On Sat, Dec 08, 2018 at 12:27:42PM +0800, Kever Yang wrote:
Hi Tom,
On 12/07/2018 10:13 PM, Tom Rini wrote:
On Fri, Dec 07, 2018 at 02:24:22PM +0100, Philipp Tomsich wrote:
Kever,
On 07.12.2018, at 02:39, Kever Yang kever.yang@rock-chips.com wrote:
Hi Philipp,
On 12/06/2018 09:50 PM, Philipp Tomsich wrote:
+Tom
On 05.12.2018, at 03:25, Kever Yang kever.yang@rock-chips.com wrote:
The U-Boot eMMC does not need to care about the power for Rockchip SoC, because if the board is using eMMC, the power will default on (for bootrom), and we do not do power management for it like kernel, so the 'vmmc', 'vqmmc' is only useful for SD in U-Boot.
This make U-Boot can boot into kernel even if the pmic driver is broken.
If the PMIC driver is broken, we should fix the PMIC driver. I would feel more comfortable w/o this statement.
The rk3288-evb dts may be used in many boards using rockchip reference schematic but with little change, so we hope it can be more robust to boot into next stage.
Again, this is not how the DTS should be used. I believe that Heiko, Fabio and I had already highlighted this in comments to the earlier thread.
Not sure if you have read my previous mail for answer all your comments,I do agree DTS should represent the hardware, but please note that the DTS is no kind of standard, and people always choose what they need and add those part in there dts, but not always add all the property and everyone use the same model. I would say there are many boards does not have this 'vmmc-supply’ in there emmc node.
That is exactly the reason why I bumped the decision up the stairs (to Tom and/or Simon): what you are saying makes sense to me (viewed through your eyes and from your specific usecase), but it directly contradicts how the DTS usage is intended.
In other words: Tom (as the top-level decision maker) or Simon (who owns the device-model and therefore will also have an opinion on DTS usage) should make the final call.
My answer is that I would strongly suspect that over in linux "we have N different close-enough boards using this one DTS" isn't acceptable. You make a dtsi and include it from the board and things that aren't common don't go into the dtsi. And yes, when starting off everyone (myself included) copies the reference platform dts and then changes it as needed, and sometimes misses a thing or two. But no, I don't think we want a wrong dts and I'm pretty sure the kernel really wouldn't want wrong dts files and the general goal is that excluding the -u-boot.dtsi files, ours are copies of the kernel.
I don't think this is a "wrong dts" after my patch, these two nodes are not mark as required property in kernel, so many dts emmc node does not have it. I check the latest kernel dtsi in arch/arm/boot/dts/rk3288-evb.dtsi [1], the emmc node do not have 'vmmc' and 'vqmmc' while the SD node have, which just like description in my commit message.
OK. So this would fall into the category of "sync with upstream dts" then, right? That is what we want.
Well, I don't know why U-Boot project is so difficult to accept a reasonable patch now, I don't want to make you unhappy, but make 'every board must have its own dts' in U-Boot to make every developer to join U-Boot does not make sense to me. The kernel need different dts for different board because they need to use/control those different feature, but U-Boot is not the case, U-Boot should work if the storage driver works.
Well, here's the thing. If you want U-Boot to then load and pass the correct DTB to the kernel, we need per-board tweaks to the code anyhow to find and load that DTB (see the various "findfdt" environment scripts for example). If you want to rework things so that you have a "generic" type board under U-Boot that's more clearly not tied to any specific board but instead runs on many, that might help clear things up too. Hope this helps!