runtime: use libc for signal functions on iOS

Also:
 - Add extra SystemStack space for darwin/arm64 just
like for darwin/arm.
 - Removed redundant stack alignment; the arm64 hardware enforces
 the 16 byte alignment.
 - Save and restore the g registers at library initialization.
 - Zero g registers since libpreinit can call libc functions
 that in turn use asmcgocall. asmcgocall requires an initialized g.
 - Change asmcgocall to work even if no g is set. The change mimics
 amd64.

Change-Id: I1b8c63b07cfec23b909c0d215b50dc229f8adbc8
Reviewed-on: https://go-review.googlesource.com/117176
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

8 files changed, 261 insertions, 203 deletions

diff --git a/src/cmd/vet/all/whitelist/darwin_arm.txt b/src/cmd/vet/all/whitelist/darwin_arm.txt
index 8e935b6ff2..1c25c6a939 100644
--- a/src/cmd/vet/all/whitelist/darwin_arm.txt
+++ b/src/cmd/vet/all/whitelist/darwin_arm.txt
@@ -1,11 +1,5 @@
 // darwin/arm-specific vet whitelist. See readme.txt for details.
 
-// False positives due to comments in assembly.
-// To be removed. See CL 27154.
-
-runtime/sys_darwin_arm.s: [arm] sigfwd: use of unnamed argument 0(FP); offset 0 is fn+0(FP)
-
-
 // Ok.
 
 runtime/asm_arm.s: [arm] sigreturn: function sigreturn missing Go declaration
diff --git a/src/cmd/vet/all/whitelist/darwin_arm64.txt b/src/cmd/vet/all/whitelist/darwin_arm64.txt
index 8cab997961..a1edb71383 100644
--- a/src/cmd/vet/all/whitelist/darwin_arm64.txt
+++ b/src/cmd/vet/all/whitelist/darwin_arm64.txt
@@ -1,5 +1,3 @@
 // darwin/arm64-specific vet whitelist. See readme.txt for details.
 
-runtime/sys_darwin_arm64.s: [arm64] sigtramp: 24(RSP) should be infostyle+8(FP)
-runtime/sys_darwin_arm64.s: [arm64] sigtramp: 24(RSP) should be infostyle+8(FP)
 runtime/asm_arm64.s: [arm64] sigreturn: function sigreturn missing Go declaration
diff --git a/src/runtime/asm_arm.s b/src/runtime/asm_arm.s
index 545e58e9b0..6722ba760f 100644
--- a/src/runtime/asm_arm.s
+++ b/src/runtime/asm_arm.s
@@ -36,25 +36,28 @@ TEXT _rt0_arm_lib(SB),NOSPLIT,$104
 	MOVW	R6, 20(R13)
 	MOVW	R7, 24(R13)
 	MOVW	R8, 28(R13)
-	MOVW	R11, 32(R13)
+	MOVW	g, 32(R13)
+	MOVW	R11, 36(R13)
 
 	// Skip floating point registers on GOARM < 6.
 	MOVB    runtime·goarm(SB), R11
 	CMP	$6, R11
 	BLT	skipfpsave
-	MOVD	F8, (32+8*1)(R13)
-	MOVD	F9, (32+8*2)(R13)
-	MOVD	F10, (32+8*3)(R13)
-	MOVD	F11, (32+8*4)(R13)
-	MOVD	F12, (32+8*5)(R13)
-	MOVD	F13, (32+8*6)(R13)
-	MOVD	F14, (32+8*7)(R13)
-	MOVD	F15, (32+8*8)(R13)
+	MOVD	F8, (40+8*0)(R13)
+	MOVD	F9, (40+8*1)(R13)
+	MOVD	F10, (40+8*2)(R13)
+	MOVD	F11, (40+8*3)(R13)
+	MOVD	F12, (40+8*4)(R13)
+	MOVD	F13, (40+8*5)(R13)
+	MOVD	F14, (40+8*6)(R13)
+	MOVD	F15, (40+8*7)(R13)
 skipfpsave:
 	// Save argc/argv.
 	MOVW	R0, _rt0_arm_lib_argc<>(SB)
 	MOVW	R1, _rt0_arm_lib_argv<>(SB)
 
+	MOVW	$0, g // Initialize g.
+
 	// Synchronous initialization.
 	CALL	runtime·libpreinit(SB)
 
@@ -77,21 +80,22 @@ rr:
 	MOVB    runtime·goarm(SB), R11
 	CMP	$6, R11
 	BLT	skipfprest
-	MOVD	(32+8*1)(R13), F8
-	MOVD	(32+8*2)(R13), F9
-	MOVD	(32+8*3)(R13), F10
-	MOVD	(32+8*4)(R13), F11
-	MOVD	(32+8*5)(R13), F12
-	MOVD	(32+8*6)(R13), F13
-	MOVD	(32+8*7)(R13), F14
-	MOVD	(32+8*8)(R13), F15
+	MOVD	(40+8*0)(R13), F8
+	MOVD	(40+8*1)(R13), F9
+	MOVD	(40+8*2)(R13), F10
+	MOVD	(40+8*3)(R13), F11
+	MOVD	(40+8*4)(R13), F12
+	MOVD	(40+8*5)(R13), F13
+	MOVD	(40+8*6)(R13), F14
+	MOVD	(40+8*7)(R13), F15
 skipfprest:
 	MOVW	12(R13), R4
 	MOVW	16(R13), R5
 	MOVW	20(R13), R6
 	MOVW	24(R13), R7
 	MOVW	28(R13), R8
-	MOVW	32(R13), R11
+	MOVW	32(R13), g
+	MOVW	36(R13), R11
 	RET
 
 // _rt0_arm_lib_go initializes the Go runtime.
@@ -582,6 +586,8 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	MOVW	arg+4(FP), R0
 
 	MOVW	R13, R2
+	CMP	$0, g
+	BEQ nosave
 	MOVW	g, R4
 
 	// Figure out if we need to switch to m->g0 stack.
@@ -590,10 +596,10 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	MOVW	g_m(g), R8
 	MOVW	m_gsignal(R8), R3
 	CMP	R3, g
-	BEQ	noswitch
+	BEQ	nosave
 	MOVW	m_g0(R8), R3
 	CMP	R3, g
-	BEQ	noswitch
+	BEQ	nosave
 	BL	gosave<>(SB)
 	MOVW	R0, R5
 	MOVW	R3, R0
@@ -602,7 +608,6 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	MOVW	(g_sched+gobuf_sp)(g), R13
 
 	// Now on a scheduling stack (a pthread-created stack).
-noswitch:
 	SUB	$24, R13
 	BIC	$0x7, R13	// alignment for gcc ABI
 	MOVW	R4, 20(R13) // save old g
@@ -624,6 +629,30 @@ noswitch:
 	MOVW	R0, ret+8(FP)
 	RET
 
+nosave:
+	// Running on a system stack, perhaps even without a g.
+	// Having no g can happen during thread creation or thread teardown
+	// (see needm/dropm on Solaris, for example).
+	// This code is like the above sequence but without saving/restoring g
+	// and without worrying about the stack moving out from under us
+	// (because we're on a system stack, not a goroutine stack).
+	// The above code could be used directly if already on a system stack,
+	// but then the only path through this code would be a rare case on Solaris.
+	// Using this code for all "already on system stack" calls exercises it more,
+	// which should help keep it correct.
+	SUB	$24, R13
+	BIC	$0x7, R13	// alignment for gcc ABI
+	// save null g in case someone looks during debugging.
+	MOVW	$0, R4
+	MOVW	R4, 20(R13)
+	MOVW	R2, 16(R13)	// Save old stack pointer.
+	BL	(R1)
+	// Restore stack pointer.
+	MOVW	16(R13), R2
+	MOVW	R2, R13
+	MOVW	R0, ret+8(FP)
+	RET
+
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize, uintptr ctxt)
 // Turn the fn into a Go func (by taking its address) and call
 // cgocallback_gofunc.
diff --git a/src/runtime/asm_arm64.s b/src/runtime/asm_arm64.s
index d1b90b056c..af389be9fe 100644
--- a/src/runtime/asm_arm64.s
+++ b/src/runtime/asm_arm64.s
@@ -863,6 +863,8 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	MOVD	arg+8(FP), R0
 
 	MOVD	RSP, R2		// save original stack pointer
+	CMP	$0, g
+	BEQ	nosave
 	MOVD	g, R4
 
 	// Figure out if we need to switch to m->g0 stack.
@@ -871,10 +873,12 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	MOVD	g_m(g), R8
 	MOVD	m_gsignal(R8), R3
 	CMP	R3, g
-	BEQ	noswitch
+	BEQ	nosave
 	MOVD	m_g0(R8), R3
 	CMP	R3, g
-	BEQ	noswitch
+	BEQ	nosave
+
+	// Switch to system stack.
 	MOVD	R0, R9	// gosave<> and save_g might clobber R0
 	BL	gosave<>(SB)
 	MOVD	R3, g
@@ -884,12 +888,10 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	MOVD	R9, R0
 
 	// Now on a scheduling stack (a pthread-created stack).
-noswitch:
 	// Save room for two of our pointers /*, plus 32 bytes of callee
 	// save area that lives on the caller stack. */
 	MOVD	RSP, R13
 	SUB	$16, R13
-	BIC	$0xf, R13	// alignment for gcc ABI
 	MOVD	R13, RSP
 	MOVD	R4, 0(RSP)	// save old g on stack
 	MOVD	(g_stack+stack_hi)(R4), R4
@@ -910,6 +912,30 @@ noswitch:
 	MOVW	R0, ret+16(FP)
 	RET
 
+nosave:
+	// Running on a system stack, perhaps even without a g.
+	// Having no g can happen during thread creation or thread teardown
+	// (see needm/dropm on Solaris, for example).
+	// This code is like the above sequence but without saving/restoring g
+	// and without worrying about the stack moving out from under us
+	// (because we're on a system stack, not a goroutine stack).
+	// The above code could be used directly if already on a system stack,
+	// but then the only path through this code would be a rare case on Solaris.
+	// Using this code for all "already on system stack" calls exercises it more,
+	// which should help keep it correct.
+	MOVD	RSP, R13
+	SUB	$16, R13
+	MOVD	R13, RSP
+	MOVD	$0, R4
+	MOVD	R4, 0(RSP)	// Where above code stores g, in case someone looks during debugging.
+	MOVD	R2, 8(RSP)	// Save original stack pointer.
+	BL	(R1)
+	// Restore stack pointer.
+	MOVD	8(RSP), R2
+	MOVD	R2, RSP	
+	MOVD	R0, ret+16(FP)
+	RET
+
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize, uintptr ctxt)
 // Turn the fn into a Go func (by taking its address) and call
 // cgocallback_gofunc.
diff --git a/src/runtime/rt0_darwin_arm64.s b/src/runtime/rt0_darwin_arm64.s
index 719944e626..d039a8e0ab 100644
--- a/src/runtime/rt0_darwin_arm64.s
+++ b/src/runtime/rt0_darwin_arm64.s
@@ -26,18 +26,21 @@ TEXT _rt0_arm64_darwin_lib(SB),NOSPLIT,$168
 	MOVD R25, 72(RSP)
 	MOVD R26, 80(RSP)
 	MOVD R27, 88(RSP)
-	FMOVD F8, 96(RSP)
-	FMOVD F9, 104(RSP)
-	FMOVD F10, 112(RSP)
-	FMOVD F11, 120(RSP)
-	FMOVD F12, 128(RSP)
-	FMOVD F13, 136(RSP)
-	FMOVD F14, 144(RSP)
-	FMOVD F15, 152(RSP)
+	MOVD g, 96(RSP)
+	FMOVD F8, 104(RSP)
+	FMOVD F9, 112(RSP)
+	FMOVD F10, 120(RSP)
+	FMOVD F11, 128(RSP)
+	FMOVD F12, 136(RSP)
+	FMOVD F13, 144(RSP)
+	FMOVD F14, 152(RSP)
+	FMOVD F15, 160(RSP)
 
 	MOVD  R0, _rt0_arm64_darwin_lib_argc<>(SB)
 	MOVD  R1, _rt0_arm64_darwin_lib_argv<>(SB)
 
+	MOVD	$0, g // initialize g to nil
+
 	// Synchronous initialization.
 	MOVD	$runtime·libpreinit(SB), R4
 	BL	(R4)
@@ -58,14 +61,16 @@ TEXT _rt0_arm64_darwin_lib(SB),NOSPLIT,$168
 	MOVD 72(RSP), R25
 	MOVD 80(RSP), R26
 	MOVD 88(RSP), R27
-	FMOVD 96(RSP), F8
-	FMOVD 104(RSP), F9
-	FMOVD 112(RSP), F10
-	FMOVD 120(RSP), F11
-	FMOVD 128(RSP), F12
-	FMOVD 136(RSP), F13
-	FMOVD 144(RSP), F14
-	FMOVD 152(RSP), F15
+	MOVD 96(RSP), g
+	FMOVD 104(RSP), F8
+	FMOVD 112(RSP), F9
+	FMOVD 120(RSP), F10
+	FMOVD 128(RSP), F11
+	FMOVD 136(RSP), F12
+	FMOVD 144(RSP), F13
+	FMOVD 152(RSP), F14
+	FMOVD 160(RSP), F15
+
 	RET
 
 TEXT _rt0_arm64_darwin_lib_go(SB),NOSPLIT,$0
diff --git a/src/runtime/stack.go b/src/runtime/stack.go
index e40fa9cc1b..d83e9d6722 100644
--- a/src/runtime/stack.go
+++ b/src/runtime/stack.go
@@ -64,8 +64,8 @@ const (
 	// StackSystem is a number of additional bytes to add
 	// to each stack below the usual guard area for OS-specific
 	// purposes like signal handling. Used on Windows, Plan 9,
-	// and Darwin/ARM because they do not use a separate stack.
-	_StackSystem = sys.GoosWindows*512*sys.PtrSize + sys.GoosPlan9*512 + sys.GoosDarwin*sys.GoarchArm*1024
+	// and iOS because they do not use a separate stack.
+	_StackSystem = sys.GoosWindows*512*sys.PtrSize + sys.GoosPlan9*512 + sys.GoosDarwin*sys.GoarchArm*1024 + sys.GoosDarwin*sys.GoarchArm64*1024
 
 	// The minimum size of stack used by Go code
 	_StackMin = 2048
diff --git a/src/runtime/sys_darwin_arm.s b/src/runtime/sys_darwin_arm.s
index a940d95732..5f6c903437 100644
--- a/src/runtime/sys_darwin_arm.s
+++ b/src/runtime/sys_darwin_arm.s
@@ -61,14 +61,12 @@ TEXT runtime·exit_trampoline(SB),NOSPLIT|NOFRAME,$0
 	MOVW	$1002, R1
 	MOVW	R0, (R1)	// fail hard
 
-TEXT runtime·raiseproc(SB),NOSPLIT,$24
-	MOVW	$SYS_getpid, R12
-	SWI	$0x80
+TEXT runtime·raiseproc_trampoline(SB),NOSPLIT,$0
+	MOVW	0(R0), R8	// signal
+	BL	libc_getpid(SB)
 	// arg 1 pid already in R0 from getpid
-	MOVW	sig+0(FP), R1	// arg 2 - signal
-	MOVW	$1, R2	// arg 3 - posix
-	MOVW	$SYS_kill, R12
-	SWI $0x80
+	MOVW	R8, R1	// arg 2 signal
+	BL	libc_kill(SB)
 	RET
 
 TEXT runtime·mmap_trampoline(SB),NOSPLIT,$0
@@ -174,91 +172,89 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-16
 	MOVW	R4, R13
 	RET
 
-// Sigtramp's job is to call the actual signal handler.
-// It is called with the following arguments on the stack:
-//	 LR  	"return address" - ignored
-//	 R0  	actual handler
-//	 R1  	siginfo style - ignored
-//	 R2   	signal number
-//	 R3   	siginfo
-//	 -4(FP)	context, beware that 0(FP) is the saved LR
 TEXT runtime·sigtramp(SB),NOSPLIT,$0
+	// Reserve space for callee-save registers and arguments.
+	SUB	$36, R13
+
+	MOVW	R4, 12(R13)
+	MOVW	R5, 16(R13)
+	MOVW	R6, 20(R13)
+	MOVW	R7, 24(R13)
+	MOVW	R8, 28(R13)
+	MOVW	R11, 32(R13)
+
+	// Save arguments.
+	MOVW	R0, 4(R13)	// sig
+	MOVW	R1, 8(R13)	// info
+	MOVW	R2, 12(R13)	// ctx
+
 	// this might be called in external code context,
 	// where g is not set.
-	// first save R0, because runtime·load_g will clobber it
-	MOVM.DB.W [R0], (R13)
 	MOVB	runtime·iscgo(SB), R0
 	CMP 	$0, R0
 	BL.NE	runtime·load_g(SB)
 
-	CMP 	$0, g
-	BNE 	cont
-	// fake function call stack frame for badsignal
-	// we only need to pass R2 (signal number), but
-	// badsignal will expect R2 at 4(R13), so we also
-	// push R1 onto stack. turns out we do need R1
-	// to do sigreturn.
-	MOVM.DB.W [R1,R2], (R13)
-	MOVW  	$runtime·badsignal(SB), R11
-	BL	(R11)
-	MOVM.IA.W [R1], (R13) // saved infostype
-	ADD		$(4+4), R13 // +4: also need to remove the pushed R0.
-	MOVW    ucontext-4(FP), R0 // load ucontext
-	B	ret
+	MOVW	R13, R6
+	CMP	$0, g
+	BEQ nog
 
-cont:
-	// Restore R0
-	MOVM.IA.W (R13), [R0]
-
-	// NOTE: some Darwin/ARM kernels always use the main stack to run the
-	// signal handler. We need to switch to gsignal ourselves.
+	// iOS always use the main stack to run the signal handler.
+	// We need to switch to gsignal ourselves.
 	MOVW	g_m(g), R11
 	MOVW	m_gsignal(R11), R5
 	MOVW	(g_stack+stack_hi)(R5), R6
-	SUB		$28, R6
 
-	// copy arguments for call to sighandler
-	MOVW	R2, 4(R6) // signal num
-	MOVW	R3, 8(R6) // signal info
-	MOVW	g, 16(R6) // old_g
-	MOVW	context-4(FP), R4
-	MOVW	R4, 12(R6) // context
+nog:
+	// Restore arguments.
+	MOVW	4(R13), R0
+	MOVW	8(R13), R1
+	MOVW	12(R13), R2
 
-	// Backup ucontext and infostyle
-	MOVW    R4, 20(R6)
-	MOVW    R1, 24(R6)
+	// Reserve space for args and the stack pointer on the
+	// gsignal stack.
+	SUB $24, R6
+	// Save stack pointer.
+	MOVW	R13, R4
+	MOVW	R4, 16(R6)
+	// Switch to gsignal stack.
+	MOVW	R6, R13
 
-	// switch stack and g
-	MOVW	R6, R13 // sigtramp is not re-entrant, so no need to back up R13.
-	MOVW	R5, g
+	// Call sigtrampgo
+	MOVW	R0, 4(R13)
+	MOVW	R1, 8(R13)
+	MOVW	R2, 12(R13)
+	BL	runtime·sigtrampgo(SB)
 
-	BL	(R0)
+	// Switch to old stack.
+	MOVW	16(R13), R5
+	MOVW	R5, R13
 
-	// call sigreturn
-	MOVW	20(R13), R0	// saved ucontext
-	MOVW	24(R13), R1	// saved infostyle
-ret:
-	MOVW	$SYS_sigreturn, R12 // sigreturn(ucontext, infostyle)
-	SWI	$0x80
+	// Restore callee-save registers.
+	MOVW	12(R13), R4
+	MOVW	16(R13), R5
+	MOVW	20(R13), R6
+	MOVW	24(R13), R7
+	MOVW	28(R13), R8
+	MOVW	32(R13), R11
 
-	// if sigreturn fails, we can do nothing but exit
-	B	runtime·exit(SB)
+	ADD $36, R13
 
-TEXT runtime·sigprocmask(SB),NOSPLIT,$0
-	MOVW	how+0(FP), R0
-	MOVW	new+4(FP), R1
-	MOVW	old+8(FP), R2
-	MOVW	$SYS_pthread_sigmask, R12
-	SWI	$0x80
-	BL.CS	notok<>(SB)
 	RET
 
-TEXT runtime·sigaction(SB),NOSPLIT,$0
-	MOVW	mode+0(FP), R0
-	MOVW	new+4(FP), R1
-	MOVW	old+8(FP), R2
-	MOVW	$SYS_sigaction, R12
-	SWI	$0x80
+TEXT runtime·sigprocmask_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 new
+	MOVW	8(R0), R2	// arg 3 old
+	MOVW	0(R0), R0	// arg 1 how
+	BL	libc_pthread_sigmask(SB)
+	CMP	$0, R0
+	BL.NE	notok<>(SB)
+	RET
+
+TEXT runtime·sigaction_trampoline(SB),NOSPLIT,$0
+	MOVW	4(R0), R1	// arg 2 new
+	MOVW	8(R0), R2	// arg 3 old
+	MOVW	0(R0), R0	// arg 1 how
+	BL	libc_sigaction(SB)
 	RET
 
 TEXT runtime·usleep_trampoline(SB),NOSPLIT,$0
@@ -387,10 +383,11 @@ TEXT runtime·closeonexec(SB),NOSPLIT,$0
 	SWI	$0x80
 	RET
 
-// sigaltstack on some darwin/arm version is buggy and will always
-// run the signal handler on the main stack, so our sigtramp has
+// sigaltstack is not supported on iOS, so our sigtramp has
 // to do the stack switch ourselves.
-TEXT runtime·sigaltstack(SB),NOSPLIT,$0
+TEXT runtime·sigaltstack_trampoline(SB),NOSPLIT,$0
+	MOVW	$43, R0
+	BL	libc_exit(SB)
 	RET
 
 // Thread related functions
diff --git a/src/runtime/sys_darwin_arm64.s b/src/runtime/sys_darwin_arm64.s
index d13e44afcf..c21a5566fa 100644
--- a/src/runtime/sys_darwin_arm64.s
+++ b/src/runtime/sys_darwin_arm64.s
@@ -61,14 +61,12 @@ TEXT runtime·exit_trampoline(SB),NOSPLIT|NOFRAME,$0
 	MOVD	$1002, R1
 	MOVD	R0, (R1)	// fail hard
 
-TEXT runtime·raiseproc(SB),NOSPLIT,$0
-	MOVW	$SYS_getpid, R16
-	SVC	$0x80
+TEXT runtime·raiseproc_trampoline(SB),NOSPLIT,$0
+	MOVD	0(R0), R19	// signal
+	BL	libc_getpid(SB)
 	// arg 1 pid already in R0 from getpid
-	MOVW	sig+0(FP), R1	// arg 2 - signal
-	MOVW	$1, R2	// arg 3 - posix
-	MOVW	$SYS_kill, R16
-	SVC	$0x80
+	MOVD	R19, R1	// arg 2 signal
+	BL	libc_kill(SB)
 	RET
 
 TEXT runtime·mmap_trampoline(SB),NOSPLIT,$0
@@ -158,95 +156,104 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 	BL	(R11)
 	RET
 
-// Sigtramp's job is to call the actual signal handler.
-// It is called with the following arguments on the stack:
-//	LR	"return address" - ignored
-//	R0	actual handler
-//	R1	siginfo style - ignored
-//	R2	signal number
-//	R3	siginfo
-//	R4	context
 TEXT runtime·sigtramp(SB),NOSPLIT,$0
+	// Reserve space for callee-save registers and arguments.
+	SUB	$(8*16), RSP
+
+	// Save callee-save registers.
+	MOVD	R19, (8*4)(RSP)
+	MOVD	R20, (8*5)(RSP)
+	MOVD	R21, (8*6)(RSP)
+	MOVD	R22, (8*7)(RSP)
+	MOVD	R23, (8*8)(RSP)
+	MOVD	R24, (8*9)(RSP)
+	MOVD	R25, (8*10)(RSP)
+	MOVD	R26, (8*11)(RSP)
+	MOVD	R27, (8*12)(RSP)
+	MOVD	g, (8*13)(RSP)
+	MOVD	R29, (8*14)(RSP)
+
+	// Save arguments.
+	MOVW	R0, (8*1)(RSP)	// sig
+	MOVD	R1, (8*2)(RSP)	// info
+	MOVD	R2, (8*3)(RSP)	// ctx
+
 	// this might be called in external code context,
 	// where g is not set.
-	// first save R0, because runtime·load_g will clobber it
-	MOVD.W	R0, -16(RSP)	// note: stack must be 16-byte aligned
 	MOVB	runtime·iscgo(SB), R0
 	CMP	$0, R0
 	BEQ	2(PC)
 	BL	runtime·load_g(SB)
 
-	CMP	$0, g
-	BNE	cont
-	// fake function call stack frame for badsignal
-	// we only need to pass R2 (signal number), but
-	// badsignal will expect R2 at 8(RSP), so we also
-	// push R1 onto stack. turns out we do need R1
-	// to do sigreturn.
-	MOVD.W	R1, -16(RSP)
-	MOVD	R2, 8(RSP)
-	MOVD	R4, 24(RSP)	// save ucontext, badsignal might clobber R4
-	MOVD	$runtime·badsignal(SB), R26
-	BL	(R26)
-	MOVD	0(RSP), R1	// saved infostype
-	MOVD	24(RSP), R0	// the ucontext
-	ADD	$(16+16), RSP
-	B	ret
-
-cont:
-	// Restore R0
-	MOVD.P	16(RSP), R0
-
-	// NOTE: some Darwin/ARM kernels always use the main stack to run the
-	// signal handler. We need to switch to gsignal ourselves.
+	MOVD RSP, R6
+	CMP	 $0, g
+	BEQ	 nog
+	// iOS always use the main stack to run the signal handler.
+	// We need to switch to gsignal ourselves.
 	MOVD	g_m(g), R11
 	MOVD	m_gsignal(R11), R5
 	MOVD	(g_stack+stack_hi)(R5), R6
-	SUB	$64, R6
 
-	// copy arguments for call to sighandler
-	MOVD	R2, 8(R6)	// signal num
-	MOVD	R3, 16(R6)	// signal info
-	MOVD	R4, 24(R6)	// context
-	MOVD	g, 32(R6)	// old_g
+nog:
+	// Restore arguments.
+	MOVW	(8*1)(RSP), R0
+	MOVD	(8*2)(RSP), R1
+	MOVD	(8*3)(RSP), R2
 
-	// Backup ucontext and infostyle
-	MOVD	R4, 40(R6)
-	MOVD	R1, 48(R6)
+	// Reserve space for args and the stack pointer on the
+	// gsignal stack.
+	SUB	$48, R6
+	// Save stack pointer.
+	MOVD	RSP, R4
+	MOVD	R4, (8*4)(R6)
+	// Switch to gsignal stack.
+	MOVD	R6, RSP
 
-	// switch stack and g
-	MOVD	R6, RSP	// sigtramp is not re-entrant, so no need to back up RSP.
-	MOVD	R5, g
+	// Call sigtrampgo.
+	MOVW	R0, (8*1)(RSP)
+	MOVD	R1, (8*2)(RSP)
+	MOVD	R2, (8*3)(RSP)
+	MOVD	$runtime·sigtrampgo(SB), R11
+	BL	(R11)
 
-	BL	(R0)
+	// Switch to old stack.
+	MOVD	(8*4)(RSP), R5
+	MOVD	R5, RSP
 
-	// call sigreturn
-	MOVD	40(RSP), R0	// saved ucontext
-	MOVD	48(RSP), R1	// saved infostyle
-ret:
-	MOVW	$SYS_sigreturn, R16 // sigreturn(ucontext, infostyle)
-	SVC	$0x80
+	// Restore callee-save registers.
+	MOVD	(8*4)(RSP), R19
+	MOVD	(8*5)(RSP), R20
+	MOVD	(8*6)(RSP), R21
+	MOVD	(8*7)(RSP), R22
+	MOVD	(8*8)(RSP), R23
+	MOVD	(8*9)(RSP), R24
+	MOVD	(8*10)(RSP), R25
+	MOVD	(8*11)(RSP), R26
+	MOVD	(8*12)(RSP), R27
+	MOVD	(8*13)(RSP), g
+	MOVD	(8*14)(RSP), R29
 
-	// if sigreturn fails, we can do nothing but exit
-	B	runtime·exit(SB)
+	ADD $(8*16), RSP
 
-TEXT runtime·sigprocmask(SB),NOSPLIT,$0
-	MOVW	how+0(FP), R0
-	MOVD	new+8(FP), R1
-	MOVD	old+16(FP), R2
-	MOVW	$SYS_pthread_sigmask, R16
-	SVC	$0x80
-	BCC	2(PC)
+	RET
+
+TEXT runtime·sigprocmask_trampoline(SB),NOSPLIT,$0
+	MOVD	8(R0), R1	// arg 2 new
+	MOVD	16(R0), R2	// arg 3 old
+	MOVW	0(R0), R0	// arg 1 how
+	BL	libc_pthread_sigmask(SB)
+	CMP $0, R0
+	BEQ	2(PC)
 	BL	notok<>(SB)
 	RET
 
-TEXT runtime·sigaction(SB),NOSPLIT,$0
-	MOVW	mode+0(FP), R0
-	MOVD	new+8(FP), R1
-	MOVD	old+16(FP), R2
-	MOVW	$SYS_sigaction, R16
-	SVC	$0x80
-	BCC	2(PC)
+TEXT runtime·sigaction_trampoline(SB),NOSPLIT,$0
+	MOVD	8(R0), R1	// arg 2 new
+	MOVD	16(R0), R2	// arg 3 old
+	MOVW	0(R0), R0	// arg 1 how
+	BL	libc_sigaction(SB)
+	CMP	$0, R0
+	BEQ	2(PC)
 	BL	notok<>(SB)
 	RET
 
@@ -375,10 +382,12 @@ TEXT runtime·closeonexec(SB),NOSPLIT,$0
 	SVC	$0x80
 	RET
 
-// sigaltstack on some darwin/arm version is buggy and will always
+// sigaltstack on iOS is not supported and will always
 // run the signal handler on the main stack, so our sigtramp has
 // to do the stack switch ourselves.
-TEXT runtime·sigaltstack(SB),NOSPLIT,$0
+TEXT runtime·sigaltstack_trampoline(SB),NOSPLIT,$0
+	MOVW	$43, R0
+	BL	libc_exit(SB)
 	RET
 
 // Thread related functions