linux-kernel - Re: [PATCH 1/7] platform/x86: fujitsu-laptop: Define constants for FUNC operations

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:   Tue, 6 Mar 2018 21:59:20 +0100
From:   Michał Kępień <kernel@...pniu.pl>
To:     Andy Shevchenko <andy.shevchenko@...il.com>
Cc:     Darren Hart <dvhart@...radead.org>,
        Jonathan Woithe <jwoithe@...t42.net>,
        Andy Shevchenko <andy@...radead.org>,
        Platform Driver <platform-driver-x86@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 1/7] platform/x86: fujitsu-laptop: Define constants for
 FUNC operations

Andy,

> What I'm trying to tell is about consistency of style.

I completely agree with all you wrote, those are all good suggestions.
But you started your reasoning with:

> So, imagine if we have two bitfields in some register, one with one
> bit and the other with two.

We are not looking at a hardware register here.  Rather, I am trying to
bring at least _some_ order to an arbitrary set of values that the
vendor assumed would be fun to scatter around a dozen of firmware
functions.  Some hardware features are controlled by setting a specific
bit in the value being passed to a function; in other cases entire
integers are taken into account; in yet another case *two* bits in a
value control state.  There is no reason or order to be found here.

In the case of OP_* constants, perhaps the simplest thing to do would be
to define them as integers, not bitfields.  But that still results in a
mess:

#define OP_GET		0x2
#define OP_GET_CAPS	0x0
#define OP_GET_EVENTS	0x1
#define OP_GET_EXT	0x4
#define OP_SET		0x1
#define OP_SET_EXT	0x5

or:

#define OP_GET_CAPS	0x0
#define OP_GET_EVENTS	0x1
#define OP_SET		0x1
#define OP_GET		0x2
#define OP_GET_EXT	0x4
#define OP_SET_EXT	0x5

Even worse, what am I supposed to do with crap like radio LED control,
where 0x20 (bit 5) is passed in one argument to select the desired
feature and 0x20 or 0 is passed as another argument to select the
desired state of that feature?  How do I name and define constants that
I can subsequently use in call_fext_func() invocations to spare the
reader the need to learn the hard way that this:

    return call_fext_func(device, FUNC_FLAGS, 0x5, RADIO_LED_ON, 0x0);

is actually supposed to turn the radio LED *off*?

This is the best I could come up with:

#define FEAT_RADIO_LED		BIT(5)
#define STATE_RADIO_LED_OFF	(0 << 0)
#define STATE_RADIO_LED_ON	BIT(5)

Yes, it is ugly.  But the resulting call is (IMHO) a lot more clear than
the original one:

    return fext_flags(device, OP_SET_EXT, FEAT_RADIO_LED, STATE_RADIO_LED_OFF);

All of the above is why I was inclined to just use alphabetic ordering
for all constants defined in fujitsu-laptop.  This approach brings at
least _some_ consistency to this mess (which only the vendor is to blame
for, of course).  If you would rather have me take a different approach,
I am all ears.

-- 
Best regards,
Michał Kępień