Re: [PATCH] rcu: Make vmalloc_dump_obj() call in a preemptible context

["Leizhen (ThunderTown)" <thunder.leizhen@huawei.com>]

On 2022/11/17 13:20, Zhang, Qiang1 wrote:
> 
> 
> On 2022/11/16 22:15, Zqiang wrote:
>> Currently, the mem_dump_obj() is invoked in call_rcu(), the
>>
>> Plan to call mem_dump_obj() in rcu_do_batch().
>>
>> call_rcu() is maybe invoked in non-preemptive code segment,
>> for object allocated from vmalloc(), the following scenarios
>> may occur:
>>
>> 	CPU 0
>> tasks context
>>    spin_lock(&vmap_area_lock)
>>           Interrupt context
>> 	      call_rcu()
>> 		mem_dump_obj
>>                   vmalloc_dump_obj
>>                     spin_lock(&vmap_area_lock) <--deadlock
>>
>> and for PREEMPT-RT kernel, the spinlock will convert to sleepable
>> lock, it also make vmap_area_lock spinlock can not acquire in
>> non-preemptive code segment. therefore, this commit make the
>> vmalloc_dump_obj() call in a preemptible context.
>>
>> Signed-off-by: Zqiang <qiang1.zhang@intel.com>
>> ---
>>  mm/util.c | 10 +++++++---
>>  1 file changed, 7 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/util.c b/mm/util.c
>> index 12984e76767e..465f8b8824ca 100644
>> --- a/mm/util.c
>> +++ b/mm/util.c
>> @@ -1124,8 +1124,12 @@ void mem_dump_obj(void *object)
>>  		return;
>>  	}
>>  
>> -	if (vmalloc_dump_obj(object))
>> -		return;
>> +	if (is_vmalloc_addr(object)) {
>> +		if (preemptible() && vmalloc_dump_obj(object))
>>
>> preemptible() --> !in_interrupt()?
> 
> Hi Leizhen 
> 
> !in_interrupt()  not include following scenario:
> 
> task context:
> 
> local_irq_save() or preempt_disable();
> call_rcu()
>       mem_dump_obj()
> 	vmalloc_dump_obj()
> 		spin_lock(&vmap_area_lock)
> local_irq_restore() or preempt_denable();
> 
> for PREEMPT-RT kernel, spinlock will convert to sleepable lock.

OK.

> 
>>
>> +			return;
>> +		type = "vmalloc memory";
>> +		goto end;
>> +	}
>>
>> Move into vmalloc_dump_obj()?
> 
> you mean move preemptible() into vmalloc_dump_obj()?

Move all? Currently, vmalloc_dump_obj() is invoked only by
mem_dump_obj(). The encapsulation is a little better in my opinion.

> 
> Thanks
> Zqiang
> 
>>
>>  
>>  	if (virt_addr_valid(object))
>>  		type = "non-slab/vmalloc memory";
>> @@ -1135,7 +1139,7 @@ void mem_dump_obj(void *object)
>>  		type = "zero-size pointer";
>>  	else
>>  		type = "non-paged memory";
>> -
>> +end:
>>  	pr_cont(" %s\n", type);
>>  }
>>  EXPORT_SYMBOL_GPL(mem_dump_obj);
>>
> 

-- 
Regards,
  Zhen Lei