fix issue with longjmp out of signal handlers and cancellation

stale state information indicating that a thread was possibly blocked
at a cancellation point could get left behind if longjmp was used to
exit a signal handler that interrupted a cancellation point.

to fix the issue, we throw away the state information entirely and
simply compare the saved instruction pointer to a range of code
addresses in the __syscall_cp_asm function. all the ugly PIC work
(which becomes minimal anyway with this approach) is defered to
cancellation time instead of happening at every syscall, which should
improve performance too.

this commit also fixes cancellation on arm, which was mildly broken
(race condition, not checking cancellation flag once inside the
cancellation point zone). apparently i forgot to implement that. the
new arm code is untested, but appears correct; i'll test and fix it
later if there are problems.

src/thread/arm/syscall_cp.s

@@ -3,20 +3,18 @@
 __syscall_cp_asm:
 	mov ip,sp
 	stmfd sp!,{r4,r5,r6,r7,lr}
-	stmfd sp!,{r0}
+.global __cp_begin
-	bl 1f
+__cp_begin:
-1:	mov r4,#(1f-.)
+	ld r0,[r0]
-	add r4,r4,lr
+	cmp r0,#0
-	str r4,[r0,#4]
+	blne __cancel
-	str sp,[r0]
 	mov r7,r1
 	mov r0,r2
 	mov r1,r3
 	ldmfd ip,{r2,r3,r4,r5,r6}
-1:	svc 0
+	svc 0
-	ldmfd sp!,{r1}
+.global __cp_end
-	mov r2,#0
+__cp_end:
-	str r2,[r1]
 	ldmfd sp!,{r4,r5,r6,r7,lr}
 	tst lr,#1
 	moveq pc,lr

src/thread/cancel_impl.c

@@ -14,19 +14,12 @@ long __syscall_cp_asm(volatile void *, long, long, long, long, long, long, long)
 long (__syscall_cp)(long nr, long u, long v, long w, long x, long y, long z)
 {
 	pthread_t self;
-	uintptr_t old_sp, old_ip;
 	long r;
 
 	if (!libc.main_thread || (self = __pthread_self())->canceldisable)
 		return __syscall(nr, u, v, w, x, y, z);
 
-	old_sp = self->cp_sp;
+	r = __syscall_cp_asm(&self->cancel, nr, u, v, w, x, y, z);
-	old_ip = self->cp_ip;
-	self->cp_sp = 0;
-	self->cp_ip = 0;
-	r = __syscall_cp_asm(&self->cp_sp, nr, u, v, w, x, y, z);
-	self->cp_ip = old_ip;
-	self->cp_sp = old_sp;
 	if (r==-EINTR && nr!=SYS_close && self->cancel && !self->canceldisable)
 		__cancel();
 	return r;
@@ -42,14 +35,14 @@ static void cancel_handler(int sig, siginfo_t *si, void *ctx)
 {
 	pthread_t self = __pthread_self();
 	ucontext_t *uc = ctx;
-	uintptr_t sp = ((uintptr_t *)&uc->uc_mcontext)[CANCEL_REG_SP];
+	const char *ip = ((char **)&uc->uc_mcontext)[CANCEL_REG_IP];
-	uintptr_t ip = ((uintptr_t *)&uc->uc_mcontext)[CANCEL_REG_IP];
+	extern const char __cp_begin[1], __cp_end[1];
 
 	if (!self->cancel || self->canceldisable) return;
 
 	_sigaddset(&uc->uc_sigmask, SIGCANCEL);
 
-	if (self->cancelasync || sp == self->cp_sp && ip <= self->cp_ip) {
+	if (self->cancelasync || ip >= __cp_begin && ip < __cp_end) {
 		self->canceldisable = 1;
 		pthread_sigmask(SIG_SETMASK, &uc->uc_sigmask, 0);
 		__cancel();

src/thread/i386/syscall_cp.s

@@ -2,34 +2,28 @@
 .global __syscall_cp_asm
 .type   __syscall_cp_asm,@function
 __syscall_cp_asm:
+	mov 4(%esp),%ecx
 	pushl %ebx
 	pushl %esi
 	pushl %edi
 	pushl %ebp
-	leal 20(%esp),%ebp
+.global __cp_begin
-	call 1f
+__cp_begin:
-1:	popl %eax
+	movl (%ecx),%eax
-	movl (%ebp),%ecx
-	addl $[1f-1b],%eax
-	movl %eax,4(%ecx)
-	movl %esp,(%ecx)
-	movl 8(%ecx),%eax
 	testl %eax,%eax
-	jnz 2f
+	jnz __cancel
-	movl 4(%ebp),%eax
+	movl 24(%esp),%eax
-	movl 8(%ebp),%ebx
+	movl 28(%esp),%ebx
-	movl 12(%ebp),%ecx
+	movl 32(%esp),%ecx
-	movl 16(%ebp),%edx
+	movl 36(%esp),%edx
-	movl 20(%ebp),%esi
+	movl 40(%esp),%esi
-	movl 24(%ebp),%edi
+	movl 44(%esp),%edi
-	movl 28(%ebp),%ebp
+	movl 48(%esp),%ebp
-1:	int $128
+	int $128
+.global __cp_end
+__cp_end:
 	popl %ebp
 	popl %edi
 	popl %esi
 	popl %ebx
-	xorl %edx,%edx
-	movl 4(%esp),%ecx
-	movl %edx,(%ecx)
 	ret
-2:	call __cancel

src/thread/x86_64/syscall_cp.s

@@ -2,12 +2,12 @@
 .global __syscall_cp_asm
 .type   __syscall_cp_asm,@function
 __syscall_cp_asm:
-	lea 1f(%rip),%rax
+
-	mov %rax,8(%rdi)
+.global __cp_begin
-	mov %rsp,(%rdi)
+__cp_begin:
-	mov 16(%rdi),%eax
+	mov (%rdi),%eax
 	test %eax,%eax
-	jnz 2f
+	jnz __cancel
 	mov %rdi,%r11
 	mov %rsi,%rax
 	mov %rdx,%rdi
@@ -17,9 +17,7 @@ __syscall_cp_asm:
 	mov 8(%rsp),%r8
 	mov 16(%rsp),%r9
 	mov %r11,8(%rsp)
-1:	syscall
+	syscall
-	xor %ecx,%ecx
+.global __cp_end
-	mov 8(%rsp),%rdi
+__cp_end:
-	mov %rcx,(%rdi)
 	ret
-2:	call __cancel