2022-03-07 09:21:27

by Jiri Slaby

[permalink] [raw]
Subject: Re: [PATCH v2 3/7] tty: serial: samsung: constify s3c24xx_serial_drv_data

On 07. 03. 22, 9:09, Krzysztof Kozlowski wrote:
> The driver data (struct s3c24xx_serial_drv_data) is only used to
> initialize the driver properly and is not modified. Make it const.
...
> @@ -2755,9 +2755,9 @@ static struct s3c24xx_serial_drv_data s5pv210_serial_drv_data = {
> },
> .fifosize = { 256, 64, 16, 16 },
> };
> -#define S5PV210_SERIAL_DRV_DATA ((kernel_ulong_t)&s5pv210_serial_drv_data)
> +#define S5PV210_SERIAL_DRV_DATA (&s5pv210_serial_drv_data)
> #else
> -#define S5PV210_SERIAL_DRV_DATA (kernel_ulong_t)NULL
> +#define S5PV210_SERIAL_DRV_DATA NULL

Yet, I still don't see why the switch from ulong->ptr happens in this
"constify it" patch?

thanks,
--
--
js
suse labs


2022-03-07 09:45:31

by Krzysztof Kozlowski

[permalink] [raw]
Subject: Re: [PATCH v2 3/7] tty: serial: samsung: constify s3c24xx_serial_drv_data

On 07/03/2022 09:33, Jiri Slaby wrote:
> On 07. 03. 22, 9:09, Krzysztof Kozlowski wrote:
>> The driver data (struct s3c24xx_serial_drv_data) is only used to
>> initialize the driver properly and is not modified. Make it const.
> ...
>> @@ -2755,9 +2755,9 @@ static struct s3c24xx_serial_drv_data s5pv210_serial_drv_data = {
>> },
>> .fifosize = { 256, 64, 16, 16 },
>> };
>> -#define S5PV210_SERIAL_DRV_DATA ((kernel_ulong_t)&s5pv210_serial_drv_data)
>> +#define S5PV210_SERIAL_DRV_DATA (&s5pv210_serial_drv_data)
>> #else
>> -#define S5PV210_SERIAL_DRV_DATA (kernel_ulong_t)NULL
>> +#define S5PV210_SERIAL_DRV_DATA NULL
>
> Yet, I still don't see why the switch from ulong->ptr happens in this
> "constify it" patch?

All these defines S5PV210_SERIAL_DRV_DATA and so on are now const and
are assigned to of_device_id.data (s3c24xx_uart_dt_match). Before, these
were assigned with a cast:

static const struct of_device_id s3c24xx_uart_dt_match[] = {
{ .compatible = "samsung,s5pv210-uart",

.data = (void *)S5PV210_SERIAL_DRV_DATA }

but since the actual data structure is const, I want to drop the cast.
Casting const via (void *) might hide some possible issues, e.g. if
of_device_id.data becomes actually non-const. There is no particular
issue here, because of_device_id.data and S5PV210_SERIAL_DRV_DATA are
const. But also because they are both const now, I want to drop the cast
via void *.

When (void *) is dropped, the S5PV210_SERIAL_DRV_DATA cannot be
kernel_ulong_t:

../drivers/tty/serial/samsung_tty.c:2753:33: warning: initialization of
‘const void *’ from ‘long unsigned int’ makes pointer from integer
without a cast [-Wint-conversion]

2753 | #define S5PV210_SERIAL_DRV_DATA (kernel_ulong_t)NULL



Best regards,
Krzysztof

2022-03-08 21:50:14

by Krzysztof Kozlowski

[permalink] [raw]
Subject: Re: [PATCH v2 3/7] tty: serial: samsung: constify s3c24xx_serial_drv_data

On 07/03/2022 09:40, Krzysztof Kozlowski wrote:
> On 07/03/2022 09:33, Jiri Slaby wrote:
>> On 07. 03. 22, 9:09, Krzysztof Kozlowski wrote:
>>> The driver data (struct s3c24xx_serial_drv_data) is only used to
>>> initialize the driver properly and is not modified. Make it const.
>> ...
>>> @@ -2755,9 +2755,9 @@ static struct s3c24xx_serial_drv_data s5pv210_serial_drv_data = {
>>> },
>>> .fifosize = { 256, 64, 16, 16 },
>>> };
>>> -#define S5PV210_SERIAL_DRV_DATA ((kernel_ulong_t)&s5pv210_serial_drv_data)
>>> +#define S5PV210_SERIAL_DRV_DATA (&s5pv210_serial_drv_data)
>>> #else
>>> -#define S5PV210_SERIAL_DRV_DATA (kernel_ulong_t)NULL
>>> +#define S5PV210_SERIAL_DRV_DATA NULL
>>
>> Yet, I still don't see why the switch from ulong->ptr happens in this
>> "constify it" patch?
>
> All these defines S5PV210_SERIAL_DRV_DATA and so on are now const and
> are assigned to of_device_id.data (s3c24xx_uart_dt_match). Before, these
> were assigned with a cast:
>
> static const struct of_device_id s3c24xx_uart_dt_match[] = {
> { .compatible = "samsung,s5pv210-uart",
>
> .data = (void *)S5PV210_SERIAL_DRV_DATA }
>
> but since the actual data structure is const, I want to drop the cast.
> Casting const via (void *) might hide some possible issues, e.g. if
> of_device_id.data becomes actually non-const. There is no particular
> issue here, because of_device_id.data and S5PV210_SERIAL_DRV_DATA are
> const. But also because they are both const now, I want to drop the cast
> via void *.
>
> When (void *) is dropped, the S5PV210_SERIAL_DRV_DATA cannot be
> kernel_ulong_t:
>
> ../drivers/tty/serial/samsung_tty.c:2753:33: warning: initialization of
> ‘const void *’ from ‘long unsigned int’ makes pointer from integer
> without a cast [-Wint-conversion]
>
> 2753 | #define S5PV210_SERIAL_DRV_DATA (kernel_ulong_t)NULL
>

I will split the casts removal to separate patch. I hope this clarifies
a bit.


Best regards,
Krzysztof