1 |
1026 |
ivang |
@c
|
2 |
|
|
@c COPYRIGHT (c) 1988-2002.
|
3 |
|
|
@c On-Line Applications Research Corporation (OAR).
|
4 |
|
|
@c All rights reserved.
|
5 |
|
|
@c
|
6 |
|
|
@c callconv.t,v 1.7 2002/07/31 00:14:11 joel Exp
|
7 |
|
|
@c
|
8 |
|
|
|
9 |
|
|
@chapter Calling Conventions
|
10 |
|
|
|
11 |
|
|
@section Introduction
|
12 |
|
|
|
13 |
|
|
Each high-level language compiler generates
|
14 |
|
|
subroutine entry and exit code based upon a set of rules known
|
15 |
|
|
as the compiler's calling convention. These rules address the
|
16 |
|
|
following issues:
|
17 |
|
|
|
18 |
|
|
@itemize @bullet
|
19 |
|
|
@item register preservation and usage
|
20 |
|
|
|
21 |
|
|
@item parameter passing
|
22 |
|
|
|
23 |
|
|
@item call and return mechanism
|
24 |
|
|
@end itemize
|
25 |
|
|
|
26 |
|
|
A compiler's calling convention is of importance when
|
27 |
|
|
interfacing to subroutines written in another language either
|
28 |
|
|
assembly or high-level. Even when the high-level language and
|
29 |
|
|
target processor are the same, different compilers may use
|
30 |
|
|
different calling conventions. As a result, calling conventions
|
31 |
|
|
are both processor and compiler dependent.
|
32 |
|
|
|
33 |
|
|
@section Processor Background
|
34 |
|
|
|
35 |
|
|
All members of the i960 architecture family support
|
36 |
|
|
two methods for performing procedure calls: a RISC-style
|
37 |
|
|
branch-and-link and an integrated call and return mechanism.
|
38 |
|
|
|
39 |
|
|
On a branch-and-link, the processor branches to the
|
40 |
|
|
invoked procedure and saves the return address in a register,
|
41 |
|
|
@code{G14}. Typically, the invoked procedure will not invoke another
|
42 |
|
|
procedure and is referred to as a leaf procedure. Many
|
43 |
|
|
high-level language compilers for the i960 family recognize leaf
|
44 |
|
|
procedures and automatically optimize them to utilize the
|
45 |
|
|
branch-and-link mechanism. Branch-and-link procedures are
|
46 |
|
|
invoked using the @code{bal} and @code{balx} instructions and return control
|
47 |
|
|
via the @code{bx} instruction. By convention, @code{G14} is zero when not in
|
48 |
|
|
a leaf procedure. It is the responsibility of the leaf
|
49 |
|
|
procedure to clear @code{G14} before returning.
|
50 |
|
|
|
51 |
|
|
The integrated call and return mechanism also
|
52 |
|
|
branches to the invoked procedure and saves the return address
|
53 |
|
|
as did the branch and link mechanism. However, the important
|
54 |
|
|
difference is that the @code{call}, @code{callx}, and @code{calls} instructions save
|
55 |
|
|
the local register set (@code{R0} through @code{R15}) before transferring
|
56 |
|
|
control to the invoked procedure. The @code{ret} instruction
|
57 |
|
|
automatically restores the previous local register set. The
|
58 |
|
|
i960CA provides a register cache which can be configured to
|
59 |
|
|
retain the last five to sixteen recent register caches. When
|
60 |
|
|
the register cache is full, the oldest cached register set is
|
61 |
|
|
written to the stack.
|
62 |
|
|
|
63 |
|
|
@section Calling Mechanism
|
64 |
|
|
|
65 |
|
|
All RTEMS directives are invoked using either a @code{call}
|
66 |
|
|
or @code{callx} instruction and return to the user via the @code{ret}
|
67 |
|
|
instruction.
|
68 |
|
|
|
69 |
|
|
@section Register Usage
|
70 |
|
|
|
71 |
|
|
As discussed above, the @code{call} and @code{callx} instructions
|
72 |
|
|
automatically save the current contents of the local register
|
73 |
|
|
set (@code{R0} through @code{R15}). The contents of the local registers will
|
74 |
|
|
be restored as part of returning to the application. The
|
75 |
|
|
contents of global registers @code{G0} through @code{G7} are not preserved by
|
76 |
|
|
RTEMS directives.
|
77 |
|
|
|
78 |
|
|
@section Parameter Passing
|
79 |
|
|
|
80 |
|
|
RTEMS uses the standard i960 family C parameter
|
81 |
|
|
passing mechanism in which @code{G0} contains the first parameter, @code{G1}
|
82 |
|
|
the second, and so on for the remaining parameters. No RTEMS
|
83 |
|
|
directive requires more than six parameters.
|
84 |
|
|
|
85 |
|
|
@section User-Provided Routines
|
86 |
|
|
|
87 |
|
|
All user-provided routines invoked by RTEMS, such as
|
88 |
|
|
user extensions, device drivers, and MPCI routines, must also
|
89 |
|
|
adhere to these calling conventions.
|
90 |
|
|
|
91 |
|
|
@section Leaf Procedures
|
92 |
|
|
|
93 |
|
|
RTEMS utilizes leaf procedures internally to improve
|
94 |
|
|
performance. This improves execution speed as well as reducing
|
95 |
|
|
stack usage and the number of register sets which must be cached.
|
96 |
|
|
|
97 |
|
|
|