On 7/10/19 9:09 PM, Bart Van Assche wrote:
> On 7/10/19 11:44 AM, Eric Dumazet wrote:
>> If anything using workqueues in dynamically allocated objects can turn off
>> lockdep,
>> we have a serious issue.
>
> As far as I know that issue is only hit by syzbot tests.
> Anyway, I see
> two possible paths forward:
> * Revert the patch that makes workqueues use dynamic lockdep keys and
> thereby reintroduce the false positives that lockdep reports if
> different workqueues share a lockdep key. I think there is agreement
> that having to analyze lockdep false positives is annoying, time
> consuming and something nobody likes.
> * Modify lockdep such that space in its fixed size arrays that is no
> longer in use gets reused. Since the stack traces saved in the
> stack_trace[] array have a variable size that array will have to be
> compacted to avoid fragmentation.
>
Can not destroy_workqueue() undo what alloc_workqueue() did ?
