The Atmel NAND controller driver depends on an SRAM pool and has an
implicit assumption that the SRAM pool has already been created.
Changing the initcall ordering can break this. Unfortunately, fw_devlink
can't save us here as there's not a standard property to track the
dependency. So it's up to deferring probe to save us.

Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
---
 drivers/mtd/nand/raw/atmel/nand-controller.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/mtd/nand/raw/atmel/nand-controller.c b/drivers/mtd/nand/raw/atmel/nand-controller.c
index 83ba4ebd02d4..e60998eb754b 100644
--- a/drivers/mtd/nand/raw/atmel/nand-controller.c
+++ b/drivers/mtd/nand/raw/atmel/nand-controller.c
@@ -2306,7 +2306,7 @@ atmel_hsmc_nand_controller_init(struct atmel_hsmc_nand_controller *nc)
 					 "atmel,nfc-sram", 0);
 	if (!nc->sram.pool) {
 		dev_err(nc->base.dev, "Missing SRAM\n");
-		return -ENOMEM;
+		return -EPROBE_DEFER;
 	}
 
 	nc->sram.virt = (void __iomem *)gen_pool_dma_alloc(nc->sram.pool,

-- 
2.51.0

Hello Rob,

> diff --git a/drivers/mtd/nand/raw/atmel/nand-controller.c b/drivers/mtd/nand/raw/atmel/nand-controller.c
> index 83ba4ebd02d4..e60998eb754b 100644
> --- a/drivers/mtd/nand/raw/atmel/nand-controller.c
> +++ b/drivers/mtd/nand/raw/atmel/nand-controller.c
> @@ -2306,7 +2306,7 @@ atmel_hsmc_nand_controller_init(struct atmel_hsmc_nand_controller *nc)
>  					 "atmel,nfc-sram", 0);
>  	if (!nc->sram.pool) {
>  		dev_err(nc->base.dev, "Missing SRAM\n");
> -		return -ENOMEM;
> +		return -EPROBE_DEFER;

I am totally fine with the idea of probe deferral, however I think the
policy is to avoid making noise when this happens. The dev_err() call is
no longer relevant there. You can either lower it to dev_dbg() manually
or, at your convenience, return with:

   return dev_err_probe(nc->base.dev, "Missing SRAM\n", -EPROBE_DEFER);

Thanks,
Miquèl