> On Jul 24, 2020, at 10:23 AM, Ira Weiny <ira.weiny@xxxxxxxxx> wrote:
>
> On Thu, Jul 23, 2020 at 10:15:17PM +0200, Thomas Gleixner wrote:
>> Thomas Gleixner <tglx@xxxxxxxxxxxxx> writes:
>>
>>> Ira Weiny <ira.weiny@xxxxxxxxx> writes:
>>>> On Fri, Jul 17, 2020 at 12:06:10PM +0200, Peter Zijlstra wrote:
>>>>>> On Fri, Jul 17, 2020 at 12:20:56AM -0700, ira.weiny@xxxxxxxxx wrote:
>>>>> I've been really digging into this today and I'm very concerned that I'm
>>>>> completely missing something WRT idtentry_enter() and idtentry_exit().
>>>>>
>>>>> I've instrumented idt_{save,restore}_pkrs(), and __dev_access_{en,dis}able()
>>>>> with trace_printk()'s.
>>>>>
>>>>> With this debug code, I have found an instance where it seems like
>>>>> idtentry_enter() is called without a corresponding idtentry_exit(). This has
>>>>> left the thread ref counter at 0 which results in very bad things happening
>>>>> when __dev_access_disable() is called and the ref count goes negative.
>>>>>
>>>>> Effectively this seems to be happening:
>>>>>
>>>>> ...
>>>>> // ref == 0
>>>>> dev_access_enable() // ref += 1 ==> disable protection
>>>>> // exception (which one I don't know)
>>>>> idtentry_enter()
>>>>> // ref = 0
>>>>> _handler() // or whatever code...
>>>>> // *_exit() not called [at least there is no trace_printk() output]...
>>>>> // Regardless of trace output, the ref is left at 0
>>>>> dev_access_disable() // ref -= 1 ==> -1 ==> does not enable protection
>>>>> (Bad stuff is bound to happen now...)
>>>
>>> Well, if any exception which calls idtentry_enter() would return without
>>> going through idtentry_exit() then lots of bad stuff would happen even
>>> without your patches.
>>>
>>>> Also is there any chance that the process could be getting scheduled and that
>>>> is causing an issue?
>>>
>>> Only from #PF, but after the fault has been resolved and the tasks is
>>> scheduled in again then the task returns through idtentry_exit() to the
>>> place where it took the fault. That's not guaranteed to be on the same
>>> CPU. If schedule is not aware of the fact that the exception turned off
>>> stuff then you surely get into trouble. So you really want to store it
>>> in the task itself then the context switch code can actually see the
>>> state and act accordingly.
>>
>> Actually thats nasty as well as you need a stack of PKRS values to
>> handle nested exceptions. But it might be still the most reasonable
>> thing to do. 7 PKRS values plus an index should be really sufficient,
>> that's 32bytes total, not that bad.
>
> I've thought about this a bit more and unless I'm wrong I think the
> idtentry_state provides for that because each nested exception has it's own
> idtentry_state doesn't it?
Only the ones that use idtentry_enter() instead of, say, nmi_enter().
>
> Ira
