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-devfreq | 143 ++++++++++++++++++++++++++ 1 file changed, 143 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-class-devfreq (limited to 'Documentation/ABI/testing/sysfs-class-devfreq') diff --git a/Documentation/ABI/testing/sysfs-class-devfreq b/Documentation/ABI/testing/sysfs-class-devfreq new file mode 100644 index 000000000..df8ba88b9 --- /dev/null +++ b/Documentation/ABI/testing/sysfs-class-devfreq @@ -0,0 +1,143 @@ +What: /sys/class/devfreq/.../ +Date: September 2011 +Contact: MyungJoo Ham +Description: + Provide a place in sysfs for the devfreq objects. + This allows accessing various devfreq specific variables. + The name of devfreq object denoted as ... is same as the + name of device using devfreq. + +What: /sys/class/devfreq/.../name +Date: November 2019 +Contact: Chanwoo Choi +Description: + The /sys/class/devfreq/.../name shows the name of device + of the corresponding devfreq object. + +What: /sys/class/devfreq/.../governor +Date: September 2011 +Contact: MyungJoo Ham +Description: + The /sys/class/devfreq/.../governor show or set the name of the + governor used by the corresponding devfreq object. + +What: /sys/class/devfreq/.../cur_freq +Date: September 2011 +Contact: MyungJoo Ham +Description: + The /sys/class/devfreq/.../cur_freq shows the current + frequency of the corresponding devfreq object. Same as + target_freq when get_cur_freq() is not implemented by + devfreq driver. + +What: /sys/class/devfreq/.../target_freq +Date: September 2012 +Contact: Rajagopal Venkat +Description: + The /sys/class/devfreq/.../target_freq shows the next governor + predicted target frequency of the corresponding devfreq object. + +What: /sys/class/devfreq/.../trans_stat +Date: October 2012 +Contact: MyungJoo Ham +Description: + This ABI shows or clears the statistics of devfreq behavior + on a specific device. It shows the time spent in each state + and the number of transitions between states. + In order to activate this ABI, the devfreq target device + driver should provide the list of available frequencies + with its profile. If need to reset the statistics of devfreq + behavior on a specific device, enter 0(zero) to 'trans_stat' + as following:: + + echo 0 > /sys/class/devfreq/.../trans_stat + + If the transition table is bigger than PAGE_SIZE, reading + this will return an -EFBIG error. + +What: /sys/class/devfreq/.../available_frequencies +Date: October 2012 +Contact: Nishanth Menon +Description: + The /sys/class/devfreq/.../available_frequencies shows + the available frequencies of the corresponding devfreq object. + This is a snapshot of available frequencies and not limited + by the min/max frequency restrictions. + +What: /sys/class/devfreq/.../available_governors +Date: October 2012 +Contact: Nishanth Menon +Description: + The /sys/class/devfreq/.../available_governors shows + currently available governors in the system. + +What: /sys/class/devfreq/.../min_freq +Date: January 2013 +Contact: MyungJoo Ham +Description: + The /sys/class/devfreq/.../min_freq shows and stores + the minimum frequency requested by users. It is 0 if + the user does not care. min_freq overrides the + frequency requested by governors. + +What: /sys/class/devfreq/.../max_freq +Date: January 2013 +Contact: MyungJoo Ham +Description: + The /sys/class/devfreq/.../max_freq shows and stores + the maximum frequency requested by users. It is 0 if + the user does not care. max_freq overrides the + frequency requested by governors and min_freq. + The max_freq overrides min_freq because max_freq may be + used to throttle devices to avoid overheating. + +What: /sys/class/devfreq/.../polling_interval +Date: September 2011 +Contact: MyungJoo Ham +Description: + The /sys/class/devfreq/.../polling_interval shows and sets + the requested polling interval of the corresponding devfreq + object. The values are represented in ms. If the value is + less than 1 jiffy, it is considered to be 0, which means + no polling. This value is meaningless if the governor is + not polling. + + A list of governors that support the node: + - simple_ondmenad + - tegra_actmon + +What: /sys/class/devfreq/.../userspace/set_freq +Date: September 2011 +Contact: MyungJoo Ham +Description: + The /sys/class/devfreq/.../userspace/set_freq shows and + sets the requested frequency for the devfreq object if + userspace governor is in effect. + + A list of governors that support the node: + - userspace + +What: /sys/class/devfreq/.../timer +Date: July 2020 +Contact: Chanwoo Choi +Description: + This ABI shows and stores the kind of work timer by users. + This work timer is used by devfreq workqueue in order to + monitor the device status such as utilization. The user + can change the work timer on runtime according to their demand + as following:: + + echo deferrable > /sys/class/devfreq/.../timer + echo delayed > /sys/class/devfreq/.../timer + + A list of governors that support the node: + - simple_ondemand + +What: /sys/class/devfreq/.../related_cpus +Date: June 2025 +Contact: Linux power management list +Description: The list of CPUs whose performance is closely related to the + frequency of this devfreq domain. + + This file is only present if a specific devfreq device is + closely associated with a subset of CPUs. -- cgit v1.3.1