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/ABI/testing/sysfs-kernel-irq | 60 ++++++++++++++++++++++++++++++ 1 file changed, 60 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-kernel-irq (limited to 'Documentation/ABI/testing/sysfs-kernel-irq') diff --git a/Documentation/ABI/testing/sysfs-kernel-irq b/Documentation/ABI/testing/sysfs-kernel-irq new file mode 100644 index 000000000..8910d0c4b --- /dev/null +++ b/Documentation/ABI/testing/sysfs-kernel-irq @@ -0,0 +1,60 @@ +What: /sys/kernel/irq +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: Directory containing information about the system's IRQs. + Specifically, data from the associated struct irq_desc. + The information here is similar to that in /proc/interrupts + but in a more machine-friendly format. This directory contains + one subdirectory for each Linux IRQ number. + +What: /sys/kernel/irq//actions +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: The IRQ action chain. A comma-separated list of zero or more + device names associated with this interrupt. + +What: /sys/kernel/irq//chip_name +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: Human-readable chip name supplied by the associated device + driver. + +What: /sys/kernel/irq//hwirq +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: When interrupt translation domains are used, this file contains + the underlying hardware IRQ number used for this Linux IRQ. + +What: /sys/kernel/irq//name +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: Human-readable flow handler name as defined by the irq chip + driver. + +What: /sys/kernel/irq//per_cpu_count +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: The number of times the interrupt has fired since boot. This + is a comma-separated list of counters; one per CPU in CPU id + order. NOTE: This file consistently shows counters for all + CPU ids. This differs from the behavior of /proc/interrupts + which only shows counters for online CPUs. + +What: /sys/kernel/irq//type +Date: September 2016 +KernelVersion: 4.9 +Contact: Craig Gallek +Description: The type of the interrupt. Either the string 'level' or 'edge'. + +What: /sys/kernel/irq//wakeup +Date: March 2018 +KernelVersion: 4.17 +Contact: Andy Shevchenko +Description: The wakeup state of the interrupt. Either the string + 'enabled' or 'disabled'. -- cgit v1.3.1