[PATCH] x86, traps: install gates using IST after cpu_init().

[PATCH] x86, traps: install gates using IST after cpu_init().

[Wang Nan]

X86_TRAP_NMI, X86_TRAP_DF and X86_TRAP_MC use their own stack. Those
stacks are invalid until cpu_init() installs TSS.

This patch moves setting of the 3 gates after cpu_init().

Signed-off-by: Wang Nan <wangnan0@huawei.com>
---

If I understand correctly, logically speaking the original code is
incorrect.  However, there is no real bug caused by it for serval years.
I'm not sure whether this fix is practical or not. Fix them only for
logical correctness.

---
 arch/x86/kernel/traps.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
index 4281988..cf7898e 100644
--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -962,7 +962,6 @@ void __init trap_init(void)
 #endif
 
 	set_intr_gate(X86_TRAP_DE, divide_error);
-	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
 	/* int4 can be called from all */
 	set_system_intr_gate(X86_TRAP_OF, &overflow);
 	set_intr_gate(X86_TRAP_BR, bounds);
@@ -970,8 +969,6 @@ void __init trap_init(void)
 	set_intr_gate(X86_TRAP_NM, device_not_available);
 #ifdef CONFIG_X86_32
 	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
-#else
-	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
 #endif
 	set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
 	set_intr_gate(X86_TRAP_TS, invalid_TSS);
@@ -981,9 +978,6 @@ void __init trap_init(void)
 	set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
 	set_intr_gate(X86_TRAP_MF, coprocessor_error);
 	set_intr_gate(X86_TRAP_AC, alignment_check);
-#ifdef CONFIG_X86_MCE
-	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
-#endif
 	set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
 
 	/* Reserve all the builtin and the syscall vector: */
@@ -1013,6 +1007,14 @@ void __init trap_init(void)
 	 */
 	cpu_init();
 
+	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
+#ifndef CONFIG_X86_32
+	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
+#endif
+#ifdef CONFIG_X86_MCE
+	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
+#endif
+
 	/*
 	 * X86_TRAP_DB and X86_TRAP_BP have been set
 	 * in early_trap_init(). However, DEBUG_STACK works only after
-- 
1.8.4

Re: [PATCH] x86, traps: install gates using IST after cpu_init().

[Andy Lutomirski]

On Wed, Feb 25, 2015 at 10:15 PM, Wang Nan <wangnan0@huawei.com> wrote:
> X86_TRAP_NMI, X86_TRAP_DF and X86_TRAP_MC use their own stack. Those
> stacks are invalid until cpu_init() installs TSS.
>
> This patch moves setting of the 3 gates after cpu_init().
>
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> ---
>
> If I understand correctly, logically speaking the original code is
> incorrect.  However, there is no real bug caused by it for serval years.
> I'm not sure whether this fix is practical or not. Fix them only for
> logical correctness.

Acked-by: Andy Lutomirski <luto@amacapital.net>

That being said, I'm pretty sure you're not fixing a bug here.
Delivery of an exception with no handler is every bit as fatal as
delivery of an exception with a non-working IST handler.

--Andy

Re: [PATCH] x86, traps: install gates using IST after cpu_init().

[Wang Nan]

On 2015/2/26 23:14, Andy Lutomirski wrote:
> On Wed, Feb 25, 2015 at 10:15 PM, Wang Nan <wangnan0@huawei.com> wrote:
>> X86_TRAP_NMI, X86_TRAP_DF and X86_TRAP_MC use their own stack. Those
>> stacks are invalid until cpu_init() installs TSS.
>>
>> This patch moves setting of the 3 gates after cpu_init().
>>
>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>> ---
>>
>> If I understand correctly, logically speaking the original code is
>> incorrect.  However, there is no real bug caused by it for serval years.
>> I'm not sure whether this fix is practical or not. Fix them only for
>> logical correctness.
> 
> Acked-by: Andy Lutomirski <luto@amacapital.net>
> 
> That being said, I'm pretty sure you're not fixing a bug here.

Agree.

> Delivery of an exception with no handler is every bit as fatal as
> delivery of an exception with a non-working IST handler.
> 

Just curious: in original code, what will happen if an NMI or MC raises after
'set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);' and before cpu_init()?
In my opinion, at that time the interrupt handler is set but IST is not ready.

In addition, why it's never happened for real? Does it means NMI is possible
to be disabled?

Thank you!

> --Andy
> 

Re: [PATCH] x86, traps: install gates using IST after cpu_init().

[Andy Lutomirski]

On Thu, Feb 26, 2015 at 8:34 PM, Wang Nan <wangnan0@huawei.com> wrote:
> On 2015/2/26 23:14, Andy Lutomirski wrote:
>> On Wed, Feb 25, 2015 at 10:15 PM, Wang Nan <wangnan0@huawei.com> wrote:
>>> X86_TRAP_NMI, X86_TRAP_DF and X86_TRAP_MC use their own stack. Those
>>> stacks are invalid until cpu_init() installs TSS.
>>>
>>> This patch moves setting of the 3 gates after cpu_init().
>>>
>>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>>> ---
>>>
>>> If I understand correctly, logically speaking the original code is
>>> incorrect.  However, there is no real bug caused by it for serval years.
>>> I'm not sure whether this fix is practical or not. Fix them only for
>>> logical correctness.
>>
>> Acked-by: Andy Lutomirski <luto@amacapital.net>
>>
>> That being said, I'm pretty sure you're not fixing a bug here.
>
> Agree.
>
>> Delivery of an exception with no handler is every bit as fatal as
>> delivery of an exception with a non-working IST handler.
>>
>
> Just curious: in original code, what will happen if an NMI or MC raises after
> 'set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);' and before cpu_init()?
> In my opinion, at that time the interrupt handler is set but IST is not ready.
>
> In addition, why it's never happened for real? Does it means NMI is possible
> to be disabled?

It means that no NMI sources are enabled (hopefully) that early.  If
they were, we'd be doomed -- AFAIK it's impossible to maintain a
continuously valid IDT all the way through the transition from real
mode to long mode.  (Maybe I'm wrong and there's some trick that would
work.)

--Andy

>
> Thank you!
>
>> --Andy
>>
>
>



-- 
Andy Lutomirski
AMA Capital Management, LLC