Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Sender: Guenter Roeck <groeck7@gmail.com>
Subject: Re: [PATCH v3 3/3] hwmon: Driver for Texas Instruments INA238
To:     Nathan Rossi <nathan@nathanrossi.com>
Cc:     linux-hwmon@vger.kernel.org, linux-doc@vger.kernel.org,
        linux-kernel@vger.kernel.org, Nathan Rossi <nathan.rossi@digi.com>,
        Jean Delvare <jdelvare@suse.com>,
        Jonathan Corbet <corbet@lwn.net>
References: <20211028081030.719265-0-nathan@nathanrossi.com>
 <20211028081030.719265-3-nathan@nathanrossi.com>
 <7b6764bf-4978-60ec-b1e6-8d59077c3023@roeck-us.net>
 <CA+aJhH1aGJXwYSCU8RC275G5=qGLyZRK94g9ic24wxKuRGCwEA@mail.gmail.com>
 <43c17bba-d4bd-1f9d-5034-1f5a9279d751@roeck-us.net>
 <CA+aJhH0EBUjQhjQhxj-kfJzmpqGN6ZMNn_M-pAae_V9yPQhs3A@mail.gmail.com>
From:   Guenter Roeck <linux@roeck-us.net>
Message-ID: <5569b2b8-ac67-757b-f7cb-302c9f663e80@roeck-us.net>
Date:   Mon, 1 Nov 2021 07:57:31 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.13.0
MIME-Version: 1.0
In-Reply-To: <CA+aJhH0EBUjQhjQhxj-kfJzmpqGN6ZMNn_M-pAae_V9yPQhs3A@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 10/31/21 10:55 PM, Nathan Rossi wrote:
> On Mon, 1 Nov 2021 at 13:48, Guenter Roeck <linux@roeck-us.net> wrote:
>>
>> On 10/31/21 7:20 PM, Nathan Rossi wrote:
>> [ ... ]
>>>>> +
>>>>> +     if (attr != hwmon_in_max && attr != hwmon_in_min)
>>>>> +             return -EOPNOTSUPP;
>>>>> +
>>>>> +     /* convert decimal to register value */
>>>>> +     switch (channel) {
>>>>> +     case 0:
>>>>> +             /* signed value, clamp to max range +/-163 mV */
>>>>> +             regval = clamp_val(val, -163, 163);
>>>>> +             regval = (regval * 1000L * (4 - (int)data->gain + 1)) /
>>>>
>>>> nit: The typecast "(int)" is not needed here.
>>>
>>> Due to the unsigned type of gain, it causes promotion of regval (and
>>> the rest of the numerator) to unsigned long which causes issues with
>>> negative numbers on the divide. It makes more sense for gain to be an
>>> int to begin with, I will change it to int to avoid the need for type
>>> casting.
>>>
>>
>> Are you sure ? I initially thought that as well and wrote a little test
>> program with that expression, but it didn't do the promotion to unsigned.
>>
> 
> It definitely calculates incorrectly at run time (on an arm 32-bit
> platform), looking at the gcc output from -fdump-tree-original reveals
> some more insight. Which is that the promotion to long overrides the
> unsigned (from the 1000L) on long=64 but not on long=32.
> 
> Where regval is int, and gain is unsigned int (u32).
> 
> regval = (regval * 1000L * (4 - gain + 1)) / 5;
> -> armv7-a (invalid)
> regval = (int) ((((long unsigned int) regval * (long unsigned int) (5
> - gain)) * 1000) / 5);
> -> x86-64 (valid result)
> regval = (int) ((unsigned int) (gain * 4294967096 + 1000) * (unsigned
> int) regval);
> 
> note: 4294967096 is -800, 1000 * (4 - gain + 1) => (-800 * gain) + 1000
> 
> Slight variation without the 1000 being long.
> 
> regval = (regval * 1000 * (4 - gain + 1)) / 5;
> -> armv7-a (invalid)
> regval = (int) ((((unsigned int) regval * (5 - gain)) * 1000) / 5);
> -> x86-64 (invalid)
> regval = (int) ((((unsigned int) regval * (5 - gain)) * 1000) / 5);
> 
> regval = (regval * 1000LL * (4 - gain + 1)) / 5;
> -> armv7-a (valid)
> regval = (int) ((unsigned int) (gain * 4294967096 + 1000) * (unsigned
> int) regval);
> -> x86-64 (valid)
> regval = (int) ((unsigned int) (gain * 4294967096 + 1000) * (unsigned
> int) regval);
> 
> I think it still makes sense to change gain to be int, and avoid the
> unsigned type issues.
> 

Thanks for the details. I agree, changing gain to int makes sense.

Thanks,
Guenter