On Sun, Feb 05, 2023 at 12:00:25AM -0500, Waiman Long wrote: > On 2/4/23 05:01, Peter Zijlstra wrote: > > On Fri, Feb 03, 2023 at 11:40:40AM -0500, Waiman Long wrote: > > > Since commit 8f9ea86fdf99 ("sched: Always preserve the user > > > requested cpumask"), relax_compatible_cpus_allowed_ptr() is calling > > > __sched_setaffinity() unconditionally. This helps to expose a bug in > > > the current cpuset hotplug code where the cpumasks of the tasks in > > > the top cpuset are not updated at all when some CPUs become online or > > > offline. It is likely caused by the fact that some of the tasks in the > > > top cpuset, like percpu kthreads, cannot have their cpu affinity changed. > > > > > > One way to reproduce this as suggested by Peter is: > > > - boot machine > > > - offline all CPUs except one > > > - taskset -p ffffffff $$ > > > - online all CPUs > > > > > > Fix this by allowing cpuset_cpus_allowed() to return a wider mask that > > > includes offline CPUs for those tasks that are in the top cpuset. For > > > tasks not in the top cpuset, the old rule applies and only online CPUs > > > will be returned in the mask since hotplug events will update their > > > cpumasks accordingly. > > So you get the task_cpu_possible_mask() interaction vs cpusets horribly > > wrong here, but given the very sorry state of task_cpu_possible_mask() > > correctness of cpuset as a whole that might just not matter at this > > point. > > > > I do very much hate how you add exceptions on exceptions instead of > > looking to do something right :-( > > > > Fixing that parition case in my patch is 1 extra line and then I think > > it fundamentally does the right thing and can serve as a basis for > > fixing cpuset as a whole. > > I am not saying that your patch is incorrect other than handling the > partition case. However, it is rather complex and is hard to understand > especially for those that are not that familiar with the cpuset code. From > the maintainability point of view, a simpler solution that is easier to > understand is better. > > If we want to get it into the next merge windows, there isn't much time left > for linux-next testing. So a lower risk solution is better from that > perspective too. This needs to land for 6.2 to fix the regression. The next merge window is too late. That's why I cooked the reverts [1] as an alternative. Will [1] https://git.kernel.org/pub/scm/linux/kernel/git/will/linux.git/log/?h=ssa-reverts