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-class-led-flash | 89 +++++++++++++++++++++++++ 1 file changed, 89 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-class-led-flash (limited to 'Documentation/ABI/testing/sysfs-class-led-flash') diff --git a/Documentation/ABI/testing/sysfs-class-led-flash b/Documentation/ABI/testing/sysfs-class-led-flash new file mode 100644 index 000000000..11e5677c3 --- /dev/null +++ b/Documentation/ABI/testing/sysfs-class-led-flash @@ -0,0 +1,89 @@ +What: /sys/class/leds//flash_brightness +Date: March 2015 +KernelVersion: 4.0 +Contact: Jacek Anaszewski +Description: read/write + Set the brightness of this LED in the flash strobe mode, in + microamperes. The file is created only for the flash LED devices + that support setting flash brightness. + + The value is between 0 and + /sys/class/leds//max_flash_brightness. + +What: /sys/class/leds//max_flash_brightness +Date: March 2015 +KernelVersion: 4.0 +Contact: Jacek Anaszewski +Description: read only + Maximum brightness level for this LED in the flash strobe mode, + in microamperes. + +What: /sys/class/leds//flash_timeout +Date: March 2015 +KernelVersion: 4.0 +Contact: Jacek Anaszewski +Description: read/write + Hardware timeout for flash, in microseconds. The flash strobe + is stopped after this period of time has passed from the start + of the strobe. The file is created only for the flash LED + devices that support setting flash timeout. + +What: /sys/class/leds//max_flash_timeout +Date: March 2015 +KernelVersion: 4.0 +Contact: Jacek Anaszewski +Description: read only + Maximum flash timeout for this LED, in microseconds. + +What: /sys/class/leds//flash_strobe +Date: March 2015 +KernelVersion: 4.0 +Contact: Jacek Anaszewski +Description: read/write + Flash strobe state. When written with 1 it triggers flash strobe + and when written with 0 it turns the flash off. + + On read 1 means that flash is currently strobing and 0 means + that flash is off. + +What: /sys/class/leds//flash_fault +Date: March 2015 +KernelVersion: 4.0 +Contact: Jacek Anaszewski +Description: read only + Space separated list of flash faults that may have occurred. + Flash faults are re-read after strobing the flash. Possible + flash faults: + + * led-over-voltage + flash controller voltage to the flash LED + has exceeded the limit specific to the flash controller + * flash-timeout-exceeded + the flash strobe was still on when + the timeout set by the user has expired; not all flash + controllers may set this in all such conditions + * controller-over-temperature + the flash controller has + overheated + * controller-short-circuit + the short circuit protection + of the flash controller has been triggered + * led-power-supply-over-current + current in the LED power + supply has exceeded the limit specific to the flash + controller + * indicator-led-fault + the flash controller has detected + a short or open circuit condition on the indicator LED + * led-under-voltage + flash controller voltage to the flash + LED has been below the minimum limit specific to + the flash + * controller-under-voltage + the input voltage of the flash + controller is below the limit under which strobing the + flash at full current will not be possible; + the condition persists until this flag is no longer set + * led-over-temperature + the temperature of the LED has exceeded + its allowed upper limit -- cgit v1.3.1