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 --- .../firmware/firmware-usage-guidelines.rst | 49 ++++++++++++++++++++++ 1 file changed, 49 insertions(+) create mode 100644 Documentation/driver-api/firmware/firmware-usage-guidelines.rst (limited to 'Documentation/driver-api/firmware/firmware-usage-guidelines.rst') diff --git a/Documentation/driver-api/firmware/firmware-usage-guidelines.rst b/Documentation/driver-api/firmware/firmware-usage-guidelines.rst new file mode 100644 index 000000000..336e91228 --- /dev/null +++ b/Documentation/driver-api/firmware/firmware-usage-guidelines.rst @@ -0,0 +1,49 @@ +=================== +Firmware Guidelines +=================== + +Users switching to a newer kernel should *not* have to install newer +firmware files to keep their hardware working. At the same time updated +firmware files must not cause any regressions for users of older kernel +releases. + +Drivers that use firmware from linux-firmware should follow the rules in +this guide. (Where there is limited control of the firmware, +i.e. company doesn't support Linux, firmwares sourced from misc places, +then of course these rules will not apply strictly.) + +* Firmware files shall be designed in a way that it allows checking for + firmware ABI version changes. It is recommended that firmware files be + versioned with at least a major/minor version. It is suggested that + the firmware files in linux-firmware be named with some device + specific name, and just the major version. The firmware version should + be stored in the firmware header, or as an exception, as part of the + firmware file name, in order to let the driver detact any non-ABI + fixes/changes. The firmware files in linux-firmware should be + overwritten with the newest compatible major version. Newer major + version firmware shall remain compatible with all kernels that load + that major number. + +* If the kernel support for the hardware is normally inactive, or the + hardware isn't available for public consumption, this can + be ignored, until the first kernel release that enables that hardware. + This means no major version bumps without the kernel retaining + backwards compatibility for the older major versions. Minor version + bumps should not introduce new features that newer kernels depend on + non-optionally. + +* If a security fix needs lockstep firmware and kernel fixes in order to + be successful, then all supported major versions in the linux-firmware + repo that are required by currently supported stable/LTS kernels, + should be updated with the security fix. The kernel patches should + detect if the firmware is new enough to declare if the security issue + is fixed. All communications around security fixes should point at + both the firmware and kernel fixes. If a security fix requires + deprecating old major versions, then this should only be done as a + last option, and be stated clearly in all communications. + +* Firmware files that affect the User API (UAPI) shall not introduce + changes that break existing userspace programs. Updates to such firmware + must ensure backward compatibility with existing userspace applications. + This includes maintaining consistent interfaces and behaviors that + userspace programs rely on. -- cgit v1.3.1