wbrawner/openssl
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
openssl/Configurations/README
Richard Levitte 9fe2bb77c4 unified build scheme: a first introduction 
The "unified" build scheme revolves around small information files,
build.info, which each describe their own bit of everything that needs
to be built, using a mini-language described in Configurations/README.

The information in build.info file contain references to source files
and final result.  Object files are not mentioned at all, they are
simply from source files.  Because of this, all the *_obj items in
Configurations/*.conf are renamed to *_asm_src and the files listed
in the values are change from object files to their corresponding
source files.  For the sake of the other build schemes, Configure
generates corresponding *_obj entries in %target.

Furthermore, the "unified" build scheme supports having a build
directory tree separate from the source directry tree.

All paths in a build.info file is assumed to be relative to its
location, either within the source tree or within the build tree.

Reviewed-by: Andy Polyakov <appro@openssl.org>
2016-02-01 12:46:58 +01:00

407 lines
18 KiB
Text
Raw Blame History

Configurations of OpenSSL target platforms
------------------------------------------
Target configurations are a collection of facts that we know about
different platforms and their capabilities. We organise them in a
hash table, where each entry represent a specific target.
In each table entry, the following keys are significant:
inherit_from => Other targets to inherit values from.
Explained further below. [1]
template => Set to 1 if this isn't really a platform
target. Instead, this target is a template
upon which other targets can be built.
Explained further below. [1]
sys_id => System identity for systems where that
is difficult to determine automatically.
cc => The compiler command, usually one of "cc",
"gcc" or "clang". This command is normally
also used to link object files and
libraries into the final program.
cflags => Flags that are used at all times when
compiling.
debug_cflags => Extra compilation flags used when making a
debug build (when Configure receives the
--debug option). Typically something like
"-g -O0".
release_cflags => Extra compilation flags used when making a
release build (when Configure receives the
--release option, or doesn't receive the
--debug option). Typically something like
"-O" or "-O3".
thread_cflags => Extra compilation flags used when
compiling with threading enabled.
Explained further below. [2]
shared_cflag => Extra compilation flags used when
compiling for shared libraries, typically
something like "-fPIC".
ld => the linker command, usually not defined
(meaning the compiler command is used
instead).
(NOTE: this is here for future use, it's
not implemented yet)
lflags => the flags that are used at all times when
linking. These can have a % sign in them
showing where the OpenSSL libraries should
appear, otherwise these flags will come
last. So in a typical links situation,
this is a quick table of results:
"-foo%-bar" > -foo -lssl -lcrypto -bar
"-foo%" > -foo -lssl -lcrypto
"-foo" > -lssl -lcrypto -foo
debug_lflags => Like debug_cflags, but used when linking.
release_lflags => Like release_cflags, but used when linking.
shared_lflags => Like shared_cflags, but used when linking.
ar => The library archive command, the default is
"ar".
(NOTE: this is here for future use, it's
not implemented yet)
arflags => Flags to be used with the library archive
command.
ranlib => The library archive indexing command, the
default is 'ranlib' it it exists.
unistd => An alternative header to the typical
'<unistd.h>'. This is very rarely needed.
shared_extension => File name extension used for shared
libraries.
obj_extension => File name extension used for object files.
On unix, this defaults to ".o" (NOTE: this
is here for future use, it's not
implemented yet)
exe_extension => File name extension used for executable
files. On unix, this defaults to "" (NOTE:
this is here for future use, it's not
implemented yet)
dso_scheme => The type of dynamic shared objects to build
for. This mostly comes into play with
engines, but can be used for other purposes
as well. Valid values are "DLFCN"
(dlopen() et al), "DLFCN_NO_H" (for systems
that use dlopen() et al but do not have
fcntl.h), "DL" (shl_load() et al), "WIN32"
and "VMS".
perlasm_scheme => The perlasm method used to created the
assembler files used when compiling with
assembler implementations.
shared_target => The shared library building method used.
This is a target found in Makefile.shared.
build_scheme => The scheme used to build up a Makefile.
In its simplest form, the value is a string
with the name of the build scheme.
The value may also take the form of a list
of strings, if the build_scheme is to have
some options. In this case, the first
string in the list is the name of the build
scheme.
Currently recognised build schemes are
"mk1mf" and "unixmake" and "unified".
For the "unified" build scheme, this item
*must* be an array with the first being the
word "unified" and the second being a word
to identify the platform family.
multilib => On systems that support having multiple
implementations of a library (typically a
32-bit and a 64-bit variant), this is used
to have the different variants in different
directories.
bn_ops => Building options (was just bignum options
in the earlier history of this option,
hence the name). This a string of words
that describe properties on the designated
target platform, such as the type of
integers used to build up the bitnum,
different ways to implement certain ciphers
and so on. To fully comprehend the
meaning, the best is to read the affected
source.
The valid words are:
BN_LLONG use 'unsigned long long' in
some bignum calculations.
This has no value when
SIXTY_FOUR_BIT or
SIXTY_FOUR_BIT_LONG is given.
RC4_CHAR makes the basic RC4 unit of
calculation an unsigned char.
SIXTY_FOUR_BIT processor registers
are 64 bits, long is
32 bits, long long is
64 bits.
SIXTY_FOUR_BIT_LONG processor registers
are 64 bits, long is
64 bits.
THIRTY_TWO_BIT processor registers
are 32 bits.
EXPORT_VAR_AS_FN for shared libraries,
export vars as
accessor functions.
cpuid_asm_src => assembler implementation of cpuid code as
well as OPENSSL_cleanse().
Default to mem_clr.c
bn_asm_src => Assembler implementation of core bignum
functions.
Defaults to bn_asm.c
ec_asm_src => Assembler implementation of core EC
functions.
des_asm_src => Assembler implementation of core DES
encryption functions.
Defaults to 'des_enc.c fcrypt_b.c'
aes_asm_src => Assembler implementation of core AES
functions.
Defaults to 'aes_core.c aes_cbc.c'
bf_asm_src => Assembler implementation of core BlowFish
functions.
Defaults to 'bf_enc.c'
md5_asm_src => Assembler implementation of core MD5
functions.
sha1_asm_src => Assembler implementation of core SHA1,
functions, and also possibly SHA256 and
SHA512 ones.
cast_asm_src => Assembler implementation of core CAST
functions.
Defaults to 'c_enc.c'
rc4_asm_src => Assembler implementation of core RC4
functions.
Defaults to 'rc4_enc.c rc4_skey.c'
rmd160_asm_src => Assembler implementation of core RMD160
functions.
rc5_asm_src => Assembler implementation of core RC5
functions.
Defaults to 'rc5_enc.c'
wp_asm_src => Assembler implementation of core WHIRLPOOL
functions.
cmll_asm_src => Assembler implementation of core CAMELLIA
functions.
Defaults to 'camellia.c cmll_misc.c cmll_cbc.c'
modes_asm_src => Assembler implementation of cipher modes,
currently the functions gcm_gmult_4bit and
gcm_ghash_4bit.
padlock_asm_src => Assembler implementation of core parts of
the padlock engine. This is mandatory on
any platform where the padlock engine might
actually be built.
[1] as part of the target configuration, one can have a key called
'inherit_from' that indicate what other configurations to inherit
data from. These are resolved recursively.
Inheritance works as a set of default values that can be overriden
by corresponding key values in the inheriting configuration.
Note 1: any configuration table can be used as a template.
Note 2: pure templates have the attribute 'template => 1' and
cannot be used as build targets.
If several configurations are given in the 'inherit_from' array,
the values of same attribute are concatenated with space
separation. With this, it's possible to have several smaller
templates for different configuration aspects that can be combined
into a complete configuration.
instead of a scalar value or an array, a value can be a code block
of the form 'sub { /* your code here */ }'. This code block will
be called with the list of inherited values for that key as
arguments. In fact, the concatenation of strings is really done
by using 'sub { join(" ",@_) }' on the list of inherited values.
An example:
"foo" => {
template => 1,
haha => "ha ha",
hoho => "ho",
ignored => "This should not appear in the end result",
},
"bar" => {
template => 1,
haha => "ah",
hoho => "haho",
hehe => "hehe"
},
"laughter" => {
inherit_from => [ "foo", "bar" ],
hehe => sub { join(" ",(@_,"!!!")) },
ignored => "",
}
The entry for "laughter" will become as follows after processing:
"laughter" => {
haha => "ha ha ah",
hoho => "ho haho",
hehe => "hehe !!!",
ignored => ""
}
[2] OpenSSL is built with threading capabilities unless the user
specifies 'no-threads'. The value of the key 'thread_cflags' may
be "(unknown)", in which case the user MUST give some compilation
flags to Configure.
Historically, the target configurations came in form of a string with
values separated by colons. This use is deprecated. The string form
looked like this:
"target" => "{cc}:{cflags}:{unistd}:{thread_cflag}:{sys_id}:{lflags}:{bn_ops}:{cpuid_obj}:{bn_obj}:{ec_obj}:{des_obj}:{aes_obj}:{bf_obj}:{md5_obj}:{sha1_obj}:{cast_obj}:{rc4_obj}:{rmd160_obj}:{rc5_obj}:{wp_obj}:{cmll_obj}:{modes_obj}:{padlock_obj}:{perlasm_scheme}:{dso_scheme}:{shared_target}:{shared_cflag}:{shared_ldflag}:{shared_extension}:{ranlib}:{arflags}:{multilib}"
Build info files
================
The build.info files that are spread over the source tree contain the
minimum information needed to build and distribute OpenSSL. It uses a
simple and yet fairly powerful language to determine what needs to be
built, from what sources, and other relationships between files.
For every build.info file, all file references are relative to the
directory of the build.info file for source files, and the
corresponding build directory for built files if the build tree
differs from the source tree.
When processed, every line is processed with the perl module
Text::Template, using the delimiters "{-" and "-}". The hashes
%config and %target are passed to the perl fragments, along with
$sourcedir and $builddir, which are the locations of the source
directory for the current build.info file and the corresponding build
directory, all relative to the top of the build tree.
To begin with, things to be built are declared by setting specific
variables:
PROGRAMS=foo bar
LIBS=libsomething
ENGINES=libeng
SCRIPTS=myhack
EXTRA=file1 file2
Note that the files mentioned for PROGRAMS, LIBS and ENGINES *must* be
without extensions. The build file templates will figure them out.
For each thing to be built, it is then possible to say what sources
they are built from:
PROGRAMS=foo bar
SOURCE[foo]=foo.c common.c
SOURCE[bar]=bar.c extra.c common.c
It's also possible to tell some other dependencies:
DEPEND[foo]=libsomething
DEPEND[libbar]=libsomethingelse
(it could be argued that 'libsomething' and 'libsomethingelse' are
source as well. However, the files given through SOURCE are expected
to be located in the source tree while files given through DEPEND are
expected to be located in the build tree)
For some libraries, we maintain files with public symbols and their
slot in a transfer vector (important on some platforms). It can be
declared like this:
ORDINALS[libcrypto]=crypto
The value is not the name of the file in question, but rather the
argument to util/mkdef.pl that indicates which file to use.
One some platforms, shared libraries come with a name that's different
from their static counterpart. That's declared as follows:
SHARED_NAME[libfoo]=cygfoo-{- $config{shlibver} -}
The example is from Cygwin, which has a required naming convention.
Sometimes, it makes sense to rename an output file, for example a
library:
RENAME[libfoo]=libbar
That lines has "libfoo" get renamed to "libbar". While it makes no
sense at all to just have a rename like that (why not just use
"libbar" everywhere?), it does make sense when it can be used
conditionally. See a little further below for an example.
For any file to be built, it's also possible to tell what extra
include paths the build of their source files should use:
INCLUDE[foo]=include
It's possible to have raw build file lines, between BEGINRAW and
ENDRAW lines as follows:
BEGINRAW[Makefile(unix)]
haha.h: {- $builddir -}/Makefile
echo "/* haha */" > haha.h
ENDRAW[Makefile(unix)]
The word withing square brackets is the build_file configuration item
or the build_file configuration item followed by the second word in the
build_scheme configuration item for the configured target within
parenthesis as shown above. For example, with the following relevant
configuration items:
build_file => "build.ninja"
build_scheme => [ "unified", "unix" ]
... these lines will be considered:
BEGINRAW[build.ninja]
build haha.h: echo "/* haha */" > haha.h
ENDRAW[build.ninja]
BEGINRAW[build.ninja(unix)]
build hoho.h: echo "/* hoho */" > hoho.h
ENDRAW[build.ninja(unix)]
See the documentation further up for more information on configuration
items.
Finally, you can have some simple conditional use of the build.info
information, looking like this:
IF[1]
something
ELSIF[2]
something other
ELSE
something else
ENDIF
The expression in square brackets is interpreted as a string in perl,
and will be seen as true if perl thinks it is, otherwise false. For
example, the above would have "something" used, since 1 is true.
Together with the use of Text::Template, this can be used as
conditions based on something in the passed variables, for example:
IF[{- $config{no_shared} -}]
LIBS=libcrypto
SOURCE[libcrypto]=...
ELSE
LIBS=libfoo
SOURCE[libfoo]=...
ENDIF
or:
# VMS has a cultural standard where all libraries are prefixed.
# For OpenSSL, the choice is 'ossl_'
IF[{- $config{target} =~ /^vms/ -}]
RENAME[libcrypto]=ossl_libcrypto
RENAME[libssl]=ossl_libssl
ENDIF
Reference in a new issue View git blame Copy permalink