for space. -Os could do this, but it was easy to hack an MD version.
This saves a whole 32 bytes in boot2, so I think it is worth using it.
(keep how much worse gcc 3.2 will compile boot2...)
Submitted by: bde (minus gcc 3.2 commentary)
The changes from GCC 2.95.3 to 3.1 are large and it is hard to figure
out how to apply our changes. Once BDE can properly test GCC 3.1 he'll
add back in our changes.
Remove our definition for FUNCTION_PROFILER as it is wrong.
Note that "jsr $28,_mcount" is a macro for
ldq $27, _mcount($29) !literal!1
jsr $28, ($27), _mcount !lituse_jsr!1
1. The call to _mcount is added by alpha_expand_prologue after we load the gp.
Our _mcount uses $27 for the incoming address, unlike OSF/1 and Linux,
which use $28. This probably doesn't matter since we probably don't use $27
within _mcount itself.
2. You can't use this insn with _mcount because it uses the PLT, which clobbers
the return address in $28. Note that the prologue_mcount pattern carefully
avoids adding the lituse_jsr relocation so that we call through the GOT
directly.
Submitted by: Richard Henderson <rth@redhat.com>
When rtld runs the .fini section in a shared lib (C++), the code in
question from .../contrib/gdb/config/alpha/crtbegin.asm first calls
__do_globals_dtors_aux and then __do_frame_takedown. Unfortunately, the
value of gp after a jsr is undefined and in this case had changed from before
the call, probably as a result of calling code in some other shared library.
The normal calling convention for alpha is to re-initialize gp using
'ldgp gp,0(ra)' after a jsr instruction but in this case no such
re-initialization is done. This leads to a bogus value being read for the
address of __do_frame_takedown and a quick segfault.
Submitted by: dfr
Obtained from: GCC 3.0
If "WANT_DWARF2_UNWIND" you get it. Note that this is a different C++ ABI
than the FreeBSD default. So you will need to recompile all your C++ apps
that uses exceptions if you turn this on. I am adding it here for Yahoo!'s
use and for those that want to be early adopters to what I will make as the
default with the GCC 3.0 import.
header to reduce the difference of our sources to the stock GNU/FSF ones.
While the mix binary format support was nice to have in the FreeBSD MI
header as a frame work, it just clutters up too much and makes the FreeBSD
MI header more different from the FSF/GNU stock one than it needs to be.
into the i386 MD FreeBSD header. Also comment tweaking, continuation line
style changes, and other minor changes to make this closer to the latest
FSF/GNU 3.0 sources (to reduce my headache when 3.0 is imported).
linked in addition to libc rather than instead of libc.
Ideally, "-pthread" would now be equivalent to adding "-lc_r" to the
end of the link command. But it is slightly different in this
implementation. Adding "-lc_r" to the link command would produce a
"ld" command line containing this:
... -lc_r /usr/lib/libgcc.a -lc /usr/lib/libgcc.a ...
but this implementation of the "-pthread" option produces this:
... /usr/lib/libgcc.a -lc_r -lc /usr/lib/libgcc.a ...
It would be possible to make them identical, but that doesn't fit
as nicely into GCC's way of doing things. I don't think the ordering
change will make any difference in practice.
This option depended on bits not part of the base system and required
people to install the LinuxThreads port in a manner non-consistent with
the workings of our Ports Collection.
The directions for properly linking with LinuxThreads are given by that
port at install time.
Requested by: jasone
to generate the special .type and .size directives which are used to set
the corresponding fields of the linker symbol table entries in the ELF
object file. As such they are not i386-specific and thus belong in our
MI header. Otherwise on the Alpha we don't properly give the type and
size of dynamic symbols. Bintuil versions past 2.9.1 warn of this and
w/o this change, `ld' generates a lot of warnings during a `make world'.
"The problem is that egcs/gcc-2.95's reorganisation of the prologue and
epilogue code to use rtl instead of output_asm_insn() completely broke our
hooks. rtl is emitted in a different order, only after optimisation, while
output_asm_insn() is emitted immediately. rtl is presumably used so that
the prologue and epilogue can be optimised.
I couldn't find any good examples to copy. gcc's own
FUNCTION_BLOCK_PROFILER still uses output_asm_insn() and seems to be
completely broken. One of the XXX comments points to this.
IIRC, the hacks here basically arrange to emit magic label names; then when
the magic names are output, they are transformed into prologue and epilogue
code."
Submitted by: bde
* Commonalities with i386, moved to a FreeBSD wide config file.
* Major additions to this file were adapted from Hidetoshi Shimokawa
<simokawa@FreeBSD.ORG>'s Alpha suppliment to the Egcs 1.1.2 port.
* be consistant on protecting "#define FOO"s with "#undef FOO".
* be consistant that macro params are upper case, and commas are followed
by a space in the macro definition
* protect macro param expansion by ()'s
* break long lines
* line continuations chars to consistant column
* remove trailing spaces
FreeBSD wide config file.
This includes "WORD_SWITCH_TAKES_ARG" & "SWITCH_TAKES_ARG". Platforms
such as FreeBSD/MIPS will need to override these two defintions, but it
can #undef them and define them approapiately.
* Support for our dual ELF/a.out building ability
* Our custom ASM_* definitions
* Our custom debugger and profiling related definitions
* Our custom STARTFILE/ENDFILE specs
* The stock EGCS 1.1.2 freebsd-elf.h file depended on
egcs-1.1.2/gcc/config/linux.h, which included "svr4.h". We will
include "svr4.h" via our "tm.h" definition. So add the few bits
from "linux.h" we actually needed.
* Using our current crtbegin.o/crtend.o we cannot support the DWARF2
unwinding mechanisms. In the future we will switch to the
non-sjlj-exceptions type exception machanism. However the `make world'
bootstrap problems with the EGCS crtstuff.c must be overcome first.
* Our a.out gas doesn't "know" to use NOP's for aligns while in the text
section. Thus the a.out alignment generation needed tweaking from
what we did with GCC 2.7.2. [from BDE]
* The definition of SUPPORTS_ONE_ONLY prevents the compiler from trying
to use "linkonce" sections for a.out. The definition of
NO_DBX_FUNCTION_END we had causes it to avoid .stabs symbols that the
assembler cannot handle for a.out. [from JDP]
* The previous "EXCEPTION_SECTION_FUNCTION" is the wrong name for EGCS.
It also needed tweaking for EGCS. [from JDP]
Also bump __FreeBSD_cc_version to 400002 in case we need to know we are
using EGCS at some point.
* Consistantly put spaces after "," in macro param lists
* Consistantly align continuation characters.
* Don't need to supply all variations of __FOO__ in CPP_PREDEFINES,
gcc will do that for us.
Our malloc can allocte pagesized blocks efficiently and the EGCS default size
of 4072 bytes is not optimal.
Submitted by: Hidetoshi Shimokawa <simokawa@sat.t.u-tokyo.ac.jp>
in libstdc++.
Until I have a chance to look at what that problem is and to carefully consider
the upgrade issues of turning it back on at a later date if we leave it turned
off for any extended peroid of time.
While I have yet to hear of any problems with us using thunks. The EGCS
mailing list notes some have problems with it and not using them are a
safer default. People wanting to use them, can set the appropiate
compiler flag.
* Turn on DEFAULT_VTABLE_THUNKS. (it is the default anyway, I'm just being
explicit about it, in case it causes us trouble it might be easier for
someone to notice it this way)
with a numeric value that describes the feature level of the
compiler. This can be used to check for the presence/absence of
FreeBSD-specific compiler features. The value is a decimal number
whose digits have the form VRRRRFF, where:
V = Compiler vendor. 0 (elided) means gcc.
RRRR = Vendor's version number, e.g., 2721 for the current
gcc version (2.7.2.1).
FF = FreeBSD-specific revision level. 00 means the stock
compiler from the vendor.
The value of "__FreeBSD_cc_version" is hard-coded in
"src/contrib/gcc/config/i386/freebsd.h" and must be incremented
when new FreeBSD-specific compiler features are added. I considered
simply picking up the value of FreeBSD_version from <osreldate.h>.
But that would break cross compiles of gcc.
PR: Part of the fix for gnu/8452
Suggested by: bde
the executable file, so it will work for both a.out and ELF format
files. I have split the object format specific code into separate
source files. It's cleaner than it was before, but it's still
pretty crufty.
Don't cheat on your make world for this update. A lot of things
have to be rebuilt for it to work, including the compiler and all
of the profiled libraries.
aout and elf support. freebsd-elf.h died as a result, so the BINFORMAT
test for elf in src/gnu/usr.bin/cc/cc_tool/Makefile will need to
use freebsd.h, not freebsd-elf.h. That means that alpha will have to
go the same way. The new alpha/freebsd.h has the alpha/freebsd-elf.h
definitions merged in.
i386-elf because that is what will be used for FreeBSD/Alpha.
Change the STARTFILE_SPEC to match the non-aout version of i386 so
that the csu files can be built in exactly the same way as i386-elf.
This means that FreeBSD/Alpha departs from NetBSD/Alpha which uses
crt0 and crtbegin/crtend. Since i386-aout uses crt0, I guess it was
decided that i386-elf should use crt1. i386-elf also references
crti and with this change, so does FreeBAS/Alpha.
I think it is important for FreeBSD to have a consistent implementation
across architectures and since ELF is quite compatible (ignoring
differences in 32 and 64 bits), gcc might as well be configured the
same.
Another change is that the entry symbol is _start, not __start as
in NetBSD.
a.out gas and the binutils gas (elf or a.out) with a single compiler.
This uses other infrastructure not yet committed, in order to support
both a.out and elf it needs to be able to get to both a.out and elf
gas, ld, libs, crt* etc. So for now, the support is pretty much dormant.
The new freebsd.h file is based on the old freebsd-elf.h file (which has a
long lineage, right back through linux and svr4 files). The change is
pretty dramatic from a gcc internals standpoint as it overrides a lot of
definitions in order to generate different output based on target mode.
There is potential for screw-ups, so please be on the lookout - gcc's
configuration mechanism wasn't really meant for this kind of thing.
It's believed to compile world etc just fine under both a.out and elf, can
handle global constructors and destructors, handles the differences in
a.out and elf stabs, and what sections things like exceptions go in.
The initial idea came from i386/osfrose.h which is a dual rose/elf format
target. These two are not as diverse as a.out and elf it would seem.
The cc front-end uses external configuration to determine default object
format (still being thrashed out, so read the source if you want to see
it so far), and has a '-aout' and '-elf' override command line switch.
There are some other internal switches that can be accessed, namely -maout,
-mno-aout, -munderscores and -mnounderscores. The underscore and local
symbol prefixing rules are controllable seperately to the output format.
(ie: it's possible to generate a.out without the _ prefixes on symbols and
also to generate elf with the _ prefixes. This isn't quite optimal, but
does seem to work pretty well, except the linkers don't always recognise
the local symbols without their normal names)
The default format is a.out (still), nobody should see any major changes.
With both elf and a.out tools and libraries installed:
[1:26pm]/tmp-223> cc -elf -o hello hello.c
peter@beast[1:27pm]/tmp-224> file hello
hello: ELF 32-bit LSB executable, Intel 80386, version 1 (FreeBSD), dynamically linked, not stripped
[1:27pm]/tmp-225> ./hello
hello world!
[1:27pm]/tmp-226> cc -aout -o hello hello.c
[1:27pm]/tmp-227> file hello
hello: FreeBSD/i386 compact demand paged dynamically linked executable not stripped
1:27pm]/tmp-228> ./hello
hello world!
Since my co-conspirators put a lot of effort into this too, I'll add them
so they can share the blame^H^H^H^H^Hglory. :-)
Reviewed by: sos, jdp
See freebsd.h and freebsd-elf.h for the silly comment that cgd@netbsd.org
wanted me to add about his claim that uncommented source files that
have been publicly available for ftp for nearly a year; that are
configuration patches to a GPL'ed program; are owned by his previous
employer who refuses to release them. Well... I did as he said. As if
that makes a difference!
At this point we've got cpp, gcc, g++ ported to FreeBSD/Alpha so all
the code that uses __FreeBSD__ is correctly pre-processed. Yay.
I'll commit the bootstrap makefile next to let others play, then on
to libc.
it can be built via BINFORMAT=elf in the environment. Most likely
some of the directory defines such as STANDARD_EXEC_PREFIX will
change again soon, as we settle on the proper locations for the
various components.
Note, the build still fails when it tries to compile libgcc2.c
using the ELF compiler, unless arrangements have been made for the
compiler to find the ELF assembler instead of the a.out assembler.
"-pg" and gprof(1) instead. FreeBSD does not support plain "-p" or
prof(1).
Plain "-p" is still allowed when just compiling. In the compile
phase, "-p" is identical "-pg". It is used by <bsd.lib.mk> for
building profiled object files.
Without this, compiled programs die with FP errors.
This is originally credited to: jlemon@netcom.com (Jonathan Lemon), and
has been forwarded to me by quite a few of people.
The symptom was an assembler warning
"GOT relocation burb: `___EXCEPTION_TABLE__' should be global"
followed (sometimes) by a core dump. The fix makes the compiler
generate the correct GOTOFF addressing for that symbol, rather than the
GOT addressing it was emitting before.
Warning: There is still at least one serious bug in the i386 exception
code for PIC. The exception code that is generated clobbers the GOT
register (%ebx) and then tries to use it later. That leads to core
dumps at program execution time. I know where the problem is, but I do
not have a fix for it at this time. Until it is fixed, exceptions will
not work in PIC code. This is a general problem for all i386 platforms;
it is not specific to FreeBSD.
non-i386, non-unix, and generatable files have been trimmed, but can easily
be added in later if needed.
gcc-2.7.2.1 will follow shortly, it's a very small delta to this and it's
handy to have both available for reference for such little cost.
The freebsd-specific changes will then be committed, and once the dust has
settled, the bmakefiles will be committed to use this code.
