From 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 Mon Sep 17 00:00:00 2001 From: Tim Chen Date: Mon, 21 Sep 2026 17:37:22 -0700 Subject: sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling bug alb_break_llc() decides whether to break LLC preference during active load balance. It does so by testing that every runnable fair task on the source rq prefers its LLC: env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable But the two counters cover different sets. nr_pref_llc_running is updated in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued, so it follows queued tasks. h_nr_runnable is updated in set_delayed()/ clear_delayed() and drops delay-dequeued tasks. So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks, alb_break_llc() returns false, and active balance is free to pull a task off its preferred LLC. Active balance only moves runnable tasks, and this is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB skips the per-task test in can_migrate_task(). The runnable set is the one we want. Fix it on the counter side. A task should be counted in nr_pref_llc_running exactly while it is both queued on its preferred LLC (pref_llc_queued) and runnable (!sched_delayed). Define that membership once in task_pref_llc_runnable(), and adjust the counter only through pref_llc_running_inc()/pref_llc_running_dec() from the four sites that change either input: account_llc_enqueue(), account_llc_dequeue(), set_delayed() and clear_delayed(). Gating every update on the same predicate keeps the delay, wake and dequeue paths from double-counting or underflowing; see the comments at those sites for the ordering. nr_llc_running and sd->llc_counts are not touched and stay on queued semantics. Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC") Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/ Reported-by: Zhan Xusheng Suggested-by: Chen Yu Signed-off-by: Tim Chen Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Kayra Cizmeci Cc: # v7.2.x Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com --- Documentation/admin-guide/cpu-load.rst | 117 +++++++++++++++++++++++++++++++++ 1 file changed, 117 insertions(+) create mode 100644 Documentation/admin-guide/cpu-load.rst (limited to 'Documentation/admin-guide/cpu-load.rst') diff --git a/Documentation/admin-guide/cpu-load.rst b/Documentation/admin-guide/cpu-load.rst new file mode 100644 index 000000000..21a984337 --- /dev/null +++ b/Documentation/admin-guide/cpu-load.rst @@ -0,0 +1,117 @@ +======== +CPU load +======== + +Linux exports various bits of information via ``/proc/stat`` and +``/proc/uptime`` that userland tools, such as top(1), use to calculate +the average time system spent in a particular state, for example:: + + $ iostat + Linux 2.6.18.3-exp (linmac) 02/20/2007 + + avg-cpu: %user %nice %system %iowait %steal %idle + 10.01 0.00 2.92 5.44 0.00 81.63 + + ... + +Here the system thinks that over the default sampling period the +system spent 10.01% of the time doing work in user space, 2.92% in the +kernel, and was overall 81.63% of the time idle. + +In most cases the ``/proc/stat`` information reflects the reality quite +closely, however due to the nature of how/when the kernel collects +this data sometimes it can not be trusted at all. + +So how is this information collected? Whenever timer interrupt is +signalled the kernel looks what kind of task was running at this +moment and increments the counter that corresponds to this tasks +kind/state. The problem with this is that the system could have +switched between various states multiple times between two timer +interrupts yet the counter is incremented only for the last state. + + +Example +------- + +If we imagine the system with one task that periodically burns cycles +in the following manner:: + + time line between two timer interrupts + |--------------------------------------| + ^ ^ + |_ something begins working | + |_ something goes to sleep + (only to be awaken quite soon) + +In the above situation the system will be 0% loaded according to the +``/proc/stat`` (since the timer interrupt will always happen when the +system is executing the idle handler), but in reality the load is +closer to 99%. + +One can imagine many more situations where this behavior of the kernel +will lead to quite erratic information inside ``/proc/stat``:: + + + /* gcc -o hog smallhog.c */ + #include + #include + #include + #include + #define HIST 10 + + static volatile sig_atomic_t stop; + + static void sighandler(int signr) + { + (void) signr; + stop = 1; + } + + static unsigned long hog (unsigned long niters) + { + stop = 0; + while (!stop && --niters); + return niters; + } + + int main (void) + { + int i; + struct itimerval it = { + .it_interval = { .tv_sec = 0, .tv_usec = 1 }, + .it_value = { .tv_sec = 0, .tv_usec = 1 } }; + sigset_t set; + unsigned long v[HIST]; + double tmp = 0.0; + unsigned long n; + signal(SIGALRM, &sighandler); + setitimer(ITIMER_REAL, &it, NULL); + + hog (ULONG_MAX); + for (i = 0; i < HIST; ++i) v[i] = ULONG_MAX - hog(ULONG_MAX); + for (i = 0; i < HIST; ++i) tmp += v[i]; + tmp /= HIST; + n = tmp - (tmp / 3.0); + + sigemptyset(&set); + sigaddset(&set, SIGALRM); + + for (;;) { + hog(n); + sigwait(&set, &i); + } + return 0; + } + + +References +---------- + +- https://lore.kernel.org/r/loom.20070212T063225-663@post.gmane.org +- Documentation/filesystems/proc.rst (1.8) + + +Thanks +------ + +Con Kolivas, Pavel Machek -- cgit v1.3.1