Re: Infinite loop on edge cases

[Alexandre Belloni <alexandre.belloni@bootlin.com>]

On 09/04/2019 12:46:20+0200, Mastro Gippo wrote:
> Hi Alexandre,
> thank you for your comments.
> Since failure behaviour is not specified in the Maxim datasheet, the
> gracious way to handle this would be a retry counter, as that would
> also solve the problem of a DS1307 that for some reason doesn't set
> the CH bit (maybe an internal oscillator failure?). A similar erratic
> behavior happened to me once, due to an unstable power supply to the
> RTC.

Ok, I had a closer look at the tissue, the solution here is to not
bother and simply not handle DS1307_BIT_CH, DS1338_BIT_OSF,
DS1340_BIT_nEOSC, DS1340_BIT_OSF, MCP794XX_BIT_VBATEN or
MCP794XX_BIT_ST. they should only be handled in the read_time and
set_time callbacks, were they really matters. I'll send a patch for
that.

> IMHO, I think that a hardware failure and/or a problematic I2C driver
> should not produce an infinite loop - be it fatal or not.

This is correct and this has to be fixed. I must admit I was too quick
and didn't see this was in the probe function.

> I already solved my boot issue with this patch, and I will ship it
> with the devices I sell, but may I ask you what approach would you
> recommend instead of hctosys? I tried looking online for solutions but
> couldn't find an "official" one. I will gladly investigate better
> approaches if available.
> 

Many distributions have initscripts that are using hwclock to set the
system time at boot and save it at shutdown. The kernel is not doing a
good job at setting the system time and saving it for technical and
historical reasons. hctosys and systohc should not be used.


-- 
Alexandre Belloni, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com