Re: [PATCH] rtc: interface^ 1969-12-31T23:59:59 is set as rtc_time if rtc_time is invalid in __rtc_read_time

[Grant Feng <von81@163.com>]

Thank you again for your patience to explain.
I get the log and now I get it.

commit 812318a094d0715194d9f686b22ee67e7dc59d93
Author: Alexandre Belloni <alexandre.belloni@bootlin.com>
Date:   Wed Feb 21 11:44:26 2018 +0100

     rtc: cmos: let the core handle invalid time

     Setting the rtc to a valid time when the time is invalid is a bad 
practice,
     because then userspace doesn't know it shouldn't trust the RTC.

     Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>

then you removed the following code fragment in "drivers/rtc/rtc-cmos.c":

-       cmos_read_time(&pdev->dev, &time);
-       ret = rtc_valid_tm(&time);
-       if (ret) {
-               struct rtc_time def_time = {
-                       .tm_year = 1,
-                       .tm_mday = 1,
-               };
-               cmos_set_time(&pdev->dev, &def_time);
-       }


On 2020-08-02 21:04, Alexandre Belloni wrote:
> On 02/08/2020 14:51:41+0800, Grant Feng wrote:
>> On 2020-08-01 21:28, Alexandre Belloni wrote:
>>> On 01/08/2020 19:20:07+0800, Grant Feng wrote:
>>>> 1969-12-31T23:59:59 is an error more clear than Invalid argument
>>> Definitively not, 1969-12-31T23:59:59 is a valid date and should not be
>>> returned when it is known the current date is not set in the RTC.
>> 'rtc_valid_tm' is used to check rtc_time and 1969-12-31T23:59:59 is invalid.
>> when the RTC clock is not set, some rtc devices always return '0' or almost
>> random data, and different rtc devices may give different return data.
>> so, I think, it's usful to return a default date when the current date is
>> not set in the RTC.
> You are not solving the issue you mention here. If the RTC doesn't know
> whether the date/time is invalid and the core think it is valid, then
> your code will not run.
>
>>>> For example, when the RTC clock is not set, it will print a kernel
>>>> error log every time someone tries to read the clock:
>>>>           ~ # hwclock -r
>>>>           hwclock: RTC_RD_TIME: Invalid argument
>>>>
>>>> It's clear and easy to understand what happened if print
>>>> 1969-12-31T23:59:59 in this situation：
>>>>           ~ # hwclock -r
>>>>           Wed Dec 31 23:59:59 1969  0.000000 seconds
>>>>
>>> How do you know this is an error an not what is actually set on the RTC?
>> 'rtc_valid_tm' will check rtc_time when someone set the RTC, the time
>> should not be earlier than 1970-1-1T00:00:00. so 1969-12-31T23:59:59
>> can not be actually set on the RTC.
>>      When someone get
>> ~ # hwclock -r
>> Wed Dec 31 23:59:59 1969  0.000000 seconds
>>      he knows: the RTC time doesn't match my watch, change it now.
>> but still lots of people don't know what happened if they see
>> ~ # hwclock -r
>> hwclock: RTC_RD_TIME: Invalid argument
>>
> This makes userspace checking for errors way worse. Think about it, first
> userspace will need to check for an error when calling the ioctl then it
> will have to check the time and consider a vlid date invalid. Seriously,
> if hwclock doesn't do what you want, you can either patch it or use
> another tool.
>