EABI's Procedure Call Standard (AAPCS) requires the stack pointer

to be 8-byte aligned. To guarantee this, align the stack pointers
passed to user processes and make sure the in-kernel stacks are
properly aligned, too.

ok jsg@

4 files changed, 23 insertions, 7 deletions

diff --git a/sys/arch/arm/arm/cpuswitch.S b/sys/arch/arm/arm/cpuswitch.S
index 98e2dbe6969..62a37e2383a 100644
--- a/sys/arch/arm/arm/cpuswitch.S
+++ b/sys/arch/arm/arm/cpuswitch.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpuswitch.S,v 1.15 2016/01/31 00:14:50 jsg Exp $	*/
+/*	$OpenBSD: cpuswitch.S,v 1.16 2016/04/24 01:31:02 patrick Exp $	*/
 /*	$NetBSD: cpuswitch.S,v 1.41 2003/11/15 08:44:18 scw Exp $	*/
 
 /*
@@ -171,6 +171,7 @@ ENTRY(cpu_idle_leave)
 
 ENTRY(cpu_switchto)
 	stmfd	sp!, {r4-r7, lr}
+	sub	sp, sp, #4
 
 #ifdef MULTIPROCESSOR
 	/* XXX use curcpu() */
@@ -451,6 +452,7 @@ ENTRY(cpu_switchto)
 	 * Pull the registers that got pushed when either savectx() or
 	 * cpu_switch() was called and return.
 	 */
+	add	sp, sp, #4
 	ldmfd	sp!, {r4-r7, pc}
 
 /* LINTSTUB: Func: void savectx(struct pcb *pcb) */
@@ -461,6 +463,7 @@ ENTRY(savectx)
 
 	/* Push registers.*/
 	stmfd	sp!, {r4-r7, lr}
+	sub	sp, sp, #4
 
 	/* Store all the registers in the process's pcb */
 #ifndef __XSCALE__
@@ -473,6 +476,7 @@ ENTRY(savectx)
 #endif
 
 	/* Pull the regs of the stack */
+	add	sp, sp, #4
 	ldmfd	sp!, {r4-r7, pc}
 
 ENTRY(proc_trampoline)
diff --git a/sys/arch/arm/arm/cpuswitch7.S b/sys/arch/arm/arm/cpuswitch7.S
index 43bf7da5906..8ebcfacc031 100644
--- a/sys/arch/arm/arm/cpuswitch7.S
+++ b/sys/arch/arm/arm/cpuswitch7.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpuswitch7.S,v 1.6 2016/04/04 09:13:44 patrick Exp $	*/
+/*	$OpenBSD: cpuswitch7.S,v 1.7 2016/04/24 01:31:02 patrick Exp $	*/
 /*	$NetBSD: cpuswitch.S,v 1.41 2003/11/15 08:44:18 scw Exp $	*/
 
 /*
@@ -158,6 +158,7 @@ ENTRY(cpu_idle_leave)
 
 ENTRY(cpu_switchto)
 	stmfd	sp!, {r4-r7, lr}
+	sub	sp, sp, #4
 
 	/* Get curcpu from TPIDRPRW. */
 	mrc	p15, 0, r3, c13, c0, 4
@@ -371,6 +372,7 @@ ENTRY(cpu_switchto)
 	 * Pull the registers that got pushed when either savectx() or
 	 * cpu_switch() was called and return.
 	 */
+	add	sp, sp, #4
 	ldmfd	sp!, {r4-r7, pc}
 
 /* LINTSTUB: Func: void savectx(struct pcb *pcb) */
@@ -381,12 +383,14 @@ ENTRY(savectx)
 
 	/* Push registers.*/
 	stmfd	sp!, {r4-r7, lr}
+	sub	sp, sp, #4
 
 	/* Store all the registers in the process's pcb */
 	add	r2, r0, #(PCB_R8)
 	stmia	r2, {r8-r13}
 
 	/* Pull the regs of the stack */
+	add	sp, sp, #4
 	ldmfd	sp!, {r4-r7, pc}
 
 ENTRY(proc_trampoline)
diff --git a/sys/arch/arm/arm/vm_machdep.c b/sys/arch/arm/arm/vm_machdep.c
index 06f217b7189..bb69e88b31e 100644
--- a/sys/arch/arm/arm/vm_machdep.c
+++ b/sys/arch/arm/arm/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.16 2015/08/15 22:20:20 miod Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.17 2016/04/24 01:31:02 patrick Exp $	*/
 /*	$NetBSD: vm_machdep.c,v 1.31 2004/01/04 11:33:29 jdolecek Exp $	*/
 
 /*
@@ -140,10 +140,11 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	*tf = *p1->p_addr->u_pcb.pcb_tf;
 
 	/*
-	 * If specified, give the child a different stack.
+	 * If specified, give the child a different stack (make sure
+	 * it's 8-byte aligned).
 	 */
 	if (stack != NULL)
-		tf->tf_usr_sp = (u_int)stack + stacksize;
+		tf->tf_usr_sp = ((vaddr_t)(stack) + stacksize) & -8;
 
 	sf = (struct switchframe *)tf - 1;
 	sf->sf_r4 = (u_int)func;
diff --git a/sys/arch/arm/include/frame.h b/sys/arch/arm/include/frame.h
index 31b2936a865..e2031486c28 100644
--- a/sys/arch/arm/include/frame.h
+++ b/sys/arch/arm/include/frame.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: frame.h,v 1.7 2016/01/31 00:14:50 jsg Exp $	*/
+/*	$OpenBSD: frame.h,v 1.8 2016/04/24 01:31:02 patrick Exp $	*/
 /*	$NetBSD: frame.h,v 1.9 2003/12/01 08:48:33 scw Exp $	*/
 
 /*
@@ -75,6 +75,7 @@ typedef struct trapframe {
 	register_t tf_svc_sp;
 	register_t tf_svc_lr;
 	register_t tf_pc;
+	register_t tf_pad;
 } trapframe_t;
 
 /* Register numbers */
@@ -137,6 +138,7 @@ typedef struct irqframe {
 	unsigned int if_svc_sp;
 	unsigned int if_svc_lr;
 	unsigned int if_pc;
+	unsigned int if_pad;
 } irqframe_t;
 
 #define clockframe irqframe
@@ -146,6 +148,7 @@ typedef struct irqframe {
  */
 
 struct switchframe {
+	u_int	sf_pad;
 	u_int	sf_r4;
 	u_int	sf_r5;
 	u_int	sf_r6;
@@ -203,6 +206,7 @@ struct frame {
  */
 
 #define PUSHFRAME							   \
+	sub	sp, sp, #4;		/* Align the stack */		   \
 	str	lr, [sp, #-4]!;		/* Push the return address */	   \
 	sub	sp, sp, #(4*17);	/* Adjust the stack pointer */	   \
 	stmia	sp, {r0-r14}^;		/* Push the user mode registers */ \
@@ -221,7 +225,8 @@ struct frame {
         ldmia   sp, {r0-r14}^;		/* Restore registers (usr mode) */ \
         mov     r0, r0;                 /* NOP for previous instruction */ \
 	add	sp, sp, #(4*17);	/* Adjust the stack pointer */	   \
- 	ldr	lr, [sp], #0x0004	/* Pull the return address */
+	ldr	lr, [sp], #0x0004;	/* Pull the return address */	   \
+	add	sp, sp, #4		/* Align the stack */
 
 /*
  * PUSHFRAMEINSVC - macro to push a trap frame on the stack in SVC32 mode
@@ -241,6 +246,8 @@ struct frame {
 	orr     r2, r2, #(PSR_SVC32_MODE);				   \
 	msr     cpsr_c, r2;		/* Punch into SVC mode */	   \
 	mov	r2, sp;			/* Save	SVC sp */		   \
+	bic	sp, sp, #7;		/* Align sp to an 8-byte address */   \
+	sub	sp, sp, #4;		/* Pad trapframe to keep alignment */ \
 	str	r0, [sp, #-4]!;		/* Push return address */	   \
 	str	lr, [sp, #-4]!;		/* Push SVC lr */		   \
 	str	r2, [sp, #-4]!;		/* Push SVC sp */		   \