ia64: fix user_stack_pointer() for ptrace()
authorSergei Trofimovich <slyfox@gentoo.org>
Fri, 9 Apr 2021 20:27:32 +0000 (13:27 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 9 Apr 2021 21:54:23 +0000 (14:54 -0700)
ia64 has two stacks:

 - memory stack (or stack), pointed at by by r12

 - register backing store (register stack), pointed at by
   ar.bsp/ar.bspstore with complications around dirty
   register frame on CPU.

In [1] Dmitry noticed that PTRACE_GET_SYSCALL_INFO returns the register
stack instead memory stack.

The bug comes from the fact that user_stack_pointer() and
current_user_stack_pointer() don't return the same register:

  ulong user_stack_pointer(struct pt_regs *regs) { return regs->ar_bspstore; }
  #define current_user_stack_pointer() (current_pt_regs()->r12)

The change gets both back in sync.

I think ptrace(PTRACE_GET_SYSCALL_INFO) is the only affected user by
this bug on ia64.

The change fixes 'rt_sigreturn.gen.test' strace test where it was
observed initially.

Link: https://bugs.gentoo.org/769614
Link: https://lkml.kernel.org/r/20210331084447.2561532-1-slyfox@gentoo.org
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Reported-by: Dmitry V. Levin <ldv@altlinux.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/ia64/include/asm/ptrace.h

index b3aa460..0817913 100644 (file)
@@ -54,8 +54,7 @@
 
 static inline unsigned long user_stack_pointer(struct pt_regs *regs)
 {
-       /* FIXME: should this be bspstore + nr_dirty regs? */
-       return regs->ar_bspstore;
+       return regs->r12;
 }
 
 static inline int is_syscall_success(struct pt_regs *regs)
@@ -79,11 +78,6 @@ static inline long regs_return_value(struct pt_regs *regs)
        unsigned long __ip = instruction_pointer(regs);                 \
        (__ip & ~3UL) + ((__ip & 3UL) << 2);                            \
 })
-/*
- * Why not default?  Because user_stack_pointer() on ia64 gives register
- * stack backing store instead...
- */
-#define current_user_stack_pointer() (current_pt_regs()->r12)
 
   /* given a pointer to a task_struct, return the user's pt_regs */
 # define task_pt_regs(t)               (((struct pt_regs *) ((char *) (t) + IA64_STK_OFFSET)) - 1)