Re: Question about qspinlock nest

[Waiman Long]

On 01/10/2019 01:25 PM, James Morse wrote:
> Hi Longman, Zhenzhong,
>
> On 10/01/2019 14:43, Waiman Long wrote:
>> On 01/10/2019 03:02 AM, Zhenzhong Duan wrote:
>>> There is a question confused me for days. Appreciate an answer.
>>>
>>> In below code, the comment says we never have more than 4 nested
>>> contexts.
>>>
>>> What happen if debug and mce exceptions nest with the four, or we
>>> ensure it never happen?
>>>
>>>
>>> /*
>>> Â* Per-CPU queue node structures; we can never have more than 4 nested
>>> Â* contexts: task, softirq, hardirq, nmi.
>>> Â*
>>> Â* Exactly fits one 64-byte cacheline on a 64-bit architecture.
>>> Â*
>>> Â* PV doubles the storage and uses the second cacheline for PV state.
>>> Â*/
>>> static DEFINE_PER_CPU_ALIGNED(struct qnode, qnodes[MAX_NODES]);
>> Yes, both debug and mce exceptions are some kind of NMIs. So
>> theoretically, it is possible to have more than four. Are you aware of
>> any debug and MCE exception handlers that need to take a spinlock for
>> synchronization?
> On arm64 if all the RAS and psuedo-NMI patches land, our worst-case interleaving
> jumps to at least 7. The culprit is APEI using spinlocks to protect fixmap slots.
>
> I have an RFC to bump the number of node bits from 2 to 3, but as this is APEI
> four times, it may be preferable to make it use something other than spinlocks.
>
>
> The worst-case order is below. Each one masks those before it:
> 1. process context
> 2. soft-irq
> 3. hard-irq
> 4. psuedo-nmi [0]
>    - using the irqchip priorities to configure some IRQs as NMI.
> 5. SError [1]
>    - a bit like an asynchronous MCE. ACPI allows this to convey CPER records,
>      requiring an APEI call.
> 6&7. SDEI [2]
>      - a firmware triggered software interrupt, only its two of them, either of
>        which could convey CPER records.
> 8. Synchronous external abort

Nested spinlocks isn't a problem as long as no more than 4 of them are
contended.

Adding one more bit for node will halving the number of CPUs supported
without going to the slower version with a single pending bit.

Cheers,
Longman