[PATCH v12 0/8] Implement vendor resets for PSCI SYSTEM_RESET2

[Florian Fainelli]

On 7/24/25 07:43, Arnd Bergmann wrote:
> On Thu, Jul 24, 2025, at 16:04, Shivendra Pratap wrote:
>> On 7/24/2025 6:18 PM, Krzysztof Kozlowski wrote:
>>> On 24/07/2025 14:24, Shivendra Pratap wrote:
> 
>>> I strongly insist using compatible as way to find your device, not node
>>> names.
>> It will look better to switch to compatible. Will define a compatible for
>> psci reboot-mode binding and align the patch to use the compatible for sysfs.
>> Current patch defines reboot-mode as a property to psci, hope its fine to
>> define a compatible for this property like "psci-vendor-reset" or
>> "psci-reboot-modes"?
>>
> 
> How about using the reboot driver name as the identifier in sysfs
> instead of the compatible string? That would make it independent of
> devicetree.

+1

> 
> I had a related idea to provide some namespacing on the actual
> reboot syscall parameter, as we have two (or more) orthogonal
> concepts here, when there is more than one reboot driver and
> drivers support multiple modes.
> 
> E.g. you could use
> 
>      syscall(__NR_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2,
>              LINUX_REBOOT_CMD_RESTART2, "watchdog");
> 
> vs
> 
>      syscall(__NR_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2,
>              LINUX_REBOOT_CMD_RESTART2, "psci");
> 
> to pick one of the drivers, or
> 
>      syscall(__NR_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2,
>              LINUX_REBOOT_CMD_RESTART2, "bootloader");
> 
>      syscall(__NR_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2,
>              LINUX_REBOOT_CMD_RESTART2, "recovery");
> 
> to ask for a reboot from any driver that supports a mode, or
> combine the two and ask a specific mode in a specific driver like
> 
>      syscall(__NR_reboot, LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2,
>              LINUX_REBOOT_CMD_RESTART2, "psci:bootloader");

Yes that seems entirely adequate as well.

If there is a single reboot driver registered say "psci" then you could 
go with the short hand of specifying "bootloader" which would be 
functionally equivalent to "psci:bootloader". That is, the driver name 
becomes optional. This would maintain user-space compatibility with 
existing systems that already support custom reboot modes.
-- 
Florian