Hi, Ridong,
On 2025/11/17 19:37, Chen Ridong wrote:
On 2025/11/17 18:00, Sun Shaojie wrote:
Certainly, this rule applies regardless of whether cs1 or cs2 is exclusive, and the current implementation already handles it this way. The following two cases cover this rule. "1. If cs1 is exclusive, cs1 and cs2 must be mutually exclusive" "3. If cs2 is exclusive, cs2's allowed CPUs cannot be a subset of cs1's exclusive CPUs"
I believe this function should return the same result regardless of whether it is called as cpus_excl_conflict(A1, B1) or cpus_excl_conflict(B1, A1), which means cs1 and cs2 should be treated symmetrically. However, since cs1 and cs2 are handled differently, it is difficult to convince me that this implementation is correct.
In patch v5, modifications to the cpus_excl_conflict interface have been avoided, along with preventing the following ineffective scenario.
Both A1 and B1 are exclusive, change B1's cpuset.cpus, avoid A1 becoming non-exclusive.
Looking forward to your feedback on patch v5. patch v5 : https://lore.kernel.org/cgroups/20251119105749.1385946-1-sunshaojie@kylinos....
Thanks, Sun Shaojie