Re: [PATCH 2/4] x86: get rid of KERNEL_STACK_OFFSET

[Denys Vlasenko <dvlasenk@redhat.com>]

On 02/24/2015 08:30 PM, Steven Rostedt wrote:
> On Tue, 24 Feb 2015 19:51:33 +0100
> Denys Vlasenko <dvlasenk@redhat.com> wrote:
> 
>> PER_CPU_VAR(kernel_stack) was set up in a way where it points
>> five stack slots below the top of stack.
>>
>> Presumably, it was done to avoid one "sub $5*8,%rsp"
>> in syscall/sysenter code paths, where iret frame needs to be
>> created by hand.
>>
>> Ironically, none of them benefit from this optimization,
>> since all of them need to allocate additional data on stack
>> (struct pt_regs), so they still have to perform subtraction.
>> And ia32_sysenter_target even needs to *undo* this optimization:
>> it constructs iret stack with pushes instead of movs,
>> so it needs to start right at the top.
>>
>> This patch eliminates KERNEL_STACK_OFFSET.
>> PER_CPU_VAR(kernel_stack) now points directly to top of stack.
>> pt_regs allocations are adjusted to allocate iret frame as well.
>>
> 
> I always thought the KERNEL_STACK_OFFSET wasn't an optimization, but a
> buffer from the real top of stack, in case we had any off by one bugs,
> it wouldn't crash the system.

I was thinking about it, but it looks unlikely. Reasons:

(1) ia32_sysenter_target does "addq $(KERNEL_STACK_OFFSET),%rsp"
on entry before saving registers with PUSHes,
this returns %rsp to the very top of kernel stack.
If that is a problem (say, a NMI at this point would do bad things),
it would be noticed by now.

(2) even ordinary 64-bit syscall path uses IRET return at times.
For one, on every execve and signal return (because they need
to load a modified %rsp). With current layout, return frame
for IRET lies exactly there, in those 5 stack slots "reserved"
via KERNEL_STACK_OFFSET thingy.

(3) There are no comments anywhere about KERNEL_STACK_OFFSET being
a safety measure.