Received-SPF: pass (google.com: domain of linux-kernel+bounces-30990-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Message-ID: <98b87792-4919-4152-8ccc-b8a731cdfd55@linaro.org>
Date: Fri, 19 Jan 2024 10:02:25 +0000
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH 17/18] tty: serial: samsung: shrink port feature flags to
 u8
Content-Language: en-US
To: Jiri Slaby <jirislaby@kernel.org>,
 Sam Protsenko <semen.protsenko@linaro.org>
Cc: krzysztof.kozlowski@linaro.org, alim.akhtar@samsung.com,
 gregkh@linuxfoundation.org, linux-arm-kernel@lists.infradead.org,
 linux-samsung-soc@vger.kernel.org, linux-kernel@vger.kernel.org,
 linux-serial@vger.kernel.org, andre.draszik@linaro.org,
 peter.griffin@linaro.org, kernel-team@android.com, willmcvicker@google.com
References: <20240110102102.61587-1-tudor.ambarus@linaro.org>
 <20240110102102.61587-18-tudor.ambarus@linaro.org>
 <CAPLW+4k091328krLB_KdHyobG-pR--Rt5WaN6c1ccpgdV8ry7Q@mail.gmail.com>
 <76e1dc42-cabe-4925-8aa1-c8f733fb36a2@linaro.org>
 <8f3f85d0-866e-4e5a-8177-05c26c08b278@kernel.org>
 <842d36c7-9452-431f-95c4-ff114484d201@linaro.org>
 <96e3d7e9-737b-484e-bc94-e95533f06ca7@kernel.org>
From: Tudor Ambarus <tudor.ambarus@linaro.org>
In-Reply-To: <96e3d7e9-737b-484e-bc94-e95533f06ca7@kernel.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit


On 1/19/24 09:54, Jiri Slaby wrote:
> Hi,
> 
> On 19. 01. 24, 10:43, Tudor Ambarus wrote:
>>>> If using unsigned int the bitfied is combined with the previous u8
>>>> fields, whereas if using u8 the bitfield will be independently defined.
>>>> So no benefit in terms of memory footprint, it's just a cosmetic change
>>>> to align the bitfield with the previous u8 fields. Allowing u32 for
>>>> just
>>>> a bit can be misleading as one would ask itself where are the other
>>>> bits. Between a u32 bitfield and a bool a u8 bitfield seems like a good
>>>> compromise.
>>>
>>> Why? What's wrong with bool? bitfields have terrible semantics wrt
>>> atomic writes for example.
>>>
>>
>> Bool occupies a byte and if more port features will ever be added we'll
>> occupy more bytes. Here's how the structure will look like with a bool:
>>
>> struct s3c24xx_uart_info {
>>     const char  *              name;                 /*     0     8 */
>>     enum s3c24xx_port_type     type;                 /*     8     4 */
>>     unsigned int               port_type;            /*    12     4 */
>>     unsigned int               fifosize;             /*    16     4 */
>>     u32                        rx_fifomask;          /*    20     4 */
>>     u32                        rx_fifoshift;         /*    24     4 */
>>     u32                        rx_fifofull;          /*    28     4 */
>>     u32                        tx_fifomask;          /*    32     4 */
>>     u32                        tx_fifoshift;         /*    36     4 */
>>     u32                        tx_fifofull;          /*    40     4 */
>>     u32                        clksel_mask;          /*    44     4 */
>>     u32                        clksel_shift;         /*    48     4 */
>>     u32                        ucon_mask;            /*    52     4 */
>>     u8                         def_clk_sel;          /*    56     1 */
>>     u8                         num_clks;             /*    57     1 */
>>     u8                         iotype;               /*    58     1 */
>>     bool                       has_divslot;          /*    59     1 */
>>
>>     /* size: 64, cachelines: 1, members: 17 */
>>     /* padding: 4 */
>> };
>>
>> What's your preference?
> 
> bool :).
> 
I'm fine with a bool too as since the introduction of this driver we
have just this flag, it's unlikey to have 4 more soon to bypass the
first cacheline. Will change to bool.

Cheers,
ta