** Tags added: kernel-daily-bug -- You received this bug notification because you are subscribed to linux in Ubuntu. Matching subscriptions: Bgg, Bmail, Nb https://bugs.launchpad.net/bugs/2171947 Title: 6.8.0-147 regression: massively increased frequency of ACPI [ECMX] EC- mutex deadlock on HP EliteBook 860 G10 (AC._PSR), -146 unaffected Status in linux package in Ubuntu: New Bug description: ## Summary Kernel 6.8.0-147 (noble, installed 2026-10-13) introduced a severe regression in how often a latent HP firmware ACPI Embedded-Controller mutex bug is triggered on an HP EliteBook 860 G10. On 6.8.0-146 the deadlock occurs roughly once every few days; on 6.8.0-147 it occurs multiple times per day, sometimes within ~10 minutes of boot, hard-locking systemd (PID 1) and requiring SysRq REISUB. Reverting to 6.8.0-146 restores normal (rare deadlock) behaviour. ## Hardware - HP EliteBook 860 16 inch G10 Notebook PC - BIOS: HP V70 Ver. 01.12.00, 2026-02-04 (latest available) - CPU: Intel (Raptor Lake), i915 integrated graphics ## Root cause vs regression - Root cause is an HP firmware/AML bug: a thread acquires the ACPI `[ECMX]` Embedded-Controller mutex via `\_SB.AC._PSR` (AC-adapter power-source read) and never releases it. This is HP's to fix and is not kernel-introduced. - The REGRESSION is that 6.8.0-147 triggers this firmware race far more often than 6.8.0-146, turning a rare annoyance into an unusable machine. ## Log signature (at moment of each deadlock) kernel: ACPI Error: Thread <A> cannot release Mutex [ECMX] acquired by thread <B> (20230628/exmutex-378) kernel: ACPI: \_SB_.AC__: Error reading AC Adapter state: AE_AML_NOT_OWNER Followed by hung EC event processing, then thermald/upowerd/kworker hangs, then systemd PID 1 hang and total D-Bus service-activation failure. ## Regression range - Good: 6.8.0-146 (and earlier: -139, -138, -137). Rare deadlocks (days apart). - Bad: 6.8.0-147. Frequent deadlocks (multiple per day). - -147 is stable patchset LP: #2168026. ## Suspected commits (not bisected) 1. PRIME: "ACPI: AC: Properly notify power management core about changes" — hits the exact AC._PSR path that grabs [ECMX]; plausibly increases _PSR evaluation frequency. 2. cpuidle menu governor cluster ("Remove iowait influence", "Select polling state in some more cases", plus a mid-series revert) — alters CPU idle/wake timing and hence EC poll frequency. 3. ACPICA lock/opregion changes (evxfregn ContextMutex, "Allow to skip Global Lock initialization", acpi_ev_address_space_dispatch NULL-deref, acpi_execute_reg_methods depth arg) — touch EC opregion/mutex machinery. ## Requested outcome Given the root cause is HP firmware, a DMI-based quirk for this model (EliteBook 860 G10) that restores -146 AC-notify / idle behaviour, or a revert of the offending change on affected hardware, would fix this for all affected users. ## Reproduction Boot 6.8.0-147 on an affected machine and use normally; deadlock typically occurs within minutes to a couple of hours. Does not occur (or very rarely) on 6.8.0-146. ## Workaround in place Pinned GRUB default to 6.8.0-146. ## Attachments (single archive: ecmx-bug-report.tar.gz) Required by the kernel bug template: - version.log (/proc/version_signature — note: shows -146, the GOOD kernel, as that is what is currently booted; -147 is the bad one, see "Regression range" above) - lspci-vnvn.log (sudo lspci -vnvn, full) Supporting evidence: - system-info.txt (model, BIOS, installed kernels, running kernel) - cmdline.txt (/proc/cmdline) - cpuinfo.txt (/proc/cpuinfo) - dmesg-boot-minus5-full.txt (full kernel log from a deadlock boot) - ecmx-logs-boot-minus5.txt (ECMX + AC._PSR + AE_AML_NOT_OWNER signatures) - ecmx-logs-boot-minus8.txt (same, a second deadlock boot) - acpi-tables/ (DSDT, SSDT1-29, FACP — raw ACPI tables containing the _PSR/[ECMX] AML; verified free of serial number / UUID) To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2171947/+subscriptions
Комментариев нет:
Отправить комментарий