kernel/sched/psi.c

11  * When CPU, memory and IO are contended, tasks experience delays that
30  * In the SOME state of a given resource, one or more tasks are
32  * perform work, but the CPU may still be executing other tasks.
34  * In the FULL state of a given resource, all non-idle tasks are
48  * FULL means all non-idle tasks in the cgroup are delayed on the CPU
62  * The more tasks and available CPUs there are, the more work can be
65  * tasks and CPUs.
67  * Consider a scenario where 257 number crunching tasks are trying to
75  * Conversely, consider a scenario of 4 tasks and 4 CPUs where at any
76  * given time *one* of the tasks is delayed due to a lack of memory.
83  * we have to base our calculation on the number of non-idle tasks in
86  * delayed tasks to possible threads, and FULL is the share of possible
99  * For the 1 out of 4 memory-delayed tasks, this yields:
221 static bool test_state(unsigned int *tasks, enum psi_states state, bool oncpu)
225 		return unlikely(tasks[NR_IOWAIT]);
227 		return unlikely(tasks[NR_IOWAIT] && !tasks[NR_RUNNING]);
229 		return unlikely(tasks[NR_MEMSTALL]);
231 		return unlikely(tasks[NR_MEMSTALL] &&
232 			tasks[NR_RUNNING] == tasks[NR_MEMSTALL_RUNNING]);
234 		return unlikely(tasks[NR_RUNNING] > oncpu);
236 		return unlikely(tasks[NR_RUNNING] && !oncpu);
238 		return tasks[NR_IOWAIT] || tasks[NR_MEMSTALL] ||
239 			tasks[NR_RUNNING];
251 	unsigned int tasks[NR_PSI_TASK_COUNTS];
267 			memcpy(tasks, groupc->tasks, sizeof(groupc->tasks));
305 			reschedule = tasks[NR_RUNNING] +
306 				     tasks[NR_IOWAIT] +
307 				     tasks[NR_MEMSTALL] > 1;
783 	 * assess the aggregate resource states this CPU's tasks
812 		if (groupc->tasks[t]) {
813 			groupc->tasks[t]--;
815 			printk_deferred(KERN_ERR "psi: task underflow! cpu=%d t=%d tasks=[%u %u %u %u] clear=%x set=%x\n",
816 					cpu, t, groupc->tasks[0],
817 					groupc->tasks[1], groupc->tasks[2],
818 					groupc->tasks[3], clear, set);
825 			groupc->tasks[t]++;
845 		if (test_state(groupc->tasks, s, state_mask & PSI_ONCPU))
851 	 * when there are no other working tasks, but also when
981 		 * any other differences between the two tasks (e.g. prev goes
1196 	 * stop percpu tasks accounting in each psi_group_cpu,
1205 	 * to get correct state mask from test_state() loop on tasks[],