initial working pthreads support from sparc64, done with drahn. there are

some `debugging features' and other optimizations (such as not saving all
registers) that we can still do perhaps

8 files changed, 298 insertions, 20 deletions

diff --git a/lib/libpthread/arch/amd64/_atomic_lock.c b/lib/libpthread/arch/amd64/_atomic_lock.c
index 087186e2858..70bdb487381 100644
--- a/lib/libpthread/arch/amd64/_atomic_lock.c
+++ b/lib/libpthread/arch/amd64/_atomic_lock.c
@@ -1,5 +1,6 @@
+/*	$OpenBSD: _atomic_lock.c,v 1.2 2004/02/25 03:48:36 deraadt Exp $	*/
 /*
- * Atomic lock for amd64
+ * Atomic lock for amd64 -- taken from i386 code.
  */
 
 #include "spinlock.h"
@@ -7,5 +8,16 @@
 int
 _atomic_lock(volatile _spinlock_lock_t *lock)
 {
-	/* dummy for now */
+	_spinlock_lock_t old;
+
+	/*
+	 * Use the eXCHanGe instruction to swap the lock value with
+	 * a local variable containing the locked state.
+	 */
+	old = _SPINLOCK_LOCKED;
+	__asm__("xchg %0,%1"
+		: "=r" (old), "=m" (*lock)
+		: "0"  (old), "1"  (*lock));
+
+	return (old != _SPINLOCK_UNLOCKED);
 }
diff --git a/lib/libpthread/arch/amd64/uthread_machdep.c b/lib/libpthread/arch/amd64/uthread_machdep.c
index d66bfbaa853..e01da1b6c58 100644
--- a/lib/libpthread/arch/amd64/uthread_machdep.c
+++ b/lib/libpthread/arch/amd64/uthread_machdep.c
@@ -2,26 +2,88 @@
 #include <pthread.h>
 #include "pthread_private.h"
 
+struct frame {
+	long	fr_gs;
+	long	fr_fs;
+	long	fr_es;
+	long	fr_ds;
+
+	long    flag;
+	long	fr_r15;
+	long	fr_r14;
+	long	fr_r13;
+	long	fr_r12;
+
+	long	fr_r11;
+	long	fr_r10;
+	long	fr_r9;
+	long	fr_r8;
+
+	long	fr_rdi;
+	long	fr_rsi;
+	long	fr_rbp;
+
+	long	fr_rbx;
+	long	fr_rdx;
+	long	fr_rcx;
+	long	fr_rax;
+
+	long	fr_rip;
+	int	fr_cs;		/* XXX unreachable? */
+	int	pad;
+};
+
+#define copyreg(reg, lval) \
+	__asm__("mov %%" #reg ", %0" : "=g"(lval))
+
 /*
  * Given a stack and an entry function, initialise a state
  * structure that can be later switched to.
  */
 void
 _thread_machdep_init(struct _machdep_state* statep, void *base, int len,
-		     void (*entry)(void))
+    void (*entry)(void))
 {
-	/* dummy */
+        struct frame *f;
+	int foo;
+
+	/* Locate the initial frame, aligned at the top of the stack */
+	f = (struct frame *)(((long)base + len - sizeof *f) & ~ALIGNBYTES);
+
+	copyreg(cs, foo);
+	f->fr_cs = foo;
+	copyreg(ds, foo);
+	f->fr_ds = foo;
+	copyreg(es, foo);
+	f->fr_es = foo;
+	copyreg(fs, foo);
+	f->fr_fs = foo;
+	copyreg(gs, foo);
+	f->fr_gs = foo;
+
+	f->fr_rbp = (long)-1;
+	f->fr_rip = (long)entry;
+
+	statep->rsp = (long)f;
+
+	_thread_machdep_save_float_state(statep);
 }
 
+#define fxsave(addr)		__asm("fxsave %0" : "=m" (*addr))
+#define fxrstor(addr)		__asm("fxrstor %0" : : "m" (*addr))
+#define fwait()			__asm("fwait")
+#define fninit()		__asm("fninit")
+
 void
 _thread_machdep_save_float_state(struct _machdep_state *ms)
 {
-	/* dummy */
+	fxsave(&ms->fpreg);
+	fninit();
+	fwait();
 }
 
 void
 _thread_machdep_restore_float_state(struct _machdep_state *ms)
 {
-	/* dummy */
+	fxrstor(&ms->fpreg);
 }
-
diff --git a/lib/libpthread/arch/amd64/uthread_machdep.h b/lib/libpthread/arch/amd64/uthread_machdep.h
index 46f5e9e87d1..43187585d17 100644
--- a/lib/libpthread/arch/amd64/uthread_machdep.h
+++ b/lib/libpthread/arch/amd64/uthread_machdep.h
@@ -1,5 +1,10 @@
-/* dummy */
+/*	$OpenBSD: uthread_machdep.h,v 1.2 2004/02/25 03:48:36 deraadt Exp $	*/
+
+#include <sys/types.h>
+#include <machine/fpu.h>
 
 struct _machdep_state {
+	long	rsp;
+	/* must be 128-bit aligned */
+	struct savefpu   fpreg __attribute__ ((aligned (16)));
 };
-
diff --git a/lib/libpthread/arch/amd64/uthread_machdep_asm.S b/lib/libpthread/arch/amd64/uthread_machdep_asm.S
index 3a5f255f285..51294f9ae5c 100644
--- a/lib/libpthread/arch/amd64/uthread_machdep_asm.S
+++ b/lib/libpthread/arch/amd64/uthread_machdep_asm.S
@@ -6,5 +6,65 @@
 
 /* void _thread_machdep_switch(new, oldsave); */
 ENTRY(_thread_machdep_switch)
-	/* dummy */
+	pushq	%rax
+	pushq	%rcx
+	pushq	%rdx
+	pushq	%rbx
+
+	pushq	%rbp
+	pushq	%rsi
+	pushq	%rdi
+
+	pushq	%r8
+	pushq	%r9
+	pushq	%r10
+	pushq	%r11
+
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	$0xdeaf12345678beaf, %r15
+	pushq	%r15
+
+	movl	%ds,%eax
+	pushq	%rax
+	movl	%es,%eax
+	pushq	%rax
+	movl	%fs,%eax
+	pushq	%rax
+	movl	%gs,%eax
+	pushq	%rax
+
+	movq	%rsp, 0(%rsi)			/* *arg2 = %rsp */
+	movq	0(%rdi), %rsp			/* %rsp = *arg1 */
+
+	popq	%rax
+	movl	%eax,%gs
+	popq	%rax
+	movl	%eax,%fs
+	popq	%rax
+	movl	%eax,%es
+	popq	%rax
+	movl	%eax,%ds
+
+	popq	%r15 	# flag word.
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+
+	popq	%r11
+	popq	%r10
+	popq	%r9
+	popq	%r8
+
+	popq	%rdi
+	popq	%rsi
+	popq	%rbp
+
+	popq	%rbx
+	popq	%rdx
+	popq	%rcx
+	popq	%rax
 	ret
diff --git a/lib/libpthread/arch/x86_64/_atomic_lock.c b/lib/libpthread/arch/x86_64/_atomic_lock.c
index 087186e2858..70bdb487381 100644
--- a/lib/libpthread/arch/x86_64/_atomic_lock.c
+++ b/lib/libpthread/arch/x86_64/_atomic_lock.c
@@ -1,5 +1,6 @@
+/*	$OpenBSD: _atomic_lock.c,v 1.2 2004/02/25 03:48:36 deraadt Exp $	*/
 /*
- * Atomic lock for amd64
+ * Atomic lock for amd64 -- taken from i386 code.
  */
 
 #include "spinlock.h"
@@ -7,5 +8,16 @@
 int
 _atomic_lock(volatile _spinlock_lock_t *lock)
 {
-	/* dummy for now */
+	_spinlock_lock_t old;
+
+	/*
+	 * Use the eXCHanGe instruction to swap the lock value with
+	 * a local variable containing the locked state.
+	 */
+	old = _SPINLOCK_LOCKED;
+	__asm__("xchg %0,%1"
+		: "=r" (old), "=m" (*lock)
+		: "0"  (old), "1"  (*lock));
+
+	return (old != _SPINLOCK_UNLOCKED);
 }
diff --git a/lib/libpthread/arch/x86_64/uthread_machdep.c b/lib/libpthread/arch/x86_64/uthread_machdep.c
index d66bfbaa853..e01da1b6c58 100644
--- a/lib/libpthread/arch/x86_64/uthread_machdep.c
+++ b/lib/libpthread/arch/x86_64/uthread_machdep.c
@@ -2,26 +2,88 @@
 #include <pthread.h>
 #include "pthread_private.h"
 
+struct frame {
+	long	fr_gs;
+	long	fr_fs;
+	long	fr_es;
+	long	fr_ds;
+
+	long    flag;
+	long	fr_r15;
+	long	fr_r14;
+	long	fr_r13;
+	long	fr_r12;
+
+	long	fr_r11;
+	long	fr_r10;
+	long	fr_r9;
+	long	fr_r8;
+
+	long	fr_rdi;
+	long	fr_rsi;
+	long	fr_rbp;
+
+	long	fr_rbx;
+	long	fr_rdx;
+	long	fr_rcx;
+	long	fr_rax;
+
+	long	fr_rip;
+	int	fr_cs;		/* XXX unreachable? */
+	int	pad;
+};
+
+#define copyreg(reg, lval) \
+	__asm__("mov %%" #reg ", %0" : "=g"(lval))
+
 /*
  * Given a stack and an entry function, initialise a state
  * structure that can be later switched to.
  */
 void
 _thread_machdep_init(struct _machdep_state* statep, void *base, int len,
-		     void (*entry)(void))
+    void (*entry)(void))
 {
-	/* dummy */
+        struct frame *f;
+	int foo;
+
+	/* Locate the initial frame, aligned at the top of the stack */
+	f = (struct frame *)(((long)base + len - sizeof *f) & ~ALIGNBYTES);
+
+	copyreg(cs, foo);
+	f->fr_cs = foo;
+	copyreg(ds, foo);
+	f->fr_ds = foo;
+	copyreg(es, foo);
+	f->fr_es = foo;
+	copyreg(fs, foo);
+	f->fr_fs = foo;
+	copyreg(gs, foo);
+	f->fr_gs = foo;
+
+	f->fr_rbp = (long)-1;
+	f->fr_rip = (long)entry;
+
+	statep->rsp = (long)f;
+
+	_thread_machdep_save_float_state(statep);
 }
 
+#define fxsave(addr)		__asm("fxsave %0" : "=m" (*addr))
+#define fxrstor(addr)		__asm("fxrstor %0" : : "m" (*addr))
+#define fwait()			__asm("fwait")
+#define fninit()		__asm("fninit")
+
 void
 _thread_machdep_save_float_state(struct _machdep_state *ms)
 {
-	/* dummy */
+	fxsave(&ms->fpreg);
+	fninit();
+	fwait();
 }
 
 void
 _thread_machdep_restore_float_state(struct _machdep_state *ms)
 {
-	/* dummy */
+	fxrstor(&ms->fpreg);
 }
-
diff --git a/lib/libpthread/arch/x86_64/uthread_machdep.h b/lib/libpthread/arch/x86_64/uthread_machdep.h
index 46f5e9e87d1..43187585d17 100644
--- a/lib/libpthread/arch/x86_64/uthread_machdep.h
+++ b/lib/libpthread/arch/x86_64/uthread_machdep.h
@@ -1,5 +1,10 @@
-/* dummy */
+/*	$OpenBSD: uthread_machdep.h,v 1.2 2004/02/25 03:48:36 deraadt Exp $	*/
+
+#include <sys/types.h>
+#include <machine/fpu.h>
 
 struct _machdep_state {
+	long	rsp;
+	/* must be 128-bit aligned */
+	struct savefpu   fpreg __attribute__ ((aligned (16)));
 };
-
diff --git a/lib/libpthread/arch/x86_64/uthread_machdep_asm.S b/lib/libpthread/arch/x86_64/uthread_machdep_asm.S
index 3a5f255f285..51294f9ae5c 100644
--- a/lib/libpthread/arch/x86_64/uthread_machdep_asm.S
+++ b/lib/libpthread/arch/x86_64/uthread_machdep_asm.S
@@ -6,5 +6,65 @@
 
 /* void _thread_machdep_switch(new, oldsave); */
 ENTRY(_thread_machdep_switch)
-	/* dummy */
+	pushq	%rax
+	pushq	%rcx
+	pushq	%rdx
+	pushq	%rbx
+
+	pushq	%rbp
+	pushq	%rsi
+	pushq	%rdi
+
+	pushq	%r8
+	pushq	%r9
+	pushq	%r10
+	pushq	%r11
+
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	$0xdeaf12345678beaf, %r15
+	pushq	%r15
+
+	movl	%ds,%eax
+	pushq	%rax
+	movl	%es,%eax
+	pushq	%rax
+	movl	%fs,%eax
+	pushq	%rax
+	movl	%gs,%eax
+	pushq	%rax
+
+	movq	%rsp, 0(%rsi)			/* *arg2 = %rsp */
+	movq	0(%rdi), %rsp			/* %rsp = *arg1 */
+
+	popq	%rax
+	movl	%eax,%gs
+	popq	%rax
+	movl	%eax,%fs
+	popq	%rax
+	movl	%eax,%es
+	popq	%rax
+	movl	%eax,%ds
+
+	popq	%r15 	# flag word.
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+
+	popq	%r11
+	popq	%r10
+	popq	%r9
+	popq	%r8
+
+	popq	%rdi
+	popq	%rsi
+	popq	%rbp
+
+	popq	%rbx
+	popq	%rdx
+	popq	%rcx
+	popq	%rax
 	ret