Re: [PATCH] rtc: pcf8563: add SMBus support

["Michael Walle" <mwalle@kernel.org>]

On Wed Sep 30, 2026 at 1:40 PM CEST, sashiko-bot wrote:
> [Severity: High]
> Could removing this check lead to time tearing on restricted SMBus adapters?
>
> While the commit message mentions SMBus controllers supporting block
> accesses, removing the check entirely delegates bus validation to regmap. If
> the adapter only supports I2C_FUNC_SMBUS_BYTE_DATA and lacks block transfer
> capabilities, regmap silently falls back to regmap_smbus_byte:

That is actually correct. But I'm not sure how this is handled in
the RTC subsystem (?). I've looked at other drivers (e.g. ds1307,
rv3028), which are using regmap and there seems to be no check
either.

-michael


> drivers/base/regmap/regmap-i2c.c:regmap_get_i2c_bus() {
>     ...
>     else if (config->val_bits == 8 && config->reg_bits == 8 &&
>              i2c_check_functionality(i2c->adapter,
>                                      I2C_FUNC_SMBUS_BYTE_DATA))
>         bus = &regmap_smbus_byte;
>     ...
> }
>
> Because of this fallback, when pcf8563_rtc_read_time() reads the time:
>
> drivers/rtc/rtc-pcf8563.c:pcf8563_rtc_read_time() {
>     ...
>     err = regmap_bulk_read(pcf8563->regmap, PCF8563_REG_ST1, buf,
>                            sizeof(buf));
>     ...
> }
>
> The bulk read executes as a loop of separate single-byte SMBus reads instead
> of an atomic transaction. Since each SMBus byte transaction issues its own
> STOP condition, the RTC's internal register latch is released between reads.
>
> Does this allow the hardware time to roll over mid-read and result in
> corrupt time data?
>
>>  	pcf8563 = devm_kzalloc(&client->dev, sizeof(struct pcf8563),
>>  				GFP_KERNEL);
>>  	if (!pcf8563)