Re: [PATCH 1/3] infiniband: hw: hfi1: verbs.c: Use built-in RCU list checking

[Dennis Dalessandro <dennis.dalessandro@intel.com>]

On 1/14/2020 12:00 PM, Dennis Dalessandro wrote:
> On 1/14/2020 11:57 AM, Jason Gunthorpe wrote:
>> On Tue, Jan 14, 2020 at 09:53:45PM +0530, 
>> madhuparnabhowmik04@gmail.com wrote:
>>> From: Madhuparna Bhowmik <madhuparnabhowmik04@gmail.com>
>>>
>>> list_for_each_entry_rcu has built-in RCU and lock checking.
>>> Pass cond argument to list_for_each_entry_rcu.
>>>
>>> Signed-off-by: Madhuparna Bhowmik <madhuparnabhowmik04@gmail.com>
>>>   drivers/infiniband/hw/hfi1/verbs.c | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/infiniband/hw/hfi1/verbs.c 
>>> b/drivers/infiniband/hw/hfi1/verbs.c
>>> index 089e201d7550..22f2d4fd2577 100644
>>> +++ b/drivers/infiniband/hw/hfi1/verbs.c
>>> @@ -515,7 +515,7 @@ static inline void hfi1_handle_packet(struct 
>>> hfi1_packet *packet,
>>>                          opa_get_lid(packet->dlid, 9B));
>>>           if (!mcast)
>>>               goto drop;
>>> -        list_for_each_entry_rcu(p, &mcast->qp_list, list) {
>>> +        list_for_each_entry_rcu(p, &mcast->qp_list, list, 
>>> lockdep_is_held(&(ibp->rvp.lock))) {
>>
>> Okay, this looks reasonable
>>
>> Mike, Dennis, is this the right lock to test?
>>
> 
> I'm looking at that right now actually, I don't think this is correct. 
> Wanted to talk to Mike before I send a response though.
> 
> -Denny

That's definitely going to throw a ton of lock dep messages. It's not 
really the right lock either. Instead what we probably need to do is 
what we do in the non-multicast part of the code and take the 
rcu_read_lock().

I'd say hold off on this and we'll fix it right. Same goes for the qib one.

The rdmavt one though looks to be OK. I'll give it a test.

-Denny