Re: [PATCH] watchdog: sbsa: Support architecture version 1

[Shaokun Zhang <zhangshaokun@hisilicon.com>]

Hi Guenter,

On 2021/5/11 11:52, Guenter Roeck wrote:
> On 5/10/21 7:49 PM, Shaokun Zhang wrote:
>> Hi Guenter,
>>
>> On 2021/5/10 21:16, Guenter Roeck wrote:
>>> On 5/10/21 1:25 AM, Shaokun Zhang wrote:
>>>> Hi Guenter,
>>>>
>>>> On 2021/5/10 12:25, Guenter Roeck wrote:
>>>>> On 5/9/21 8:41 PM, Shaokun Zhang wrote:
>>>>>> Arm Base System Architecture 1.0[1] has introduced watchdog
>>>>>> revision 1 that increases the length the watchdog offset
>>>>>
>>>>> Is that how they call the watchdog count register ?
>>>>>
>>>>
>>>> I think yes.
>>>>
>>>>> Also, doesn't that mean that the maximum timeout supported
>>>>> by the hardware is now larger ?
>>>>
>>>> No, maximum timeout is the same. But the clock can be higher than
>>>> before. For Armv8.6, The frequency of CNTFRQ_EL0 is standardized to
>>>> a frequency of 1GHz which will set gwdt->clk. If the timeout is
>>>> greater than 4(second), the 32-bit counter(WOR) is not enough.
>>>>
>>>
>>> The maximuma timeout is limited with
>>>
>>> wdd->max_hw_heartbeat_ms = U32_MAX / gwdt->clk * 1000;
>>>> You did not update that calculation. That means that the maximuma
>>> timeout is still U32_MAX / gwdt->clk * 1000, which still fits
>>> into 32 bit.
>>
>> Correct, I will fix this in next version.
>>
>>>
>>> Please correct me if I am missing something.
>>>
>>
>> My bad, you are right. The maximum timeout shall be
>> 0xFFFF.FFFF.FFFF / 1000.000.000(1GHz) which is larger than before
>> 0xFFFF.FFFF / 100.000.000(100MHz) by the hardware.
>>
>> Can I do like this, after the version is got and check the version?
>>          sbsa_gwdt_set_timeout(wdd, wdd->timeout);
>> +       sbsa_gwdt_get_version(wdd);
>> +       if (wdd->version > 0)
>> +               wdd->max_hw_heartbeat_ms = GENMASK_ULL(47, 0) / gwdt->clk * 1000;
>>
> 
> I would suggest to set max_hw_heartbeat_ms in one place
> to avoid confusion. Either check the version earlier,
> or move setting both max_hw_heartbeat_ms values
> after the call to sbsa_gwdt_get_version().
> 

Got it, I will follow the former that many members in @wdd are
initialized closely.

Thanks again,
Shaokun

> Thanks,
> Guenter
> 
> .