Re: [RFC 2/3] x86: Switch all C consumers of kernel_stack to this_cpu_sp0

[Andy Lutomirski]

On Feb 27, 2015 1:12 PM, "Denys Vlasenko" <dvlasenk@xxxxxxxxxx> wrote:
>
> On 02/27/2015 08:56 PM, Andy Lutomirski wrote:
> > On Feb 27, 2015 8:13 AM, "Denys Vlasenko" <dvlasenk@xxxxxxxxxx> wrote:
> >>
> >> On 02/27/2015 01:07 AM, Andy Lutomirski wrote:
> >>> This will make modifying the semantics of kernel_stack easier.
> >>>
> >>> Cc: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
> >>> Cc: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
> >>> Cc: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
> >>> Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
> >>> ---
> >>>  arch/x86/include/asm/thread_info.h | 3 +--
> >>>  arch/x86/kernel/traps.c            | 2 +-
> >>>  2 files changed, 2 insertions(+), 3 deletions(-)
> >>>
> >>> diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h
> >>> index e82e95abc92b..92549053d86d 100644
> >>> --- a/arch/x86/include/asm/thread_info.h
> >>> +++ b/arch/x86/include/asm/thread_info.h
> >>> @@ -163,8 +163,7 @@ DECLARE_PER_CPU(unsigned long, kernel_stack);
> >>>  static inline struct thread_info *current_thread_info(void)
> >>>  {
> >>>       struct thread_info *ti;
> >>> -     ti = (void *)(this_cpu_read_stable(kernel_stack) +
> >>> -                   KERNEL_STACK_OFFSET - THREAD_SIZE);
> >>> +     ti = (void *)(this_cpu_sp0() - THREAD_SIZE);
> >>>       return ti;
> >>>  }
> >>>
> >>> diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
> >>> index c74f2f5652da..d287ea779728 100644
> >>> --- a/arch/x86/kernel/traps.c
> >>> +++ b/arch/x86/kernel/traps.c
> >>> @@ -174,7 +174,7 @@ void ist_begin_non_atomic(struct pt_regs *regs)
> >>>        * will catch asm bugs and any attempt to use ist_preempt_enable
> >>>        * from double_fault.
> >>>        */
> >>> -     BUG_ON(((current_stack_pointer() ^ this_cpu_read_stable(kernel_stack))
> >>> +     BUG_ON(((current_stack_pointer() ^ (this_cpu_sp0() - 1))
> >>>               & ~(THREAD_SIZE - 1)) != 0);
> >>
> >> While we are at it, I propose a more readable version of this check:
> >>
> >> BUG_ON(this_cpu_sp0() - current_stack_pointer() >= THREAD_SIZE);
> >>
> >> Yes, I am aware that it is not equivalent to the existing condition
> >> - it uses the fact that this_cpu_sp0(), previous check
>              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> Oops...
> I meant to say "...the fact that this_cpu_sp0() points to the very top
> of the stack, previous check ..."
>
> >> wasn't making that assumption. But that assumption is true,
> >> so shouldn't be a problem.
> >
> > You're missing an absolute value in here, though.  This isn't a check
> > for overflow; it's a check that we aren't on an IST or other per cpu
> > stack.
>
> Yes, that's exactly what the condition checks for. It reads
>
> "is current stack pointer below task's kernel stack by no more
> than THREAD_SIZE?"
>
> which is only possible if we are on task's kernel stack:
>
> If current_stack_pointer() is elsewhere, it is either
> (a) much smaller than this_cpu_sp0(), and BUG_ON condition
>     obviously triggers; or
> (b) it is somewhere above this_cpu_sp0(), in which case subtraction
>     overflows and condition triggers too.
>

Right, it's unsigned.  I like yours better if I add a comment.  It
also avoids relying on the stack being aligned to its full size.

--Andy

> How would you write this condition so that it's easily readable?
> Evidently, my version isn't as readable sa I hoped...
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/