[PATCH v2] arm: mvebu: Espressobin: Set environment variable fdtfile

["Pali Rohár" <pali@kernel.org>]

On Monday 05 October 2020 18:20:17 Andre Heider wrote:
> On 03/10/2020 10:50, Pali Roh?r wrote:
> > On Saturday 03 October 2020 09:17:35 Andre Heider wrote:
> > > On 02/10/2020 12:49, Pali Roh?r wrote:
> > > > On Saturday 26 September 2020 11:09:59 Andre Heider wrote:
> > > > > On 25/09/2020 09:46, Pali Roh?r wrote:
> > > > > > On Friday 11 September 2020 06:35:10 Andre Heider wrote:
> > > > > ...
> > > > > > > +#ifdef CONFIG_BOARD_LATE_INIT
> > > > > > > +int board_late_init(void)
> > > > > > > +{
> > > > > > > +	bool ddr4, emmc;
> > > > > > > +
> > > > > > > +	if (env_get("fdtfile"))
> > > > > > > +		return 0;
> > > > > > > +
> > > > > > > +	if (!of_machine_is_compatible("globalscale,espressobin"))
> > > > > > > +		return 0;
> > > > > > > +
> > > > > > > +	/* If the memory controller has been configured for DDR4, we're running on v7 */
> > > > > > > +	ddr4 = ((readl(A3700_CH0_MC_CTRL2_REG) >> A3700_MC_CTRL2_SDRAM_TYPE_OFFS)
> > > > > > > +		& A3700_MC_CTRL2_SDRAM_TYPE_MASK) == A3700_MC_CTRL2_SDRAM_TYPE_DDR4;
> > > > > > > +
> > > > > > > +	emmc = of_machine_is_compatible("globalscale,espressobin-emmc");
> > > > > > 
> > > > > > Maybe stupid question, but are not we able to detect if eMMC is
> > > > > > connected or not at runtime? That could simplify setup and avoid usage
> > > > > > of additional special DTS file for eMMC in U-Boot.
> > > > > 
> > > > > At some point I wondered about the same.
> > > > > 
> > > > > IIRC armbian just enables it and uses one u-boot binary everywhere. A
> > > > > non-existing chip won't get detected, so that shouldn't be a problem.
> > > > > 
> > > > > But I think it has more to do with enabling/powering up hardware blocks,
> > > > > which is not wanted or may even problematic.
> > > > 
> > > > In U-Boot such functionality may be implemented in board_fix_fdt()
> > > > function which allows U-Boot to modify its Device tree prior using it.
> > > > 
> > > > You have already wrote code which is doing V5 vs V7 detection, so
> > > > detecting eMMC is the last thing which is missing to have just one
> > > > U-Boot DTS file and therefore only one U-Boot binary for Espressobin.
> > > 
> > > That implies setting status=okay for the emmc node, which then powers up
> > > that block. I don't know if that might be problematic. Can we just do that?
> > 
> > No. I mean to detect presence of eMMC in board_fix_fdt() and then set
> > tatus=okay only if check passed.
> 
> Yes, but how do you detect the emmc then? Enabling it in u-boot's dts and
> calling mmc_init() on it would be the easy way, but open coding all the
> required parts to do that with status=disabled could get rather big and/or
> unmaintanable (I didn't check what exactly would be required)?

Hm... seems that it would be really easier to enable eMMC in U-Boot DTS,
let U-Boot to try initialize eMMC and in case it fails, then disable
eMMC drivers &?devices and disable it also in FTB blob.

For powering up eMMC there is function comphy_emmc_power_up(), so it
would be needed to write power_down equivalent. And for xenon_sdhci.c
would be needed to write .remove callback (for disabling eMMC).

> > 
> > > The emmc detection would also add some delay to the boot process.
> > > 
> > > At least it should be powered down if no emmc chip has been detected, but I
> > > didn't check if that happens right now.
> > > 
> > > Regards,
> > > Andre
>