On Tue, Jul 19, 2016 at 3:47 PM, Alexandre Belloni <alexandre.belloni@xxxxxxxxxxxxxxxxxx> wrote: > On 21/06/2016 at 19:34:56 -0700, Andrey Smirnov wrote : >> On Tue, Jun 21, 2016 at 2:07 PM, Alexandre Belloni >> <alexandre.belloni@xxxxxxxxxxxxxxxxxx> wrote: >> > On 21/06/2016 at 15:49:04 -0500, Rob Herring wrote : >> >> So wouldn't you want to set one mode while running and the lower power >> >> mode while suspended? I'm trying to understand the frequency of changing >> >> this. If it is always one setting for a board, then yes it belongs in >> >> DT. If it is a user decision, then it probably shouldn't be in DT. >> >> >> >> Seeing as these are reused, I've probably already had this discussion... >> >> >> > >> > I would agree with Rob here. It may be better to provide a sysfs >> > interface to configure that particular behavior. >> >> I don't see any value in doing that, could you give me a realistic >> example of a scenario in which a user would want to spend some of >> uptime with RTC oscillator fault detection/glitch filtering disabled >> and then enable it? >> > > Well, the issue is not being dynamic, it is differentiating between > hardware description and user configuration. Configuration must not be in > DT. Why? And I don't mean in a generic sense, but in this particular case. What is gained by not having this bit of configuration, whose only consumer is the driver, in the device tree file? > And this choice is definitively not hardware related (as opposed to > the trickle charging that depends on the battery that is used on the > board). There's most certainly plenty of precedents of non hardware-related in device tree, first two that come to mind are "chosen" node and "local-mac-address" property and, granted, those might be exceptions/legacy bindings that are just there to stay, but even if you look at RTC subsystem rtc-cmos.txt, atmel,at91sam-rtc.txt and possibly rtc-st-lpc.txt are providing bindings that are not exactly hardware related. Rtc-cmos.txt is especially noticeable example since it literally does what I am trying to do -- allows the user to specify initial values to certain registers that would be initialized by the driver. > >> > This is usually ok because the use case is: >> > - the RTC is not configured, time has never been set >> > - time is set for the first time >> > - the user can set the oscillator mode/detection/... >> >> Unfortunately exposing that feature using sysfs gives you a leaky >> abstraction and your userspace instead of using a generic RTC starts >> using DS1341 RTC. So to accommodate for that a user would have to: >> >> a) Write + integrate a userspace tool to set the mode (which IMHO is >> decided upon once and doesn't change) >> b) If a board design is new and there's a chance of moving this chip >> to a different I2C bus, the code above would have to account for that >> and not hardcore sysfs path >> c) If board's BSP is intended to be used in multiple generations of a >> product, not all of which would use DS1341, it would be necessary to >> accommodate for that by either more code in the tool or an additional >> BSP build configuration variant >> > > Well, it doesn't leak abstraction as long as all the RTC that are able > to disable the oscillator failure detection use the same ABI. Correct me if I am wrong, but no such, at least semi-standardized, ABI exist as of today, correct? If that is so, then what you are saying is that the abstraction doesn't leak as long as you put it inside of a new abstraction that doesn't leak. I am not arguing that it is impossible to create a new one that would allow to hide hardware differences, I am positive it is, what I am arguing is that to do so is a lot of work for as far as I can see for no gain. > >> > - on subsequent reboots, the mode is kept alongside the time and date >> > >> >> This assumes that your bootloader leaves those mode bits alone. >> > > Well, if that is not the case, the bootloader as to be fixed anyway and > silently changing the configuration back using DT is probably worse. > How so? Consider the following two scenarios with assumption that the bootloader is broken: - Bits set wrong by bootloader, then corrected by kernel, device is powered off RTC consumes expected amount of current - Bits set wrong by bootloader, kernel does nothing, device is powered off RTC consumes more than anticipated and we drain the power storage device and loose time. What do you you think former is worse than latter? Andrey -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html