linux-kernel - Re: [Regression] cached max_state breaks ACPI processor cooling device

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Thu, 2 Mar 2023 20:29:08 +0100
From:   "Rafael J. Wysocki" <rafael@...nel.org>
To:     "Zhang, Rui" <rui.zhang@...el.com>
Cc:     "viresh.kumar@...aro.org" <viresh.kumar@...aro.org>,
        "Wysocki, Rafael J" <rafael.j.wysocki@...el.com>,
        "daniel.lezcano@...aro.org" <daniel.lezcano@...aro.org>,
        "srinivas.pandruvada@...ux.intel.com" 
        <srinivas.pandruvada@...ux.intel.com>,
        "linux-pm@...r.kernel.org" <linux-pm@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [Regression] cached max_state breaks ACPI processor cooling device

On Sat, Feb 25, 2023 at 6:30 AM Zhang, Rui <rui.zhang@...el.com> wrote:
>
> Hi, All,
>
> Starting from commit c408b3d1d9bb("thermal: Validate new state in
> cur_state_store()") and commit a365105c685c("thermal: sysfs: Reuse
> cdev->max_state"), the cdev->get_max_state() is evaluated only once
> during cooling device registration.
>
> This is done to fix the below Smatch static checker warning:
>         drivers/thermal/thermal_sysfs.c:656 thermal_cooling_device_stats_update()
>         warn: potential integer overflow from user 'stats->state *
> stats->max_states + new_state'
> reported here https://lore.kernel.org/all/Y0ltRJRjO7AkawvE@kili/.
>
> But this actually breaks cooling devices which could have dynamic max
> cooling state, like ACPI processor cooling device.
>
> acpi_processor_driver_init
>         driver_register(&acpi_processor_driver);
>                 acpi_processor_start
>                         acpi_processor_thermal_init
>                                 thermal_cooling_device_register
>                                         processor_get_max_state
>         acpi_processor_cpufreq_init = true
> The driver doesn't count cpufreq as cooling state until
> acpi_processor_cpufreq_init is set later.
>
> As a result, without the commits above,
> /sys/class/thermal/cooling_device10/cur_state:0
> /sys/class/thermal/cooling_device10/max_state:3
> /sys/class/thermal/cooling_device10/type:Processor
> after the commits above,
> /sys/class/thermal/cooling_device10/cur_state:0
> /sys/class/thermal/cooling_device10/max_state:0
> /sys/class/thermal/cooling_device10/type:Processor
>
> In order to fix this, there are something worth clarification IMO.
> 1. should we support dynamic max_state or not?
>    ACPI processor cooling state is a combination of processor p-states
>    and t-states.
>    t-states are static, but p-states can vary depends on processor
>    frequency driver loaded or not.
>    I'm not sure if there is any other user like this, but still this
>    is a valid use case of dynamic max_state.
> 2. how to handle dynamic max_state for cooling device statistics in
>    sysfs?
>    IMO, when max_state changes, the definition of previous cooling
>    device is changed as well, thus we should abandon the previous
>    statistics and restart counting.
> 3. anything else needs to be handled when max_state changes?
>    Say, as the definition of each cooling state is changed when
>    max_state changes, should we invalidate and re-update the
>    thermal instances of this cdev in each thermal zone device?
> 4. how to detect/handle max cooling states changes?
>    Should we do this as before, which invokes .get_max_state()
>    everywhere and do updates when necessary ,

No.

>  or
>    a. cache max_state like we do today
>    b. introduce a new API for max_state change
>    c. invoke the new API in the cooling device driver explicitly when
>       max_state changes
>    d. update cached max_state value, statistics sysfs and
>       thermal_instances in the API
>    e. remove .get_max_state() callback as we register and update the
>       max_state with a fixed value each time.

Yes, IMO.