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/driver-api/hw-recoverable-errors.rst | 60 ++++++++++++++++++++++ 1 file changed, 60 insertions(+) create mode 100644 Documentation/driver-api/hw-recoverable-errors.rst (limited to 'Documentation/driver-api/hw-recoverable-errors.rst') diff --git a/Documentation/driver-api/hw-recoverable-errors.rst b/Documentation/driver-api/hw-recoverable-errors.rst new file mode 100644 index 000000000..fc526c345 --- /dev/null +++ b/Documentation/driver-api/hw-recoverable-errors.rst @@ -0,0 +1,60 @@ +.. SPDX-License-Identifier: GPL-2.0 + +================================================= +Recoverable Hardware Error Tracking in vmcoreinfo +================================================= + +Overview +-------- + +This feature provides a generic infrastructure within the Linux kernel to track +and log recoverable hardware errors. These are hardware recoverable errors +visible that might not cause immediate panics but may influence health, mainly +because new code path will be executed in the kernel. + +By recording counts and timestamps of recoverable errors into the vmcoreinfo +crash dump notes, this infrastructure aids post-mortem crash analysis tools in +correlating hardware events with kernel failures. This enables faster triage +and better understanding of root causes, especially in large-scale cloud +environments where hardware issues are common. + +Benefits +-------- + +- Facilitates correlation of hardware recoverable errors with kernel panics or + unusual code paths that lead to system crashes. +- Provides operators and cloud providers quick insights, improving reliability + and reducing troubleshooting time. +- Complements existing full hardware diagnostics without replacing them. + +Data Exposure and Consumption +----------------------------- + +- The tracked error data consists of per-error-type counts and timestamps of + last occurrence. +- This data is stored in the `hwerror_data` array, categorized by error source + types like CPU, memory, PCI, CXL, and others. +- It is exposed via vmcoreinfo crash dump notes and can be read using tools + like `crash`, `drgn`, or other kernel crash analysis utilities. +- There is no other way to read these data other than from crash dumps. +- These errors are divided by area, which includes CPU, Memory, PCI, CXL and + others. + +Typical usage example (in drgn REPL): + +.. code-block:: python + + >>> prog['hwerror_data'] + (struct hwerror_info[HWERR_RECOV_MAX]){ + { + .count = (int)844, + .timestamp = (time64_t)1752852018, + }, + ... + } + +Enabling +-------- + +- This feature is enabled when CONFIG_VMCORE_INFO is set. + -- cgit v1.3.1