riscv: Stop using kstack_end()
kstack_end() is a hack that is not correct on all arches. The ancient
comment suggests that it masterfully works fully correctly on x86, for
example, but it's actually wrong on x86.
Since it could plausibly be correct on riscv, copy the implementation to
sh near its only call site.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
diff --git a/arch/riscv/kernel/stacktrace.c b/arch/riscv/kernel/stacktrace.c
index a4b1d943..e5142f3 100644
--- a/arch/riscv/kernel/stacktrace.c
+++ b/arch/riscv/kernel/stacktrace.c
@@ -75,6 +75,12 @@ static void notrace walk_stackframe(struct task_struct *task,
#else /* !CONFIG_FRAME_POINTER */
+static bool end_of_stack(void *addr)
+{
+ return !(((unsigned long)addr+sizeof(void*)-1) &
+ (THREAD_SIZE-sizeof(void*)));
+}
+
static void notrace walk_stackframe(struct task_struct *task,
struct pt_regs *regs, bool (*fn)(unsigned long, void *), void *arg)
{
@@ -98,7 +104,7 @@ static void notrace walk_stackframe(struct task_struct *task,
return;
ksp = (unsigned long *)sp;
- while (!kstack_end(ksp)) {
+ while (!end_of_stack(ksp)) {
if (__kernel_text_address(pc) && unlikely(fn(pc, arg)))
break;
pc = (*ksp++) - 0x4;
