| [ << ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
This chapter describes the C library usually provided with vbcc.
To execute code compiled by vbcc, a library is needed. It
provides basic interfaces to the underlying operating system or
hardware as well as a set of often used functions.
A big part of the library is portable across all architectures. However, some functions (e.g. for input/output or memory allocation) are naturally dependent on the operating system or hardware. There are several sections in this chapter dealing with different versions of the library.
The library itself often is split into several parts. A startup-code will do useful initializations, like setting up IO, parsing the command line or initializing variables and hardware.
The biggest part of the functions will usually be stored in one library file. The name and format of this file depends on the conventions of the underlying system (e.g. ‘vc.lib’ or ‘libvc.a’).
Often, floating point code (if available) is stored in a different file (e.g. ‘m.lib’ or ‘libm.a’). If floating point is used in an application, it might be necessary to explicitly link with this library (e.g. by specifying ‘-lm’).
In many cases, the include files provide special inline-code or similar optimizations. Therefore, it is recommended to always include the corresponding include file when using a library function. Even if it is not necessary in all cases, it may affect the quality of the generated code.
The library implements the functions specified by ISO9899:1989 as well as a small number of the new functions from ISO9899:1999.
Most parts of this library are public domain. However, for some systems, parts may be under a different license. Please consult the system specific documentation. Usually, linking against this library will not put any restrictions on the created executable unless otherwise mentioned.
Parts of the math library (e.g. transcendental functions) are derived from Sun’s free math library:
* ==================================================== * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. * * Developed at SunPro, a Sun Microsystems, Inc. business. * Permission to use, copy, modify, and distribute this * software is freely granted, provided that this notice * is preserved. * ==================================================== |
The softfloat functions, used by some targets, are derived from John Hauser’s IEC/IEEE Floating-point Artithmetic Package:
This C source file is part of the SoftFloat IEC/IEEE Floating-point Arithmetic Package, Release 2. Written by John R. Hauser. This work was made possible in part by the International Computer Science Institute, located at Suite 600, 1947 Center Street, Berkeley, California 94704. Funding was partially provided by the National Science Foundation under grant MIP-9311980. The original version of this code was written as part of a project to build a fixed-point vector processor in collaboration with the University of California at Berkeley, overseen by Profs. Nelson Morgan and John Wawrzynek. More information is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/ arithmetic/softfloat.html'. THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. Derivative works are acceptable, even for commercial purposes, so long as (1) they include prominent notice that the work is derivative, and (2) they include prominent notice akin to these three paragraphs for those parts of this code that are retained. |
This section describes specifics of the C library for embedded systems.
The startup is usually split into two parts. The first part is done by assembly code that produces the object file ‘lib/startup.o’. This assembly code is usually provided with vbcc and may have to be adapted to the hardware you are using. The key actions that have to be performed by this code are:
It may be necessary to perform some hardware initialization right at the beginning, e.g. to configure the memory system. This has to be modified by the user.
When running code from ROM, some memory sections have to be initialized. Usually, the init-values of initialized variables have to be copied from ROM to the data segment and the values of un-initialized variables have to be cleared in the bss segment. This code is usually provided in the startup code.
The stack pointer has to be set to a suitable memory area.
The startup code
will set the stack pointer to the value of the pointer __stack.
There is a default stack provided in the C library which will be used
unless the application defines its own stack using, for example, the
following code (assuming that the stack grows downwards):
#define STACKSIZE <whatever> static long mystack[STACKSIZE/sizeof(long)]; char *__stack=((char*)mystack)+STACKSIZE; |
__main After all the above initializations have been performed, the function
__main() has to be called. This function is provided by the
library and performs high-level initializations, if necessary (mainly
it calls constructors created by the linker) and will then call the
user main() function. Note that the library
may not work correctly if the user main() function is called
directly from the startup code.
When dynamic memory management is used (e.g. by using the malloc()
function), a heap memory area is needed to allocate memory from. The
malloc() function assumes that __heapptr is a variable pointing
to the beginning of the heap memory and that __heapsize specifies
the size of the heap area in bytes. The library will provide a default heap
memory area that can be replaced by adding, for example, the following file
to the application:
#define HEAPSIZE <whatever> char __heap[HEAPSIZE],*__heapptr=__heap; size_t __heapsize=HEAPSIZE; |
The standard C input/output functions are provided also for embedded systems. Reading/writing to a stream will be directed to void unless the following low-level I/O-functions are provided by the application:
int __open(const char *name,const char *mode); void __close(int h); size_t __read(int h,char *p,size_t l); size_t __write(int h,const char *p,size_t l); |
The __open() function receives a name and a mode string (as in the C
fopen() function) as arguments and has to return a file-descriptor if
it is possible to open this file. The other functions are equivalent to the
corresponding POSIX functions.
Also, stdin, stdout and stderr can be used with the standard
descriptors.
Whether floating point is supported, depends on the target architecture and chip. If it is supported, there will usually be a math-library that has to be linked (using option ‘-lm’) when floating point is used.
Of course, some of the C library functions can not be implemented reasonably on embedded systems. These functions are contained in the library but will always return an error value. Mainly affected are:
Depending on the hardware provided by a system it is possible to implement these functions and add them to the application. In this case, the new functions will be used rather than the default ones returning only error values.
To produce ROM images (e.g. in the form of absolute ELF executables, Intel Hex files or Motorola S-Records), the linker is called with a linker script. This script can be used to join together different sections of the input files and locate them to suitable absolute memory areas. Also, this linker script can be used to set symbols that may be used by the application or the startup code, e.g. addresses of data sections, initialization values or small data pointers.
Code or data that has to reside at special locations can be put into a special
section using the __section attribute. This section can then be
placed at the desired location using the linker script.
Usually, an example linker script will be provided. While this is often not suitable for different chips, it may serve as a starting point.
This section describes specifics of the C library for AmigaOS/68k. The relevant files are ‘startup.o’, ‘minstart.o’, ‘vc.lib’, ‘vcs.lib’, ‘mieee.lib’, ‘mieees.lib’, ‘m881.lib’, ‘m881s.lib’, ‘m040.lib’, ‘m040s.lib’ ‘amiga.lib’, ‘amigas.lib’, ‘auto.lib’ and ‘autos.lib’.
Note that ‘extra.lib’ is no longer part of the vbcc distribution. It was replaced by ’PosixLib’, available on Aminet ‘dev/c/vbcc_PosixLib.lha’, which has a much more comprehensive support for POSIX and Unix functions.
The startup code currently consists of a slightly modified standard Amiga startup (‘startup.o’). The startup code sets up some global variables and initializes stdin, stdout and stderr. The exit code closes all open files and frees all memory. If you link with a math library the startup/exit code will be taken from there if necessary.
Note that you have to link with a math library if you want to use floating point. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only. At the moment there are three math libraries:
This one uses the C= math libraries. The startup code will always open MathIeeeSingBas.library, MathIeeeDoubBas.library and MathIeeeDoubTrans.library. Float return values are passed in d0, double return values are passed in d0/d1. A 68000 is sufficient to use this library. You must not specify ‘-fpu=...’ when you use this library.
This one uses direct FPU instructions and function return values are passed in fp0. You must have a 68020 or higher and an FPU to use this library. You also have to specify ‘-fpu=68881’. Several FPU instructions that have to be emulated on 040/060 may be used.
This one uses only direct FPU instructions that do not have to be emulated on a 040/060. Other functions use the Motorola emulation routines modified by Aki M Laukkanen. It should be used for programs compiled for 040 or 060 with FPU. Return values are passed in fp0.
Depending on the CPU/FPU selected, #including ‘math.h’ will cause inline-code generated for certain math functions.
An application can specify the stack-size needed by defining a variable
__stack (of type size_t) with external linkage, e.g.
size_t __stack=65536; /* 64KB stack-size */ |
The startup code will check whether the stack-size specified is larger than the default stack-size (as set in the shell) and switch to a new stack of appropriate size, if necessary.
If the ‘-stack-check’ option is specified when compiling, the library will check for a stack overflow and abort the program, if the stack overflows. Note, however, that only code compiled with this option will be checked. Calls to libraries which have not been compiled with ‘-stack-check’ or calls to OS function may cause a stack overflow which is not noticed.
Additionally, if ‘-stack-check’ is used, the maximum stack-size
used can be read by querying the external variable __stack_usage.
#include <stdio.h>
extern size_t __stack_usage;
main()
{
do_program();
printf("stack used: %lu\n",(unsigned long)__stack_usage);
}
|
Like above, the stack used by functions not compiled using ‘-stack-check’ or OS functions is ignored.
When using the small data model of the 68k series CPUs, you also have to link with appropriate libraries. Most libraries documented here are also available as small data versions (with an ’s’ attached to the file name). Exceptions are the math libraries.
To compile and link a program using the small data model a command like
vc test.c -o test -sd -lvcs -lamigas |
might be used.
The following list contains some restrictions of this version of the library:
tmpfile()The tmpfile() function always returns an error.
clock()The clock() function always returns -1. This is correct,
according to the C standard, because on AmigaOS it is not possible to
obtain the time used by the calling process.
If you want to write programs that use only Amiga functions and none from vc.lib you can use ‘minstart.o’ instead of ‘startup.o’ and produce smaller executables.
This is only useful for people who know enough about the Amiga shared libraries, the stubs in amiga.lib etc. If you do not know enough about those things better forget minstart at all.
This startup code does not set up all the things needed by vc.lib, so you
must not use most of those functions (string and ctype funtions are ok,
but most other functions - especially I/O and memory handling - must not
be used).
exit() is supplied by minstart and can be used.
The command line is not parsed, but passed to main() as a single
string,
so you can declare main as
int main(char *command) or int main(void).
Also no Amiga libraries are opened (but SysBase ist set up), so you
have to define and open DOSBase yourself if you need it.
If you want to use floating point with the IEEE libraries you have to
define and open MathIeeeSingBas.library, MathIeeeDoubBas.library and
MathIeeeDoubTrans.library (in this order!) and link with mieee.lib (if
compiled for FPU this is not needed).
A hello world using minstart could look like this:
#include <proto/exec.h>
#include <proto/dos.h>
struct DosLibrary *DOSBase;
main()
{
if(DOSBase=(struct DosLibrary *)OpenLibrary("dos.library",0)){
Write(Output(),"Hello, world!\n",14);
CloseLibrary((struct Library *)DOSBase);
}
return 0;
}
|
This can yield an executable of under 300 bytes when compiled with
‘-sc -sd -O2’ and linked with ‘minstart.o’ and amigas.lib
(using vlink - may not work with other linkers).
To write programs accessing AmigaOS (rather than standard C functions
only), a replacement for the original (copyrighted) ‘amiga.lib’
is provided with vbcc. This replacement is adapted to vbcc,
does not cause collisions with some functions (e.g. sprintf)
provided by the original ‘amiga.lib’ and is available in
small data. It is recommended to always use this library rather than
the original version.
Additionally, there are header files (in the ‘proto’- and ‘inline’-subdirectories) which cause inlined calls to Amiga library functions.
Specify ‘-lamiga’ to link with ‘amiga.lib’.
To link with ‘auto.lib’ (or the small data version
‘autos.lib’) specify
the ‘-lauto’ or ‘-lautos’ option to vc.
When you are calling a standard Amiga library function and do not have defined the corresponding library base then the library base as well as code to open/close it will be taken from ‘auto.lib’.
By default, ‘auto.lib’ will try to open any library version. If you need at least a certain version you can define and set a variable _<library-base>Ver with external linkage, e.g. (on file-scope):
int _IntuitionBaseVer = 39; |
Note that your program will abort before reaching main() if one
of the libraries cannot be opened. Also note that the dos.library
will be openened by the standard startup code, not by auto.lib.
This means you have to open dos.library yourself, when linking
with ‘minstart.o’.
This section describes specifics of the C library for PowerUp/PPC. The relevant files are ‘startup.o’, ‘minstart.o’, ‘libvc.a’, ‘libvcs.a’, ‘libm.a’, ‘libms.a’ ‘libamiga.a’, ‘libamigas.a’, ‘libauto.a’ and ‘libautos.a’.
Note that ‘libextra.a’ is no longer part of the vbcc distribution. It was replaced by ’PosixLib’, available on Aminet ‘dev/c/vbcc_PosixLib.lha’, which has a much more comprehensive support for POSIX and Unix functions.
The startup code ‘startup.o’ sets up some global variables and initializes stdin, stdout and stderr. The exit code closes all open files and frees all memory. If you link with a math library the startup/exit code will be taken from there if necessary.
Note that you have to link with a math library if you want to use floating point. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only.
The math library (‘libm.a’) is linked against the floating point library libmoto by Motorola.
Depending on the CPU/FPU selected, #including ‘math.h’ will cause inline-code generated for certain math functions.
Stack-handling is similar to AmigaOS/68k (See section Stack).
The only difference is that stack-swapping cannot be done. If the
default stack-size is less than the stack-size specified with
__stack the program will abort.
When using the small data model of the PPC series CPUs, you also have to link with appropriate libraries. Most libraries documented here are also available as small data versions (with an ’s’ attached to the file name). Exceptions are the math libraries.
To compile and link a program using the small data model a command like
vc test.c -o test -sd -lvcs -lamigas |
might be used.
The following list contains some restrictions of this version of the library:
tmpfile()The tmpfile() function always returns an error.
clock()The clock() function always returns -1. This is correct,
according to the C standard, because on AmigaOS it is not possible to
obtain the time used by the calling process.
The provided minimal startup code (‘minstart.o’) is used similarly like the one for 68k (See section Minimal startup). Only use it if you know what you are doing.
To write programs accessing AmigaOS (rather than standard C functions
only), a replacement for the original (copyrighted) ‘amiga.lib’
is provided with vbcc. This replacement (‘libamiga.a’)
automatically performs a necessary context switch to the 68k to execute
the system call. Furthermore, it is adapted to vbcc,
does not cause collisions with some functions (e.g. sprintf)
provided by the original ‘amiga.lib’ and is available in
small data.
Specify ‘-lamiga’ to link with ‘libamiga.a’.
This library corresponds to the AmigaOS/68k version (See section auto.lib).
This section describes specifics of the C library for WarpOS/PPC. The relevant files are ‘startup.o’, ‘vc.lib’, ‘m.lib’, ‘amiga.lib’ and ‘auto.lib’.
Note that ‘extra.lib’ is no longer part of the vbcc distribution. It was replaced by ’PosixLib’, available on Aminet ‘dev/c/vbcc_PosixLib.lha’, which has a much more comprehensive support for POSIX and Unix functions.
The startup code ‘startup.o’ sets up some global variables and initializes stdin, stdout and stderr. The exit code closes all open files and frees all memory. If you link with a math library the startup/exit code will be taken from there if necessary.
Note that you have to link with a math library if you want to use floating point. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only.
The math library (‘m.lib’) contains functions from Sun’s
portable floating point library. Additionally, there is a
vbcc version of Andreas Heumann’s ‘ppcmath.lib’.
These routines are linked against Motorola’s floating point
routines optimized for PowerPC and therefore are much faster.
To make use of this library, link with ‘ppcmath.lib’ before ‘m.lib’, e.g.
vc test.c -lppcmath -lm |
Depending on the CPU/FPU selected, #including ‘math.h’ will cause inline-code generated for certain math functions.
Stack-handling is similar to AmigaOS/68k (See section Stack).
The following list contains some restrictions of this version of the library:
tmpfile()The tmpfile() function always returns an error.
clock()The clock() function always returns -1. This is correct,
according to the C standard, because on AmigaOS it is not possible to
obtain the time used by the calling process.
To write programs accessing AmigaOS (rather than standard C functions
only), a replacement for the original (copyrighted) ‘amiga.lib’
is provided with vbcc. This replacement
automatically performs a necessary context switch to the 68k to execute
the system call. Furthermore, it is adapted to vbcc,
does not cause collisions with some functions (e.g. sprintf)
provided by the original ‘amiga.lib’ and is available in
small data.
Specify ‘-lamiga’ to link with ‘amiga.lib’.
This library corresponds to the AmigaOS/68k version (See section auto.lib).
This section describes specifics of the C library for MorphOS/PPC. The relevant files are ‘startup.o’, ‘minstart.o’, ‘libvc.a’, ‘libvcs.a’, ‘libm.a’, ‘libms.a’ ‘libamiga.a’, ‘libamigas.a’, ‘libauto.a’ and ‘libautos.a’.
Note that ‘libextra.a’ is no longer part of the vbcc distribution. It was replaced by ’PosixLib’, available on Aminet ‘dev/c/vbcc_PosixLib.lha’, which has a much more comprehensive support for POSIX and Unix functions.
The startup code ‘startup.o’ sets up some global variables and initializes stdin, stdout and stderr. The exit code closes all open files and frees all memory. If you link with a math library the startup/exit code will be taken from there if necessary.
Note that you have to link with a math library if you want to use floating point. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only.
The math library (‘libm.a’) is linked against the floating point library libmoto by Motorola.
Depending on the CPU/FPU selected, #including ‘math.h’ will cause inline-code generated for certain math functions.
Stack-handling is similar to AmigaOS/68k (See section Stack).
When using the small data model of the PPC series CPUs, you also have to link with appropriate libraries. Most libraries documented here are also available as small data versions (with an ’s’ attached to the file name). Exceptions are the math libraries.
To compile and link a program using the small data model a command like
vc test.c -o test -sd -lvcs -lamigas |
might be used.
The following list contains some restrictions of this version of the library:
tmpfile()The tmpfile() function always returns an error.
clock()The clock() function always returns -1. This is correct,
according to the C standard, because on MorphOS it is not possible to
obtain the time used by the calling process.
To write programs using AmigaOS compatible functions, a replacement for
the original (copyrighted) ‘amiga.lib’
is provided with vbcc. This replacement (‘libamiga.a’)
will invoke the MorphOS 68k emulator to execute the system function.
Furthermore, it is adapted to vbcc and
does not cause collisions with some functions (e.g. sprintf)
and is available in small data.
Specify ‘-lamiga’ to link with ‘libamiga.a’.
This library corresponds to the AmigaOS/68k version (See section auto.lib).
This section describes specifics of the C library for AmigaOS4/PPC. The relevant files are ‘startup.o’, ‘minstart.o’, ‘libvc.a’, ‘libvcs.a’, ‘libm.a’, ‘libms.a’ ‘libamiga.a’, ‘libamigas.a’, ‘libauto.a’ and ‘libautos.a’.
Note that ‘libextra.a’ is no longer part of the vbcc distribution. It was replaced by ’PosixLib’, available on Aminet ‘dev/c/vbcc_PosixLib.lha’, which has a much more comprehensive support for POSIX and Unix functions.
The startup code ‘startup.o’ sets up some global variables and initializes stdin, stdout and stderr. Then it runs all constructors of dynamically linked libraries, before entering the main program. The exit code runs all destructors of dynamically linked libraries, closes all open files and frees all memory. If you link with a math library the startup/exit code will be taken from there if necessary.
Note that you have to link with a math library if you want to use floating point. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only.
The math library (‘libm.a’) is linked against the floating point library libmoto by Motorola.
Depending on the CPU/FPU selected, #including ‘math.h’ will cause inline-code generated for certain math functions.
There is no automatic stack extension for AmigaOS 4! This should be done automatically by the operating system.
When using the small data model of the PPC series CPUs, you also have to link with appropriate libraries. Most libraries documented here are also available as small data versions (with an ’s’ attached to the file name). Exceptions are the math libraries.
To compile and link a program using the small data model a command like
vc test.c -o test -sd -lvcs -lamigas |
might be used.
Since ‘elf.library’ V52.2 AmigaOS4 supports dynamic linking with
shared object files (‘.so’ extension), similar to Unix. The default
behaviour is to prefer linking against a shared object over a static
library. To force static linking you might want to give the
‘-static’ option to the ‘vc’ frontend.
The following list contains some restrictions of this version of the library:
tmpfile()The tmpfile() function always returns an error.
clock()The clock() function always returns -1. This is correct,
according to the C standard, because on AmigaOS it is not possible to
obtain the time used by the calling process.
Small data in dynamically linked executablesThere is a bug in ‘elf.library’ V52.4 (and earlier), which
doesn’t load .sdata and .sbss as a contiguous block into
memory, when the executable requires dynamic linking. I decided against
writing a workaround, as the bug should be fixed in OS4.
In contrast to other amigalibs the OS4 ‘libamiga.a’ doesn’t contain any stubs for calling system functions. AmigaOS 4 system calls are done through special macros in the SDK’s interface header files.
The library only includes some remaining amigalib functions, not already
integrated into the OS, like CreateIO(), but its use is discouraged.
Specify ‘-lamiga’ to link with ‘libamiga.a’.
Auto-open -close functions for the following libraries are included:
Asl, CyberGfx, DataTypes, Dos, GadTools, Graphics, Icon, IFFParse,
Intuition, Locale, LowLevel, Picasso96, BSDSocket, Utility, Workbench
Note that gcc’s ‘libauto.a’ doesn’t include CyberGfx.
newlib.library is a shared AmigaOS4 library, which is covered by several BSD like licenses, and includes standard ANSI and POSIX functions as well as some functions common in Unix, BSD and similar operating systems. It is part of the OS4 SDK.
The config file ‘newlib’ will be created on installation to use the paths for header files and libraries pointing to the newlib from the SDK.
What are the main differences between vclib and newlib?
Things you should note:
To compile a program to use newlib for OS4 you must make sure the proper config-file (‘newlib’) is used, e.g.
vc +newlib hello.c |
With a new SDK this will usually generate a dynamically linked executable, which requires ‘libc.so’. To force a statically linked executable:
vc +newlib -static hello.c |
This section describes specifics of the C library for Atari TOS and MiNT. M680x0 processors are supported by the target ‘m68k-atari’, while ColdFire processors are supported by the target ‘cf-atari’. Both share the same startup-code and are based on common library sources and header files. Executables linked with this C library run on plain TOS as well as on MiNT, without modifications.
The relevant files are ‘startup.o’, ‘minstart.o’, ‘libvc.a’, ‘libvc16.a’, ‘libm.a’, ‘libm16.a’.
The following config files are available:
tos M68k 32-bit int
tos16 M68k 16-bit int
toscf ColdFire 32-bit int
The startup code ‘startup.o’ sets up some global variables and initializes stdin, stdout and stderr and returns the unneeded memory to the system. The exit code closes all open files and frees all memory.
Note that you have to link with a math library if you want to use floating point. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only.
At the moment (‘libm.a’) is a soft-float library, which is compatible with all the Atari models without an FPU. There may be real FPU support in future updates.
The default stack size is 64k. There is a MiNT tool called ‘stack’
which can adjust the stack size of an executable to any value, by looking
for a symbol named __stksize (defined by vclib’s startup code).
Additionally the required stack size can be specified by defining a
variable __stack (of type size_t) with external linkage, as
in other vbcc targets.
The default libraries use 32-bit int types, but you may want to
use 16-bit int types for compatibility reasons. In this case you
have to specify the config file tos16 and link with the appropriate
16-bit libraries (which have a ’‘16’’ attached to their name).
To compile and link a program using 16-bit integers a command like
vc +tos16 test.c -o test -lm16 -lvc16 |
may be used. There are no 16-bit versions for ColdFire targets, because this is strictly a 32-bit CPU.
The following list contains some restrictions of this version of the library:
tmpfile()The tmpfile() function always returns an error.
clock()The clock() function always returns -1. This is correct,
according to the C standard, because neither under TOS nor under MiNT it
is possible to obtain the time used by the calling process.
This section describes specifics of the C library for VideoCore under Linux.
The relevant files are vcload, ‘startup.o’,
‘libvc.a’, ‘libm.a’, ‘libms.a’.
The config file vc4-linux is part of the library.
The startup code ‘startup.o’ sets up stack and heap and provides
a function __armcall() to transfer control to the loader on
the ARM side.
The startup process calls constructors to set up some
global variables and initialize stdin, stdout and stderr if needed.
Note that you have to link with a math library if you want to use floating point operations that are not natively implemented. All math functions, special startup code and printf/scanf functions which support floating point are contained in the math libraries only.
The library contains a default stack of 32KB. If another size is needed, you can add the following to your project:
.align 4
.space <desired-size, suitably aligned>
___stackend:
.global ___stackend
|
Currently, a global variable of 16KB is used to get memory for malloc() etc. If another size is needed, you can add the following to your project:
#define HEAPSIZE <desired size> char __heap[HEAPSIZE],*__heapptr=__heap; size_t __heapsize=HEAPSIZE; |
Note that this mechanism will likely be changed in the future!
To access system functions from the VideoCore-side, the function
__armcall() can be used. It will save the current context and return
to the loader. Registers r0-r5 (the function arguments) will be saved
and are available to the loader. The loader can then execute the system
call and resume execution, passing the return value of the system
function.
Resuming is done by calling the image with offset 2.
This functionality can also be used for debugging purposes.
A loader is required to execute VideoCore code from the ARM side. For standalone VideoCore code, the provided loader can be used. Usually, it will be necessary to adapt the loader to communicate between ARM and VideoCore side during runtime.
Currently, the loader loads an simple binary image that must be pc-relative and located to address 0x00000000. Additionally, if present, a file with extension ‘.reltext’ will be loaded for some limited relocation. This file contains a 32bit word containing the number of relocations followed by n 32bit words containing an offset. For each offset, the address will be relocated to the image load address.
vcload [-debug] [-cache] [-offset] <image-name>
The loader currently has the following options:
-debugThe loader will enter debug mode (see below).
-cacheThe loader will set the LSB in the start address when executing code. This is supposed to inhibit a cache flush.
Just for testing!
-offsetThe loader will allocate 1 KB more memory than required and leaves this space unused at the beginning of the allocated memory.
Just for testing!
In debug mode, the loader will wait for user input before starting the
VideoCore code as well as after every __armcall.
The following commands are available:
w <addr> [<num>]Display <num> 32bit words starting at <addr>. <addr> must be the offset into the image. If <num> is omitted, one unit is displayed.
If one word is displayed, it is additionally displayed translated as an offset into the image.
h <addr> [<num>]Display <num> 16bit halfwords starting at <addr>. <addr> must be the offset into the image. If <num> is omitted, one unit is displayed.
b <addr> [<num>]Display <num> 8bit bytes starting at <addr>. <addr> must be the offset into the image. If <num> is omitted, one unit is displayed.
cStart/continue execution.
qQuit.
bp <addr>Set a breakpoint at <addr>.
This is currently a very crude implementation. It will just write
a branch to __armcall() to <addr>. If everything works well,
you will end in the debugger if <addr> is reached. However, the
arguments passed are random (and might be dangerous syscalls by
accident). Also, the old code at this address is currently not
restored.
As a result, you must not continue execution after hitting a breakpoint!
The following list contains some restrictions of this version of the library:
| [ << ] | [ >> ] | [Top] | [Contents] | [Index] | [ ? ] |
This document was generated by vb on January 3, 2015 using texi2html 1.82.