On Fri, Jun 4, 2021 at 5:15 AM Hillf Danton [off-list ref] wrote:
Good material to be shoehorned into commit message though without sheding
any light first on the reasons why CTRL is so special its work wont bother
being canceled.
Not sure what you mean. If it's that CTRL work isn't being cancelled,
but OFLD is being cancelled twice, thanks for pointing it. I'm sending
a new patch.
How long? Long enough for the kworker to become a CPU hog?
What is it in your opinion to cut the chance for that risk?
I don't think so, not by its own, but the maybe yes the sum of many
tasks in softirq context (it will depend on what else is being
executed in the system). As far as I know, moving work from softirq to
threads to reduce latencies is seen as a good thing, unless it's
really necessary.
--
Íñigo Huguet