Re: [PATCH v2] mm/hugetlb: defer free_huge_page() to a workqueue

[Waiman Long <longman@redhat.com>]

On 12/16/19 8:37 AM, Michal Hocko wrote:
> On Thu 12-12-19 11:04:27, Davidlohr Bueso wrote:
>> There have been deadlock reports[1, 2] where put_page is called
>> from softirq context and this causes trouble with the hugetlb_lock,
>> as well as potentially the subpool lock.
>>
>> For such an unlikely scenario, lets not add irq dancing overhead
>> to the lock+unlock operations, which could incur in expensive
>> instruction dependencies, particularly when considering hard-irq
>> safety. For example PUSHF+POPF on x86.
>>
>> Instead, just use a workqueue and do the free_huge_page() in regular
>> task context.
> I am afraid that work_struct is too large to be stuffed into the struct
> page array (because of the lockdep part).
>
> I think that it would be just safer to make hugetlb_lock irq safe. Are
> there any other locks that would require the same?

Currently, free_huge_page() can be called from the softIRQ context. The
hugetlb_lock will be acquired during that call. The subpool lock may
conditionally be acquired as well.

I am still torn between converting both locks to be irq-safe or
deferring the freeing to a workqueue.

Cheers,
Longman