[PATCH] xilinx: zynqmp: Do not use null as spl bss start address

[Michal Simek]

On 6/1/22 16:03, Stefan Herbrechtsmeier wrote:
> Hi,
> 
> Am 01.06.2022 um 14:59 schrieb Michal Simek:
>>
>>
>> On 6/1/22 14:27, Stefan Herbrechtsmeier wrote:
>>> Hi,
>>>
>>> Am 01.06.2022 um 13:59 schrieb Michal Simek:
>>>> Hi,
>>>>
>>>> first of all subject is not accurate. We are not using null as start but 
>>>> address 0.
>>>
>>> I will replace null with 0.
>>>
>>>>
>>>> On 6/1/22 12:55, Stefan Herbrechtsmeier wrote:
>>>>> [CAUTION: External Email]
>>>>>
>>>>> From: Stefan Herbrechtsmeier <stefan.herbrechtsmeier at weidmueller.com>
>>>>>
>>>>> Do not use null as address for memory because of the special meaning for
>>>>> pointers. Change the spl bss start address to the second page.
>>>>>
>>>>> Signed-off-by: Stefan Herbrechtsmeier <stefan.herbrechtsmeier at weidmueller.com>
>>>>>
>>>>> ---
>>>>> The problem was discovered with a static char array initialized with an
>>>>> empty string.
>>>>
>>>> It means your code is doing wrong pointer arithmeticians which pointed to 
>>>> BSS section and overwrites there something. What is that code doing?
>>>
>>> I like to call the zynqmp_get_silicon_idcode_name function from my board code 
>>> and therefore rework the function to return static memory to avoid a memory 
>>> allocation on each call.
>>>
>>> -    char name[ZYNQMP_VERSION_SIZE];
>>> +    static char name[ZYNQMP_VERSION_SIZE] = "";
>>
>> Try to remove = "".
> 
> The = "" is needed to skip the processing on the second run.

I looked if this really change anything. Static variable is placed to BSS 
section which is zeroed by default at start. It means before you jump to this 
function name array is already zeroed.


> 
>> Undefined variables should be placed to bss section.
> 
> A = "zu" will also move the variable to an other section.
> 
>>
>>>
>>> +    if (name[0])
>>> +        return name;
>>>
>>> -    return strdup(name);
>>> +    return name;

this function is called only once now that's why I personally can't see to do 
any change there unless you have good reason to call it more times.

But that being said I think that would make sense to move the whole 
zynqmp_get_silicon_idcode_name() to soc_xilinx_zynqmp.c and do silicon detection 
there. Only option there is soc_get_machine() which can return silicon type.


>>>
>>> The name variable gets the address 0 which means that snprintf and strncat do 
>>> nothing because the dest pointer and NULL are equal.
>>
>> Ok. I see. It become the first variable in bss section.
> 
> Should I resend the patch with null replaced by 0?

Yes please.
M