[U-Boot] PATCH mtd CFI flash: timeout calculation underflow if imprecise 1kHz timer: fix

[Renato Andreola <renato.andreola@imagos.it>]

This patch is necessary to fix the common case in which CONFIG_SYS_HZ is 
just below 1000 (e.g. 999 that results from an integer division between 
a system clock frequency like 83333333Hz and a divisor like 83334).
In that case the CONFIG_SYS_HZ/1000 expression returns 0 and the tout *= 
CONFIG_SYS_HZ/1000 multiplication is always 0.
As you said, an incorrect Common Flash Interface interrogation of the 
flash registers can lead to a tout of 0 but this misbehavior should have 
been caught by the first stages of the flash interrogation code.
Note that the AMD/Spansion chips does not work correctly in a Nios2 
environment due to "spurious" read cycles generated when flash bus width 
is lower than 32.
If the flash chip you used is not an Intel/Numoyx than you should have 
to do some work to make them behave correctly (e.g. add extra wait state 
into the procedures or some other tricks).
 
Renato

Shinya Kuribayashi wrote:
> Renato Andreola wrote:
>   
>>  From 21d84ab72266f118794233176bd356d8b1cfdf35 Mon Sep 17 00:00:00 2001
>> From: Renato Andreola <renato.andreola@imagos.it>
>> Date: Fri, 21 Aug 2009 18:05:51 +0200
>> Subject: [PATCH] drivers/mtd/cfi_flash: precision and underflow problem in tout calculation
>>
>> With old configuration it could happen tout=0 if CONFIG_SYS_HZ<1000.
>>
>> Signed-off-by: Alessandro Rubini <rubini@gnudd.com> Renato Andreola <renato.andreola@imagos.it>
>> ---
>>   drivers/mtd/cfi_flash.c |    8 +++++---
>>   1 files changed, 5 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/mtd/cfi_flash.c b/drivers/mtd/cfi_flash.c
>> index 81ac5d3..0d8fc54 100644
>> --- a/drivers/mtd/cfi_flash.c
>> +++ b/drivers/mtd/cfi_flash.c
>> @@ -660,9 +660,11 @@ static int flash_status_check (flash_info_t * info, flash_sect_t sector,
>>   	ulong start;
>>
>>   #if CONFIG_SYS_HZ != 1000
>> -	tout *= CONFIG_SYS_HZ/1000;
>> -#endif
>> -
>> +	if ((ulong)CONFIG_SYS_HZ > 100000)
>> +		tout *= (ulong)CONFIG_SYS_HZ/1000;  /* for a big HZ, avoid overflow */
>> +	else
>> +		tout = DIV_ROUND_UP(tout*(ulong)CONFIG_SYS_HZ, 1000);
>> +#endif	
>>   	/* Wait for command completion */
>>   	start = get_timer (0);
>>   	while (flash_is_busy (info, sector)) {
>>     
>
> What should to be fixed first in this case, would be your CONFIG_SYS_HZ
> setting, that is NIOS2? timer implementation, yeah really.  But I would
> also point out that there is another case flash_status_check() doensn't
> work as expected.
>
> One of my colleagues found that with some flash device(s) (I don't
> recall precisely, sorry), 'tout' would be probed to be zero.  In that
> case, a workaround something like above still doesn't work.
>
> We have not sorted out where the problem is; it might be in cfi_flash.c,
> or in the flash device itself.  This is observed with v2009.03 release,
> and we've been having a workaround for it for months.  I'd like to have
> a look someday.
>
> Anyway, checking to see if 'tout' is zero or not would be sometimes
> worth a try, when you think cfi_flash.c doesn't work as expected.
>
> Just for your information,
>
>