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

From: Grant Feng <von81@163.com>
To: Alexandre Belloni <alexandre.belloni@bootlin.com>
Cc: a.zummo@towertech.it, linux-rtc@vger.kernel.org
Subject: Re: [PATCH] rtc: interface^ 1969-12-31T23:59:59 is set as rtc_time if rtc_time is invalid in __rtc_read_time
Date: Sun, 2 Aug 2020 21:35:07 +0800	[thread overview]
Message-ID: <6775e458-d563-5e2a-91ee-9b6e96ab965c@163.com> (raw)
In-Reply-To: <20200802130439.GF3679@piout.net>

Thank you for explaining that.

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.
>