lkml.org 
[lkml]   [2022]   [Mar]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/4] hwmon: (asus-ec-sensors) introduce ec_board_info struct for board data
On 3/29/22 12:22, Eugene Shalygin wrote:
> On Tue, 29 Mar 2022 at 15:44, Guenter Roeck <linux@roeck-us.net> wrote:
>>>
>>> struct ec_sensors_data {
>>> - unsigned long board_sensors;
>>> + struct ec_board_info board_info;
>>
>> Please explain why this needs to be the entire structure and not
>> just a pointer to it.
>
> I marked the board_info array as __initconst assuming that this large
> array will be unloaded from memory after the init phase, while we keep
> only a single element. Is that assumption incorrect?
>

What happens if you build the driver into the kernel and then instantiate
and de-instantiate it multiple times ?

>>> +static int sensor_count(const struct ec_board_info *board)
>>> +{
>>> + return hweight_long(board->sensors);
>>> +}
>>
>> This function is called several times. Does it really make sense, or is it
>> necessary, to re-calculate the number of sensors over and over again
>> instead of keeping it in ec->nr_sensors as before ? What are the benefits ?
>> Unless there is a good explanation I see that as unrelated and unnecessary
>> change.
>
> This had something to do with data deduplication. However, I need the
> count value only for looping over the sensor array, thus I can as well
> add an invalid element to the end of the array. I rushed to submit
> this driver to replace the wmi one, and it still has an artifact for
> the WMI code I'd like to get rid of eventually, which is the read
> buffer and the registers array. This will remove all the nr_ variables
> and two dynamically allocated arrays. I will understand, of course, if
> you ask to submit that refactoring separately.
>

The rule of "one logical change per patch" still applies. If you start
intermixing parts of future clean-up efforts into current patches, you'll
see a very unhappy maintainer - especially since this change makes up
a significant part of this patch, complicates review significantly,
and makes me wonder if other unrelated changes are included that I don't
see right now due to all the noise.

Besides, at least in this patch, I don't buy the "deduplication" argument.
Keeping a single additional variable in a data structure is much simpler
and straightforward than calling hweight_long() several times. I'd call
that "complification".

Guenter

\
 
 \ /
  Last update: 2022-03-29 22:27    [W:0.129 / U:0.124 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site