of the constructor that
defines them. Branching into the middle of some completely different
instruction is not possible.
</p>
<p>
Internally, branches to labels are encoded as a relative index. Each
p-code operation is assigned an index corresponding to the operation&#8217;s
position within the entire translation of the instruction. Then the
branch can be expressed as a relative offset between the branch
operation&#8217;s index and the destination operation&#8217;s index. The SLEIGH
compiler encodes this offset as a constant varnode that is used as
input to
the <span class="emphasis"><em>BRANCH</em></span>, <span class="emphasis"><em>CBRANCH</em></span>,
or <span class="emphasis"><em>CALL</em></span> operation.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="sleigh_bitrange_assign"></a>7.7.2.7. Bit Range Assignments</h5></div></div></div>
<p>
The bit range operator can appear on the left-hand side of an
assignment. But as with the &#8216;*&#8217; operator, its meaning is slightly
different when used on this side. The bit range is specified in square
brackets, as before, by giving the integer specifying the least
significant bit of the range, followed by the number of bits in the
range. In contrast with its use on the right however (see
<a class="xref" href="sleigh_constructors.html#sleigh_bitrange_operator" title="7.7.1.5. Bit Range Operator">Section 7.7.1.5, &#8220;Bit Range Operator&#8221;</a>), the indicated bit range
is filled rather than extracted. Bits obtained from evaluating the
expression on the right are extracted and spliced into the result at
the indicated bit offset.
</p>
<div class="informalexample"><pre class="programlisting">
:bitset3 r1 is op=0x7d &amp; r1 { r1[3,1] = 1; }
</pre></div>
<p>
In this example, bit 3 of varnode <span class="emphasis"><em>r1</em></span> is set to 1,
leaving all other bits unaffected.
</p>
<p>
As in the right-hand case, the desired insertion is achieved by
piecing together some combination of the p-code
operations <span class="emphasis"><em>INT_LEFT</em></span>, <span class="emphasis"><em>INT_ZEXT</em></span>, <span class="emphasis"><em>INT_AND</em></span>,
and <span class="emphasis"><em>INT_OR</em></span>.
</p>
<p>
In terms of the rest of the assignment expression, the bit range
operator is again assumed to have a size equal to the minimum number
of bytes needed to hold the bit range. In particular, in order to
satisfy size restrictions (see
<a class="xref" href="sleigh_constructors.html#sleigh_varnode_sizes" title="7.7.3. Varnode Sizes">Section 7.7.3, &#8220;Varnode Sizes&#8221;</a>), the right-hand side must
match this size. Furthermore, it is assumed that any extra bits in the
right-hand side expression are already set to zero.
</p>
</div>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_varnode_sizes"></a>7.7.3. Varnode Sizes</h4></div></div></div>
<p>
All statements within the semantic section must be specified up to the
point where the sizes of all varnodes are unambiguously
determined. Most specific symbols, like registers, must have their
size defined by definition, but there are two sources of size
ambiguity.
</p>
<div class="informalexample"><div class="itemizedlist"><ul class="itemizedlist compact" style="list-style-type: bullet; ">
<li class="listitem" style="list-style-type: disc">
Constants
</li>
<li class="listitem" style="list-style-type: disc">
Temporary Variables
</li>
</ul></div></div>
<p>
</p>
<p>
The SLEIGH compiler does not make assumptions about the size of a
constant variable based on the constant value itself. This is true of
values occurring explicitly in the specification and of values that
are calculated dynamically in a disassembly action. As described in
<a class="xref" href="sleigh_constructors.html#sleigh_assign_statements" title="7.7.2.1. Assignment Statements and Temporary Variables">Section 7.7.2.1, &#8220;Assignment Statements and Temporary Variables&#8221;</a>, temporary variables do not
need to have their size given explicitly.
</p>
<p>
The SLEIGH compiler can usually fill in the required size by examining
these situations in the context of the entire semantic section. Most
p-code operations have size restrictions on their inputs and outputs,
which when put together can uniquely determine the unspecified
sizes. Referring to <a class="xref" href="sleigh_ref.html#syntaxref.htmltable" title="Table 5. Semantic Expression Operators and Syntax">Table 5, &#8220;Semantic Expression Operators and Syntax&#8221;</a> in the
Appendix, all arithmetic and logical operations, both integer and
floating point, must have inputs and outputs all of the same size. The
only exceptions are as follows. The overflow
operators, <span class="emphasis"><em>INT_CARRY</em></span>, <span class="emphasis"><em>INT_SCARRY</em></span>, <span class="emphasis"><em>INT_SBORROW</em></span>,
and <span class="emphasis"><em>FLOAT_NAN</em></span> have a boolean output. The shift
operators, <span class="emphasis"><em>INT_LEFT</em></span>, <span class="emphasis"><em>INT_RIGHT</em></span>,
and <span class="emphasis"><em>INT_SRIGHT</em></span>, currently place no restrictions
on the <span class="emphasis"><em>shift amount</em></span> operand. All the comparison
operators, both integer and floating point, insist that their inputs
are all the same size, and the output must be a boolean variable, with
a size of 1 byte.
</p>
<p>
The operators without a size constraint are the load and store
operators, the extension and truncation operators, and the conversion
operators. As discussed in <a class="xref" href="sleigh_constructors.html#sleigh_star_operator" title="7.7.1.2. The '*' Operator">Section 7.7.1.2, &#8220;The '*' Operator&#8221;</a>, the
&#8216;*&#8217; operator cannot get size information for the dynamic (pointed-to)
object from the pointer itself. The other operators by definition
involve a change of size from input to output.
</p>
<p>
If the SLEIGH compiler cannot discover the sizes of constants and
temporaries, it will report an error stating that it could not resolve
variable sizes for that constructor. This can usually be fixed rapidly
by appending the size &#8216;:&#8217; modifier to either the &#8216;*&#8217; operator, the
temporary variable definition, or to an explicit integer. Here are
three examples of statements that generate a size resolution error,
each followed by a variation which corrects the error.
</p>
<div class="informalexample"><pre class="programlisting">
:sta [r1],imm is opcode=0x3a &amp; r1 &amp; imm {
*r1 = imm; #Error
}
:sta [r1],imm is opcode=0x3a &amp; r1 &amp; imm {
*:4 r1 = imm; #Correct
}
:inc [r1] is opcode=0x3b &amp; r1 {
tmp = *r1 + 1; *r1 = tmp; # Error
}
:inc [r1] is opcode=0x3b &amp; r1 {
tmp:4 = *r1 + 1; *r1 = tmp; # Correct
}
:clr [r1] is opcode=0x3c &amp; r1 {
* r1 = 0; # Error
}
:clr [r1] is opcode=0x3c &amp; r1 {
* r1 = 0:4; # Correct
}
</pre></div>
<p>
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920360336"></a>7.7.4. Unimplemented Semantics</h4></div></div></div>
<p>
The semantic section must be present for every constructor in the
specification. But the designer can leave the semantics explicitly
unimplemented if the keyword <span class="bold"><strong>unimpl</strong></span>
is put in the constructor definition in place of the curly
braces. This serves as a placeholder if a specification is still in
development or if the designer does not intend to model data flow for
portions of the instruction set. Any instruction involving a
constructor that is unimplemented in this way will still be
disassembled properly, but the basic data flow routines will report an
error when analyzing the instruction. Analysis routines then can
choose whether or not to intentionally ignore the error, effectively
treating the unimplemented portion of the instruction as if it does
nothing.
</p>
<div class="informalexample"><pre class="programlisting">
:cache r1 is opcode=0x45 &amp; r1 unimpl
:nop is opcode=0x0 { }
</pre></div>
<p>
</p>
</div>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_tables"></a>7.8. Tables</h3></div></div></div>
<p>
A single constructor does not form a new specific
symbol. The <span class="emphasis"><em>table</em></span> that the constructor is
associated with via its table header is the actual symbol that can be
reused to build up more complicated elements. With all the basic
building blocks now in place, we outline the final elements for
building symbols that represent larger and larger portions of the
disassembly and p- code translation process.
</p>
<p>
The best analogy here is with grammar specifications and Regular
Language parsers. Those who have
used <span class="emphasis"><em>yacc</em></span>, <span class="emphasis"><em>bison</em></span>, or
otherwise looked at BNF grammars should find the concepts here
familiar.
</p>
<p>
With SLEIGH, there are in some sense two separate grammars being
parsed at the same time. A display grammar and a semantic grammar. To
the extent that the two grammars breakdown in the same way, SLEIGH can
exploit the similarity to produce an extremely concise description.
</p>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="sleigh_matching"></a>7.8.1. Matching</h4></div></div></div>
<p>
If a table contains exactly one constructor, the meaning of the table
as a specific symbol is straightforward. The display meaning of the
symbol comes from the <span class="emphasis"><em>display section</em></span> of the
constructor, and the symbol&#8217;s semantic meaning comes from the
constructor&#8217;s <span class="emphasis"><em>semantic section</em></span>.
</p>
<div class="informalexample"><pre class="programlisting">
mode1: (r1) is addrmode=1 &amp; r1 { export r1; }
</pre></div>
<p>
</p>
<p>
The table symbol in this example
is <span class="emphasis"><em>mode1</em></span>. Assuming this is the only constructor,
the display meaning of the symbol are the literal characters &#8216;(&#8216;, and
&#8216;)&#8217; concatenated with the display meaning of <span class="emphasis"><em>r1</em></span>,
presumably a register name that has been attached. The semantic
meaning of <span class="emphasis"><em>mode1</em></span>, because of the export
statement, becomes whatever register is matched by
the <span class="emphasis"><em>r1</em></span>.
</p>
<div class="informalexample"><pre class="programlisting">
mode1: (r1) is addrmode=1 &amp; r1 { export r1; }
mode1: [r2] is addrmode=2 &amp; r2 { export r2; }
</pre></div>
<p>
</p>
<p>
If there are two or more constructors defined for the same table,
the <span class="emphasis"><em>bit pattern section</em></span> is used to select between
the constructors in context. In the above example,
the <span class="emphasis"><em>mode1</em></span> table is now defined with two
constructors and the distinguishing feature of their bit patterns is
that in one the <span class="emphasis"><em>addrmode</em></span> field must be 1 and in
the other it must be 2. In the context of a particular instruction,
the matching constructor can be determined uniquely based on this
field, and the <span class="emphasis"><em>mode1</em></span> symbol takes on the display
and semantic characteristics of the matching constructor.
</p>
<p>
The bit patterns for constructors under a single table must be built
so that a constructor can be uniquely determined in context. The above
example shows the easiest way to accomplish this. The two sets of
instruction encodings, which match one or the other of the
two <span class="emphasis"><em>addrmode</em></span> constraints, are disjoint. In
general, if each constructor has a set of instruction encodings
associated with it, and if the sets for any two constructors are
disjoint, then no two constructors can match at the same time.
</p>
<p>
It is possible for two sets to intersect, if one of the two sets
properly contains the other. In this situation, the constructor
corresponding to the smaller (contained) set is considered
a <span class="emphasis"><em>special case</em></span> of the other constructor. If an
instruction encoding matches the special case, that constructor is
used to define the symbol, even though the other constructor will also
match. If the special case does not match but the other more general
constructor does, then the general constructor is used to define the
symbol.
</p>
<div class="informalexample"><pre class="programlisting">
zA: r1 is addrmode=3 &amp; r1 { export r1; }
zA: &#8220;0&#8221; is addrmode=3 &amp; r1=0 { export 0:4; } # Special case
</pre></div>
<p>
</p>
<p>
In this example, the symbol <span class="emphasis"><em>zA</em></span> takes on the same
display and semantic meaning as <span class="emphasis"><em>r1</em></span>, except in the
special case when the field <span class="emphasis"><em>r1</em></span> equals 0. In this
case, <span class="emphasis"><em>zA</em></span> takes on the display and semantic
meaning of the constant zero. Notice that the first constructor has
only the one constraint on <span class="emphasis"><em>addrmode</em></span>, which is
also a constraint for the second constructor. So any instruction that
matches the second must also match the first.
</p>
<p>
The same exact rules apply when there are more than two
constructors. Any two sets defined by the bit patterns must be either
disjoint or one contained in the other. It is entirely possible to
have one general case with many special cases, or a special case of a
special case, and so on.
</p>
<p>
If the patterns for two constructors intersect, but one pattern does
not properly contain the other, this is generally an error in the
specification. Depending on the flags given to the SLEIGH compiler, it
may be more or less lenient with this kind of situation however. In
the case where an intersection is not flagged as an error,
the <span class="emphasis"><em>first</em></span> constructor that matches, in the order
that the constructors appear in the specification, is used.
</p>
<p>
If two constructors intersect, but there is a third constructor whose
pattern is exactly equal to the intersection, then the third pattern
is said to <span class="emphasis"><em>resolve</em></span> the conflict produced by the
first two constructors. An instruction in the intersection will match
the third constructor, as a specialization, and the remaining pieces
in the patterns of the first two constructors are disjoint. A resolved
conflict like this is not flagged as an error even with the strictest
checking. Other types of intersections, in combination with lenient
checking, can be used for various tricks in the specification but
should generally be avoided.
</p>
</div>
<div class="sect3">
<div class="titlepage"><div><div><h4 class="title">
<a name="idm140526920333184"></a>7.8.2. Specific Symbol Trees</h4></div></div></div>
<p>
When the SLEIGH parser analyzes an instruction, it starts with the
root symbol <span class="emphasis"><em>instruction</em></span>, and decides which of the
constructors defined under it match. This particular constructor is
likely to be defined in terms of one or more other family symbols. The
parsing process recurses at this point. Each of the unresolved family
symbols is analyzed in the same way to find the matching specific
symbol. The matching is accomplished either with a table lookup, as
with a field with attached registers, or with the matching algorithm
described in <a class="xref" href="sleigh_constructors.html#sleigh_matching" title="7.8.1. Matching">Section 7.8.1, &#8220;Matching&#8221;</a>. By the end of the
parsing process, we have a tree of specific symbols representing the
parsed instruction. We present a small but complete SLEIGH
specification to illustrate this hierarchy.
</p>
<p>
</p>
<div class="informalexample"><pre class="programlisting">
define endian=big;
define space ram type=ram_space size=4 default;
define space register type=register_space size=4;
define register offset=0 size=4 [ r0 r1 r2 r3 r4 r5 r6 r7 ];
define token instr(16)
op=(10,15) mode=(6,9) reg1=(3,5) reg2=(0,2) imm=(0,2)
;
attach variables [ reg1 reg2 ] [ r0 r1 r2 r3 r4 r5 r6 r7 ];
op2: reg2 is mode=0 &amp; reg2 { export reg2; }
op2: imm is mode=1 &amp; imm { export *[const]:4 imm; }
op2: [reg2] is mode=2 &amp; reg2 { tmp = *:4 reg2; export tmp;}
:and reg1,op2 is op=0x10 &amp; reg1 &amp; op2 { reg1 = reg1 &amp; op2; }
:xor reg1,op2 is op=0x11 &amp; reg1 &amp; op2 { reg1 = reg1 ^ op2; }
:or reg1,op2 is op=0x12 &amp; reg1 &amp; op2 { reg1 = reg1 | op2; }
</pre></div>
<p>
</p>
<p>
This processor has 16 bit instructions. The high order 6 bits are the
main <span class="emphasis"><em>opcode</em></span> field, selecting between logical
operations, <span class="emphasis"><em>and</em></span>, <span class="emphasis"><em>or</em></span>,
and <span class="emphasis"><em>xor</em></span>. The logical operations each take two
operands, <span class="emphasis"><em>reg1</em></span> and <span class="emphasis"><em>op2</em></span>. The
operand <span class="emphasis"><em>reg1</em></span> selects between the 8 registers of
the processor, <span class="emphasis"><em>r0</em></span>
through <span class="emphasis"><em>r7</em></span>. The operand <span class="emphasis"><em>op2</em></span>
is a table built out of more complicated addressing modes, determined
by the field <span class="emphasis"><em>mode</em></span>. The addressing mode can either
be direct, in which <span class="emphasis"><em>op2</em></span> is really just the
register selected by <span class="emphasis"><em>reg2</em></span>, it can be immediate,
in which case the same bits are interpreted as a constant
value <span class="emphasis"><em>imm</em></span>, or it can be an indirect mode, where
the register <span class="emphasis"><em>reg2</em></span> is interpreted as a pointer to
the actual operand. In any case, the two operands are combined by the
logical operation and the result is stored back
in <span class="emphasis"><em>reg1</em></span>.
</p>
<p>
The parsing proceeds from the root symbol down. Once a particular
matching constructor is found, any disassembly action associated with
that constructor is executed. After that, each operand of the
constructor is resolved in turn.
</p>
<div class="figure">
<a name="sleigh_encoding_image"></a><div class="figure-contents"><div class="mediaobject" align="center"><table border="0" summary="manufactured viewport for HTML img" style="cellpadding: 0; cellspacing: 0;" width="100%"><tr><td align="center"><img src="Diagram1.png" align="middle" width="540" height="225" alt="Two Encodings and the Resulting Specific Symbol Trees"></td></tr></table></div></div>
<p class="title"><b>Figure 1. Two Encodings and the Resulting Specific Symbol Trees</b></p>
</div>
<br class="figure-break"><p>
In <a class="xref" href="sleigh_constructors.html#sleigh_encoding_image" title="Figure 1. Two Encodings and the Resulting Specific Symbol Trees">Figure 1, &#8220;Two Encodings and the Resulting Specific Symbol Trees&#8221;</a>, we can see the break down
of two typical instructions in the example instruction set. For each
instruction, we see the how the encodings split into the relevant
fields and the resulting tree of specific symbols. Each node in the
trees are labeled with the base family symbol, the portion of the bit
pattern that matches, and then the resulting specific symbol or
constructor. Notice that the use of the overlapping
fields, <span class="emphasis"><em>reg2</em></span> and <span class="emphasis"><em>imm</em></span>, is
determined by the matching constructor for
the <span class="emphasis"><em>op2</em></span> table. SLEIGH generates the disassembly
and p-code for these encodings by walking the trees.
</p>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920314640"></a>7.8.2.1. Disassembly Trees</h5></div></div></div>
<p>
If the nodes of each tree are replaced with the display information of
the corresponding specific symbol, we see how the disassembly
statement is built.
</p>
<div class="figure">
<a name="sleigh_disassembly_image"></a><div class="figure-contents"><div class="mediaobject" align="center"><table border="0" summary="manufactured viewport for HTML img" style="cellpadding: 0; cellspacing: 0;" width="100%"><tr><td align="center"><img src="Diagram2.png" align="middle" width="310" height="151" alt="Two Disassembly Trees"></td></tr></table></div></div>
<p class="title"><b>Figure 2. Two Disassembly Trees</b></p>
</div>
<br class="figure-break"><p>
<a class="xref" href="sleigh_constructors.html#sleigh_disassembly_image" title="Figure 2. Two Disassembly Trees">Figure 2, &#8220;Two Disassembly Trees&#8221;</a>, shows the resulting
disassembly trees corresponding to the specific symbol trees in
<a class="xref" href="sleigh_constructors.html#sleigh_encoding_image" title="Figure 1. Two Encodings and the Resulting Specific Symbol Trees">Figure 1, &#8220;Two Encodings and the Resulting Specific Symbol Trees&#8221;</a>. The display information comes
from constructor display sections, the names of attached registers, or
the integer interpretation of fields. The identifiers in a constructor
display section serves as placeholders for the subtrees below them. By
walking the tree, SLEIGH obtains the final illustrated assembly
statements corresponding to the original instruction encodings.
</p>
</div>
<div class="sect4">
<div class="titlepage"><div><div><h5 class="title">
<a name="idm140526920308256"></a>7.8.2.2. P-code Trees</h5></div></div></div>
<p>
A similar procedure produces the resulting p-code translation of the
instruction. If each node in the specific symbol tree is replaced with
the corresponding p-code, we see how the final translation is built.
</p>
<div class="figure">
<a name="sleigh_pcode_image"></a><div class="figure-contents"><div class="mediaobject" align="center"><table border="0" summary="manufactured viewport for HTML img" style="cellpadding: 0; cellspacing: 0;" width="100%"><tr><td align="center"><img src="Diagram3.png" align="middle" width="405" height="149" alt="Two P-code Trees"></td></tr></table></div></div>
<p class="title"><b>Figure 3. Two P-code Trees</b></p>
</div>
<br class="figure-break"><p>
<a class="xref" href="sleigh_constructors.html#sleigh_pcode_image" title="Figure 3. Two P-code Trees">Figure 3, &#8220;Two P-code Trees&#8221;</a> lists the final p-code
translation for our example instructions and shows the trees from
which the translation is derived. Symbol names within the p-code for a
particular node, as with the disassembly tree, are placeholders for
the subtree below them. The final translation is put together by
concatenating the p-code from each node, traversing the nodes in a
depth-first order. Thus the p-code of a child tends to come before the
p-code of the parent node (but see
<a class="xref" href="sleigh_constructors.html#sleigh_macros" title="7.9. P-code Macros">Section 7.9, &#8220;P-code Macros&#8221;</a>). Placeholders are filled in with the
appropriate varnode, as determined by the export statement of the root
of the corresponding subtree.
</p>
</div>
</div>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="sleigh_macros"></a>7.9. P-code Macros</h3></div></div></div>
<p>
SLEIGH supports a macro facility for encapsulating semantic
actions. The syntax, in effect, allows the designer to define p-code
subroutines which can be invoked as part of a constructor&#8217;s semantic
action. The subroutine is expanded automatically at compile time.
</p>
<p>
A macro definition is started with
the <span class="bold"><strong>macro</strong></span> keyword, which can occur
anywhere in the file before its first use. This is followed by the
global identifier for the new macro and a parameter list, comma
separated and in parentheses. The body of the definition comes next,
surrounded by curly braces. The body is a sequence of semantic
statements with the same syntax as a constructor&#8217;s semantic
section. The identifiers in the macro&#8217;s parameter list are local in
scope. The macro can refer to these and any global specific symbol.
</p>
<div class="informalexample"><pre class="programlisting">
macro resultflags(op) {
zeroflag = (op == 0);
signflag = (op1 s&lt; 0);
}
:add r1,r2 is opcode=0xba &amp; r1 &amp; r2 { r1 = r1 + r2; resultflags(r1); }
</pre></div>
<p>
</p>
<p>
The macro is invoked in the semantic section of a constructor by using
the identifier with a functional syntax, listing the varnodes which
are to be passed into the macro. In the example above, the
macro <span class="emphasis"><em>resultflags</em></span> calculates the value of two
global flags by comparing its parameter to zero.
The <span class="emphasis"><em>add</em></span> constructor invokes the macro so that
the <span class="emphasis"><em>r1</em></span> is used in the comparisons. Parameters are
passed by <span class="emphasis"><em>reference</em></span>, so the value of varnodes
passed into the macro can be changed. Currently, there is no syntax
for returning a value from the macro, except by writing to a parameter
or global symbol.
</p>
<p>
Almost any statement that can be used in a constructor can also be
used in a macro. This includes assignment statements, branching
statements, <span class="bold"><strong>delayslot</strong></span> directives, and
calls to other macros. A <span class="bold"><strong>build</strong></span>
directive however should not be used in a macro.
</p>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="idm140526920290640"></a>7.10. Build Directives</h3></div></div></div>
<p>
Because the nodes of a specific symbol tree are traversed in a
depth-first order, the p-code for a child node in general comes before
the p-code of the parent. Furthermore, without special intervention,
the specification designer has no control over the order in which the
children of a particular node are
traversed. The <span class="bold"><strong>build</strong></span> directive is
used to affect these issues in the rare cases where it is
necessary. The <span class="bold"><strong>build</strong></span> directive occurs
as another form of statement in the semantic section of a
constructor. The keyword <span class="bold"><strong>build</strong></span> is
followed by one of the constructor&#8217;s operand identifiers. Then,
instead of filling in the operand&#8217;s associated p-code based on an
arbitrary traversal of the symbol tree, the directive specifies that
the operand&#8217;s p-code must occur at that point in the p-code for the
parent constructor.
</p>
<p>
This directive is useful in situations where an instruction supports
prefixes or addressing modes with side-effects that must occur in a
particular order. Suppose for example that many instructions support a
condition bit in their encoding. If the bit is set, then the
instruction is executed only if a status flag is set. Otherwise, the
instruction always executes. This situation can be implemented by
treating the instruction variations as distinct constructors. However,
if many instructions support the same variation, it is probably more
efficient to treat the condition bit which distinguishes the variants
as a special operand.
</p>
<div class="informalexample"><pre class="programlisting">
cc: &#8220;c&#8221; is condition=1 { if (flag==1) goto inst_next; }
cc: is condition=0 { }
:and^cc r1,r2 is opcode=0x67 &amp; cc &amp; r1 &amp; r2 {
build cc;
r1 = r1 &amp; r2;
}
</pre></div>
<p>
</p>
<p>
In this example, the conditional variant is distinguished by a &#8216;c&#8217;
appended to the assembly mnemonic. The <span class="emphasis"><em>cc</em></span> operand
performs the conditional side-effect, checking a flag in one case, or
doing nothing in the other. The two forms of the instruction can now
be implemented with a single constructor. To make sure that the flag
is checked first, before the action of the instruction,
the <span class="emphasis"><em>cc</em></span> operand is forced to evaluate first with
a <span class="bold"><strong>build</strong></span> directive, followed by the
normal action of the instruction.
</p>
</div>
<div class="sect2">
<div class="titlepage"><div><div><h3 class="title">
<a name="idm140526920281024"></a>7.11. Delay Slot Directives</h3></div></div></div>
<p>
For processors with a pipe-lined architecture, multiple instructions
are typically executing simultaneously. This can lead to processor
conventions where certain pairs of instructions do not seem to execute
sequentially. The standard examples are branching instructions that
execute the instruction in the <span class="emphasis"><em>delay
slot</em></span>. Despite the fact that execution of the branch
instruction does not fall through, the following instruction is
executed anyway. Such semantics can be implemented in SLEIGH with
the <span class="bold"><strong>delayslot</strong></span> directive.
</p>
<p>
This directive appears as a standalone statement in the semantic
section of a constructor. When p- code is generated for a matching
instruction, at the point where the directive occurs, p-code for the
following instruction(s) will be generated and inserted. The directive
takes a single integer argument, indicating the minimum number of
bytes in the delay slot. Additional machine instructions will be
parsed and p-code generated, until at least that many bytes have been
disassembled. Typically the value of 1 is used to indicate that there
is exactly one more instruction in the delay slot.
</p>
<div class="informalexample"><pre class="programlisting">
:beq r1,r2,dest is op=0xbc &amp; r1 &amp; r2 &amp; dest { flag=(r1==r2);
delayslot(1);
if flag goto dest; }
</pre></div>
<p>
</p>
<p>
This is an example of a conditional branching instruction with a delay
slot. The p-code for the following instruction is inserted before the
final <span class="emphasis"><em>CBRANCH</em></span>. Notice that
the <span class="bold"><strong>delayslot</strong></span> directive can appear
anywhere in the semantic section. In this example, the condition
governing the branch is evaluated before the directive because the
following instruction could conceivably affect the registers checked
by the condition.
</p>
<p>
Because the <span class="bold"><strong>delayslot</strong></span> directive
combines two or more instructions into one, the meaning of the
symbol <span class="emphasis"><em>inst_next</em></span> becomes ambiguous. It is not
clear anymore what exactly the &#8220;next instruction&#8221; is. SLEIGH uses the
following conventions for interpreting
an <span class="emphasis"><em>inst_next</em></span> symbol. If it is used in the
semantic section, the symbol refers to the address of the instruction
after any instructions in the delay slot. However, if it is used in a
disassembly action, the <span class="emphasis"><em>inst_next</em></span> symbol refers
to the address of the instruction immediately after the first
instruction, even if there is a delay slot.
</p>
</div>
</div>
<div class="navfooter">
<hr>
<table width="100%" summary="Navigation footer">
<tr>
<td width="40%" align="left">
<a accesskey="p" href="sleigh_tokens.html">Prev</a> </td>
<td width="20%" align="center"> </td>
<td width="40%" align="right"> <a accesskey="n" href="sleigh_context.html">Next</a>
</td>
</tr>
<tr>
<td width="40%" align="left" valign="top">6. Tokens and Fields </td>
<td width="20%" align="center"><a accesskey="h" href="sleigh.html">Home</a></td>
<td width="40%" align="right" valign="top"> 8. Using Context</td>
</tr>
</table>
</div>
</body>
</html>