[PING] Optionally: Set the serial# environment variable on the i.MX7.

Enabling this new option allows the kernel to obtain the unique ID of the 
CPU when not using ATAGS.

Signed-off-by: Mark G <mark at novtech.com>
---

Hi Mark,

patch was hidden in the flood of other patches and I am unsure if this
belongs to i.MX:

On 19.02.20 22:01, Mark G wrote:
> Enabling this new option allows the kernel to obtain the unique ID of
> the CPU when not using ATAGS.
> 
> Signed-off-by: Mark G <mark at novtech.com>
> ---
> diff --git a/arch/arm/include/asm/bootm.h b/arch/arm/include/asm/bootm.h
> index a2131ca07c..64ceb36ed8 100644
> --- a/arch/arm/include/asm/bootm.h
> +++ b/arch/arm/include/asm/bootm.h
> @@ -39,7 +39,7 @@ extern void udc_disconnect(void);
> ?#endif
> 
> ?struct tag_serialnr;
> -#ifdef CONFIG_SERIAL_TAG
> +#if defined(CONFIG_SERIAL_TAG) || defined(CONFIG_SET_SERIAL_ENV)
> ? #define BOOTM_ENABLE_SERIAL_TAG??? 1
> ?void get_board_serial(struct tag_serialnr *serialnr);
> ?#else
> diff --git a/arch/arm/mach-imx/mx7/Kconfig b/arch/arm/mach-imx/mx7/Kconfig
> index 286d36589d..4cf14d43c0 100644
> --- a/arch/arm/mach-imx/mx7/Kconfig
> +++ b/arch/arm/mach-imx/mx7/Kconfig
> @@ -71,6 +71,13 @@ config TARGET_COLIBRI_IMX7
> 
> ?endchoice
> 
> +config SET_SERIAL_ENV
> +??? bool "Set serial number variable from the OCOTP"
> +??? help
> +????? Selecting this option will populate the serial# environment
> +????? variable with a unique identifier from the i.MX7 CPU. The
> +????? Linux kernel will use this as the serial number of the machine.
> +

I do not see the need of *another* CONFIG_ option. To make this
available, CONFIG_SERIAL_TAG should also be on, else you cannot call
get_board_serial() to get the serial from OCOTP. It think SET_SERIAL_ENV
is quite superflous.

> ?config SYS_SOC
> ???? default "mx7"
> 
> diff --git a/arch/arm/mach-imx/mx7/soc.c b/arch/arm/mach-imx/mx7/soc.c
> index 4aafeed188..208400a4a7 100644
> --- a/arch/arm/mach-imx/mx7/soc.c
> +++ b/arch/arm/mach-imx/mx7/soc.c
> @@ -344,11 +344,24 @@ int arch_misc_init(void)
> ???? sec_init();
> ?#endif
> 
> +#ifdef CONFIG_SET_SERIAL_ENV
> +??? {
> +??????? struct tag_serialnr serialnr;
> +??????? char serialbuf[sizeof(serialnr) * 2 + 8];
> +
> +??????? get_board_serial(&serialnr);
> +??????? snprintf(serialbuf, sizeof(serialbuf),
> +???????????? "%08lx%08lx",
> +???????????? (ulong)serialnr.high, (ulong)serialnr.low);
> +??????? env_set("serial#", serialbuf);
> +??? }

If this should be done global, local solution should be moved too. I
mean the warp board doing exactly this in board code.

So which is the idea of this ? To move warp code and making global, or
what ?

Regards,
Stefano Babic

> +#endif
> +
> ???? return 0;
> ?}
> ?#endif
> 
> -#ifdef CONFIG_SERIAL_TAG
> +#if defined(CONFIG_SERIAL_TAG) || defined(CONFIG_SET_SERIAL_ENV)
> ?/*
> ? * OCOTP_TESTER
> ? * i.MX 7Solo Applications Processor Reference Manual, Rev. 0.1, 08/2016

On Mon, 11 May 2020, Stefano Babic wrote:

> patch was hidden in the flood of other patches and I am unsure if this
> belongs to i.MX:

No worries. To clarify the patch applies to all i.MX7 boards.

> On 19.02.20 22:01, Mark G wrote:
>> +????? Linux kernel will use this as the serial number of the machine.
>> +
>
> I do not see the need of *another* CONFIG_ option. To make this
> available, CONFIG_SERIAL_TAG should also be on, else you cannot call
> get_board_serial() to get the serial from OCOTP. It think SET_SERIAL_ENV
> is quite superflous.

The intent here is to pass in the serial number via the environemnt 
(i.e. when not using ATAGS). My patch changes the preprocessor machinery 
so that get_board_serial() is callable if this option is set.

This then lets fdt_root() put it in the FDT passed to the kernel. It also 
means that it's easily changed for testing by being in the environment. 
The origins of this patch were for a customer switching to the Meerkat96 
board that utilized the CPU serial number in Linux for provisioning and 
testing.

> If this should be done global, local solution should be moved too. I
> mean the warp board doing exactly this in board code.

Ah. To be fair I never looked at the port for the Warp7 board as I do not 
have one. This really should be done globally; there isn't anything board 
specific for this behavior. It should apply to all i.MX7 boards, at a 
minimum.

Well - to clarify in theory there is nothing board specific. Could there 
be a board out there with an EEPROM that contains a different serial 
number that a customer would want to pass to the kernel? Possibly but 
unlikely.

> So which is the idea of this ? To move warp code and making global, or
> what ?

I would say this behavior should be global but the Warp7 port seems to add 
a board specific prefix to the CPU serial number. So I am hesitant to 
remove it. If you like I can #ifdef around it.

Additionally the Warp7 code isn't quite the same as it relies on ATAGS 
being enabled where as my patch does not.

Sincerely,
Mark G.