linux-kernel - Re: [PATCH] PM: runtime: Use pmruntime sync variant to put suppliers

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

Date:   Thu, 8 Oct 2020 04:08:06 +0300
From:   Stanimir Varbanov <stanimir.varbanov@...aro.org>
To:     "Rafael J. Wysocki" <rafael@...nel.org>
Cc:     Linux PM <linux-pm@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        "Rafael J . Wysocki" <rjw@...ysocki.net>
Subject: Re: [PATCH] PM: runtime: Use pmruntime sync variant to put suppliers

Hi Rafael,

On 10/7/20 5:37 PM, Rafael J. Wysocki wrote:
> On Wed, Oct 7, 2020 at 2:20 AM Stanimir Varbanov
> <stanimir.varbanov@...aro.org> wrote:
>>
>> Calling pm_runtime_put_sync over a device with suppliers with device
>> link flags PM_RUNTIME | RPM_ACTIVE it is observed that the supplier
>> is not put (turned off) at the end, but instead put asynchronously.
> 
> Yes, that's by design.
> 
>> In some case This could lead to issues for the callers which expects
>> that the pmruntime sync variants should also put the suppliers
>> synchronously.
> 
> Why would anyone expect that to happen?

It is logical to me that when I call pm_runtime_put_sync the device and
its suppliers are put synchronously. If I want to put device and its
suppliers asynchronously I'd use pm_runtime_put. Is that wrong assumption?

> 
>> Also the opposite rpm_get_suppliers is already using pmruntime _sync
>> variant of the API.
> 
> Yes, it does, because that is necessary.
> 
>> Correct this by changing pmruntime_put to pmruntime_put_sync in
>> rpm_put_suppliers.
> 
> It is not a correction, but a change in behavior without good enough
> rationale, as it stands.

In my driver case I want to deal with a recovery of a crash in remote
processor (the remote processor is used to control and program hardware
blocks and also to communicate with host processor through shared
memory). To restart the remote processor I have to disable clocks and
turn off few power domains (one of the power domains is made a supplier
of my main device) in order to complete the cold-boot.

The problem I'm facing with this design is that when I call
runtime_put_sync (to disable device's clocks and turn off power domain)
the clocks are disabled (part of pmruntime_suspend callback) but the
pmdomain (the device supplier) is not turned synchronously. I workaround
this by checking the supplier device via pm_runtime_active() until it
becomes inactive and the continue with rest of the steps.

>From my point of view this check for supplier activity should be part of
pmruntime API.

> 
> Thanks!
> 

-- 
regards,
Stan