linux-kernel - Re: [PATCH] PM / OPP: Initialize regulator pointer to an error value

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]

Message-ID: <20160216131108.GK18327@sirena.org.uk>
Date:	Tue, 16 Feb 2016 13:11:08 +0000
From:	Mark Brown <broonie@...nel.org>
To:	Arnd Bergmann <arnd@...db.de>
Cc:	linaro-kernel@...ts.linaro.org,
	Viresh Kumar <viresh.kumar@...aro.org>,
	Nishanth Menon <nm@...com>, linux-pm@...r.kernel.org,
	Viresh Kumar <vireshk@...nel.org>,
	Rafael Wysocki <rjw@...ysocki.net>,
	linux-kernel@...r.kernel.org, Jon Hunter <jonathanh@...dia.com>,
	Stephen Boyd <sboyd@...eaurora.org>
Subject: Re: [PATCH] PM / OPP: Initialize regulator pointer to an error value

On Tue, Feb 16, 2016 at 10:10:44AM +0100, Arnd Bergmann wrote:
> On Tuesday 16 February 2016 01:56:16 Mark Brown wrote:

> > No, NULL is explicitly not something you can substitute in,
> > essentially all the users are just not bothering to implement error
> > checking and we don't want to encourage that.  The set of use cases
> > where we legitimately have optional supplies is very small, much smaller
> > than clocks, because it makes the electrical engineering a lot harder.

> I must have misinterpreted the idea behind that API as well then.

> From this function definition:

> static inline struct regulator *__must_check regulator_get(struct device *dev,
>         const char *id)
> {       
>         /* Nothing except the stubbed out regulator API should be
>          * looking at the value except to check if it is an error
>          * value. Drivers are free to handle NULL specifically by
>          * skipping all regulator API calls, but they don't have to.
>          * Drivers which don't, should make sure they properly handle
>          * corner cases of the API, such as regulator_get_voltage()
>          * returning 0.
>          */             
>         return NULL;
> }

This is the stubbed regulator API which is only ever used with the stub
regulator API, it uses NULL to give a non-error pointer it can return to
well written callers so they don't know they are running with the stubs.
We are explicitly using NULL because callers should treat it as a valid
regulator.

> my reading was that the expected behavior in any driver was:

> * call regulator_get()
> * if IS_ERR(), fail device probe function, never use invalid
>   pointer other than PTR_ERR()
> * if NULL, and regulator is required, fail probe so we never
>   use the regulator

No, drivers should never look at the value of the pointer other than to
check it for error.  If there is a problem of any kind an error will be
returned.

> * if NULL, and regulators are optional, continue with the NULL
>   value.

No, we always return an error pointer if we fail to get a regulator.
The difference with optional regulators is in how we handle the
situation where we have full constraints and a regulator is not mapped
in, normally we assume there must be one with no software control but we
need to work around buggy bindings as the device would be non-functional
without power.

> * drivers never look into the regulator pointer, and only
>   pass it into regulator APIs which can cope with the NULL
>   value when CONFIG_REGULATOR is disabled.

> That would be similar to what we have for clocks. Which part of
> my interpretation is wrong?

See above.

Download attachment "signature.asc" of type "application/pgp-signature" (474 bytes)