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

[Waiman Long]

On 12/16/19 8:26 AM, Michal Hocko wrote:
> On Thu 12-12-19 15:52:20, Waiman Long wrote:
>> On 12/12/19 2:22 PM, Mike Kravetz wrote:
>>> On 12/12/19 11:04 AM, 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.
>>>>
>>>> [1] https://lore.kernel.org/lkml/20191211194615.18502-1-longman@xxxxxxxxxx/
>>>> [2] https://lore.kernel.org/lkml/20180905112341.21355-1-aneesh.kumar@xxxxxxxxxxxxx/
>>>>
>>>> Reported-by: Waiman Long <longman@xxxxxxxxxx>
>>>> Reported-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxx>
>>>> Signed-off-by: Davidlohr Bueso <dbueso@xxxxxxx>
>>> Thank you Davidlohr.
>>>
>>> The patch does seem fairly simple and straight forward.  I need to brush up
>>> on my workqueue knowledge to provide a full review.
>>>
>>> Longman,
>>> Do you have a test to reproduce the issue?  If so, can you try running with
>>> this patch.
>> Yes, I do have a test that can reproduce the issue. I will run it with
>> the patch and report the status tomorrow.
> Can you extract guts of the testcase and integrate them into hugetlb
> test suite?

The test case that I used is the Red Hat internal "Fork vs. fast GUP
race test" written by Jarod Wilson. I would have to ask him if he is OK
with that.

Cheers,
Longman