Tejun Heo 於 3/25/2020 2:23 AM 寫道:
Hello,
On Fri, Mar 20, 2020 at 04:08:47PM +0800, Wright Feng wrote:
So two things changed, 1) WQ_HIGHPRI flag added *and* 2) use
allow_workqueue basically dropping __WQ_ORDERED. Not sure which one
contributes to the behavior described in the commit message.
The combination of Unbound and max_active==1 implies ordered, so I suppose
we don't need to set __WQ_ORDERED bit in flags.
It doesn't on NUMA setups. If you need ordered execution, you need the flag.
Thanks.
Hi Tejun,
In kernel/workqueue.c, it helps set __WQ_ORDERED flag for the
condition(Unbound && max_active == 1) to keep the previous behavior to
avoid subtle breakages on NUMA.
Does it mean we don't have to set the flags in alloc_workqueue? or I
misunderstand something?
If that's incorrect, would you please give me a hint how to set
__WQ_ORDERED(internal use flag) and WQ_HIGHPRI flags at the same time?
---
/*
* Unbound && max_active == 1 used to imply ordered, which is no
* longer the case on NUMA machines due to per-node pools. While
* alloc_ordered_workqueue() is the right way to create an ordered
* workqueue, keep the previous behavior to avoid subtle breakages
* on NUMA.
*/
if ((flags & WQ_UNBOUND) && max_active == 1)
flags |= __WQ_ORDERED;
---
Regards,
Wright
