code review 152142: FreeBSD-specific porting work.

FreeBSD-specific porting work.

cgo/libmach remain unimplemented. However, compilers, runtime,
and packages are 100%. I still need to go through and implement
missing syscalls (at least make sure they're all listed), but
for all shipped functionality, this is done. Ship! ;)

[dho, 2009-11-14 17:50:16]

Hello rsc,

I'd like you to review the following change.

[VenkateshSrinivas, 2009-11-14 19:29:15]

Perhaps uint32 xadd(int *val, int delta) in src/pkg/runtime/freebsd/thread.c
should use XADD instead of CMPXCHG?

[rsc, 2009-11-14 19:57:14]

On 2009/11/14 19:29:15, VenkateshSrinivas wrote:
> Perhaps uint32 xadd(int *val, int delta) in src/pkg/runtime/freebsd/thread.c
> should use XADD instead of CMPXCHG?

Maybe, though I'm trying to keep the amount of assembly to a minimum.

[dho, 2009-11-16 00:24:51]

Hello rsc, VenkateshSrinivas,

I'd like you to review the following change.

[rsc, 2009-11-16 21:26:54]

http://codereview.appspot.com/152142/diff/1006/1007
File src/cmd/6l/asm.c (right):

http://codereview.appspot.com/152142/diff/1006/1007#newcode288
src/cmd/6l/asm.c:288: if(HEADTYPE != 7 && HEADTYPE != 9)
I wonder if we should let -H7 be ELF and
then use $GOOS to distinguish the OS.

http://codereview.appspot.com/152142/diff/1006/1007#newcode321
src/cmd/6l/asm.c:321: if(HEADTYPE == 7)
make this a switch; there will be more.

http://codereview.appspot.com/152142/diff/1006/1013
File src/libmach/freebsd.c (right):

http://codereview.appspot.com/152142/diff/1006/1013#newcode65
src/libmach/freebsd.c:65: // Linux gets very upset if a debugger forgets the
reported state
This is pretty clearly a copy of the Linux file.
I don't know how much really applies to FreeBSD.
I would rather see stub versions of these functions
that return error codes or call sysfatal.

http://codereview.appspot.com/152142/diff/1006/1014
File src/pkg/debug/proc/proc_freebsd.go (right):

http://codereview.appspot.com/152142/diff/1006/1014#newcode9
src/pkg/debug/proc/proc_freebsd.go:9: // Process tracing is not supported on OS
X yet.
s/OS X/FreeBSD/

http://codereview.appspot.com/152142/diff/1006/1014#newcode12
src/pkg/debug/proc/proc_freebsd.go:12: return nil, os.NewError("debug/proc not
implemented on OS X")
s/OS X/FreeBSD/

http://codereview.appspot.com/152142/diff/1006/1016
File src/pkg/net/fd_freebsd.go (right):

http://codereview.appspot.com/152142/diff/1006/1016#newcode43
src/pkg/net/fd_freebsd.go:43: flags := syscall.EV_ADD;
I think you want to add EV_RECEIPT here like in Darwin

http://codereview.appspot.com/152142/diff/1006/1016#newcode74
src/pkg/net/fd_freebsd.go:74: syscall.SetKevent(ev, fd, kmode,
syscall.EV_DELETE);
Same thing: add EV_RECEIPT so that Kevent doesn't block.

http://codereview.appspot.com/152142/diff/1006/1017
File src/pkg/os/dir_freebsd.go (right):

http://codereview.appspot.com/152142/diff/1006/1017#newcode16
src/pkg/os/dir_freebsd.go:16: // Readdirnames reads the contents of the
directory associated with file and
drop comment; already one on dir_darwin and godoc
gets confused when there are two.

http://codereview.appspot.com/152142/diff/1006/1019
File src/pkg/os/sys_freebsd.go (right):

http://codereview.appspot.com/152142/diff/1006/1019#newcode5
src/pkg/os/sys_freebsd.go:5: // Darwin-specific
Delete comment (unnecessary and incorrect)

http://codereview.appspot.com/152142/diff/1006/1023
File src/pkg/runtime/freebsd/amd64/sys.s (right):

http://codereview.appspot.com/152142/diff/1006/1023#newcode6
src/pkg/runtime/freebsd/amd64/sys.s:6: // System calls and other sys.stuff for
AMD64, Darwin
Delete or revise comment.

http://codereview.appspot.com/152142/diff/1006/1023#newcode49
src/pkg/runtime/freebsd/amd64/sys.s:49: MOVQ 0, AX
// crash (not reached)

http://codereview.appspot.com/152142/diff/1006/1027
File src/pkg/runtime/freebsd/thread.c (right):

http://codereview.appspot.com/152142/diff/1006/1027#newcode20
src/pkg/runtime/freebsd/thread.c:20: /*if(u&0xffffffff00000000ULL > 0)
please keep this.  s/>/!=/

http://codereview.appspot.com/152142/diff/1006/1027#newcode21
src/pkg/runtime/freebsd/thread.c:21: throw("top bits set?");*/
throw("runtime/freebsd: semaphore too large")

http://codereview.appspot.com/152142/diff/1006/1027#newcode23
src/pkg/runtime/freebsd/thread.c:23: *sem = (uint32)u&0xffffffffUL;
Cannot use *sem = here.  There's a race.

if(!cas(sem, 0, u)){
    // Someone else filled it in.  Use theirs.
    ksem_close(u);
    return;
}

http://codereview.appspot.com/152142/diff/1006/1027#newcode53
src/pkg/runtime/freebsd/thread.c:53: if(l->sema == 0) initsema(&l->sema);
move if body onto next line

http://codereview.appspot.com/152142/diff/1006/1027#newcode64
src/pkg/runtime/freebsd/thread.c:64: if (m->locks < 0)
through this file, change "if (" to "if(" to match
the prevailing style.

http://codereview.appspot.com/152142/diff/1006/1027#newcode123
src/pkg/runtime/freebsd/thread.c:123: if (fn){} //USED
you can say

USED(fn);  // assume fn is mstart

this is the Plan 9 compiler

http://codereview.appspot.com/152142/diff/1006/1029
File src/pkg/syscall/asm_freebsd_amd64.s (right):

http://codereview.appspot.com/152142/diff/1006/1029#newcode6
src/pkg/syscall/asm_freebsd_amd64.s:6: // System call support for AMD64, Darwin
s/Darwin/FreeBSD/

http://codereview.appspot.com/152142/diff/1006/1030
File src/pkg/syscall/mksysnum_freebsd.sh (right):

http://codereview.appspot.com/152142/diff/1006/1030#newcode7
src/pkg/syscall/mksysnum_freebsd.sh:7: # (for example,
xnu-1228/bsd/kern/syscalls.master).
update example

[dho, 2009-11-17 03:27:53]

Hello rsc, VenkateshSrinivas,

I'd like you to review the following change.

[dho, 2009-11-17 06:08:37]

On 2009/11/17 03:27:53, dho wrote:
> Hello rsc, VenkateshSrinivas,
> 
> I'd like you to review the following change.

Just added a new change that incorporates some files I didn't have on the CL.
This is a complete CL.

[rsc, 2009-11-17 07:46:56]

This looks really great.  Thanks.

A few nits below.  Please fix and hg upload and I'll submit it.

http://codereview.appspot.com/152142/diff/2263/1198
File src/cmd/6l/obj.c (right):

http://codereview.appspot.com/152142/diff/2263/1198#newcode155
src/cmd/6l/obj.c:155: debug['e'] = 1; /* generate names always */
please delete.
this makes the binaries much larger.

http://codereview.appspot.com/152142/diff/2263/1203
File src/libmach/freebsd.c (right):

http://codereview.appspot.com/152142/diff/2263/1203#newcode37
src/libmach/freebsd.c:37: int procthreadpids(int pid, int *p, int np)
break line after int

http://codereview.appspot.com/152142/diff/2263/1212
File src/pkg/runtime/darwin/amd64/rt0.s (right):

http://codereview.appspot.com/152142/diff/2263/1212#newcode9
src/pkg/runtime/darwin/amd64/rt0.s:9: MOVQ	SP,DI
space after comma, please

http://codereview.appspot.com/152142/diff/2263/1220
File src/pkg/runtime/freebsd/thread.c (right):

http://codereview.appspot.com/152142/diff/2263/1220#newcode21
src/pkg/runtime/freebsd/thread.c:21: prints("umtx_wait addr=");
This can be

printf("umtx_wait addr=%p val=%d ret=%d\n", addr, val, ret);

http://codereview.appspot.com/152142/diff/2263/1220#newcode40
src/pkg/runtime/freebsd/thread.c:40: prints("umtx_wake addr=");
printf("umtx_wake addr=%p ret=%d\n", addr, ret);

http://codereview.appspot.com/152142/diff/2263/1220#newcode48
src/pkg/runtime/freebsd/thread.c:48: static void
// See linux.c for comments about algorithm.

http://codereview.appspot.com/152142/diff/2263/1220#newcode139
src/pkg/runtime/freebsd/thread.c:139: USED(fn);
USED(fn);  // thr_start assumes fn == mstart
USED(g);  // thr_start assumes g == m->g0

http://codereview.appspot.com/152142/diff/2263/1221
File src/pkg/runtime/linux/amd64/rt0.s (right):

http://codereview.appspot.com/152142/diff/2263/1221#newcode14
src/pkg/runtime/linux/amd64/rt0.s:14: MOVQ	SP,DI
space after comma, please

[dho, 2009-11-17 13:36:00]

Hello rsc, VenkateshSrinivas,

I'd like you to review the following change.

[rsc, 2009-11-17 16:05:04]

LGTM

Thanks very much.
Will submit soon.

[rsc, 2009-11-17 16:21:00]

*** Submitted as http://code.google.com/p/go/source/detail?r=b23b2cd39e85 ***

FreeBSD-specific porting work.

cgo/libmach remain unimplemented. However, compilers, runtime,
and packages are 100%. I still need to go through and implement
missing syscalls (at least make sure they're all listed), but
for all shipped functionality, this is done. Ship! ;)

R=rsc, VenkateshSrinivas
http://codereview.appspot.com/152142

Committer: Russ Cox <rsc@golang.org>