Hi, On Thu, Jan 16, 2020 at 05:47:12PM +0100, Jernej Škrabec wrote: > Dne četrtek, 16. januar 2020 ob 09:06:52 CET je Maxime Ripard napisal(a): > > Hi Jernej, > > > > On Mon, Jan 13, 2020 at 07:07:20PM +0100, Jernej Skrabec wrote: > > > Tanix TX6 doesn't have external 32 kHz oscillator, so switch RTC clock > > > to internal one. > > > > > > Signed-off-by: Jernej Skrabec <jernej.skrabec@xxxxxxxx> > > > --- > > > > > > While this patch gives one possible solution, I mainly want to start > > > discussion why Allwinner SoC dtsi reference external 32 kHz crystal > > > although some boards don't have it. My proposal would be to make clock > > > property optional, based on the fact if external crystal is present or > > > not. However, I'm not sure if that is possible at this point or not. > > > > It's probably a bit of a dumb question but.. are you sure the crystal > > is missing? > > Although I don't have schematic, I'm pretty sure. Without this patch or one at > [1], RTC gives a lot of errors in dmesg. I think that unpopulated XC2 pads > near SoC (see [2]) are probably reserved for crystal. > > With patch in [1], which enables automatic switching in case of error, I saw > that on this box RTC always switched to internal RC. > > > > > The H6 datasheet mentions that the 32kHz crystal needs to be there, > > and it's part of the power sequence, so I'd expect all boards to have > > it. > > Can you be more specific where it is stated that crystal is mandatory? I was mostly referring to the power sequence mentionned in the H6 Datasheet (not the user manual, the smaller one). https://linux-sunxi.org/images/5/5c/Allwinner_H6_V200_Datasheet_V1.1.pdf Page 74 > Note that schematic of some boards, like OrangePi PC2 (H5) or OrangePi Zero > (H3) don't even have 32K crystal in them. And we can't use the compatible for these.. > > > > > Driver also considers missing clock property as deprecated (old DT) [1], > > > so this might complicate things even further. > > > > > > What do you think? > > > > I'm pretty sure (but that would need to be checked) that we never got > > a node without the clocks property on the H6. If that's the case, then > > we can add a check on the compatible. > > Yes, that would be nice solution. I can work something out if you agree that > this is the way. So if we want to have something that works for the H3 too, then I guess we need to revert the patch that switches the 32kHz clock source to the external one all the time, and do it only if we have a clock provided. If we don't, we would run from the internal oscillator (which would work for both the H3 and H6 boards you have I guess?) and if we do we will still use the better, more accurate, clock. That would change a bit the behaviour of the old DTs again and revert to the old behaviour we had, but we didn't hear anything the first time we did, so I wouldn't be overly concerned. Does that make sense? Maxime
Attachment:
signature.asc
Description: PGP signature
