arm64: stackleak: fix current_top_of_stack()

[ Upstream commit e85094c ]

Due to some historical confusion, arm64's current_top_of_stack() isn't
what the stackleak code expects. This could in theory result in a number
of problems, and practically results in an unnecessary performance hit.
We can avoid this by aligning the arm64 implementation with the x86
implementation.

The arm64 implementation of current_top_of_stack() was added
specifically for stackleak in commit:

  0b3e336 ("arm64: Add support for STACKLEAK gcc plugin")

This was intended to be equivalent to the x86 implementation, but the
implementation, semantics, and performance characteristics differ
wildly:

* On x86, current_top_of_stack() returns the top of the current task's
  task stack, regardless of which stack is in active use.

  The implementation accesses a percpu variable which the x86 entry code
  maintains, and returns the location immediately above the pt_regs on
  the task stack (above which x86 has some padding).

* On arm64 current_top_of_stack() returns the top of the stack in active
  use (i.e. the one which is currently being used).

  The implementation checks the SP against a number of
  potentially-accessible stacks, and will BUG() if no stack is found.

The core stackleak_erase() code determines the upper bound of stack to
erase with:

| if (on_thread_stack())
|         boundary = current_stack_pointer;
| else
|         boundary = current_top_of_stack();

On arm64 stackleak_erase() is always called on a task stack, and
on_thread_stack() should always be true. On x86, stackleak_erase() is
mostly called on a trampoline stack, and is sometimes called on a task
stack.

Currently, this results in a lot of unnecessary code being generated for
arm64 for the impossible !on_thread_stack() case. Some of this is
inlined, bloating stackleak_erase(), while portions of this are left
out-of-line and permitted to be instrumented (which would be a
functional problem if that code were reachable).

As a first step towards improving this, this patch aligns arm64's
implementation of current_top_of_stack() with x86's, always returning
the top of the current task's stack. With GCC 11.1.0 this results in the
bulk of the unnecessary code being removed, including all of the
out-of-line instrumentable code.

While I don't believe there's a functional problem in practice I've
marked this as a fix since the semantic was clearly wrong, the fix
itself is simple, and other code might rely upon this in future.

Fixes: 0b3e336 ("arm64: Add support for STACKLEAK gcc plugin")
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Popov <alex.popov@linux.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Will Deacon <will@kernel.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20220427173128.2603085-2-mark.rutland@arm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>

arch/arm64/include/asm/processor.h

@@ -381,12 +381,10 @@ long get_tagged_addr_ctrl(struct task_struct *task);
  * of header definitions for the use of task_stack_page.
  */
 
-#define current_top_of_stack()								\
-({											\
-	struct stack_info _info;							\
-	BUG_ON(!on_accessible_stack(current, current_stack_pointer, 1, &_info));	\
-	_info.high;									\
-})
+/*
+ * The top of the current task's task stack
+ */
+#define current_top_of_stack()	((unsigned long)current->stack + THREAD_SIZE)
 #define on_thread_stack()	(on_task_stack(current, current_stack_pointer, 1, NULL))
 
 #endif /* __ASSEMBLY__ */