Hi Arnd:
On 2015?12?28? 23:28, Arnd Bergmann wrote:
> On Wednesday 23 December 2015 17:31:45 Andy Yan wrote:
>> I also have the idea to put is in drivers/power/reset, But considering
>> this driver have not bind anything about power, so I put it in
>> driver/misc
>> at last. So I hope if some people can give more suggestions here.
> I vote for drivers/power/reset as well. On some platforms, the two things
> are related, on others they are not, but it's good to have it all in one
> place I think.
Okay, I will move to drivers/power/reset next version.
>>>> drivers/soc/rockchip/Kconfig | 9 ++
>>>> drivers/soc/rockchip/Makefile | 1 +
>>>> drivers/soc/rockchip/reboot.c | 68 +++++++++++++++
>>> And maybe that part could be made generic instead of rockchip specific.
>>> It simply uses a regmap to do the accesses, I guess a lot of other
>>> platforms will do that. We have syscon-reboot and syscon-poweroff for example.
>>>
>>> I think then we can extend the "framework" by having generic drivers to
>>> store the value in eeprom or nvram for example.
>>>
>> I also hope the write interface can be generic. But I found some
>> platform
>> use different hardware to store the value. For example, John's patch
>> use
>> SRAM on qcom apq8064 to store value for nexus7. It seems there also have
>> some platform use dram or nvram to store it. And these different
>> hardware use
>> different write method. I don't have a generic way to handle this.
>>
>> I have a idea to handle it like this:
>>
>> +static const struct of_device_id reboot_mode_dt_match[] = {
>> + { .compatible = "linux,reboot-mode-sfr", /*for magic value
>> stored in special function register, which can be accessed by regmap*/
>> + .data = (void *)&reboot-mode-sfr },
> I'd make this one syscon-reboot-mode, to go along with syscon-poweroff
> as implemented by drivers/power/reset/syscon-poweroff.c. The syscon concept
> is already generic enough that we don't need a linux prefix for that.
Okay, syscon is better.
>> + { .compatible = "linux,reboot-mode-sram", /*for magic value
>> stored in
>> + .data = (void *)&reboot-mode-sram },
> the sram mode should probably follow the generic SRAM binding and make
> the location that stores the reboot mode a separate partition, unless
> we want to store other data in the same partition, in which case we might
> want to use the same implementation as for the nvram.
>
>> + { .compatible = "linux,reboot-mode-sdram",
>> + .data = (void *)&reboot-mode-sdram }, /*for magic value
>> stored
> I think "sdram" is not an appropriate name here, as the main system memory
> might use some other technology, e.g. DRAM or DDR2-SDRAM.
>
>> + { .compatible = "rockchip,reboot-mode-nvram",
>> + .data = (void *)&reboot-mode-nvram },
>> + {},
>> +};
> nvram is a complex topic by itself, because there are so many ways to do it.
> I think what you are referring to here is a battery-backed memory that uses
> one or more bytes at a fixed offset to store a particular piece of information,
> as the drivers/char/nvram.c driver does. Maybe we should put the reboot mode
> into that driver then?
>
> There are other nvram drivers at various places in the kernel, and each may
> be slightly different, or completely different, like the EFIVARs driver on
> UEFI firmware or the key/value store on Open Firmware, these probably need
> their own methods and not share the generic driver.
>
>> the data point to different hardware access method.
>>
>> Hope to see more suggestions from you.
> It's probably best to leave these four examples as separate drivers and we can
> add further ones when needed.
>
> Arnd
>
>
Okay, thanks for your suggestion. I will add the reboot-mode.c as
the core library, syscon-reboot-mode.c as one example first. As for
sram/dram/nvram case, I am not familiar with them, so hope there are
some hero will extend them when needed.
