Re: [PATCH] x86, entry: Check for syscall exit work with IRQs disabled

[Andy Lutomirski]

On Mon, Mar 23, 2015 at 12:27 PM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> On Mon, Mar 23, 2015 at 12:21 PM, Denys Vlasenko <dvlasenk@xxxxxxxxxx> wrote:
>> On 03/23/2015 08:10 PM, Andy Lutomirski wrote:
>>> We currently have a race: if we're preempted during syscall exit, we
>>> can fail to process syscall return work that is queued up while
>>> we're preempted in ret_from_sys_call after checking ti.flags.
>>>
>>> Fix it by disabling interrupts before checking ti.flags.
>>>
>>> Fixes: 96b6352c1271 x86_64, entry: Remove the syscall exit audit and schedule optimizations
>>> Reported-by: Stefan Seyfried <stefan.seyfried@xxxxxxxxxxxxxx>
>>> Reported-by: Takashi Iwai <tiwai@xxxxxxx>
>>> Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxx>
>>> ---
>>>
>>> Ingo, I don't understand the LOCKDEP_SYS_EXIT stuff.  Can you take a quick
>>> look to confirm that it's okay to call it more than once?
>>>
>>> arch/x86/kernel/entry_64.S | 18 ++++++++++++++----
>>>  1 file changed, 14 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
>>> index 1d74d161687c..2babb393915e 100644
>>> --- a/arch/x86/kernel/entry_64.S
>>> +++ b/arch/x86/kernel/entry_64.S
>>> @@ -364,12 +364,21 @@ system_call_fastpath:
>>>   * Has incomplete stack frame and undefined top of stack.
>>>   */
>>>  ret_from_sys_call:
>>> -     testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
>>> -     jnz int_ret_from_sys_call_fixup /* Go the the slow path */
>>> -
>>>       LOCKDEP_SYS_EXIT
>>>       DISABLE_INTERRUPTS(CLBR_NONE)
>>>       TRACE_IRQS_OFF
>>> +
>>> +     /*
>>> +      * We must check ti flags with interrupts (or at least preemption)
>>> +      * off because we must *never* return to userspace without
>>> +      * processing exit work that is enqueued if we're preempted here.
>>> +      * In particular, returning to userspace with any of the one-shot
>>> +      * flags (TIF_NOTIFY_RESUME, TIF_USER_RETURN_NOTIFY, etc) set is
>>> +      * very bad.
>>> +      */
>>> +     testl $_TIF_ALLWORK_MASK,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
>>> +     jnz int_ret_from_sys_call_fixup /* Go the the slow path */
>>                                            ^^^^^^^^^^^^^^^^^^^^
>>
>> typo here; s/the the/to the/
>
> Whoops.
>
>>
>>
>>> +
>>>       CFI_REMEMBER_STATE
>>>       /*
>>>        * sysretq will re-enable interrupts:
>>> @@ -386,7 +395,7 @@ ret_from_sys_call:
>>>
>>>  int_ret_from_sys_call_fixup:
>>>       FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
>>> -     jmp int_ret_from_sys_call
>>> +     jmp int_ret_from_sys_call_irqs_off
>>>
>>>       /* Do syscall tracing */
>>>  tracesys:
>>> @@ -432,6 +441,7 @@ tracesys_phase2:
>>>  GLOBAL(int_ret_from_sys_call)
>>>       DISABLE_INTERRUPTS(CLBR_NONE)
>>>       TRACE_IRQS_OFF
>>> +int_ret_from_sys_call_irqs_off:
>>>       movl $_TIF_ALLWORK_MASK,%edi
>>>       /* edi: mask to check */
>>>  GLOBAL(int_with_check)
>>
>>
>> You can avoid having to know LOCKDEP_SYS_EXIT    :)
>> Just set %edi = $_TIF_ALLWORK_MASK, and jump a bit farther:
>>
>>
>>         movl $_TIF_ALLWORK_MASK,%edi
>>         testl %edi,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
>>         jnz int_ret_from_sys_call_fixup /* Go to the slow path */
>> ...
>> ...
>> GLOBAL(int_ret_from_sys_call)
>>         DISABLE_INTERRUPTS(CLBR_NONE)
>>         TRACE_IRQS_OFF
>>         movl $_TIF_ALLWORK_MASK,%edi
>>         /* edi: mask to check */
>> GLOBAL(int_with_check)
>>         LOCKDEP_SYS_EXIT_IRQ
>> int_ret_from_sys_call_irqs_off:  <========== HERE
>>
>
> I didn't want to do that, because I really want to rewrite
> int_ret_from_sys_call in C.
>

To say that better: I don't want to further spread the %edi garbage
around entry_64.S.  Saving a single load on the slow path isn't worth
any of this complexity, and, if we're going to rewrite it in C anyway,
then maybe we could consider microoptimizations like that later on.

--Andy
--
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/