On Thu, Oct 05, 2023 at 05:25:29PM +0800, Zhu Yanjun wrote: > 在 2023/10/5 2:38, Jason Gunthorpe 写道: > > On Wed, Oct 04, 2023 at 10:43:20AM -0700, Bart Van Assche wrote: > > > > > Thank you for having reported this. I'm OK with integrating the > > > above change in my patch. However, code changes must be > > > motivated. Do you perhaps have an explanation of why WQ_HIGHPRI > > > makes the issue disappear that you observed? > > > > I think it is clear there are locking bugs in all this, so it is not > > surprising that changing the scheduling behavior can make locking bugs > > hide > > > > Jason > > With the flag WQ_HIGHPRI, an ordered workqueue with high priority is > allocated. With this workqueue, to now, the test has run for several days. > And the problem did not appear. So to my test environment, this problem > seems fixed with the above commit. > > I doubt, without the flag WQ_HIGHPRI, the workqueue is allocated with normal > priority. The work item in this workqueue is more likely preempted than high > priority work queue. Which is why it shows there are locking problems in this code. Jason
