Re: [PATCH 1/6] dt-bindings: nvmem: add cell-type to nvmem cells

From: Srinivas Kandagatla
Date: Wed Sep 22 2021 - 08:23:56 EST




On 22/09/2021 12:34, Ahmad Fatoum wrote:
Hi,

On 08.09.21 12:02, Joakim Zhang wrote:
From: Srinivas Kandagatla <srinivas.kandagatla@xxxxxxxxxx>

Some of the nvmem providers encode data for certain type of nvmem cell,
example mac-address is stored in ascii or with delimiter or in reverse order.

This is much specific to vendor, so having a cell-type would allow nvmem
provider drivers to post-process this before using it.

I don't agree with this assessment. Users of the OCOTP so far
used this specific encoding. Bootloaders decode the OCOTP this way, but this
encoding isn't really an inherent attribute of the OCOTP. A new NXP SoC
with a different OTP IP will likely use the same format. Users may even
use the same format on an EEPROM to populate a second off-SoC interface, .. etc.


That is okay.

I'd thus prefer to not make this specific to the OCOTP as all:

* #define NVMEM_CELL_ENCODING_MAC_ADDRESS_IMX /* ... */

* cell-type = <NVMEM_CELL_ENCODING_MAC_ADDRESS_IMX>;


No, this is not okay, cell-type is just representing what is the cell type in a generic way, and its not intended to have any encoding information.

Encoding information should be derived from the provider specific bindings. If we start adding this info in generic binding we will endup with a mess.
And this has been discussed in various instances.

* and then the decoder is placed into some generic location, e.g.
drivers/nvmem/encodings.c for Linux

This is fine, we could have a library to do these post-processing, but only if we have enough users. For now its better to keep it within provider drivers till we have more consumers of these functions.


--srini

That way, we can reuse this and future encodings across nvmem providers.
It's also more extendable: e.g. big endian fields on EEPROMs. Just stick
the cell-type in, document it in the binding and drivers supporting it
will interpret bytes appropriately.

It's still a good idea to record the type as well as the encoding,
e.g. split the 32 bit encoding constant into two 16-bit values.
One is an enum of possible types (unknown, mac_address, IP address ... etc.)
and one is an enum of the available encodings.

What do you think?


Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@xxxxxxxxxx>
Signed-off-by: Joakim Zhang <qiangqing.zhang@xxxxxxx>
---
Documentation/devicetree/bindings/nvmem/nvmem.yaml | 11 +++++++++++
include/dt-bindings/nvmem/nvmem.h | 8 ++++++++
2 files changed, 19 insertions(+)
create mode 100644 include/dt-bindings/nvmem/nvmem.h

diff --git a/Documentation/devicetree/bindings/nvmem/nvmem.yaml b/Documentation/devicetree/bindings/nvmem/nvmem.yaml
index b8dc3d2b6e92..8cf6c7e72b0a 100644
--- a/Documentation/devicetree/bindings/nvmem/nvmem.yaml
+++ b/Documentation/devicetree/bindings/nvmem/nvmem.yaml
@@ -60,6 +60,11 @@ patternProperties:
- minimum: 1
description:
Size in bit within the address range specified by reg.
+ cell-type:
+ $ref: /schemas/types.yaml#/definitions/uint32
+ maxItems: 1
+ description:
+ Type of nvmem, Use defines in dt-bindings/nvmem/nvmem.h.
required:
- reg
@@ -69,6 +74,7 @@ additionalProperties: true
examples:
- |
#include <dt-bindings/gpio/gpio.h>
+ #include <dt-bindings/nvmem/nvmem.h>
qfprom: eeprom@700000 {
#address-cells = <1>;
@@ -98,6 +104,11 @@ examples:
reg = <0xc 0x1>;
bits = <2 3>;
};
+
+ mac_addr: mac-addr@90{
+ reg = <0x90 0x6>;
+ cell-type = <NVMEM_CELL_TYPE_MAC_ADDRESS>;
+ };
};
...
diff --git a/include/dt-bindings/nvmem/nvmem.h b/include/dt-bindings/nvmem/nvmem.h
new file mode 100644
index 000000000000..eed0478f6bfd
--- /dev/null
+++ b/include/dt-bindings/nvmem/nvmem.h
@@ -0,0 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __DT_NVMMEM_H
+#define __DT_NVMMEM_H
+
+#define NVMEM_CELL_TYPE_UNKNOWN 0
+#define NVMEM_CELL_TYPE_MAC_ADDRESS 1
+
+#endif /* __DT_NVMMEM_H */