Re: [PATCH 2/7] irqchip/irq-qcom-mpm: Register MPM under CPU cluster power domain

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] The `cpus_in_pm` counter is permanently leaked if `handle_rpm_notification()` fails during CPU PM entry.
- [High] The return value of `devm_pm_runtime_enable(dev)` is ignored, potentially breaking cluster power down.
--

--- Patch [2]: [PATCH 2/7] irqchip/irq-qcom-mpm: Register MPM under CPU cluster power domain ---
commit 1e6fda1a51a8d1d0370b919f0b149226ff3d825a
Author: Sneh Mankad <sneh.mankad@oss.qualcomm.com>
Subject: irqchip/irq-qcom-mpm: Register MPM under CPU cluster power domain

This commit resolves a hard probe dependency by moving RPM notification
handling to the GENPD_NOTIFY_PRE_OFF callback and registering MPM under the
CPU cluster power domain. It also provides a CPU PM notification fallback if
the MPM hasn't registered with the CPU cluster power domain.

Fixes: a6199bb514d8 ("irqchip: Add Qualcomm MPM controller driver")

(Note: The standard inline-quoted diff format has been omitted due to automated
recitation restrictions. The regressions found are described using free-form
text below.)

[Severity: High]
In drivers/irqchip/irq-qcom-mpm.c:mpm_cpu_pm_callback(), during the
CPU_PM_ENTER case, atomic_inc_return(&priv->cpus_in_pm) is executed at the
beginning of the transition.

If handle_rpm_notification(priv) subsequently fails, the function returns
NOTIFY_BAD. Because notifier_call_chain_robust() rolls back successfully
executed notifiers but does not send CPU_PM_ENTER_FAILED to the notifier
that failed, the atomic increment is never reverted.

Does this leave cpus_in_pm permanently incremented, causing all subsequent
CPU PM transitions to incorrectly evaluate cpus_in_pm < num_online_cpus()
and permanently preventing the cluster from entering low power mode?

[Severity: High]
In drivers/irqchip/irq-qcom-mpm.c:qcom_mpm_probe(), the return value of
devm_pm_runtime_enable(dev) is not checked.

If devm_pm_runtime_enable() fails, for instance due to a memory allocation
failure in devm_add_action_or_reset(), runtime PM will not be enabled for
the device.

Can this unhandled failure prevent the genpd from managing the device's
power state properly, preventing the CPU cluster power domain from
entering low power mode? Should the return value be checked and handled?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260713-b4-shikra_lpm_addition-v1-0-3d858df2cbbf@oss.qualcomm.com?part=2