>2) find out how long the tlb batches actually are and make them smaller

PAGE_SIZE is 16k other than 4k in my kernel configuration, batch count 
max is 2015.
currently MAX_GATHER_BATCH depends on PAGE_SIZE, I will make batch's max 
size smaller without depending on PAGE_SIZE.

#define MAX_GATHER_BATCH   ((PAGE_SIZE - sizeof(struct 
mmu_gather_batch)) / sizeof(void *))

On 03/10/2022 05:11 PM, Vlastimil Babka wrote:
> On 3/10/22 04:29, Andrew Morton wrote:
>> On Thu, 10 Mar 2022 10:48:41 +0800 wangjianxing<wangjianxing@loongson.cn>  wrote:
>>
>>> spin_lock will preempt_disable(), interrupt context will
>>> __irq_enter/local_bh_disable and also add preempt count with offset.
>>>
>>> cond_resched check whether if preempt_count == 0 in first and won't
>>> schedule in previous context.
>>>
>>> Is this right?
>>>
>>>
>>> With another way, could we add some condition to avoid call cond_resched
>>> in interrupt context or spin_lock()?
>>>
>>> + if (preemptible())
>>> +       cond_resched();
>>>
>> None of this works with CONFIG_PREEMPTION=n.
> Yeah I think we have at least two options.
>
> 1) check all callers, maybe realize all have enabled interrupts anyway,
> rewrite the locking to only assume those
>
> 2) find out how long the tlb batches actually are and make them smaller