diff --git a/sys/amd64/amd64/genassym.c b/sys/amd64/amd64/genassym.c
index 47856455f099..92915b6a116d 100644
--- a/sys/amd64/amd64/genassym.c
+++ b/sys/amd64/amd64/genassym.c
@@ -157,6 +157,9 @@ ASSYM(SIGF_HANDLER, offsetof(struct sigframe, sf_ahu.sf_handler));
 ASSYM(SIGF_SC, offsetof(struct osigframe, sf_siginfo.si_sc));
 #endif
 ASSYM(SIGF_UC, offsetof(struct sigframe, sf_uc));
+#ifdef COMPAT_FREEBSD4
+ASSYM(SIGF_UC4, offsetof(struct sigframe4, sf_uc));
+#endif
 #ifdef COMPAT_43
 ASSYM(SC_PS, offsetof(struct osigcontext, sc_ps));
 ASSYM(SC_FS, offsetof(struct osigcontext, sc_fs));
diff --git a/sys/amd64/amd64/locore.S b/sys/amd64/amd64/locore.S
index 8f1d801268b3..19a6ac7117c1 100644
--- a/sys/amd64/amd64/locore.S
+++ b/sys/amd64/amd64/locore.S
@@ -394,32 +394,36 @@ begin:
  * Signal trampoline, copied to top of user stack
  */
 NON_GPROF_ENTRY(sigcode)
-	call	*SIGF_HANDLER(%esp)	/* call signal handler */
-	lea	SIGF_UC(%esp),%eax	/* get ucontext_t */
+	calll	*SIGF_HANDLER(%esp)
+	leal	SIGF_UC(%esp),%eax	/* get ucontext */
 	pushl	%eax
 	testl	$PSL_VM,UC_EFLAGS(%eax)
-	jne	9f
+	jne	1f
 	movl	UC_GS(%eax),%gs		/* restore %gs */
-9:
+1:
 	movl	$SYS_sigreturn,%eax
 	pushl	%eax			/* junk to fake return addr. */
 	int	$0x80			/* enter kernel with args */
-0:	jmp	0b
+					/* on stack */
+1:
+	jmp	1b
 
 #ifdef COMPAT_FREEBSD4
 	ALIGN_TEXT
 freebsd4_sigcode:
-	call	*SIGF_HANDLER(%esp)	/* call signal handler */
-	lea	SIGF_UC(%esp),%eax	/* get ucontext_t */
+	calll	*SIGF_HANDLER(%esp)
+	leal	SIGF_UC4(%esp),%eax	/* get ucontext */
 	pushl	%eax
 	testl	$PSL_VM,UC4_EFLAGS(%eax)
-	jne	9f
+	jne	1f
 	movl	UC4_GS(%eax),%gs	/* restore %gs */
-9:
+1:
 	movl	$344,%eax		/* 4.x SYS_sigreturn */
 	pushl	%eax			/* junk to fake return addr. */
 	int	$0x80			/* enter kernel with args */
-0:	jmp	0b
+					/* on stack */
+1:
+	jmp	1b
 #endif
 
 #ifdef COMPAT_43
diff --git a/sys/amd64/amd64/locore.s b/sys/amd64/amd64/locore.s
index 8f1d801268b3..19a6ac7117c1 100644
--- a/sys/amd64/amd64/locore.s
+++ b/sys/amd64/amd64/locore.s
@@ -394,32 +394,36 @@ begin:
  * Signal trampoline, copied to top of user stack
  */
 NON_GPROF_ENTRY(sigcode)
-	call	*SIGF_HANDLER(%esp)	/* call signal handler */
-	lea	SIGF_UC(%esp),%eax	/* get ucontext_t */
+	calll	*SIGF_HANDLER(%esp)
+	leal	SIGF_UC(%esp),%eax	/* get ucontext */
 	pushl	%eax
 	testl	$PSL_VM,UC_EFLAGS(%eax)
-	jne	9f
+	jne	1f
 	movl	UC_GS(%eax),%gs		/* restore %gs */
-9:
+1:
 	movl	$SYS_sigreturn,%eax
 	pushl	%eax			/* junk to fake return addr. */
 	int	$0x80			/* enter kernel with args */
-0:	jmp	0b
+					/* on stack */
+1:
+	jmp	1b
 
 #ifdef COMPAT_FREEBSD4
 	ALIGN_TEXT
 freebsd4_sigcode:
-	call	*SIGF_HANDLER(%esp)	/* call signal handler */
-	lea	SIGF_UC(%esp),%eax	/* get ucontext_t */
+	calll	*SIGF_HANDLER(%esp)
+	leal	SIGF_UC4(%esp),%eax	/* get ucontext */
 	pushl	%eax
 	testl	$PSL_VM,UC4_EFLAGS(%eax)
-	jne	9f
+	jne	1f
 	movl	UC4_GS(%eax),%gs	/* restore %gs */
-9:
+1:
 	movl	$344,%eax		/* 4.x SYS_sigreturn */
 	pushl	%eax			/* junk to fake return addr. */
 	int	$0x80			/* enter kernel with args */
-0:	jmp	0b
+					/* on stack */
+1:
+	jmp	1b
 #endif
 
 #ifdef COMPAT_43
diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c
index 1aedc3db1f25..27b3bcac7b5e 100644
--- a/sys/amd64/amd64/machdep.c
+++ b/sys/amd64/amd64/machdep.c
@@ -506,6 +506,7 @@ sendsig(catcher, sig, mask, code)
 	struct proc *p;
 	struct thread *td;
 	struct sigacts *psp;
+	char *sp;
 	struct trapframe *regs;
 	int oonstack;
 
@@ -544,13 +545,15 @@ sendsig(catcher, sig, mask, code)
 	/* Allocate space for the signal handler context. */
 	if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
-		sfp = (struct sigframe *)(p->p_sigstk.ss_sp +
-		    p->p_sigstk.ss_size - sizeof(struct sigframe));
+		sp = p->p_sigstk.ss_sp +
+		    p->p_sigstk.ss_size - sizeof(struct sigframe);
 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
 		p->p_sigstk.ss_flags |= SS_ONSTACK;
 #endif
 	} else
-		sfp = (struct sigframe *)regs->tf_esp - 1;
+		sp = (char *)regs->tf_esp - sizeof(struct sigframe);
+	/* Align to 16 bytes. */
+	sfp = (struct sigframe *)((unsigned int)sp & ~0xF);
 	PROC_UNLOCK(p);
 
 	/* Translate the signal if appropriate. */
diff --git a/sys/amd64/include/signal.h b/sys/amd64/include/signal.h
index efc7ef045888..fb2d82c4d489 100644
--- a/sys/amd64/include/signal.h
+++ b/sys/amd64/include/signal.h
@@ -100,7 +100,7 @@ struct osigcontext {
 struct sigcontext {
 	struct __sigset sc_mask;	/* signal mask to restore */
 	int	sc_onstack;		/* sigstack state to restore */
-	int	sc_gs;			/* machine state (struct trapframe): */
+	int	sc_gs;			/* machine state (struct trapframe) */
 	int	sc_fs;
 	int	sc_es;
 	int	sc_ds;
@@ -127,7 +127,7 @@ struct sigcontext {
 	int	sc_fpformat;
 	int	sc_ownedfp;
 	int	sc_spare1[1];
-	int	sc_fpstate[128];
+	int	sc_fpstate[128] __aligned(16);
 	int	sc_spare2[8];
 };
 
diff --git a/sys/amd64/include/ucontext.h b/sys/amd64/include/ucontext.h
index ea6f0b6b1b8f..ec3e0eefaea2 100644
--- a/sys/amd64/include/ucontext.h
+++ b/sys/amd64/include/ucontext.h
@@ -38,7 +38,7 @@ typedef struct __mcontext {
 	 * and ucontext_t at the same time.
 	 */
 	int	mc_onstack;		/* XXX - sigcontext compat. */
-	int	mc_gs;			/* machine state (trapframe) */
+	int	mc_gs;			/* machine state (struct trapframe) */
 	int	mc_fs;
 	int	mc_es;
 	int	mc_ds;
@@ -68,15 +68,17 @@ typedef struct __mcontext {
 #define	_MC_FPOWNED_PCB		0x20002	/* FP state came from PCB */
 	int	mc_ownedfp;
 	int	mc_spare1[1];		/* align next field to 16 bytes */
-	int	mc_fpstate[128];	/* must be multiple of 16 bytes */
+	/*
+	 * See <machine/npx.h> for the internals of mc_fpstate[].
+	 */
+	int	mc_fpstate[128] __aligned(16);
 	int	mc_spare2[8];
 } mcontext_t;
 
 #if defined(_KERNEL) && defined(COMPAT_FREEBSD4)
-/* For 4.x binaries */
 struct mcontext4 {
 	int	mc_onstack;		/* XXX - sigcontext compat. */
-	int	mc_gs;
+	int	mc_gs;			/* machine state (struct trapframe) */
 	int	mc_fs;
 	int	mc_es;
 	int	mc_ds;
diff --git a/sys/i386/i386/genassym.c b/sys/i386/i386/genassym.c
index 47856455f099..92915b6a116d 100644
--- a/sys/i386/i386/genassym.c
+++ b/sys/i386/i386/genassym.c
@@ -157,6 +157,9 @@ ASSYM(SIGF_HANDLER, offsetof(struct sigframe, sf_ahu.sf_handler));
 ASSYM(SIGF_SC, offsetof(struct osigframe, sf_siginfo.si_sc));
 #endif
 ASSYM(SIGF_UC, offsetof(struct sigframe, sf_uc));
+#ifdef COMPAT_FREEBSD4
+ASSYM(SIGF_UC4, offsetof(struct sigframe4, sf_uc));
+#endif
 #ifdef COMPAT_43
 ASSYM(SC_PS, offsetof(struct osigcontext, sc_ps));
 ASSYM(SC_FS, offsetof(struct osigcontext, sc_fs));
diff --git a/sys/i386/i386/locore.s b/sys/i386/i386/locore.s
index 8f1d801268b3..19a6ac7117c1 100644
--- a/sys/i386/i386/locore.s
+++ b/sys/i386/i386/locore.s
@@ -394,32 +394,36 @@ begin:
  * Signal trampoline, copied to top of user stack
  */
 NON_GPROF_ENTRY(sigcode)
-	call	*SIGF_HANDLER(%esp)	/* call signal handler */
-	lea	SIGF_UC(%esp),%eax	/* get ucontext_t */
+	calll	*SIGF_HANDLER(%esp)
+	leal	SIGF_UC(%esp),%eax	/* get ucontext */
 	pushl	%eax
 	testl	$PSL_VM,UC_EFLAGS(%eax)
-	jne	9f
+	jne	1f
 	movl	UC_GS(%eax),%gs		/* restore %gs */
-9:
+1:
 	movl	$SYS_sigreturn,%eax
 	pushl	%eax			/* junk to fake return addr. */
 	int	$0x80			/* enter kernel with args */
-0:	jmp	0b
+					/* on stack */
+1:
+	jmp	1b
 
 #ifdef COMPAT_FREEBSD4
 	ALIGN_TEXT
 freebsd4_sigcode:
-	call	*SIGF_HANDLER(%esp)	/* call signal handler */
-	lea	SIGF_UC(%esp),%eax	/* get ucontext_t */
+	calll	*SIGF_HANDLER(%esp)
+	leal	SIGF_UC4(%esp),%eax	/* get ucontext */
 	pushl	%eax
 	testl	$PSL_VM,UC4_EFLAGS(%eax)
-	jne	9f
+	jne	1f
 	movl	UC4_GS(%eax),%gs	/* restore %gs */
-9:
+1:
 	movl	$344,%eax		/* 4.x SYS_sigreturn */
 	pushl	%eax			/* junk to fake return addr. */
 	int	$0x80			/* enter kernel with args */
-0:	jmp	0b
+					/* on stack */
+1:
+	jmp	1b
 #endif
 
 #ifdef COMPAT_43
diff --git a/sys/i386/i386/machdep.c b/sys/i386/i386/machdep.c
index 1aedc3db1f25..27b3bcac7b5e 100644
--- a/sys/i386/i386/machdep.c
+++ b/sys/i386/i386/machdep.c
@@ -506,6 +506,7 @@ sendsig(catcher, sig, mask, code)
 	struct proc *p;
 	struct thread *td;
 	struct sigacts *psp;
+	char *sp;
 	struct trapframe *regs;
 	int oonstack;
 
@@ -544,13 +545,15 @@ sendsig(catcher, sig, mask, code)
 	/* Allocate space for the signal handler context. */
 	if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
-		sfp = (struct sigframe *)(p->p_sigstk.ss_sp +
-		    p->p_sigstk.ss_size - sizeof(struct sigframe));
+		sp = p->p_sigstk.ss_sp +
+		    p->p_sigstk.ss_size - sizeof(struct sigframe);
 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
 		p->p_sigstk.ss_flags |= SS_ONSTACK;
 #endif
 	} else
-		sfp = (struct sigframe *)regs->tf_esp - 1;
+		sp = (char *)regs->tf_esp - sizeof(struct sigframe);
+	/* Align to 16 bytes. */
+	sfp = (struct sigframe *)((unsigned int)sp & ~0xF);
 	PROC_UNLOCK(p);
 
 	/* Translate the signal if appropriate. */
diff --git a/sys/i386/include/signal.h b/sys/i386/include/signal.h
index efc7ef045888..fb2d82c4d489 100644
--- a/sys/i386/include/signal.h
+++ b/sys/i386/include/signal.h
@@ -100,7 +100,7 @@ struct osigcontext {
 struct sigcontext {
 	struct __sigset sc_mask;	/* signal mask to restore */
 	int	sc_onstack;		/* sigstack state to restore */
-	int	sc_gs;			/* machine state (struct trapframe): */
+	int	sc_gs;			/* machine state (struct trapframe) */
 	int	sc_fs;
 	int	sc_es;
 	int	sc_ds;
@@ -127,7 +127,7 @@ struct sigcontext {
 	int	sc_fpformat;
 	int	sc_ownedfp;
 	int	sc_spare1[1];
-	int	sc_fpstate[128];
+	int	sc_fpstate[128] __aligned(16);
 	int	sc_spare2[8];
 };
 
diff --git a/sys/i386/include/ucontext.h b/sys/i386/include/ucontext.h
index ea6f0b6b1b8f..ec3e0eefaea2 100644
--- a/sys/i386/include/ucontext.h
+++ b/sys/i386/include/ucontext.h
@@ -38,7 +38,7 @@ typedef struct __mcontext {
 	 * and ucontext_t at the same time.
 	 */
 	int	mc_onstack;		/* XXX - sigcontext compat. */
-	int	mc_gs;			/* machine state (trapframe) */
+	int	mc_gs;			/* machine state (struct trapframe) */
 	int	mc_fs;
 	int	mc_es;
 	int	mc_ds;
@@ -68,15 +68,17 @@ typedef struct __mcontext {
 #define	_MC_FPOWNED_PCB		0x20002	/* FP state came from PCB */
 	int	mc_ownedfp;
 	int	mc_spare1[1];		/* align next field to 16 bytes */
-	int	mc_fpstate[128];	/* must be multiple of 16 bytes */
+	/*
+	 * See <machine/npx.h> for the internals of mc_fpstate[].
+	 */
+	int	mc_fpstate[128] __aligned(16);
 	int	mc_spare2[8];
 } mcontext_t;
 
 #if defined(_KERNEL) && defined(COMPAT_FREEBSD4)
-/* For 4.x binaries */
 struct mcontext4 {
 	int	mc_onstack;		/* XXX - sigcontext compat. */
-	int	mc_gs;
+	int	mc_gs;			/* machine state (struct trapframe) */
 	int	mc_fs;
 	int	mc_es;
 	int	mc_ds;