Swap the syscall caller frame info (i.e. the return pointer and

frame marker) and the syscall stub frame info in the trap frame.
Previously we stored the stub frame info in (rp,pfs) and the
caller frame info in (iip,cfm). This ends up being suboptimal
for the following reasons:
1. When we create a new context, such as for an execve(2), we had
   to set the (rp,pfs) pair for the entry point when using the
   syscall path out of the kernel but we need to set the (iip,cfm)
   pair when we take the interrupt way out. This is mostly just
   an inconsistency from the kernel's point of view, but an ugly
   irregularity from gdb(1)'s point of view.
2. The getcontext(2) and setcontext(2) syscalls had to swap the
   (rp,pfs) and (iip,cfm) pairs to make the context compatible
   with one created purely in userland.

Swapping the (rp,pfs) and (iip,cfm) pairs is visible to signal
handlers that actually peek at the mcontext_t and to gdb(1).
Since this change is made for gdb(1) and we don't care about
signal handlers that peek at the mcontext_t because we're still
a tier 2 platform, this ABI breakage is academic at this moment
in time.

Note that there was no real reason to save the caller frame info
in (iip,cfm) and the stub frame info in (rp,pfs).

3 files changed, 37 insertions, 40 deletions

diff --git a/sys/ia64/ia64/machdep.c b/sys/ia64/ia64/machdep.c
index 064fe0f..765f85f 100644
--- a/sys/ia64/ia64/machdep.c
+++ b/sys/ia64/ia64/machdep.c
@@ -933,7 +933,7 @@ sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
 		tf->tf_special.iip = ia64_get_k5() +
 		    ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
 	} else
-		tf->tf_special.rp = ia64_get_k5() +
+		tf->tf_special.iip = ia64_get_k5() +
 		    ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
 
 	/*
@@ -1095,9 +1095,6 @@ get_mcontext(struct thread *td, mcontext_t *mc, int clear_ret)
 		s.bspstore = (uintptr_t)ustk;
 	}
 	if (tf->tf_flags & FRAME_SYSCALL) {
-		s.pfs = s.cfm;
-		s.rp = s.iip;
-		s.cfm = s.iip = 0;
 		/*
 		 * Put the syscall return values in the context. We need this
 		 * for swapcontext() to work. Note that we don't use gr11 in
@@ -1152,10 +1149,8 @@ set_mcontext(struct thread *td, const mcontext_t *mc)
 		/* XXX High FP */
 	} else {
 		KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
-		s.cfm = s.pfs;
-		s.pfs = tf->tf_special.pfs;
-		s.iip = s.rp;
-		s.rp = tf->tf_special.rp;
+		s.cfm = tf->tf_special.cfm;
+		s.iip = tf->tf_special.iip;
 		if ((mc->mc_flags & _MC_FLAGS_SCRATCH_VALID) == 0) {
 			if (mc->mc_flags & _MC_FLAGS_RETURN_VALID) {
 				tf->tf_scratch.gr8 = mc->mc_scratch.gr8;
@@ -1204,7 +1199,6 @@ exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
 	if ((tf->tf_flags & FRAME_SYSCALL) == 0) {	/* break syscalls. */
 		bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
 		bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
-		tf->tf_special.iip = entry;
 		tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
 		tf->tf_special.bspstore = td->td_md.md_bspstore;
 		/*
@@ -1224,8 +1218,7 @@ exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
 		*kst = stack;
 		tf->tf_special.ndirty = (ksttop - kst) << 3;
 	} else {				/* epc syscalls (default). */
-		tf->tf_special.rp = entry;
-		tf->tf_special.pfs = (3UL<<62) | (3UL<<7) | 3UL;
+		tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
 		tf->tf_special.bspstore = td->td_md.md_bspstore + 24;
 		/*
 		 * Write values for out0, out1 and out2 to the user's backing
@@ -1238,6 +1231,7 @@ exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
 		suword((caddr_t)tf->tf_special.bspstore -  8, 0);
 	}
 
+	tf->tf_special.iip = entry;
 	tf->tf_special.sp = (stack & ~15) - 16;
 	tf->tf_special.rsc = 0xf;
 	tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
diff --git a/sys/ia64/ia64/syscall.S b/sys/ia64/ia64/syscall.S
index 6323ab7..e66d12b 100644
--- a/sys/ia64/ia64/syscall.S
+++ b/sys/ia64/ia64/syscall.S
@@ -93,11 +93,11 @@ gw_ret:
 	mov		ar.rnat=r22
 	;;
 	mov		ar.rsc=r24
-	mov		ar.pfs=r28
+	mov		ar.pfs=r20
 }
 {	.mib
 	mov		ar.fpsr=r25
-	mov		b0=r29
+	mov		b0=r18
 	br.sptk		b6
 	;;
 }
@@ -196,24 +196,24 @@ ENTRY(epc_sigtramp, 0)
 {	.mmi
 	mov		r17=ar.bsp
 	mov		r18=ar.rnat
-	add		r14=32,r16
+	add		r14=14*8,r16
 	;;
 }
 {	.mmi
 (p15)	mov		ar.bspstore=gp
-	ld8		r19=[r14],8
-	add		r15=48,r16
+	ld8		r19=[r14],-9*8		// cfm
+	add		r15=6*8,r16
 	;;
 }
 {	.mmi
-	st8		[r14]=r17,64		// bspstore
-	st8		[r15]=r18,-16		// rnat
+	st8		[r14]=r17,8*8		// bspstore
+	st8		[r15]=r18,8*8		// rnat
 	dep		r19=r19,r19,7,7
 	;;
 }
 {	.mmi
 	st8		[r14]=r0		// ndirty
-	st8		[r15]=r19		// pfs
+	st8		[r15]=r19		// cfm
 	nop		0
 	;;
 }
@@ -300,13 +300,13 @@ ENTRY(epc_syscall, 8)
 	;;
 }
 {	.mmi
-	st8		[r30]=r24,16		// rp (syscall stub)
+	st8		[r30]=r10,16		// rp (syscall caller)
 	st8		[r31]=r25,16		// pr
 	mov		r26=ar.pfs
 	;;
 }
 {	.mmi
-	st8		[r30]=r26,16		// pfs (syscall stub)
+	st8		[r30]=r9,16		// pfs (syscall caller)
 	st8		[r31]=r18,16		// bspstore
 	sub		r27=r23,r15
 	;;
@@ -336,8 +336,8 @@ ENTRY(epc_syscall, 8)
 	;;
 }
 {	.mmi
-	st8		[r30]=r9,16		// pfs (syscall caller)
-	st8		[r31]=r10,16		// rp (syscall caller)
+	st8		[r30]=r26,16		// pfs (syscall stub)
+	st8		[r31]=r24,16		// rp (syscall stub)
 	nop		0
 	;;
 }
@@ -425,14 +425,14 @@ epc_syscall_return:
 }
 {	.mmi
 	ld8		r17=[r15],16		// unat (before)
-	ld8		r18=[r14],16		// rp (syscall stub)
+	ld8		r18=[r14],16		// rp (syscall caller)
 	add		r31=r31,sp
 	;;
 }
-{	.mmi
+{	.mmb
 	ld8		r19=[r15],16		// pr
-	ld8		r20=[r14],16		// pfs (syscall stub)
-	mov		b6=r18
+	ld8		r20=[r14],16		// pfs (syscall caller)
+	nop		0
 	;;
 }
 {	.mmi
@@ -450,7 +450,7 @@ epc_syscall_return:
 {	.mmi
 	ld8		r25=[r15],16		// fpsr
 	ld8		r26=[r14],16		// psr
-	mov		ar.pfs=r20
+	nop		0
 	;;
 }
 {	.mmi
@@ -460,15 +460,15 @@ epc_syscall_return:
 	;;
 }
 {	.mmi
-	ld8		r28=[r15],56		// pfs (syscall caller)
-	ld8		r29=[r14],56		// rp (syscall caller)
+	ld8		r28=[r15],56		// pfs (syscall stub)
+	ld8		r29=[r14],56		// rp (syscall stub)
 	shl		r27=r27,16
 	;;
 }
-{	.mmb
+{	.mmi
 	ld8		r8=[r15],16		// r8
 	mov		ar.rsc=r27
-	nop		0
+	mov		b6=r29
 	;;
 }
 {	.mmb
@@ -485,21 +485,25 @@ epc_syscall_return:
 }
 {	.mmi
 	mov		r30=ar.bspstore
-	;;
 	mov		r14=ar.k5
-	dep		r30=0,r30,0,13		// 8KB aligned.
+	mov		ar.pfs=r28
 	;;
 }
 {	.mmi
-	mov		ar.k6=r30
 	mov		ar.bspstore=r21
-	mov		r13=r23
+	add		r14=gw_ret-ia64_gateway_page,r14
+	dep		r30=0,r30,0,13		// 8KB aligned.
 	;;
 }
+{	.mib
+	mov		ar.k6=r30
+	mov		r13=r23
+	nop		0
+}
 {	.mmi
 	mov		psr.l=r26
 	mov		ar.unat=r17
-	add		r14=gw_ret-ia64_gateway_page,r14
+	nop		0
 	;;
 }
 {	.mib
diff --git a/sys/ia64/ia64/vm_machdep.c b/sys/ia64/ia64/vm_machdep.c
index 2025951..4e02b8a 100644
--- a/sys/ia64/ia64/vm_machdep.c
+++ b/sys/ia64/ia64/vm_machdep.c
@@ -191,6 +191,7 @@ cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku)
 	stack = (uint64_t)ku->ku_stack.ss_sp;
 
 	bzero(&tf->tf_special, sizeof(tf->tf_special));
+	tf->tf_special.iip = fuword(&fd->func);
 	tf->tf_special.gp = fuword(&fd->gp);
 	tf->tf_special.sp = (stack + ku->ku_stack.ss_size - 16) & ~15;
 	tf->tf_special.rsc = 0xf;
@@ -200,12 +201,10 @@ cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku)
 	    IA64_PSR_CPL_USER;
 
 	if (tf->tf_flags & FRAME_SYSCALL) {
-		tf->tf_special.rp = fuword(&fd->func);
-		tf->tf_special.pfs = (3UL<<62) | (1UL<<7) | 1UL;
+		tf->tf_special.cfm = (3UL<<62) | (1UL<<7) | 1UL;
 		tf->tf_special.bspstore = stack + 8;
 		suword((caddr_t)stack, (uint64_t)ku->ku_mailbox);
 	} else {
-		tf->tf_special.iip = fuword(&fd->func);
 		tf->tf_special.cfm = (1UL<<63) | (1UL<<7) | 1UL;
 		tf->tf_special.bspstore = stack;
 		tf->tf_special.ndirty = 8;