> On 13 Sep 2021, at 10:27, Leon Romanovsky <leon@xxxxxxxxxx> wrote:
>
> On Mon, Sep 13, 2021 at 08:17:00AM +0000, Haakon Bugge wrote:
>>
>>
>>> On 13 Sep 2021, at 10:04, Leon Romanovsky <leon@xxxxxxxxxx> wrote:
>>>
>>> From: Leon Romanovsky <leonro@xxxxxxxxxx>
>>>
>>> Usnic VF doesn't need lock in atomic context to create QPs, so it is safe
>>> to use mutex instead of spinlock. Such change fixes the following smatch
>>> error.
>>
>> s/GFP_ATOMIC/GFP_KERNEL/ in find_free_vf_and_create_qp_grp() as well?
>
> Do you mean in usnic_uiom_get_dev_list()?
Sorry, my comments was v5.14.1 centric. For latest upstream, its alloc_res_chunk_list(),
find_free_vf_and_create_qp_grp()
-> usnic_ib_qp_grp_create
-> alloc_res_chunk_list()
> That GFP_ATOMIC existed before my patch while we are holding usdev_lock mutex.
True.
> Anyway, I prefer to touch that driver as less as possible.
>
> The allocations can continue to be with GFP_ATOMIC while we use mutex.
> It is bad thing, but not a necessary to fix bug. We just wasting atomic
> memory and instruct kernel do not sleep while
> doing allocations.
I assume this driver is not very much maintained by Cisco, hence, I agree with your rationale above,
Reviewed-by: Håkon Bugge <haakon.bugge@xxxxxxxxxx>
Thxs, Håkon
