No subject

initialised. Now the AMD SC520 which I am using (which is a 12 year old
part) has 16kB which is way more than enough for initial Global Data, stack
and pre-console buffer

Now I'm not saying (like you so rightly point out) that every CPU will have
the required functionality. It would be interesting to know how much
'temporary memory' each of the various CPUs have.

> 
>> diff --git a/arch/x86/include/asm/global_data.h b/arch/x86/include/asm/global_data.h
>> index 2902e61..4ebc5bd 100644
>> --- a/arch/x86/include/asm/global_data.h
>> +++ b/arch/x86/include/asm/global_data.h
>> @@ -44,6 +44,7 @@ typedef	struct global_data {
>>  	unsigned long	env_addr;	/* Address  of Environment struct */
>>  	unsigned long	cpu_clk;	/* CPU clock in Hz!		*/
>>  	unsigned long	bus_clk;
>> +	unsigned long	con_buf_idx;	/* Console buffer index */
>>  	unsigned long	relocaddr;	/* Start address of U-Boot in RAM */
>>  	unsigned long	start_addr_sp;	/* start_addr_stackpointer */
>>  	phys_size_t	ram_size;	/* RAM size */
>> @@ -65,13 +66,14 @@ extern gd_t *gd;
>>  #define GD_ENV_ADDR		5
>>  #define GD_CPU_CLK		6
>>  #define GD_BUS_CLK		7
>> -#define GD_RELOC_ADDR		8
>> -#define GD_START_ADDR_SP	9
>> -#define GD_RAM_SIZE		10
>> -#define GD_RESET_STATUS		11
>> -#define GD_JT			12
>> +#define GD_CON_BUF_IDX		8
>> +#define GD_RELOC_ADDR		9
>> +#define GD_START_ADDR_SP	10
>> +#define GD_RAM_SIZE		11
>> +#define GD_RESET_STATUS		12
>> +#define GD_JT			13
>>
>> -#define GD_SIZE			13
>> +#define GD_SIZE			14
> 
> Argh... your whole "Word Offsets into Global Data" needs to be
> removed.  This should be auto-generated as asm-offsets.

Agreed - I did some work on this but found the rabbit hole gets rather deep
as asm-offsets needs to be build in the 'depend' target to avoid file not
found errors while building the dependencies. Also, the temporary files
need to be cleaned up in the clean target. So I started digging and hit
rules.mk - I'm still working on it ;)

>> --- a/common/console.c
>> +++ b/common/console.c
>> @@ -323,6 +323,28 @@ int tstc(void)
>>  	return serial_tstc();
>>  }
> 
> Hm... this adds a lot of code, unconditionally.  In thos form this is
> not acceptable, especially as many boards cannot make use of this, or
> eventually don't want to make use of it.

As I said, rough-and-ready. Yes, I would expect in the board config file:

#define CONFIG_PRE_CONSOLE_BUFFER

>> +	if (gd->flags & GD_FLG_HAVE_CONSOLE) {
>> +		if (gd->flags & GD_FLG_DEVINIT) {
>> +			/* Send to the standard output */
>> +			fputc(stdout, c);
>> +		} else {
>> +			/* Send directly to the handler */
>> +			serial_putc(c);
>> +		}
>>  	} else {
>> -		/* Send directly to the handler */
>> -		serial_putc(c);
>> +		pre_console_putc(c);
>>  	}
>>  }
> 
> And this is actually wrong.  If we can use the serial console for
> output, we definitely don;t want to use your buffer any more.

Look closer - It's a diff weirdness - The actual code is:

	if (gd->flags & GD_FLG_HAVE_CONSOLE) {
		if (gd->flags & GD_FLG_DEVINIT) {
			/* Send to the standard output */
			fputc(stdout, c);
		} else {
			/* Send directly to the handler */
			serial_putc(c);
		}
	} else {
		pre_console_putc(c);
	}

So if we have console, it goes there, otherwise it goes to the buffer

>> diff --git a/include/configs/eNET.h b/include/configs/eNET.h
>> index 548d52c..4fb971f 100644
>> --- a/include/configs/eNET.h
>> +++ b/include/configs/eNET.h
> 
> ...and compilation for all other boards breaks?

See above.

Also, I have noticed in drivers/i2c/ppc4xx_i2c.c:

		if (gd->flags & GD_FLG_HAVE_CONSOLE) {
			printf("I2C %s: failed %d\n",
			       read ? "read" : "write", ret);
		}

and in drivers/i2c/soft_i2c.c:

#ifdef DEBUG_I2C
#define PRINTD(fmt,args...)	do {		\
	if (gd->flags & GD_FLG_HAVE_CONSOLE)	\
		printf (fmt ,##args);		\
	} while (0)
#else
#define PRINTD(fmt,args...)
#endif

So there are drivers that anticipate generating output before console is
initialised - I think we should not put the onus on the driver and move
these conditions to printf() in console.c - Unfortunately this could lead
to head-scratching when one _thinks_ a printf() should generate an output,
but it is squelched (which is what the pre-console buffer is for)

Regards,

Graeme